1
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Cognitive Models in Recommender
Systems
Christoph Trattner
Know-Center & NTNU
ctrattner@know-center.at
@Graz University of Technology, Austria
2
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Before start in this presentation I will talk a bit about
myself, my background…
3
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Where do I come from (Austria)?
4
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Graz
5
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Academic Back-Ground?
 Studied Computer Science at Graz University of
Technology & University of Pittsburgh
 Worked since 2009 as scientific researcher at the KMI &
IICM (BSc 2008, MSc 2009)
 My PhD thesis was on the Search & Navigation in Social
Tagging Systems (defended 2012)
 Since Feb. 2013 @ Know-Center
 Leading the Social Computing Area
 At TUG:
 WebScience
 Semantic Technologies
6
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
My team
2 Post-Docs, 5 Pre-Docs (2 more to join soon )
2 MSc student
2 BSc student
DI. Dieter
Theiler
DI. Dominik
Kowald
Dr. Peter
Kraker
Mag. Sebastian
Dennerlein
Dr. Elisabeth
Lex
Mag. Matthias
Rella
DI. Emanuel
Lacic
DI. Ilire Hasani
7
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Thanks to my Collaborators
8
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
What is my group doing?
… we research on novel methods and tools that exploit
social data to generate a greater value for the
individual, communities, companies and the society as
whole.
Our competences:
• Network & Web Science
• Science 2.0
• Predictive Modeling
• Social Network Analysis
• Information Quality Assessment
• User Modeling
• Machine Learning and Data Mining
• Collaborative Systems
Our Services:
• Social Analytics: Hub-, Expert -, Community -
, Influencer -, Information Flow-, Trend
(Event) Detection, etc.
• Information Quality Assessment
• Social & Location-based Recommander
Systems
• Customer Segmentation
• Social Systems Design
9
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Some industry partners...
10
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Current projects
BlancNoir - “Towards a Big Data recommender engine for offline
and online marketplaces”
I2F - “Towards a Social Media and Online Marketing Manager
Seminar”
Automation-X - “Towards a scalable Graph-based Visual search
solution”
Styria - “Towards a scalable crowd-based hierarchical cluster
labeling approach for willhaben.at”
TripRebel - “Towards an engaging hybrid hotel recommender
solution for triprebel.com”
CDS - “Towards a scalable Entity & Graph-based Visual search
solution for cds.at”
Exthex - “Towards an efficient viral social media marketing
champagne in Facebook and Twitter”
11
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
The Projects
Project: Tallinn University – Interested in the problem of
recommending tags and items to users in social information
systems.
12
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Ok, let’s start….
13
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Research Question:
To what extent is human cognition theory applicable to
the problem of predicting tags and items to users?
Externals involved:
• PUC - Chile, UFCG – Brazil
14
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
What are social tags?
Where can we find them?
15
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
16
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
17
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
18
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
19
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Why are social tags good?
20
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
 They help you to classify Web content better [Zubiaga 2012]
 They help people to navigate large knowledge repositories better
[Helic et al. 2012]
 They help people to search for information faster [Trattner et al. 2012]
However, there is an issue with social tags…
People are typically lazy to apply social tags(!!)
Zubiaga, A. (2012). Harnessing Folksonomies for Resource Classification. arXiv preprint arXiv:1204.6521.
Helic, D., Körner, C., Granitzer, M., Strohmaier, M., & Trattner, C. (2012, June). Navigational efficiency of broad vs.
narrow folksonomies. In Proceedings of the 23rd ACM conference on Hypertext and social media (pp. 63-72). ACM.
Trattner, C., Lin, Y. L., Parra, D., Yue, Z., Real, W., & Brusilovsky, P. (2012, June). Evaluating tag-based information
access in image collections. In Proceedings of the 23rd ACM conference on Hypertext and social media (pp. 113-
122). ACM.
