SlideShare a Scribd company logo
1 of 76
Download to read offline
Networks Navigability: Theory and Applications

                                     Denis Helic & Christoph Trattner

                                                       KMI, TU Graz


                                                   August 31, 2011




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   1 / 75
Internet of Things




   http://www.youtube.com/watch?v=sfEbMV295Kk




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   2 / 75
Internet of Things




           We are heading towards a completely interconnected society
           Where people, devices, sensors are all connected to each other
           producing billions of billions of data each day...




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   3 / 75
Internet of Things




           One big challenge in this context is how we can find relevant
           information in such a networked world of data
           Hence, in this presentation:
           Latest research results on the navigability of such networks will be
           shown




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   4 / 75
Internet of Things



   In particular I will show:
           what are structural clues that make such networks
           navigable/searchable?
           In addition to this, I will present a framework that is able to measure
           network navigability.
           and I will present two algorithms to generate efficient navigational
           tools for that networks.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   5 / 75
Networks




   What are networks?
           Basically a network is a system that can be modeled with graphs.
           Graphs are mathematical structures consisting of vertices and edges
           connecting the vertices
           When we observe large graphs that exist in nature, societies, or
           systems we refer to them as networks




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   6 / 75
Networks


   What are popular examples of such networks?
           Social networks. Nodes are people and links are acquaintances,
           friendship, and so on.
           Communication networks. Internet: nodes are computers and links
           are cables connecting computers
           Biological networks. Metabolism: nodes are substances and links are
           metabolic reactions
           Information networks. Web: nodes are Web pages and links are
           hyperlinks connecting pages




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   7 / 75
Networks
                            6                                                      How to search in a small world




                                                                                                             Pajek




                            Figure 2: HP Labs’ email communication (light grey lines) mapped onto the organizational
   Figure: Social network of lines). Note that communication tends to “cling” toof formal organizational
                    hierarchy (black HP Labs constructed out the e-mail communication.
                    chart.
   From: How to search a social network, Adamic, 2005.
                            with one another. The h-distance, used to navigate the network, is computed as
                            follows: individuals have h-distance one to their manager and to everyone they share
                            a manager with. Distances are then recursively assigned, so that each individual
                            has h-distance 2 to their first neighbor’s neighbors, and h-distance 3 to their second
Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                             August 31, 2011   8 / 75
Networks




   Figure: Network of pages and hyperlinks on a Website. From: Networks, Mark
   Newman, 2011.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   9 / 75
Structure and Function of Networks



           One of the most important research questions in the study of
           networks: what is the relation between structure and function of
           networks
           For example, the Internet – how should the link structure of the
           Internet look like that supports efficient routing?
           Or how should the link structure of the Web look like to be efficient
           navigable?
           In this presentation we will focus on network navigability!




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   10 / 75
Network Navigability



   Definition
   Put simple, a network is navigable if and only if there is a short path
   between all or almost all pairs of nodes in the network.

   Formally:
       1   There exist a giant component
       2   The effective diameter is low – bounded by log (n), where n is the
           number of nodes in the network




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   11 / 75
Network Navigability
  Knowledge Management Institute




                                      Navigability: Examples
   Example 1:
                 Example 1:


            Not navigable:           No giant component
   Figure: Network is not navigable because there is no giant component, i.e. the
   network is not connected.
                 Example 2:


                 Not navigable:                        giant component, BUT
                                                       eff.diam: 7 > log2(8)
Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   12 / 75
Example 1:
   Network Navigability


                  Not navigable:                       No giant component
   Example 2:
                  Example 2:


                  Not navigable:                       giant component, BUT
   Figure: Now, there is a giant component, i.e. the network is connected. However
   the network is not navigable because eff .diam = > log26 > log2 (8).
                                     eff.diam: 7 6, and (8)


    Denis Helic                                             2010
                                                                                                                7




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   13 / 75
Network Navigability
   Knowledge Management Institute




                                    Navigability: Examples
   Example 3:




   Figure: The network is navigable because there is AND component and
             Navigable: Giant component a giant
   eff .diam = 2. Effective diamater is boundedlog2(10)
                              eff.diam: 2 < by log2 (10).

                   Is this efficiently navigable?
                   There are short paths between all nodes, but can an
                        agent or algorithm find them with local knowledge
                        only?
Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications August 31, 2011 14 / 75
Global Network Navigability



           We discussed so far global network navigability
           Suppose that the network is navigable and we have global knowledge
           of network
           Then it is easy to design efficient procedures to find an arbitrary
           target node from an arbitrary start node
           For example, breadth-first search is such an algorithm that has linear
           time complexity O(n + m), where m is the number of links
           Such procedures are called centralized search




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   15 / 75
Local Network Navigability



           Let us now discuss local network navigability
           Suppose that the network is navigable but we have only local
           knowledge of network
           That means when we arrive at a particular node we know only
           outgoing links from that node and nothing beyond that
           For instance on Facebook we only know our friends or the friends of
           of our friends.
           These procedure are typically called decentralized search




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   16 / 75
Local Network Navigability



           But, how efficient are people in such social search?
           As shown by Millgram’s experiment, people are very efficient in social
           search.
           As shown, people are able to find each other in less than seven hops
           (friends), ∝ log (n)
           Hence, people have an extremely efficient decentralized search
           procedure




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   17 / 75
Local Network Navigability




           How we are able to find other people efficiently?
           Or in other words, what are the properties of social networks, or
           networks in general that make efficient decentralized search possible?
           Are there some structural clues in the network which allows us to
           design sub-linear algorithms?
           And if yes, what are these structural clues?




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   18 / 75
Efficiently navigable
    Local Network Navigability
                   A network is efficiently navigable iff:
                   If there is an algorithm that can find a short path with
                       only local knowledge, and the delivery time of the
                       algorithm is bounded polynomially by logk(n).
     Example:

                                                                             B                                                                  D



                       A                                                                                                       C

                   Efficiently navigable, if the algorithm knows it needs to
                             Figure: A is start node and D is target node.
                      go through A         B      C
J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Denis Helic                                                               2010
                                                                                                                                                                 9




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                           August 31, 2011             19 / 75
Local A networkNavigability navigable iff:
          Network is efficiently
                   If there is an algorithm that can find a short path with
                       only local knowledge, and the delivery time of the
                       algorithm is bounded polynomially by logk(n).
     Step 1:

                                                                             B                                                                  D



                       A                                                                                                       C


     Figure: Efficiently navigable, if the algorithm knows it needs to
             There are two possible paths from A. Obviously, the optimal path leads to
     B. What is the structuralA
                go through property that can guide us in selecting B?
                                    B      C
J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Denis Helic                                                               2010
                                                                                                                                                                10




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                           August 31, 2011             20 / 75
Local A networkNavigability navigable iff:
          Network is efficiently
                   If there is an algorithm that can find a short path with
                       only local knowledge, and the delivery time of the
                       algorithm is bounded polynomially by logk(n).
     Step 1:

                                                                             B                                                                  D



                       A                                                                                                       C


     Figure: Efficiently navigable, if the algorithm knows it needs to
             There are two possible paths from A. Obviously, the optimal path leads to
     B. What is the structuralA
                go through property that can guide us in selecting B?
                                    B      C
J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Denis Helic                                                               2010
     Nodes degree                                                                                                                                               10




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                           August 31, 2011             20 / 75
Local A networkNavigability navigable iff:
          Network is efficiently
                   If there is an algorithm that can find a short path with
                       only local knowledge, and the delivery time of the
                       algorithm is bounded polynomially by logk(n).
     Step 2:

                                                                             B                                                                  D



                       A                                                                                                       C


     Figure: Efficiently navigable,paths from B. Obviously, the it needs to leads
             There are seven possible if the algorithm knows optimal path
     to C. What is through A property that can guide us in selecting C?
                go the structural B        C
J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Denis Helic                                                               2010
                                                                                                                                                                11




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                           August 31, 2011             21 / 75
Local A networkNavigability navigable iff:
          Network is efficiently
                   If there is an algorithm that can find a short path with
                       only local knowledge, and the delivery time of the
                       algorithm is bounded polynomially by logk(n).
     Step 2:

                                                                             B                                                                  D



                       A                                                                                                       C


     Figure: Efficiently navigable,paths from B. Obviously, the it needs to leads
             There are seven possible if the algorithm knows optimal path
     to C. What is through A property that can guide us in selecting C?
                go the structural B        C
J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Denis Helic                                                               2010
     Nodes clustering                                                                                                                                           11




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                           August 31, 2011             21 / 75
Local Network Navigability




   Summarizing, local network navigability requires:
       1   Existence of network hubs that are connected to many nodes
       2   Existence of network clusters where nodes are highly interlinked




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   22 / 75
Local Network Navigability




   Formally:
       1   Power-low degree distribution with exponent γ
       2   High clustering coefficient C




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   23 / 75
γ=2.2
             γ=2.6                                                                            α=1.5          free nor clustered




                                    success pro
γ=2.3
γ=2.4
             γ=2.7                                                                            α=2.0
             γ=2.8                                                                            α=3.0
   Local Network Navigability
γ=2.5
             γ=2.9
             α=1.1
                   0.2
                                                                                              α=5.0
             γ=3.0
                                                            0                                                 IV.      AIR TRAV
         3            4          5                              2      2.2   2.4    2.6       2.8        3
        10
   network size (N)
                     10        10
                                                                        degree exponent (γ)                                  A
                                                            3
                                                                    non-navigable region



                                      degree exponent (γ)
                                                                       We illustrate th
                              2.5                                   structure of netwo    Web of trust

