1. Using a Trust Network To Improve Top-N Recommendation
Mohsen Jamali, Martin Ester
Simon Fraser University
Vancouver, Canada
ACM RecSys 2009
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

2. Outline Introduction Collaborative Filtering Approaches
User based CF
Item based CF
Trust based Approaches
Random Walk
Combined
Experiments
Conclusion
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

3. Introduction Tasks of Recommender Top-N Recommendation
Predicting an unknown rating
Recommending a list of items
Top-N Recommendation
Given a user u recommend a list of items
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

4. Collaborative Filtering for Top-N Recommendation
Item-based CF for top-N recommendation
[Deshpande 2004]
Binary rating matrix
Items: vectors in the user space.
Top similar items for each item purchased by the user
Most frequent items  top-N recommended items
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

5. Collaborative Filtering for Top-N Recommendation
User-based CF for top-N recommendation
Very few in the literature
[McLaughlin 2004] used as baseline
Find Similar users to user u according to Pearson correlation.
Aggregate the items rated by similar users to compute top-N recommendation
Most frequent items [McLaughlin 2004].
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

6. Using Trust Network for Top-N Recommendation
Issues with CF
Requires Enough Ratings
Cold Start Users
Cold Start Items
Social Networks Emerged Recently
Independent source of information
Motivations of Trust-based RS
Social Influence: users adopt the behavior of their friends
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

7. Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation
TrustWalker 2 4 5 4 3 2
item i 4 3 1 3 4 5 1 5 3 ? 2 4 2 4 3 1
user u 3 2 5 4 1 3 5 2 3 2 5 3 1 3 4 2 3 1 4 4 2
[Jamali 2009]
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

8. Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation
TrustWalker 2 4 5 4 3 2
item i 4 3 1 3 4 5 1 5 3 ? 2 4 2 4 3 1
user u 3 2 5 4 Φ 1 3 5 2 3 2 5 3 1 3 4 2 3 1 4 4 2
[Jamali 2009]
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

9. Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation
TrustWalker 2 4 5 4 3 2
item i 4 3 1 3 4 5 1 5 3 ? 2 4 2 4 3 1
user u 3 2 5 4 1 3 5
1-Φ 2 3 2 5 3 1 3 4 2 3 1 4 4 2
[Jamali 2009]
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

10. Random Walk Approach Extension of Trust Walker [Jamali 2009]
The random walk stops at a certain user v.
All items rated by v but not by u will be considered as recommended items
Several random walks.
Rank items by aggregate rating according to different random walks.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

11. Random Walk Approach (cont)
Single Random Walk
Start from the user u.
At each user v, with probability Φu,v,k stops and returns items rated by v.
With 1- Φu,v,k continue the random walk
We consider the current step k of random walk to avoid noisy data.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

12. Combined Approach
When a user u trusts another users v, it does not necessarily mean that they rate the same items.
Users who are similar according to Pearson correlation are more likely to have one more item in common.
Using Leave-one-out, CF may beat Random Walk based Approach by recommending the withheld item.
The list recommended by random walk approach may not contain the exact withheld item
Recommended items could still be related to the user
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

13. Combined Approach Recommended Movies Rated Movies Withheld Movie
Images from
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

14. Combined Approach K2 trusted users K1 similar users
Find top k1 similar users and top k2 trusted users for u.
Find the top-N items according to each set of users
Merge the two sets of items to have one set of N items.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

15. Combined Approach (cont)
Finding Top k1 Similar Users
Similar to CF approaches
Finding Top k2 Trusted Users
Breadth First Search
Random Walk
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

16. Combined Approach (cont)
Aggregated rating for an item i
CFu , TRu :
items recommended using similar (trusted) users of u
: aggregate rating according to k2 trusted users
: aggregate rating according to k1 similar users
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

17. Experiments We use Epinions.com Dataset 49k users (24k cold start)
104k items
575k ratings, 508k trust relations.
N=100
Evaluation Metric: Recall
L: number of queries.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