Motivation
21
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
To overcome that issue some smart people started to invent mechanisms that
should help the user in applying tags:
 Collaborative Filtering
 User based- and item-based CF [Marinho et al. 2008]
 Matrix Factorization
 FM, PITF [Rendle et al. 2010, 2011, 2012]
 Graph Structures
 Adapted PageRank and FolkRank [Hotho et al. 2006]
 Topic Models
 Latent Dirichlet Allocation (LDA) [Krestel et al. 2009, 2010, 2011]
Motivation
22
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Why do we need cognitive models in
recommender systems?
23
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Why do we need cognitive models?
First answer: We do not like data data driven approaches…
Me: OK
Second answer: We can understand things better…
…why is something happening and how…
24
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
ACT-R
25
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Cognitive Models
• In principle, there are several approaches to model
cognitive processes (memory, speech, ...) in the
human brain
• Most popular one: ACT-R (Adaptive
Control of Thought-Rational) theory
by J.R. Anderson (1998)
• American Psychologist (CMU)
J.R. Anderson
CMU
26
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
27
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
ACT-R
28
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Interestingly, when looking into the literatur of tagging
systems - temporal processes are typically modeled
with an exponential function...
D. Yin, L. Hong, and B. D. Davison. Exploiting session-like behaviors in tag prediction. In
Proceedings of the 20th international conference companion on World wide web, pages
167–168. ACM, 2011.
L. Zhang, J. Tang, and M. Zhang. Integrating temporal usage pattern into personalized tag
prediction. In Web Technologies and Applications, pages 354–365. Springer, 2012
29
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Empirical Analysis: BibSonomy (1)
29
 Linear distribution with log-
scale on Y-axis 
exponential function
 Linear distribution with log-
scale on X- and Y-axes 
power function
30
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Empirical Analysis: BibSonomy (2)
30
Exponential distribution
R² = 31%
Power distribution
R² = 89%
31
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results:
Decay factor is better modeled as
power-function rather than an ex-
function
32
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Experiment 1: Predicting re-use of tags
33
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: Predicting re-use of tags
BLLAC
BLL
MPU
GIRP
34
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: Recall / Precision
Results:
BLLAC performs fairly well in
predicting the re-use of tags
35
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Experiment 2: Recommending Tags
36
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: Recall-Precision plots
36
 The time-depended
approaches outperform the
state-of-the-art
 BLL+MPr reaches the
highest level of accuracy
CiteULike
37
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: Recall/Precision
Results:
BLL approaches outperform current
state-of-the-art tag recommender
approaches.
38
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
...how about runtime?
39
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: Runtime
BLL+C needs only around 1s to generate tag-
recommendations for 5,500 users in BibSonomy
40
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: Runtime
41
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
...predicting items with ACT-R
42
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Our Approach
= CIRTT  2 main steps
First step:
– User-based Collaborative Filtering (CF) to get
candidate items of similar users
Second step:
– Item-based CF to rank these candidate items using
the BLL equation to integrate tag and time
information:
42
43
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Example
big fruit (t=10d)
small fruit (t=10d)
big fruit (t=2d)
small fruit (t=10d)
Recommendation:
Rank@1 =
Rank@2 =
44
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
What are the results?
44
45
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results: nDCG plots
45
CIRTT reaches the highest level of accuracy
46
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Results
Results:
CIRTT works quite well compared to
the current state-of-the-art in tag-
based item recommender systems
47
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
... ok that‘s basically it 
48
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Code & Framework
https://github.com/learning-layers/TagRec/
49
. Christoph Trattner 28.1.2015 – Yahoo!, Trondheim
Thank you!
Christoph Trattner
Email: trattner.christoph@gmail.com
Web: christophtrattner.info
Twitter: @ctrattner
Sponsors:

Cognitive Models in Recommender Systems

  • 1.
    1 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Cognitive Models in Recommender Systems Christoph Trattner Know-Center & NTNU ctrattner@know-center.at @Graz University of Technology, Austria
  • 2.