                                                         Metabolic  an example of pa
                                                Internet            to travel from Tok
                   α=5.0
                                2
                                                           Airports the public air tran
                                         navigable region
                                                                    work are airports,
         3            4          50 0.1 0.2 0.3 0.4 0.5 0.6 0.7
    10        10         10                                         is at least one flig
  network size (N)                  clustering coefficient (C)
                                                                    ing to the greedy
 Success probability of greedy routing. Left                        the underlying me
                        Figure: Navigable networks in γ, C space. the next-hop airpo
ccess probability ps as a function of network size N
ent values of γ with weak (top) and strong (bottom)                 nation. Under th
g. The top-right plot shows ps as a function of γ                   Bethel, then to An
r networks of fixed size N ≈ 105 . In the bottom-                    to Paris, then to V
t, parameter α is mapped to Navigability: Theorycoefficient
                                           Networks clustering
 Denis Helic & Christoph Trattner (KMI, TU Graz)               and Applications                               August 31, 2011   24 / 75
A network is efficiently navigable iff:
   Local If there is an algorithm that can find a short path with
          Network Navigability
             only local knowledge, and the delivery time of the
             algorithm is bounded polynomially by logk(n).
    Revisiting Step 2:

                                                                             B                                                                  D



                       A                                                     E                                                 C




J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Figure: There are seven possible paths from B. Obviously, the optimal path leads
    Denis Helic                           2010

    to C. What is an additional hint that can guide us in selecting C over E?     12




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                          August 31, 2011             25 / 75
A network is efficiently navigable iff:
   Local If there is an algorithm that can find a short path with
          Network Navigability
             only local knowledge, and the delivery time of the
             algorithm is bounded polynomially by logk(n).
    Revisiting Step 2:

                                                                             B                                                                  D



                       A                                                     E                                                 C




J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science
Technical Report 99-1776 (October 1999)
    Figure: There are seven possible paths from B. Obviously, the optimal path leads
    Denis Helic                           2010

    to C. What is an additional hint that can guide us in selecting C over E?     12




    Nodes similarity

Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                          August 31, 2011             25 / 75
Local Network Navigability




           Nodes similarity is external to the network
           It is derived from some additional information that we have about
           network nodes
           In Millgram’s experiment people selected the next person according to
           their occupation or geography




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   26 / 75
Local Network Navigability




           All of this information, i.e. degrees, clustering, similarity can be
           understood as a kind of our background knowledge about the network
           We use this background knowledge to guide us in our search for a
           target node
           When we have more than one link to follow we consult the
           background knowledge and ask which of the links will lead us with
           highest probability to a given target node




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   27 / 75
Greedy Decentralized Search



           On the next abstraction level we can say that background knowledge
           defines a notion of distance between nodes
           In other words, background knowledge is a metric space where each
           node has unique coordinates and we can calculate the distance
           between nodes
           Or in other words, we can abstract background knowledge as a
           black-box executing a simple function:
           getDistance(node, target node)




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   28 / 75
Greedy Decentralized Search



           Let us now take an algorithmic perspective on decentralized search
           We start at an arbitrary node and need to find as fast as possible a
           target node having only local knowledge of the network
           In addition, we have background knowledge represented through
           getDistance(node, target node) function
           At each search step we have to make a decision which of the available
           links to follow




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   29 / 75
Greedy Decentralized Search



           In order to maximize the probability of finding the target node we
           always select a node which has the smallest distance to the target
           node
           It has been shown that the greedy algorithm is very efficient, i.e. the
           number of steps to reach an arbitrary target node is ∝ log (n)
           Kleinberg proved it theoretically, Watts by simulation
           Watts was able to reproduce Millgram’s experiment with proper
           selection of parameters: Identity and Search in Social Networks, 2002




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   30 / 75
Background Knowledge



           Now, how does our background knowledge of people typically look
           like?
           It is a metric space, e.g. 1-D spaces, 2-D vector spaces, 3-D Euclidean
           spaces, hyperbolic spaces, ... or does it look like completely different?
           Actually, it was observed by Kleinberg and also by Watts that a
           hierarchy of nodes is also a very good approximation of how people
           think
           Hence, we will also use hierarchical background knowledge




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   31 / 75
Hierarchy as a Metric Space



                                                                                     2    3


                                                                                                                2
                                                              3                           3
                                                                   23                24                    21


                                         4                              4        4        4            1
                                                 1                 15       22       25                    3


                           5                 5       5             5                               1            1
                               11   12                   13   14                              31           32       33




                                         Figure: Node distances in a hierarchy.


   Distance: d(i, j) = h(i) + h(j) − 2h(lca(i, j)) − 1



Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                            August 31, 2011   32 / 75
Example of a Greedy Navigation

                                                                                                               3
                                                                                                          2

                                                                                                                                               2
                                                                          3                                    3
                                                                               23                         24                          21
                                                                                                2                                                  3
                                                    4                                  4             4         4                  1
                                                            1                  15               22        25                          3
                                                                                                                                                       4
                                                                     1
                                 5                      5       5                  5                                         1                 1
                                      11       12                   13    14                                            31            32                   33




                                                        11                     2                     21             3                     31


                                     12

                                           1            1                                  22        2                                     3           4


                                     13                                                    23                  25            32                            33

                                                        14               15                          24




                                                                         Figure: Greedy search.



Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                                                        August 31, 2011   33 / 75
Calculating Network Navigability



           Now in order to measure network navigability, we developed a
           theoretical framework to estimate network navigability by simulations
           As input we take a network, e.g. information network like Wikipedia,
           or Delicious
           and a suitable hierarchy that models background knowledge
           For example, Wikipedia categories or Delicious folksonomy
           and simulate decentralized search on 106 start and target node pairs




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   34 / 75
Network Navigability Simulation Framework



           The metrics we measure by our framework are
           success rate s
           and stretch τ
           For both metrics we calculate distributions over global shortest path

   Definition
   Stretch: τ = h , where h is the number of simulator steps and l is the
                 l
   global shortest path.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   35 / 75
Evaluating hierarchies



           The framework lets you e.g. estimate the quality of a hierarchy to
           serve as background knowledge
           A hierarchy with better navigational properties will have better
           success rate and stretch in comparison with other hierarchies
           For example, Wikipedia categories versus Delicious tags
           For example, different folksonomies for navigating social tagging
           systems, see Helic et al.: Pragmatic Evaluation of Folksonomies, 2011




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   36 / 75
Evaluating Navigational Tools



           But we can use framework to estimate the effects of changes in the
           network on its navigational properties
           For example, how navigable is Wikipedia now?
           How navigable will be Wikipedia if we include Delicious tags?
           How navigable will be Wikipedia if we include breadcrumbs?
           We take Wikipedia as the starting network and create new links in the
           network to emulate Delicious tags, breadcrumbs, etc.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   37 / 75
Evaluating folksonomies




           A folksonomy is a hierarchy that is automatically generated from a
           tagging system
           Today there are several folksonomy algorithms, see e.g. Heymann
           2008, or Benz 2010
           In addition, you can produce folksonomies by using standard
           hierarchical clustering methods such as K-Means or Affinity
           Propagation




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   38 / 75
Evaluating folksonomies




           In Helic et al.: Pragmatic Evaluation of Folksonomies, WWW2011 we
           took 5 tagging datasets and 5 different folksonomy algorithms
           We produced 5x5 folksonomies and simulated (100.000 samples)
           greedy decentralized search on the datasets
           We measured the success rate and stretch to see if some folksonomies
           perform better than the other ones.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   39 / 75
Evaluating folksonomies


                                                                              Greedy Search Success Rate: BibSonomy
                                                                    100
                                                                                                                Folksonomy
                                                                                                               Random
                                                                                                               Aff.Prop.


                                        Success Rate (Percentage)
                                                                    80                                         K-Means
                                                                                                              Deg/Cooc
                                                                                                                Clo/Cos
                                                                    60


                                                                    40


                                                                    20


                                                                     0
                                                                          1   2       3         4         5           6      7
                                                                                          Shortest path




                  Figure: Success Rate of different folksonomies in BibSonomy




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                         August 31, 2011   40 / 75
Evaluating folksonomies


                                                                                  Greedy Search Success Rate: CiteULike
                                                                    100
                                                                                                                    Folksonomy
                                                                                                                   Random
                                                                                                                   Aff.Prop.


                                        Success Rate (Percentage)
                                                                    80                                             K-Means
                                                                                                                  Deg/Cooc
                                                                                                                    Clo/Cos
                                                                    60


                                                                    40


                                                                    20


                                                                     0
                                                                          1   2           3         4         5           6      7
                                                                                              Shortest path




                   Figure: Success Rate of different folksonomies in CiteULike




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                             August 31, 2011   41 / 75
Evaluating folksonomies


                                                                                  Greedy Search Success Rate: Delicious
                                                                    100
                                                                                                                    Folksonomy
                                                                                                                   Random
                                                                                                                   Aff.Prop.


                                        Success Rate (Percentage)
                                                                    80                                             K-Means
                                                                                                                  Deg/Cooc
                                                                                                                    Clo/Cos
                                                                    60


                                                                    40


                                                                    20


                                                                     0
                                                                          1   2           3         4         5           6      7
                                                                                              Shortest path




                    Figure: Success Rate of different folksonomies in Delicious




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                             August 31, 2011   42 / 75
Evaluating folksonomies


                                                                                  Greedy Search Success Rate: Flickr
                                                                    100
                                                                                                                Folksonomy
                                                                                                               Random
                                                                                                               Aff.Prop.


                                        Success Rate (Percentage)
                                                                    80                                         K-Means
                                                                                                              Deg/Cooc
                                                                                                                Clo/Cos
                                                                    60


                                                                    40


                                                                    20


                                                                     0
                                                                          1   2             3            4             5     6
                                                                                            Shortest path




                       Figure: Success Rate of different folksonomies in Flickr




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                         August 31, 2011   43 / 75
Evaluating folksonomies


                                                                              Greedy Search Success Rate: LastFM
                                                                    100
                                                                                                            Folksonomy
                                                                                                           Random
                                                                                                           Aff.Prop.


                                        Success Rate (Percentage)
                                                                    80                                     K-Means
                                                                                                          Deg/Cooc
                                                                                                            Clo/Cos
                                                                    60


                                                                    40


                                                                    20


                                                                     0
                                                                          1     2              3             4           5
                                                                                         Shortest path




                     Figure: Success Rate of different folksonomies in LastFM




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                     August 31, 2011   44 / 75
Evaluating folksonomies




           Centrality-based algorithms such as Heymann 2008 or Benz 2010
           outperform traditional methods
           However, these are all theoretical results
           Because, what is if we wanted to embed folksonomies in the user
           interface (UI) to support users in their navigation tasks
           and the space in user interface is limited?