18. Comparison Partners
CF-User: User based Collaborative Filtering Approach.
CF-Item: Item based Collaborative Filtering Approach.
TrustWalkerList D2: RandomWalk approach with maxDepth=2.
TrustWalkerList D2-pure: The random Walk approach ignoring similarities.
TrustWalkerList D3: RandomWalk approach with maxDepth=3.
Trust-CF (k1=k2): The combined approach with k1 = k2.
Trust-CF (k1=70). k1 is fixed. The top k2 trusted users are computed using BFS.
Trust-CF-RW : same as Trust-CF (k1=70), but the top k2 trusted users are computed using a random walk approach.
Trust-CF Weighted. Combined approach with a fixed k1 and weighted merge of results of CF and trust-based approach. The trusted neighborhood is computed using BFS.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

19. Experimental Results
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

20. Experimental Results
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

21. Experimental Results
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

22. Conclusion Future Work Addressing top-N Recommendation
Exploit both trust network and rating profiles
Experiments demonstrate that exploiting the social network improves the recommendations.
Future Work
More sophisticated combined models
Better suitable evaluation metrics
Distributed rating repositories
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

23. Thank you!
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

24. References
[Deshpande 2004] M. Deshpande and G. Karypis. Item based top-n recommendation algorithms. ACM Transactions on Information Systems, 22:143–177, 2004.
[Golbeck 2005] J. Golbeck. Computing and Applying Trust in Web-based Social Networks. PhD thesis, University of Maryland College Park, 2005.
[Goldberg 1992] D. Goldberg, D. Nichols, B. M. Oki, and D. Terry. Using collaborative filtering to weave an information tapestry. Communications of the ACM, 35(12), 1992.
[Jamali 2009] M. Jamali and M. Ester. Trustwalker: A random walk model for combining trust- based and item-based recommendation. In KDD’09: The 15th ACM SIGKDD conference on Knowledge Discovery andData Mining, 2009.
[Karypis 2001] G. Karypis. Evaluation of item-based top-n recommendation algorithms. In CIKM ’01: Proceedings of the tenth international conference on Information and knowledge management, pages 247–254, New York, NY, USA, 2001.
[Kim 2007] H.-N. Kim, A.-T. Ji, H.-J. Kim, and G.-S. Jo. Error-based collaborative filtering algorithm for top-n recommendation. In The Joint International Conferences on Asia-Pacific Web Conference and Web-Age Information Management (APWeb/WAIM), pages 594–605, Huang Shan, China, June 2007.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

25. References (cont)
[Kwon 2008] Y. Kwon. Improving top-n recommendation techniques using rating variance. In RecSys’08: Proceedings of the 2008 ACM conference on Recommender systems, pages 307–310, New York, NY, USA, 2008.
[Massa 2007] P. Massa and P. Avesani. Trust-aware recommender systems. In RecSys’07: ACM Recommender Systems Conference, USA, 2007.
[McLaughlin 2004] M. R. McLaughlin and J. L. Herlocker. A collaborative filtering algorithm and evaluation metric that accurately model the user experience. In SIGIR ’04: Proceedings of the 27th international ACM SIGIR conference on Information Retrieval, pages 329–336, New York, NY, USA, 2004.
[Richardson 2002] M. Richardson and P. Domingos. Mining knowledge-sharing sites for viral marketing. In KDD’02: The 8th ACM SIGKDD conference on Knowledge Discovery andData Mining, 2002.
[Sarwar 2001] B. Sarwar, G. Karypis, J. Konstan, and J. Riedl. Item-based collaborative filtering recommendation algorithms. In WWW’01: 10th International World Wide Web Conference, 2001.
Mohsen Jamali. Using Trust Networks to Improve Top-N Recommendation

26. Using Collaborative Filtering for Top-N Recommendation
User-based CF for top-N recommendation
Find top K similar users to u  Nu.
Aggregate the list of items rated by all vє Nu.
Similarity measure = Revised Pearson Correlation [4]

27. Using Collaborative Filtering for Top-N Recommendation
Item-based CF for top-N recommendation
Consider items rated highly by u  I’u.
Find top K similar items to all items jєI’u.
Aggregate these items to compute top-N items
Similarity measure = Revised Pearson Correlation

28. Combined Approach (cont)
Finding k2 trusted users:
Breadth First Search
Random Walk
Perform several random walks to find k2 users.
Estimated rating for an item would be mean of ratings expressed by trusted friends on item i.

29. Experimental Results – Cold Start Users

30. Experimental Results – Cold Start Users

31. Experimental Results – Cold Start Users

32. Experimental Results
Mixture weights are not tuned