    2 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Before start in this presentation I will talk a bit about myself, my background…
  • 3.
    3 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Where do I come from (Austria)?
  • 4.
    4 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Graz
  • 5.
    5 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Academic Back-Ground?  Studied Computer Science at Graz University of Technology & University of Pittsburgh  Worked since 2009 as scientific researcher at the KMI & IICM (BSc 2008, MSc 2009)  My PhD thesis was on the Search & Navigation in Social Tagging Systems (defended 2012)  Since Feb. 2013 @ Know-Center  Leading the Social Computing Area  At TUG:  WebScience  Semantic Technologies
  • 6.
    6 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim My team 2 Post-Docs, 5 Pre-Docs (2 more to join soon ) 2 MSc student 2 BSc student DI. Dieter Theiler DI. Dominik Kowald Dr. Peter Kraker Mag. Sebastian Dennerlein Dr. Elisabeth Lex Mag. Matthias Rella DI. Emanuel Lacic DI. Ilire Hasani
  • 7.
    7 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Thanks to my Collaborators
  • 8.
    8 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim What is my group doing? … we research on novel methods and tools that exploit social data to generate a greater value for the individual, communities, companies and the society as whole. Our competences: • Network & Web Science • Science 2.0 • Predictive Modeling • Social Network Analysis • Information Quality Assessment • User Modeling • Machine Learning and Data Mining • Collaborative Systems Our Services: • Social Analytics: Hub-, Expert -, Community - , Influencer -, Information Flow-, Trend (Event) Detection, etc. • Information Quality Assessment • Social & Location-based Recommander Systems • Customer Segmentation • Social Systems Design
  • 9.
    9 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Some industry partners...
  • 10.
    10 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Current projects BlancNoir - “Towards a Big Data recommender engine for offline and online marketplaces” I2F - “Towards a Social Media and Online Marketing Manager Seminar” Automation-X - “Towards a scalable Graph-based Visual search solution” Styria - “Towards a scalable crowd-based hierarchical cluster labeling approach for willhaben.at” TripRebel - “Towards an engaging hybrid hotel recommender solution for triprebel.com” CDS - “Towards a scalable Entity & Graph-based Visual search solution for cds.at” Exthex - “Towards an efficient viral social media marketing champagne in Facebook and Twitter”
  • 11.
    11 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim The Projects Project: Tallinn University – Interested in the problem of recommending tags and items to users in social information systems.
  • 12.
    12 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Ok, let’s start….
  • 13.
    13 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Research Question: To what extent is human cognition theory applicable to the problem of predicting tags and items to users? Externals involved: • PUC - Chile, UFCG – Brazil
  • 14.
    14 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim What are social tags? Where can we find them?
  • 15.
    15 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim
  • 16.
    16 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim
  • 17.
    17 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim
  • 18.
    18 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim
  • 19.
    19 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Why are social tags good?
  • 20.
    20 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim  They help you to classify Web content better [Zubiaga 2012]  They help people to navigate large knowledge repositories better [Helic et al. 2012]  They help people to search for information faster [Trattner et al. 2012] However, there is an issue with social tags… People are typically lazy to apply social tags(!!) Zubiaga, A. (2012). Harnessing Folksonomies for Resource Classification. arXiv preprint arXiv:1204.6521. Helic, D., Körner, C., Granitzer, M., Strohmaier, M., & Trattner, C. (2012, June). Navigational efficiency of broad vs. narrow folksonomies. In Proceedings of the 23rd ACM conference on Hypertext and social media (pp. 63-72). ACM. Trattner, C., Lin, Y. L., Parra, D., Yue, Z., Real, W., & Brusilovsky, P. (2012, June). Evaluating tag-based information access in image collections. In Proceedings of the 23rd ACM conference on Hypertext and social media (pp. 113- 122). ACM. Motivation
  • 21.