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   45 / 75
Embedding folksonomies in UI     Google Directory - Computers > Internet > On the Web > Online Communities




                                                                                                                                                   Directory Help

                                   Online Communities
                                      Computers > Internet > On the Web > Online Communities                                                Go to Directory Home

                                   Categories

                                     Bulletin Board                           Directories (9)                   PowerPets (6)
                                     Systems (132)                            Events (1)                        Second Life (119)
                                     By Region (8)                            Mailing Lists (85)                Social Networking
                                     By Subject (204)                         Message Boards                    (222)
                                     Chat (745)                               (154)                             Software and
                                     Community                                MySpace (28)                      Services (27)
                                     Management (36)                          Neopets (171)                     The Palace (51)
                                     Community                                                                  Zetapets (3)
                                     Providers (14)

                                   Related Categories:
                                     Society > Activism > Community Building (26)
                                     Society > Organizations (16987)
                                     Society > People > Personal Homepages (8890)
                                     Society > Relationships > Cyber Relationships (59)
                                     Society > Subcultures > Cyberculture (162)

                                   Web Pages                  Viewing in Google PageRank order                                    View in alphabetical order
                                             Talk City - http://www.talkcity.com/
                                      Figure: Directory Based Navigation technology, health and other
                                         Participate in discussions about relationships, hobbies, business,
                                           topics. Socialize with friends, or start your own chat group.
                                           Whyville - http://www.whyville.net/
                                           A virtual 3-D world for curious minds where you can own land, build your own house, play
                                           simulation games, win prizes, chat, and help the community grow.
                                           Buzznet - http://www.buzznet.com/
Denis Helic & Christoph Trattner (KMI,   TU Graz) create communitiesTheory and Applications
                                           Users can
                                             Networks Navigability: and share blogs and photographs.                 August 31, 2011                                46 / 75
Embedding folksonomies in UI




           We have breadcrumbs connecting each node all the way up to the
           root node
           We have limited number of subcategories (n)
           We have limited number of related categories (m)
           Now we embed folksonomy as in Benz 2010 and apply different
           restrictions




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   47 / 75
Embedding folksonomies in UI



                                                        -       Greedy Navigator (1000000 Runs)
                                                                    -
                                                        l=3.585123, h=5.936013, sg=0.005548, τg=1.655735

                                                                                       Success Rate (s)
                                               3                                             Stretch (τ)

                                              2.5

                                               2
                                       s, τ



                                              1.5

                                               1

                                              0.5

                                               0
                                                    1    2      3       4         5       6      7         8   9
                                                                            Shortest Path




         Figure: Success Rate and stretch in BibSonomy with n = 20 and m = 20




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                           August 31, 2011   48 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.634634, h=6.536937, sg=0.001110, τg=1.798513

                                              9                                      Success Rate (s)
                                                                                           Stretch (τ)
                                              8
                                              7
                                              6
                                              5
                                       s, τ




                                              4
                                              3
                                              2
                                              1
                                              0
                                                  1   2      3       4         5       6      7          8   9
                                                                         Shortest Path




          Figure: Success Rate and stretch in CiteULike with n = 20 and m = 20




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                         August 31, 2011   49 / 75
Embedding folksonomies in UI



                                                          -       Greedy Navigator (1000000 Runs)
                                                                      -
                                                          l=3.518932, h=5.557032, sg=0.000903, τg=1.579181

                                              7                                         Success Rate (s)
                                                                                              Stretch (τ)
                                              6

                                              5

                                              4
                                       s, τ




                                              3

                                              2

                                              1

                                              0
                                                  1   2      3     4     5      6     7    8    9    10      11   12
                                                                             Shortest Path




           Figure: Success Rate and stretch in Delicious with n = 20 and m = 20




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                               August 31, 2011   50 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.467684, h=4.162304, sg=0.000382, τg=1.200312

                                                                                     Success Rate (s)
                                              7                                            Stretch (τ)

                                              6

                                              5
                                       s, τ


                                              4

                                              3

                                              2

                                              1

                                              0
                                                  1   2      3       4         5       6      7          8   9
                                                                         Shortest Path




             Figure: Success Rate and stretch in Flickr with n = 20 and m = 20




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                         August 31, 2011   51 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.197477, h=6.662900, sg=0.001062, τg=2.083799

                                                                                    Success Rate (s)
                                              6                                           Stretch (τ)

                                              5

                                              4
                                       s, τ



                                              3

                                              2

                                              1

                                              0
                                                  1       2            3            4            5       6
                                                                       Shortest Path




            Figure: Success Rate and stretch in LastFM with n = 20 and m = 20




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                     August 31, 2011   52 / 75
Embedding folksonomies in UI




   Under this restriction the navigator in Considering practical user interface
   restriction folksonomies are useless for supporting navigation. The success
   rate drops below 1%.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   53 / 75
Embedding folksonomies in UI


           Thus, folksonomies (unlimited) are useful theoretically but useless
           practically
           The problem is that top nodes have many children (possibly
           thousands) and UI restrictions cut to many children nodes off
           Hence, we need a new algorithm that takes into account these UI
           restrictions
           Technically, we need to able to determine the branching factor for the
           hierarchy
           We developed such an algorithm and published in CIKM2011. Helic
           et al. Building Directories for Social Tagging Systems
           We were able to almost recover theoretical navigability


Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   54 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.585123, h=8.691685, sg=1.000000, τg=2.424376
                                              7
                                                                                     Success Rate (s)
                                                                                           Stretch (τ)
                                              6

                                              5

                                              4
                                       s, τ




                                              3

                                              2

                                              1

                                              0
                                                  1   2      3       4         5       6      7          8   9
                                                                         Shortest Path




    Figure: Success Rate and stretch in BibSonomy with new folksonomy algorithm




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                         August 31, 2011   55 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.634634, h=9.163688, sg=1.000000, τg=2.521213

                                              7                                      Success Rate (s)
                                                                                           Stretch (τ)
                                              6

                                              5

                                              4
                                       s, τ




                                              3

                                              2

                                              1

                                              0
                                                  1   2      3       4         5       6      7          8   9
                                                                         Shortest Path




      Figure: Success Rate and stretch in CiteULike with new folksonomy algorithm




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                         August 31, 2011   56 / 75
Embedding folksonomies in UI



                                                          -       Greedy Navigator (1000000 Runs)
                                                                      -
                                                          l=3.518932, h=9.720769, sg=1.000000, τg=2.762420

                                              6                                         Success Rate (s)
                                                                                              Stretch (τ)
                                              5

                                              4
                                       s, τ



                                              3

                                              2

                                              1

                                              0
                                                  1   2      3     4     5      6     7    8    9    10      11   12
                                                                             Shortest Path




      Figure: Success Rate and stretch in Delicious with new folksonomy algorithm




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                               August 31, 2011   57 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.467684, h=8.886960, sg=0.996066, τg=2.562794

                                                                                     Success Rate (s)
                                              6                                            Stretch (τ)

                                              5

                                              4
                                       s, τ



                                              3

                                              2

                                              1

                                              0
                                                  1   2      3       4         5       6      7          8   9
                                                                         Shortest Path




         Figure: Success Rate and stretch in Flickr with new folksonomy algorithm




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                         August 31, 2011   58 / 75
Embedding folksonomies in UI



                                                      -       Greedy Navigator (1000000 Runs)
                                                                  -
                                                      l=3.197477, h=9.830726, sg=1.000000, τg=3.074526
                                              6
                                                                                    Success Rate (s)
                                                                                          Stretch (τ)
                                              5

                                              4
                                       s, τ


                                              3

                                              2

                                              1

                                              0
                                                  1       2            3            4            5       6
                                                                       Shortest Path




       Figure: Success Rate and stretch in LastFM with new folksonomy algorithm




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                     August 31, 2011   59 / 75
Why usefulness of folksonomies for navigation is limited?


           Even if folksonomies allow the user to navigate to related concepts in
           an efficient manner navigation to a particular resource is still a
           problem
           As shown related work, in tagging systems the tag-resource
           distribution follows a power-law function, i.e. there are many tags
           that refer to a large number of resources.
           In BibSonomy or CiteULike for instance there are tags, which refer to
           hundreds or even thousands of resources.
           To display such long resource lists, developers typically paginate the
           resource lists in a tagging system by a certain factor k




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   60 / 75
Why usefulness of folksonomies for navigation is limited?




             (a) Austria-Forum                       (b) BibSonomy                         (c) CiteULike

                                             Figure: Tag distributions.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications      August 31, 2011   61 / 75
Creating tag-resource Taxonomies

           To support the user not only to navigate to related tags in efficient
           manner but also to the resources of a tagging system, we invented
           the approach of the so-called tag-resource taxonomies.

                                       Car

                                                                                     Car
                              Tire                Motor



                                                                           Tire                 Motor
                   Mercedes    VOLVO         VW           BMW




                                                                 VW           BMW          VW             BMW



                              (a) Folksonomy                          (b) Tag-Resource Taxonomy

   Figure: Folksonomy vs. Tag-Resource Taxonomy. In a Folksonomy tags appear
   only once. However, resources can be referred by different tags. In a tag-resource
   taxonomy on the other hand resources can occur only once while tags can appear
   on multiple and on different levels.

Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                August 31, 2011   62 / 75
Why usefulness of folksonomies for navigation is limited?



           In the worst case a user would have to click max{click(Ttag )} times
           to reach a desired resource with the help of a Folksonomy.

                                                         c1 |r |
                       max{click(Ttag )} =                       + logb/2 (c2 · |r |), b ≥ 2                 (1)
                                                           k
   or
                                                                            c1 · |r |
                                          max{click(Ttag )} ≈                                                (2)
                                                                               k
                                                          c1 ·|r |
   supposing that logb/2 (c2 · |r |)                        k




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   63 / 75
Why usefulness of folksonomies for navigation is limited?



           The worst case scenario of a tag-resource taxonomy is significantly
           better. In the worst case a user would have to click max{click(Tres )}
           times to reach a desired target resource.
                max{click(Tres )} = max{depth(Tres )} = logk/2 |r | , k ≥ 2                                  (3)
   Then for large values of |r | we have:

                                                                     c1 · |r |
                                                logk/2 |r |                                                  (4)
                                                                        k




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   64 / 75
Why usefulness of folksonomies for navigation is
   limited?xxx




                                               Austria-Forum               BibSonomy       CiteULike
              max{click(Ttag )}                     184                       5,278         20,799
              max{click(Tres )}                     6.1                        7.7            8.5
   Table: Tag Taxonomy vs. Tag-Resource Taxonomy: Maximum number of clicks
   for k = 10.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   65 / 75
Why usefulness of folksonomies for navigation is limited?


           To calculate the number of tags suffering from the so-called
           pagination effect, we can user the following equation:
                                                                       1
                                                                 α 1 (α)
                                            |tp | = |t| ·          −                                          (5)
                                                                 k   k

                                    Austria-Forum               BibSonomy               CiteULike
                  |tp | (%)          5079 (38%)                 7401 (28%)             51748 (32%)
                              Table: Number of paginated tags for k = 10.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications    August 31, 2011   66 / 75
Why usefulness of folksonomies for navigation is limited?

   The mean number of clicks is calculated as follows:
   Tag-Resource Taxonomy: mean{click(Tres )} = logk (|r |)
                                                       |t|
   Folksonomy: mean{click(Ttag )} = logk (|t|) + |t| i=1 ri
                                                  1
                                                           k

                                                k      Austria-Forum         BibSonomy     CiteULike
                   mean{click(Tres )}           2          14.2                 17.8          19.8
                   mean{click(Ttag )}           2           29.5                22.4          30.7
                   mean{click(Tres )}           5           6.1                  7.6          8.5
                   mean{click(Ttag )}           5           11.6                 9.2          12.3
                   mean{click(Tres )}          10           4.3                  5.3          5.9
                   mean{click(Ttag )}          10            6.4                 5.6          7.3

   Table: Tag Taxonomy vs. Tag-Resource Taxonomy: Mean number of clicks for
   different branching factors k.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications    August 31, 2011   67 / 75
Creating tag-resource Taxonomies


           1. Computer Degree centrality of the resource-to-resource tag network
           2. Take most general resource as root an attach max. b resources as
           childs. Child-nodes are selected according their cos-sim values.
           3. After that we take the resource taxonomy and apply labels (tags)
           to the resource (top-down, in left-order)
           3.1 We calculate candidate labels by the method of co-occurance, i.e.
           we take the tags of the related resources into account to rank the
           actual tags of the currently processed resource.
           3.2. If the candidate tag has already been applied to one of the
           parent resources of the currently processed resource we take the next
           candidate tag from the co-occurance vector and try to apply it.



Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   68 / 75
Evaluating Tag-Resource Taxonomies



           In the first experiment we measured the average and maximum
           number of clicks and the drop rate

                       Name         b        n          max{click(Tres )}         mean{click(Tres )}
                       Res2         2      19,430             17                       12.45
                       Res5         5      19,430             10                       5.93
                       Res10        10     19,430              8                       4.44

   Table: max{click(Tres )} and mean{click(Tres )} for different branching factors b.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications       August 31, 2011   69 / 75
Evaluating Tag-Resource Taxonomies



           In the second experiment we measured the number of collisions

                                           Name        b        n          CR (%)
                                           Res2        2      19,430        0.1%
                                           Res5        5      19,430        0.2%
                                           Res10       10     19,430        0.2%

   Table: Collision Rates (CR) for different resource taxonomies with different
   branching factor b.




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   70 / 75
Evaluating Tag-Resource Taxonomies
           In the third experiment we measured the semantic structure of the
           tag-resource taxonomy compared to popular folksonomy induction
           algorithms such as Heymann, K-Means, Affinity Propagation and
           Co-Occurance
           As measure for this experiment we used Taxonomic Recall/Prec. and
           overlap.
           Ground truth: Germanet ontholoy
                                                   0.4
                                                         Taxonomic F−Measure
                                                  0.35   Taxonomic Overlap


                                                   0.3
                               Count (1 = 100%)




                                                  0.25


                                                   0.2


                                                  0.15


                                                   0.1


                                                  0.05


                                                    0
                                                         Res2     Res5     Res10   Deg/Cooc Aff. Prop   K−Means Heymann


Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications                                  August 31, 2011   71 / 75
Evaluating Tag-Resource Taxonomies

           In the fourth and last experiment a user study was conducted to test
           weather the approach is also useful for humans or not
           As ground truth for the experiment the best so far known folksonomy
           generation approach was used
           All over we had 9 test users who had to judge 200 tag trails extracted
           from both hierarchies


     Name                    b       Correct (%)           Related (%)           Equivalent (%)    Not Related (%
     Deg/Cooc10              10         33.2                   27.3                    13               21.9
     Res10                   10         27.3                   36.2                   12.3              19.8
   Table: Results of the empirical analysis of the tag-resource taxonomy with
   branching factor b = 10 compared to a Deg/Cooc folksonomy with branching
   factor b = 10.

Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   72 / 75
End of presentation




                               Thank you very much for your attention!
                               Christoph Trattner (ctrattner@iicm.edu)




Denis Helic & Christoph Trattner (KMI, TU Graz)
                                          Networks Navigability: Theory and Applications   August 31, 2011   73 / 75

More Related Content

What's hot

Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
Leslie
 
Networking Theories
Networking TheoriesNetworking Theories
Networking Theories
Leslie
 
Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
Leslie
 
ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...
ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...
ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...
eswcsummerschool
 
VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...
VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...
VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...
ijwmn
 
SCID Final meeting - Staging Abstraction
SCID Final meeting - Staging AbstractionSCID Final meeting - Staging Abstraction
SCID Final meeting - Staging Abstraction
Bruce Edmonds
 
Hybrid MPI: Efficient Message Passing for Multi-core Systems
Hybrid MPI: Efficient Message Passing for Multi-core Systems Hybrid MPI: Efficient Message Passing for Multi-core Systems
Hybrid MPI: Efficient Message Passing for Multi-core Systems
TakuyaFukuoka2
 
Pervasive computing write up
Pervasive computing write upPervasive computing write up
Pervasive computing write up
WhoGoesThere
 

What's hot (18)

Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
 
Networking Theories
Networking TheoriesNetworking Theories
Networking Theories
 
Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
 
Networking Updated 4.12.10
Networking Updated 4.12.10Networking Updated 4.12.10
Networking Updated 4.12.10
 
Visualizing Networks
Visualizing NetworksVisualizing Networks
Visualizing Networks
 
Structure Function Fsi Wignall
Structure Function Fsi WignallStructure Function Fsi Wignall
Structure Function Fsi Wignall
 
Does artificial neural network support connectivism’s assumptions?
Does artificial neural network support connectivism’s assumptions?Does artificial neural network support connectivism’s assumptions?
Does artificial neural network support connectivism’s assumptions?
 
ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...
ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...
ESWC SS 2013 - Monday Keynote Stefan Decker: From Linked Data to Networked Kn...
 
DireWolf Goes Pack Hunting: A Peer-to-Peer Approach for Secure Low Latency Wi...
DireWolf Goes Pack Hunting: A Peer-to-Peer Approach for Secure Low Latency Wi...DireWolf Goes Pack Hunting: A Peer-to-Peer Approach for Secure Low Latency Wi...
DireWolf Goes Pack Hunting: A Peer-to-Peer Approach for Secure Low Latency Wi...
 
cloud and edge networking technologies
cloud and edge networking technologiescloud and edge networking technologies
cloud and edge networking technologies
 
UK e-Infrastructure: Widening Access, Increasing Participation
UK e-Infrastructure: Widening Access, Increasing ParticipationUK e-Infrastructure: Widening Access, Increasing Participation
UK e-Infrastructure: Widening Access, Increasing Participation
 
Convoluional neural networks
Convoluional neural networksConvoluional neural networks
Convoluional neural networks
 
A comparative study between cloud computing and fog
A comparative study between cloud computing and fog A comparative study between cloud computing and fog
A comparative study between cloud computing and fog
 
VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...
VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...
VIRTUAL ARCHITECTURE AND ENERGYEFFICIENT ROUTING PROTOCOLS FOR 3D WIRELESS SE...
 
SCID Final meeting - Staging Abstraction
SCID Final meeting - Staging AbstractionSCID Final meeting - Staging Abstraction
SCID Final meeting - Staging Abstraction
 
Atul
AtulAtul
Atul
 
Hybrid MPI: Efficient Message Passing for Multi-core Systems
Hybrid MPI: Efficient Message Passing for Multi-core Systems Hybrid MPI: Efficient Message Passing for Multi-core Systems
Hybrid MPI: Efficient Message Passing for Multi-core Systems
 
Pervasive computing write up
Pervasive computing write upPervasive computing write up
Pervasive computing write up
 

Similar to Networks Navigability: Theory and Applications

On the Navigability of Social Tagging Systems
On the Navigability of Social Tagging SystemsOn the Navigability of Social Tagging Systems
On the Navigability of Social Tagging Systems
Christoph Trattner
 
Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
Leslie
 
Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
Leslie
 
On the Utility of Tags for Search and Navigation in Online Information Systems
On the Utility of Tags for Search and Navigation in Online Information SystemsOn the Utility of Tags for Search and Navigation in Online Information Systems
On the Utility of Tags for Search and Navigation in Online Information Systems
Christoph Trattner
 
COMMUNICATIONS OF THE ACM November 2004Vol. 47, No. 11 15.docx
COMMUNICATIONS OF THE ACM November  2004Vol. 47, No. 11 15.docxCOMMUNICATIONS OF THE ACM November  2004Vol. 47, No. 11 15.docx
COMMUNICATIONS OF THE ACM November 2004Vol. 47, No. 11 15.docx
monicafrancis71118
 
Temporal Networks of Human Interaction
Temporal Networks of Human InteractionTemporal Networks of Human Interaction
Temporal Networks of Human Interaction
Petter Holme
 
THIC MedIX Summer 2015 Poster
THIC MedIX Summer 2015 PosterTHIC MedIX Summer 2015 Poster
THIC MedIX Summer 2015 Poster
Diana Zajac
 

Similar to Networks Navigability: Theory and Applications (20)

On the Navigability of Social Tagging Systems
On the Navigability of Social Tagging SystemsOn the Navigability of Social Tagging Systems
On the Navigability of Social Tagging Systems
 
On the Navigability of Social Tagging Systems
On the Navigability of Social Tagging SystemsOn the Navigability of Social Tagging Systems
On the Navigability of Social Tagging Systems
 
A ROUTING MECHANISM BASED ON SOCIAL NETWORKS AND BETWEENNESS CENTRALITY IN DE...
A ROUTING MECHANISM BASED ON SOCIAL NETWORKS AND BETWEENNESS CENTRALITY IN DE...A ROUTING MECHANISM BASED ON SOCIAL NETWORKS AND BETWEENNESS CENTRALITY IN DE...
A ROUTING MECHANISM BASED ON SOCIAL NETWORKS AND BETWEENNESS CENTRALITY IN DE...
 
EVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKS
EVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKSEVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKS
EVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKS
 
Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
 
Networking Theories Presentation
Networking Theories PresentationNetworking Theories Presentation
Networking Theories Presentation
 
Wanted: a larger, different kind of box
Wanted: a larger, different kind of boxWanted: a larger, different kind of box
Wanted: a larger, different kind of box
 
Wanted: a larger, different kind of box
Wanted: a larger, different kind of boxWanted: a larger, different kind of box
Wanted: a larger, different kind of box
 
EVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKS
EVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKSEVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKS
EVOLUTIONARY CENTRALITY AND MAXIMAL CLIQUES IN MOBILE SOCIAL NETWORKS
 
SIX DEGREES OF SEPARATION TO IMPROVE ROUTING IN OPPORTUNISTIC NETWORKS
SIX DEGREES OF SEPARATION TO IMPROVE ROUTING IN OPPORTUNISTIC NETWORKSSIX DEGREES OF SEPARATION TO IMPROVE ROUTING IN OPPORTUNISTIC NETWORKS
SIX DEGREES OF SEPARATION TO IMPROVE ROUTING IN OPPORTUNISTIC NETWORKS
 
Six Degrees of Separation to Improve Routing in Opportunistic Networks
Six Degrees of Separation to Improve Routing in Opportunistic NetworksSix Degrees of Separation to Improve Routing in Opportunistic Networks
Six Degrees of Separation to Improve Routing in Opportunistic Networks
 
On the Utility of Tags for Search and Navigation in Online Information Systems
On the Utility of Tags for Search and Navigation in Online Information SystemsOn the Utility of Tags for Search and Navigation in Online Information Systems
On the Utility of Tags for Search and Navigation in Online Information Systems
 
Node similarity
Node similarityNode similarity
Node similarity
 
Class 1_ Introduction.pdf
Class 1_ Introduction.pdfClass 1_ Introduction.pdf
Class 1_ Introduction.pdf
 
COMMUNICATIONS OF THE ACM November 2004Vol. 47, No. 11 15.docx
COMMUNICATIONS OF THE ACM November  2004Vol. 47, No. 11 15.docxCOMMUNICATIONS OF THE ACM November  2004Vol. 47, No. 11 15.docx
COMMUNICATIONS OF THE ACM November 2004Vol. 47, No. 11 15.docx
 
MODELING SOCIAL GAUSS-MARKOV MOBILITY FOR OPPORTUNISTIC NETWORK
MODELING SOCIAL GAUSS-MARKOV MOBILITY FOR OPPORTUNISTIC NETWORK MODELING SOCIAL GAUSS-MARKOV MOBILITY FOR OPPORTUNISTIC NETWORK
MODELING SOCIAL GAUSS-MARKOV MOBILITY FOR OPPORTUNISTIC NETWORK
 
Temporal Networks of Human Interaction
Temporal Networks of Human InteractionTemporal Networks of Human Interaction
Temporal Networks of Human Interaction
 
DTN
DTNDTN
DTN
 
THIC MedIX Summer 2015 Poster
THIC MedIX Summer 2015 PosterTHIC MedIX Summer 2015 Poster
THIC MedIX Summer 2015 Poster
 
A Comprehensive Study on Vehicular Ad-Hoc Delay Tolerant Networking for Infra...
A Comprehensive Study on Vehicular Ad-Hoc Delay Tolerant Networking for Infra...A Comprehensive Study on Vehicular Ad-Hoc Delay Tolerant Networking for Infra...
A Comprehensive Study on Vehicular Ad-Hoc Delay Tolerant Networking for Infra...
 

More from Christoph Trattner

Towards health (aware) food recommender systems
Towards health (aware) food recommender systemsTowards health (aware) food recommender systems
Towards health (aware) food recommender systems
Christoph Trattner
 
Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...
Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...
Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...
Christoph Trattner
 
Cognitive Models in Recommender Systems
Cognitive Models in Recommender SystemsCognitive Models in Recommender Systems
Cognitive Models in Recommender Systems
Christoph Trattner
 
Crowd-Based Evaluation Methods
Crowd-Based Evaluation MethodsCrowd-Based Evaluation Methods
Crowd-Based Evaluation Methods
Christoph Trattner
 
Recommending Tags with a Model of Human Categorization
Recommending Tags with a Model of Human CategorizationRecommending Tags with a Model of Human Categorization
Recommending Tags with a Model of Human Categorization
Christoph Trattner
 
Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...
Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...
Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...
Christoph Trattner
 
Why Social Network Research is good for you!
Why Social Network Research is good for you!Why Social Network Research is good for you!
Why Social Network Research is good for you!
Christoph Trattner
 
Enhancing the Navigability of Social Tagging Systems with Tag Taxonomies
Enhancing the Navigability of Social Tagging Systems with Tag TaxonomiesEnhancing the Navigability of Social Tagging Systems with Tag Taxonomies
Enhancing the Navigability of Social Tagging Systems with Tag Taxonomies
Christoph Trattner
 

More from Christoph Trattner (17)

Towards health (aware) food recommender systems
Towards health (aware) food recommender systemsTowards health (aware) food recommender systems
Towards health (aware) food recommender systems
 
Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...
Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...
Investigating the Healthiness of Internet-Sourced Recipes: Implications for M...
 
Understanding the Impact of Weather for POI Recommendations
Understanding the Impact of Weather for POI RecommendationsUnderstanding the Impact of Weather for POI Recommendations
Understanding the Impact of Weather for POI Recommendations
 
Studying Online Food Consumption and Production Patterns: Recent Trends and C...
Studying Online Food Consumption and Production Patterns: Recent Trends and C...Studying Online Food Consumption and Production Patterns: Recent Trends and C...
Studying Online Food Consumption and Production Patterns: Recent Trends and C...
 
Cognitive Models in Recommender Systems
Cognitive Models in Recommender SystemsCognitive Models in Recommender Systems
Cognitive Models in Recommender Systems
 
Crowd-Based Evaluation Methods
Crowd-Based Evaluation MethodsCrowd-Based Evaluation Methods
Crowd-Based Evaluation Methods
 
From Search to Predictions in Tagged Information Spaces
From Search to Predictions in Tagged Information SpacesFrom Search to Predictions in Tagged Information Spaces
From Search to Predictions in Tagged Information Spaces
 
Towards a Big Data Recommender Engine for Online and Offline Marketplaces
Towards a Big Data Recommender Engine for Online and Offline MarketplacesTowards a Big Data Recommender Engine for Online and Offline Marketplaces
Towards a Big Data Recommender Engine for Online and Offline Marketplaces
 
Recommending Items in Social Tagging Systems Using Tag and Time Information
Recommending Items in Social Tagging Systems Using Tag and Time InformationRecommending Items in Social Tagging Systems Using Tag and Time Information
Recommending Items in Social Tagging Systems Using Tag and Time Information
 
Social Computing in the area of Big Data at the Know-Center Austria's leading...
Social Computing in the area of Big Data at the Know-Center Austria's leading...Social Computing in the area of Big Data at the Know-Center Austria's leading...
Social Computing in the area of Big Data at the Know-Center Austria's leading...
 
Recommending Tags with a Model of Human Categorization
Recommending Tags with a Model of Human CategorizationRecommending Tags with a Model of Human Categorization
Recommending Tags with a Model of Human Categorization
 
Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...
Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...
Je t’aime… moi non plus: reporting on the opportunities, expectations and cha...
 
The Smart Pharma Wiki
The Smart Pharma WikiThe Smart Pharma Wiki
The Smart Pharma Wiki
 
Why Social Network Research is good for you!
Why Social Network Research is good for you!Why Social Network Research is good for you!
Why Social Network Research is good for you!
 
Pragmatic Evaluation of Concept Hierarchies
Pragmatic Evaluation of Concept HierarchiesPragmatic Evaluation of Concept Hierarchies
Pragmatic Evaluation of Concept Hierarchies
 
Enhancing the Navigability of Social Tagging Systems with Tag Taxonomies
Enhancing the Navigability of Social Tagging Systems with Tag TaxonomiesEnhancing the Navigability of Social Tagging Systems with Tag Taxonomies
Enhancing the Navigability of Social Tagging Systems with Tag Taxonomies
 
Evaluating Tag-Based Information Access in Image Collections
Evaluating Tag-Based Information Access in Image CollectionsEvaluating Tag-Based Information Access in Image Collections
Evaluating Tag-Based Information Access in Image Collections
 

Recently uploaded

1029 - Danh muc Sach Giao Khoa 10 . pdf
1029 -  Danh muc Sach Giao Khoa 10 . pdf1029 -  Danh muc Sach Giao Khoa 10 . pdf
1029 - Danh muc Sach Giao Khoa 10 . pdf
QucHHunhnh
 
Spellings Wk 3 English CAPS CARES Please Practise
Spellings Wk 3 English CAPS CARES Please PractiseSpellings Wk 3 English CAPS CARES Please Practise
Spellings Wk 3 English CAPS CARES Please Practise
AnaAcapella
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
heathfieldcps1
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdf
QucHHunhnh
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
ciinovamais
 
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
ZurliaSoop
 

Recently uploaded (20)

1029 - Danh muc Sach Giao Khoa 10 . pdf
1029 -  Danh muc Sach Giao Khoa 10 . pdf1029 -  Danh muc Sach Giao Khoa 10 . pdf
1029 - Danh muc Sach Giao Khoa 10 . pdf
 
Spellings Wk 3 English CAPS CARES Please Practise
Spellings Wk 3 English CAPS CARES Please PractiseSpellings Wk 3 English CAPS CARES Please Practise
Spellings Wk 3 English CAPS CARES Please Practise
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
 
Unit-V; Pricing (Pharma Marketing Management).pptx
Unit-V; Pricing (Pharma Marketing Management).pptxUnit-V; Pricing (Pharma Marketing Management).pptx
Unit-V; Pricing (Pharma Marketing Management).pptx
 
SKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptx
SKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptxSKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptx
SKILL OF INTRODUCING THE LESSON MICRO SKILLS.pptx
 
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxBasic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
 
PROCESS RECORDING FORMAT.docx
PROCESS      RECORDING        FORMAT.docxPROCESS      RECORDING        FORMAT.docx
PROCESS RECORDING FORMAT.docx
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdf
 
Application orientated numerical on hev.ppt
Application orientated numerical on hev.pptApplication orientated numerical on hev.ppt
Application orientated numerical on hev.ppt
 
Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...
 
Third Battle of Panipat detailed notes.pptx
Third Battle of Panipat detailed notes.pptxThird Battle of Panipat detailed notes.pptx
Third Battle of Panipat detailed notes.pptx
 
Holdier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfHoldier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdf
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
 
Understanding Accommodations and Modifications
Understanding  Accommodations and ModificationsUnderstanding  Accommodations and Modifications
Understanding Accommodations and Modifications
 
ICT role in 21st century education and it's challenges.
ICT role in 21st century education and it's challenges.ICT role in 21st century education and it's challenges.
ICT role in 21st century education and it's challenges.
 
psychiatric nursing HISTORY COLLECTION .docx
psychiatric  nursing HISTORY  COLLECTION  .docxpsychiatric  nursing HISTORY  COLLECTION  .docx
psychiatric nursing HISTORY COLLECTION .docx
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The Basics
 
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
 
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
 
Dyslexia AI Workshop for Slideshare.pptx
Dyslexia AI Workshop for Slideshare.pptxDyslexia AI Workshop for Slideshare.pptx
Dyslexia AI Workshop for Slideshare.pptx
 

Networks Navigability: Theory and Applications

  • 1. Networks Navigability: Theory and Applications Denis Helic & Christoph Trattner KMI, TU Graz August 31, 2011 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 1 / 75
  • 2. Internet of Things http://www.youtube.com/watch?v=sfEbMV295Kk Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 2 / 75
  • 3. Internet of Things We are heading towards a completely interconnected society Where people, devices, sensors are all connected to each other producing billions of billions of data each day... Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 3 / 75
  • 4. Internet of Things One big challenge in this context is how we can find relevant information in such a networked world of data Hence, in this presentation: Latest research results on the navigability of such networks will be shown Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 4 / 75
  • 5. Internet of Things In particular I will show: what are structural clues that make such networks navigable/searchable? In addition to this, I will present a framework that is able to measure network navigability. and I will present two algorithms to generate efficient navigational tools for that networks. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 5 / 75
  • 6. Networks What are networks? Basically a network is a system that can be modeled with graphs. Graphs are mathematical structures consisting of vertices and edges connecting the vertices When we observe large graphs that exist in nature, societies, or systems we refer to them as networks Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 6 / 75
  • 7. Networks What are popular examples of such networks? Social networks. Nodes are people and links are acquaintances, friendship, and so on. Communication networks. Internet: nodes are computers and links are cables connecting computers Biological networks. Metabolism: nodes are substances and links are metabolic reactions Information networks. Web: nodes are Web pages and links are hyperlinks connecting pages Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 7 / 75
  • 8. Networks 6 How to search in a small world Pajek Figure 2: HP Labs’ email communication (light grey lines) mapped onto the organizational Figure: Social network of lines). Note that communication tends to “cling” toof formal organizational hierarchy (black HP Labs constructed out the e-mail communication. chart. From: How to search a social network, Adamic, 2005. with one another. The h-distance, used to navigate the network, is computed as follows: individuals have h-distance one to their manager and to everyone they share a manager with. Distances are then recursively assigned, so that each individual has h-distance 2 to their first neighbor’s neighbors, and h-distance 3 to their second Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 8 / 75
  • 9. Networks Figure: Network of pages and hyperlinks on a Website. From: Networks, Mark Newman, 2011. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 9 / 75
  • 10. Structure and Function of Networks One of the most important research questions in the study of networks: what is the relation between structure and function of networks For example, the Internet – how should the link structure of the Internet look like that supports efficient routing? Or how should the link structure of the Web look like to be efficient navigable? In this presentation we will focus on network navigability! Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 10 / 75
  • 11. Network Navigability Definition Put simple, a network is navigable if and only if there is a short path between all or almost all pairs of nodes in the network. Formally: 1 There exist a giant component 2 The effective diameter is low – bounded by log (n), where n is the number of nodes in the network Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 11 / 75
  • 12. Network Navigability Knowledge Management Institute Navigability: Examples Example 1: Example 1: Not navigable: No giant component Figure: Network is not navigable because there is no giant component, i.e. the network is not connected. Example 2: Not navigable: giant component, BUT eff.diam: 7 > log2(8) Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 12 / 75
  • 13. Example 1: Network Navigability Not navigable: No giant component Example 2: Example 2: Not navigable: giant component, BUT Figure: Now, there is a giant component, i.e. the network is connected. However the network is not navigable because eff .diam = > log26 > log2 (8). eff.diam: 7 6, and (8) Denis Helic 2010 7 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 13 / 75
  • 14. Network Navigability Knowledge Management Institute Navigability: Examples Example 3: Figure: The network is navigable because there is AND component and Navigable: Giant component a giant eff .diam = 2. Effective diamater is boundedlog2(10) eff.diam: 2 < by log2 (10). Is this efficiently navigable? There are short paths between all nodes, but can an agent or algorithm find them with local knowledge only? Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 14 / 75
  • 15. Global Network Navigability We discussed so far global network navigability Suppose that the network is navigable and we have global knowledge of network Then it is easy to design efficient procedures to find an arbitrary target node from an arbitrary start node For example, breadth-first search is such an algorithm that has linear time complexity O(n + m), where m is the number of links Such procedures are called centralized search Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 15 / 75
  • 16. Local Network Navigability Let us now discuss local network navigability Suppose that the network is navigable but we have only local knowledge of network That means when we arrive at a particular node we know only outgoing links from that node and nothing beyond that For instance on Facebook we only know our friends or the friends of of our friends. These procedure are typically called decentralized search Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 16 / 75
  • 17. Local Network Navigability But, how efficient are people in such social search? As shown by Millgram’s experiment, people are very efficient in social search. As shown, people are able to find each other in less than seven hops (friends), ∝ log (n) Hence, people have an extremely efficient decentralized search procedure Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 17 / 75
  • 18. Local Network Navigability How we are able to find other people efficiently? Or in other words, what are the properties of social networks, or networks in general that make efficient decentralized search possible? Are there some structural clues in the network which allows us to design sub-linear algorithms? And if yes, what are these structural clues? Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 18 / 75
  • 19. Efficiently navigable Local Network Navigability A network is efficiently navigable iff: If there is an algorithm that can find a short path with only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Example: B D A C Efficiently navigable, if the algorithm knows it needs to Figure: A is start node and D is target node. go through A B C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Denis Helic 2010 9 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 19 / 75
  • 20. Local A networkNavigability navigable iff: Network is efficiently If there is an algorithm that can find a short path with only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Step 1: B D A C Figure: Efficiently navigable, if the algorithm knows it needs to There are two possible paths from A. Obviously, the optimal path leads to B. What is the structuralA go through property that can guide us in selecting B? B C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Denis Helic 2010 10 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 20 / 75
  • 21. Local A networkNavigability navigable iff: Network is efficiently If there is an algorithm that can find a short path with only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Step 1: B D A C Figure: Efficiently navigable, if the algorithm knows it needs to There are two possible paths from A. Obviously, the optimal path leads to B. What is the structuralA go through property that can guide us in selecting B? B C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Denis Helic 2010 Nodes degree 10 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 20 / 75
  • 22. Local A networkNavigability navigable iff: Network is efficiently If there is an algorithm that can find a short path with only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Step 2: B D A C Figure: Efficiently navigable,paths from B. Obviously, the it needs to leads There are seven possible if the algorithm knows optimal path to C. What is through A property that can guide us in selecting C? go the structural B C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Denis Helic 2010 11 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 21 / 75
  • 23. Local A networkNavigability navigable iff: Network is efficiently If there is an algorithm that can find a short path with only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Step 2: B D A C Figure: Efficiently navigable,paths from B. Obviously, the it needs to leads There are seven possible if the algorithm knows optimal path to C. What is through A property that can guide us in selecting C? go the structural B C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Denis Helic 2010 Nodes clustering 11 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 21 / 75
  • 24. Local Network Navigability Summarizing, local network navigability requires: 1 Existence of network hubs that are connected to many nodes 2 Existence of network clusters where nodes are highly interlinked Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 22 / 75
  • 25. Local Network Navigability Formally: 1 Power-low degree distribution with exponent γ 2 High clustering coefficient C Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 23 / 75
  • 26. γ=2.2 γ=2.6 α=1.5 free nor clustered success pro γ=2.3 γ=2.4 γ=2.7 α=2.0 γ=2.8 α=3.0 Local Network Navigability γ=2.5 γ=2.9 α=1.1 0.2 α=5.0 γ=3.0 0 IV. AIR TRAV 3 4 5 2 2.2 2.4 2.6 2.8 3 10 network size (N) 10 10 degree exponent (γ) A 3 non-navigable region degree exponent (γ) We illustrate th 2.5 structure of netwo Web of trust Metabolic an example of pa Internet to travel from Tok α=5.0 2 Airports the public air tran navigable region work are airports, 3 4 50 0.1 0.2 0.3 0.4 0.5 0.6 0.7 10 10 10 is at least one flig network size (N) clustering coefficient (C) ing to the greedy Success probability of greedy routing. Left the underlying me Figure: Navigable networks in γ, C space. the next-hop airpo ccess probability ps as a function of network size N ent values of γ with weak (top) and strong (bottom) nation. Under th g. The top-right plot shows ps as a function of γ Bethel, then to An r networks of fixed size N ≈ 105 . In the bottom- to Paris, then to V t, parameter α is mapped to Navigability: Theorycoefficient Networks clustering Denis Helic & Christoph Trattner (KMI, TU Graz) and Applications August 31, 2011 24 / 75
  • 27. A network is efficiently navigable iff: Local If there is an algorithm that can find a short path with Network Navigability only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Revisiting Step 2: B D A E C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Figure: There are seven possible paths from B. Obviously, the optimal path leads Denis Helic 2010 to C. What is an additional hint that can guide us in selecting C over E? 12 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 25 / 75
  • 28. A network is efficiently navigable iff: Local If there is an algorithm that can find a short path with Network Navigability only local knowledge, and the delivery time of the algorithm is bounded polynomially by logk(n). Revisiting Step 2: B D A E C J. Kleinberg. The small-world phenomenon: An algorithmic perspective. Proc. 32nd ACM Symposium on Theory of Computing, 2000. Also appears as Cornell Computer Science Technical Report 99-1776 (October 1999) Figure: There are seven possible paths from B. Obviously, the optimal path leads Denis Helic 2010 to C. What is an additional hint that can guide us in selecting C over E? 12 Nodes similarity Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 25 / 75
  • 29. Local Network Navigability Nodes similarity is external to the network It is derived from some additional information that we have about network nodes In Millgram’s experiment people selected the next person according to their occupation or geography Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 26 / 75
  • 30. Local Network Navigability All of this information, i.e. degrees, clustering, similarity can be understood as a kind of our background knowledge about the network We use this background knowledge to guide us in our search for a target node When we have more than one link to follow we consult the background knowledge and ask which of the links will lead us with highest probability to a given target node Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 27 / 75
  • 31. Greedy Decentralized Search On the next abstraction level we can say that background knowledge defines a notion of distance between nodes In other words, background knowledge is a metric space where each node has unique coordinates and we can calculate the distance between nodes Or in other words, we can abstract background knowledge as a black-box executing a simple function: getDistance(node, target node) Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 28 / 75
  • 32. Greedy Decentralized Search Let us now take an algorithmic perspective on decentralized search We start at an arbitrary node and need to find as fast as possible a target node having only local knowledge of the network In addition, we have background knowledge represented through getDistance(node, target node) function At each search step we have to make a decision which of the available links to follow Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 29 / 75
  • 33. Greedy Decentralized Search In order to maximize the probability of finding the target node we always select a node which has the smallest distance to the target node It has been shown that the greedy algorithm is very efficient, i.e. the number of steps to reach an arbitrary target node is ∝ log (n) Kleinberg proved it theoretically, Watts by simulation Watts was able to reproduce Millgram’s experiment with proper selection of parameters: Identity and Search in Social Networks, 2002 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 30 / 75
  • 34. Background Knowledge Now, how does our background knowledge of people typically look like? It is a metric space, e.g. 1-D spaces, 2-D vector spaces, 3-D Euclidean spaces, hyperbolic spaces, ... or does it look like completely different? Actually, it was observed by Kleinberg and also by Watts that a hierarchy of nodes is also a very good approximation of how people think Hence, we will also use hierarchical background knowledge Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 31 / 75
  • 35. Hierarchy as a Metric Space 2 3 2 3 3 23 24 21 4 4 4 4 1 1 15 22 25 3 5 5 5 5 1 1 11 12 13 14 31 32 33 Figure: Node distances in a hierarchy. Distance: d(i, j) = h(i) + h(j) − 2h(lca(i, j)) − 1 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 32 / 75
  • 36. Example of a Greedy Navigation 3 2 2 3 3 23 24 21 2 3 4 4 4 4 1 1 15 22 25 3 4 1 5 5 5 5 1 1 11 12 13 14 31 32 33 11 2 21 3 31 12 1 1 22 2 3 4 13 23 25 32 33 14 15 24 Figure: Greedy search. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 33 / 75
  • 37. Calculating Network Navigability Now in order to measure network navigability, we developed a theoretical framework to estimate network navigability by simulations As input we take a network, e.g. information network like Wikipedia, or Delicious and a suitable hierarchy that models background knowledge For example, Wikipedia categories or Delicious folksonomy and simulate decentralized search on 106 start and target node pairs Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 34 / 75
  • 38. Network Navigability Simulation Framework The metrics we measure by our framework are success rate s and stretch τ For both metrics we calculate distributions over global shortest path Definition Stretch: τ = h , where h is the number of simulator steps and l is the l global shortest path. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 35 / 75
  • 39. Evaluating hierarchies The framework lets you e.g. estimate the quality of a hierarchy to serve as background knowledge A hierarchy with better navigational properties will have better success rate and stretch in comparison with other hierarchies For example, Wikipedia categories versus Delicious tags For example, different folksonomies for navigating social tagging systems, see Helic et al.: Pragmatic Evaluation of Folksonomies, 2011 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 36 / 75
  • 40. Evaluating Navigational Tools But we can use framework to estimate the effects of changes in the network on its navigational properties For example, how navigable is Wikipedia now? How navigable will be Wikipedia if we include Delicious tags? How navigable will be Wikipedia if we include breadcrumbs? We take Wikipedia as the starting network and create new links in the network to emulate Delicious tags, breadcrumbs, etc. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 37 / 75
  • 41. Evaluating folksonomies A folksonomy is a hierarchy that is automatically generated from a tagging system Today there are several folksonomy algorithms, see e.g. Heymann 2008, or Benz 2010 In addition, you can produce folksonomies by using standard hierarchical clustering methods such as K-Means or Affinity Propagation Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 38 / 75
  • 42. Evaluating folksonomies In Helic et al.: Pragmatic Evaluation of Folksonomies, WWW2011 we took 5 tagging datasets and 5 different folksonomy algorithms We produced 5x5 folksonomies and simulated (100.000 samples) greedy decentralized search on the datasets We measured the success rate and stretch to see if some folksonomies perform better than the other ones. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 39 / 75
  • 43. Evaluating folksonomies Greedy Search Success Rate: BibSonomy 100 Folksonomy Random Aff.Prop. Success Rate (Percentage) 80 K-Means Deg/Cooc Clo/Cos 60 40 20 0 1 2 3 4 5 6 7 Shortest path Figure: Success Rate of different folksonomies in BibSonomy Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 40 / 75
  • 44. Evaluating folksonomies Greedy Search Success Rate: CiteULike 100 Folksonomy Random Aff.Prop. Success Rate (Percentage) 80 K-Means Deg/Cooc Clo/Cos 60 40 20 0 1 2 3 4 5 6 7 Shortest path Figure: Success Rate of different folksonomies in CiteULike Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 41 / 75
  • 45. Evaluating folksonomies Greedy Search Success Rate: Delicious 100 Folksonomy Random Aff.Prop. Success Rate (Percentage) 80 K-Means Deg/Cooc Clo/Cos 60 40 20 0 1 2 3 4 5 6 7 Shortest path Figure: Success Rate of different folksonomies in Delicious Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 42 / 75
  • 46. Evaluating folksonomies Greedy Search Success Rate: Flickr 100 Folksonomy Random Aff.Prop. Success Rate (Percentage) 80 K-Means Deg/Cooc Clo/Cos 60 40 20 0 1 2 3 4 5 6 Shortest path Figure: Success Rate of different folksonomies in Flickr Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 43 / 75
  • 47. Evaluating folksonomies Greedy Search Success Rate: LastFM 100 Folksonomy Random Aff.Prop. Success Rate (Percentage) 80 K-Means Deg/Cooc Clo/Cos 60 40 20 0 1 2 3 4 5 Shortest path Figure: Success Rate of different folksonomies in LastFM Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 44 / 75
  • 48. Evaluating folksonomies Centrality-based algorithms such as Heymann 2008 or Benz 2010 outperform traditional methods However, these are all theoretical results Because, what is if we wanted to embed folksonomies in the user interface (UI) to support users in their navigation tasks and the space in user interface is limited? Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 45 / 75
  • 49. Embedding folksonomies in UI Google Directory - Computers > Internet > On the Web > Online Communities Directory Help Online Communities Computers > Internet > On the Web > Online Communities Go to Directory Home Categories Bulletin Board Directories (9) PowerPets (6) Systems (132) Events (1) Second Life (119) By Region (8) Mailing Lists (85) Social Networking By Subject (204) Message Boards (222) Chat (745) (154) Software and Community MySpace (28) Services (27) Management (36) Neopets (171) The Palace (51) Community Zetapets (3) Providers (14) Related Categories: Society > Activism > Community Building (26) Society > Organizations (16987) Society > People > Personal Homepages (8890) Society > Relationships > Cyber Relationships (59) Society > Subcultures > Cyberculture (162) Web Pages Viewing in Google PageRank order View in alphabetical order Talk City - http://www.talkcity.com/ Figure: Directory Based Navigation technology, health and other Participate in discussions about relationships, hobbies, business, topics. Socialize with friends, or start your own chat group. Whyville - http://www.whyville.net/ A virtual 3-D world for curious minds where you can own land, build your own house, play simulation games, win prizes, chat, and help the community grow. Buzznet - http://www.buzznet.com/ Denis Helic & Christoph Trattner (KMI, TU Graz) create communitiesTheory and Applications Users can Networks Navigability: and share blogs and photographs. August 31, 2011 46 / 75
  • 50. Embedding folksonomies in UI We have breadcrumbs connecting each node all the way up to the root node We have limited number of subcategories (n) We have limited number of related categories (m) Now we embed folksonomy as in Benz 2010 and apply different restrictions Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 47 / 75
  • 51. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.585123, h=5.936013, sg=0.005548, τg=1.655735 Success Rate (s) 3 Stretch (τ) 2.5 2 s, τ 1.5 1 0.5 0 1 2 3 4 5 6 7 8 9 Shortest Path Figure: Success Rate and stretch in BibSonomy with n = 20 and m = 20 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 48 / 75
  • 52. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.634634, h=6.536937, sg=0.001110, τg=1.798513 9 Success Rate (s) Stretch (τ) 8 7 6 5 s, τ 4 3 2 1 0 1 2 3 4 5 6 7 8 9 Shortest Path Figure: Success Rate and stretch in CiteULike with n = 20 and m = 20 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 49 / 75
  • 53. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.518932, h=5.557032, sg=0.000903, τg=1.579181 7 Success Rate (s) Stretch (τ) 6 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 Shortest Path Figure: Success Rate and stretch in Delicious with n = 20 and m = 20 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 50 / 75
  • 54. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.467684, h=4.162304, sg=0.000382, τg=1.200312 Success Rate (s) 7 Stretch (τ) 6 5 s, τ 4 3 2 1 0 1 2 3 4 5 6 7 8 9 Shortest Path Figure: Success Rate and stretch in Flickr with n = 20 and m = 20 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 51 / 75
  • 55. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.197477, h=6.662900, sg=0.001062, τg=2.083799 Success Rate (s) 6 Stretch (τ) 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 Shortest Path Figure: Success Rate and stretch in LastFM with n = 20 and m = 20 Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 52 / 75
  • 56. Embedding folksonomies in UI Under this restriction the navigator in Considering practical user interface restriction folksonomies are useless for supporting navigation. The success rate drops below 1%. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 53 / 75
  • 57. Embedding folksonomies in UI Thus, folksonomies (unlimited) are useful theoretically but useless practically The problem is that top nodes have many children (possibly thousands) and UI restrictions cut to many children nodes off Hence, we need a new algorithm that takes into account these UI restrictions Technically, we need to able to determine the branching factor for the hierarchy We developed such an algorithm and published in CIKM2011. Helic et al. Building Directories for Social Tagging Systems We were able to almost recover theoretical navigability Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 54 / 75
  • 58. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.585123, h=8.691685, sg=1.000000, τg=2.424376 7 Success Rate (s) Stretch (τ) 6 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 7 8 9 Shortest Path Figure: Success Rate and stretch in BibSonomy with new folksonomy algorithm Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 55 / 75
  • 59. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.634634, h=9.163688, sg=1.000000, τg=2.521213 7 Success Rate (s) Stretch (τ) 6 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 7 8 9 Shortest Path Figure: Success Rate and stretch in CiteULike with new folksonomy algorithm Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 56 / 75
  • 60. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.518932, h=9.720769, sg=1.000000, τg=2.762420 6 Success Rate (s) Stretch (τ) 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 Shortest Path Figure: Success Rate and stretch in Delicious with new folksonomy algorithm Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 57 / 75
  • 61. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.467684, h=8.886960, sg=0.996066, τg=2.562794 Success Rate (s) 6 Stretch (τ) 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 7 8 9 Shortest Path Figure: Success Rate and stretch in Flickr with new folksonomy algorithm Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 58 / 75
  • 62. Embedding folksonomies in UI - Greedy Navigator (1000000 Runs) - l=3.197477, h=9.830726, sg=1.000000, τg=3.074526 6 Success Rate (s) Stretch (τ) 5 4 s, τ 3 2 1 0 1 2 3 4 5 6 Shortest Path Figure: Success Rate and stretch in LastFM with new folksonomy algorithm Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 59 / 75
  • 63. Why usefulness of folksonomies for navigation is limited? Even if folksonomies allow the user to navigate to related concepts in an efficient manner navigation to a particular resource is still a problem As shown related work, in tagging systems the tag-resource distribution follows a power-law function, i.e. there are many tags that refer to a large number of resources. In BibSonomy or CiteULike for instance there are tags, which refer to hundreds or even thousands of resources. To display such long resource lists, developers typically paginate the resource lists in a tagging system by a certain factor k Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 60 / 75
  • 64. Why usefulness of folksonomies for navigation is limited? (a) Austria-Forum (b) BibSonomy (c) CiteULike Figure: Tag distributions. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 61 / 75
  • 65. Creating tag-resource Taxonomies To support the user not only to navigate to related tags in efficient manner but also to the resources of a tagging system, we invented the approach of the so-called tag-resource taxonomies. Car Car Tire Motor Tire Motor Mercedes VOLVO VW BMW VW BMW VW BMW (a) Folksonomy (b) Tag-Resource Taxonomy Figure: Folksonomy vs. Tag-Resource Taxonomy. In a Folksonomy tags appear only once. However, resources can be referred by different tags. In a tag-resource taxonomy on the other hand resources can occur only once while tags can appear on multiple and on different levels. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 62 / 75
  • 66. Why usefulness of folksonomies for navigation is limited? In the worst case a user would have to click max{click(Ttag )} times to reach a desired resource with the help of a Folksonomy. c1 |r | max{click(Ttag )} = + logb/2 (c2 · |r |), b ≥ 2 (1) k or c1 · |r | max{click(Ttag )} ≈ (2) k c1 ·|r | supposing that logb/2 (c2 · |r |) k Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 63 / 75
  • 67. Why usefulness of folksonomies for navigation is limited? The worst case scenario of a tag-resource taxonomy is significantly better. In the worst case a user would have to click max{click(Tres )} times to reach a desired target resource. max{click(Tres )} = max{depth(Tres )} = logk/2 |r | , k ≥ 2 (3) Then for large values of |r | we have: c1 · |r | logk/2 |r | (4) k Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 64 / 75
  • 68. Why usefulness of folksonomies for navigation is limited?xxx Austria-Forum BibSonomy CiteULike max{click(Ttag )} 184 5,278 20,799 max{click(Tres )} 6.1 7.7 8.5 Table: Tag Taxonomy vs. Tag-Resource Taxonomy: Maximum number of clicks for k = 10. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 65 / 75
  • 69. Why usefulness of folksonomies for navigation is limited? To calculate the number of tags suffering from the so-called pagination effect, we can user the following equation: 1 α 1 (α) |tp | = |t| · − (5) k k Austria-Forum BibSonomy CiteULike |tp | (%) 5079 (38%) 7401 (28%) 51748 (32%) Table: Number of paginated tags for k = 10. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 66 / 75
  • 70. Why usefulness of folksonomies for navigation is limited? The mean number of clicks is calculated as follows: Tag-Resource Taxonomy: mean{click(Tres )} = logk (|r |) |t| Folksonomy: mean{click(Ttag )} = logk (|t|) + |t| i=1 ri 1 k k Austria-Forum BibSonomy CiteULike mean{click(Tres )} 2 14.2 17.8 19.8 mean{click(Ttag )} 2 29.5 22.4 30.7 mean{click(Tres )} 5 6.1 7.6 8.5 mean{click(Ttag )} 5 11.6 9.2 12.3 mean{click(Tres )} 10 4.3 5.3 5.9 mean{click(Ttag )} 10 6.4 5.6 7.3 Table: Tag Taxonomy vs. Tag-Resource Taxonomy: Mean number of clicks for different branching factors k. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 67 / 75
  • 71. Creating tag-resource Taxonomies 1. Computer Degree centrality of the resource-to-resource tag network 2. Take most general resource as root an attach max. b resources as childs. Child-nodes are selected according their cos-sim values. 3. After that we take the resource taxonomy and apply labels (tags) to the resource (top-down, in left-order) 3.1 We calculate candidate labels by the method of co-occurance, i.e. we take the tags of the related resources into account to rank the actual tags of the currently processed resource. 3.2. If the candidate tag has already been applied to one of the parent resources of the currently processed resource we take the next candidate tag from the co-occurance vector and try to apply it. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 68 / 75
  • 72. Evaluating Tag-Resource Taxonomies In the first experiment we measured the average and maximum number of clicks and the drop rate Name b n max{click(Tres )} mean{click(Tres )} Res2 2 19,430 17 12.45 Res5 5 19,430 10 5.93 Res10 10 19,430 8 4.44 Table: max{click(Tres )} and mean{click(Tres )} for different branching factors b. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 69 / 75
  • 73. Evaluating Tag-Resource Taxonomies In the second experiment we measured the number of collisions Name b n CR (%) Res2 2 19,430 0.1% Res5 5 19,430 0.2% Res10 10 19,430 0.2% Table: Collision Rates (CR) for different resource taxonomies with different branching factor b. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 70 / 75
  • 74. Evaluating Tag-Resource Taxonomies In the third experiment we measured the semantic structure of the tag-resource taxonomy compared to popular folksonomy induction algorithms such as Heymann, K-Means, Affinity Propagation and Co-Occurance As measure for this experiment we used Taxonomic Recall/Prec. and overlap. Ground truth: Germanet ontholoy 0.4 Taxonomic F−Measure 0.35 Taxonomic Overlap 0.3 Count (1 = 100%) 0.25 0.2 0.15 0.1 0.05 0 Res2 Res5 Res10 Deg/Cooc Aff. Prop K−Means Heymann Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 71 / 75
  • 75. Evaluating Tag-Resource Taxonomies In the fourth and last experiment a user study was conducted to test weather the approach is also useful for humans or not As ground truth for the experiment the best so far known folksonomy generation approach was used All over we had 9 test users who had to judge 200 tag trails extracted from both hierarchies Name b Correct (%) Related (%) Equivalent (%) Not Related (% Deg/Cooc10 10 33.2 27.3 13 21.9 Res10 10 27.3 36.2 12.3 19.8 Table: Results of the empirical analysis of the tag-resource taxonomy with branching factor b = 10 compared to a Deg/Cooc folksonomy with branching factor b = 10. Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 72 / 75
  • 76. End of presentation Thank you very much for your attention! Christoph Trattner (ctrattner@iicm.edu) Denis Helic & Christoph Trattner (KMI, TU Graz) Networks Navigability: Theory and Applications August 31, 2011 73 / 75