3. Motivation
• Social network can be split into two categories.
o Explicit Network.
o Implicit Network.
• Explicit Network
o Reinforces Existing Real World Connections
• Implicit Network
o Forms Dynamic communities based on mutual interest,
common activities, places etc.
• The idea is to bring about a convergence
between implicit and explicit networks.
5. Problem:
Combining explicit and implicit social networks
has huge cost.
Enormous amount of information necessary to
be managed.
One Possible solution is to use internet-based
Gossip overlays.
6. Social Market
System model:
Consider system of interconnected users
exchanging information.
Each user has a profile associated
Profile is vector of strings
Each string is referred to as "Keyword"
Every keyword has a counter and a weight
associated.
7. Social Market
Weigth measure of relavance between a
given keyword to other keywords in the
profile.
𝑢 ∈ 𝑈 where U is universe of all profiles.
And u is denotes user or user profile.
Cosine Similarity :
𝑆𝑖𝑚 𝑢1, 𝑢2 = cos 𝑢1, 𝑢2 =
𝑢1 𝑢2
𝑢1 ∗ | 𝑢2 |
8. Social Market
Items:
• User interact with social market by creating items.
• Every item has a profile and is stored in a similar fashion as User
profiles.
Once a item is created, goal of social market is to lead this item to meet
other user who
• Are interested in the item
• Can be trusted and can trust the creator of the item
• Can be reached through a trusted path on the social network
9. Social Market
SM uses a feature called trust to build this trust
path.
The trust between users are provided by the users
themselves.
For instance, User A can assign 0 trust on user B.
0 trust doesn’t mean, User A distrusts B, simply
means that A does not know B enough.
11. Trust Aware Peer Sampling
A novel protocol that operates by directly
incorporating trust relationships.
Extracted from an explicit social network into
the gossip-based overlay.
• Goal:
o Create TAPS view with ever changing set of
reference to other nodes
o Periodically, nodes contact to exchange information
of their views
12. Trust Aware Peer Sampling
• In standard peer sampling contains:
o Contact information of other nodes
o Timestamp indicating last update.
• TAPS contain information:
o User profile
o Inferred trusts value.
13. Trust propogation
• Each edge in the trusted path associates
uncertainty about the trustworthiness.
• To model inferred trust.
o Trust path as product of trust values of its edges,
weighted by trust transitivity co efficient 𝜁.
o Given path u1, u2, … un with trust values t1,2 , t2,3, .. tn-
1,n
o 𝑡1,𝑛 = 𝜁 𝑛−2
𝑖=1
𝑖=𝑛−1
𝑡1,𝑖+1
o Lower 𝜁 values causes trust to decay faster with path
length.
14. View Exchanges
o Views are initialized with agreed upon trust value
during explicit friendship relationships.
o Initialize TAPS view by inserting one entry of each
explicit neighbors.
o These views are exchanged with other nodes.
o View are exchanged between friends, friends of
friends of friends.
15. View Exchanges
o As gossip process evolves nodes collaborate
computing inferred trust.
o Let trust of Nodes A and X be tA,X and trust of A and B
be tA,B , to compute tB,X
tBX = τtABtAX.
.
16. View Exchanges
o A node might receive views from multiple nodes.
A node always selects the largest trust value for
any node.
o To enchance trust inference, nodes initiate gossip
exchanges with nodes in TAPS view and explicit
neighbours.
o The trust path values are kept up to date and
maximum trust path is chosen to provide shortest
path.
17. Evaluation
o Dataset of 300 users where taken from facebook and Digg.
Binary Trace
Multivalued Trace
Impact of trust density
19. Reference
Frey, Davide, Arnaud Jégou, and Anne-Marie Kermarrec. "Social market: combining explicit and
impBertier, Marin et al. "The gossple anonymous social network." Middleware 2010 (2010): 191-
211.licit social networks." Stabilization, Safety, and Security of Distributed Systems (2011): 193-
207.
Bertier, Marin et al. "The gossple anonymous social network." Middleware 2010 (2010): 191-211.
Questions