    21 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim To overcome that issue some smart people started to invent mechanisms that should help the user in applying tags:  Collaborative Filtering  User based- and item-based CF [Marinho et al. 2008]  Matrix Factorization  FM, PITF [Rendle et al. 2010, 2011, 2012]  Graph Structures  Adapted PageRank and FolkRank [Hotho et al. 2006]  Topic Models  Latent Dirichlet Allocation (LDA) [Krestel et al. 2009, 2010, 2011] Motivation
  • 22.
    22 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Why do we need cognitive models in recommender systems?
  • 23.
    23 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Why do we need cognitive models? First answer: We do not like data data driven approaches… Me: OK Second answer: We can understand things better… …why is something happening and how…
  • 24.
    24 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim ACT-R
  • 25.
    25 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Cognitive Models • In principle, there are several approaches to model cognitive processes (memory, speech, ...) in the human brain • Most popular one: ACT-R (Adaptive Control of Thought-Rational) theory by J.R. Anderson (1998) • American Psychologist (CMU) J.R. Anderson CMU
  • 26.
    26 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim
  • 27.
    27 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim ACT-R
  • 28.
    28 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Interestingly, when looking into the literatur of tagging systems - temporal processes are typically modeled with an exponential function... D. Yin, L. Hong, and B. D. Davison. Exploiting session-like behaviors in tag prediction. In Proceedings of the 20th international conference companion on World wide web, pages 167–168. ACM, 2011. L. Zhang, J. Tang, and M. Zhang. Integrating temporal usage pattern into personalized tag prediction. In Web Technologies and Applications, pages 354–365. Springer, 2012
  • 29.
    29 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Empirical Analysis: BibSonomy (1) 29  Linear distribution with log- scale on Y-axis  exponential function  Linear distribution with log- scale on X- and Y-axes  power function
  • 30.
    30 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Empirical Analysis: BibSonomy (2) 30 Exponential distribution R² = 31% Power distribution R² = 89%
  • 31.
    31 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Decay factor is better modeled as power-function rather than an ex- function
  • 32.
    32 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Experiment 1: Predicting re-use of tags
  • 33.
    33 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Predicting re-use of tags BLLAC BLL MPU GIRP
  • 34.
    34 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Recall / Precision Results: BLLAC performs fairly well in predicting the re-use of tags
  • 35.
    35 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Experiment 2: Recommending Tags
  • 36.
    36 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Recall-Precision plots 36  The time-depended approaches outperform the state-of-the-art  BLL+MPr reaches the highest level of accuracy CiteULike
  • 37.
    37 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Recall/Precision Results: BLL approaches outperform current state-of-the-art tag recommender approaches.
  • 38.
    38 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim ...how about runtime?
  • 39.
    39 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Runtime BLL+C needs only around 1s to generate tag- recommendations for 5,500 users in BibSonomy
  • 40.
    40 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: Runtime
  • 41.
    41 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim ...predicting items with ACT-R
  • 42.
    42 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Our Approach = CIRTT  2 main steps First step: – User-based Collaborative Filtering (CF) to get candidate items of similar users Second step: – Item-based CF to rank these candidate items using the BLL equation to integrate tag and time information: 42
  • 43.
    43 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Example big fruit (t=10d) small fruit (t=10d) big fruit (t=2d) small fruit (t=10d) Recommendation: Rank@1 = Rank@2 =
  • 44.
    44 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim What are the results? 44
  • 45.
    45 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results: nDCG plots 45 CIRTT reaches the highest level of accuracy
  • 46.
    46 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Results Results: CIRTT works quite well compared to the current state-of-the-art in tag- based item recommender systems
  • 47.
    47 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim ... ok that‘s basically it 
  • 48.
    48 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Code & Framework https://github.com/learning-layers/TagRec/
  • 49.
    49 . Christoph Trattner28.1.2015 – Yahoo!, Trondheim Thank you! Christoph Trattner Email: trattner.christoph@gmail.com Web: christophtrattner.info Twitter: @ctrattner Sponsors: