Analyzing Weighting Schemes in Collaborative Filtering: Cold Start, Post Cold Start and Power Users
Download as ODP, PDF2 likes1,155 views
The document analyzes different weighting schemes in collaborative filtering to address the problem of popularity bias. It describes how standard collaborative filtering approaches can over-recommend popular items. The paper presents experiments testing two similarity weighting strategies (linear inverse and inverse user frequency) in different scenarios using two movie rating datasets. The results show that popularity weighting improves precision for users with a moderate number of ratings when the rating scale is compact, but not when the number of ratings is very low or very high.
Analyzing Weighting Schemes in Collaborative Filtering: Cold Start, Post Cold Start and Power Users
- 1. Analyzing Weighting Schemes in Collaborative Filtering: Cold Start, Post Cold Start and Power Users ACM SAC 2012 Alan Said, Brijnesh J. Jain, Sahin Albayrak {alan, jain, sahin}@dai-lab.de TU-Berlin
- 2. Outline Movie Recommendation Problem: Popularity Bias Collaborative Filtering Similarity Weighting Schemes Experiments Results Conclusion 30.03.2012 2
- 3. Recommender Systems What they should do: Find items which should be of interest to users Find items which should be useful to users What they often do instead: Find items which are known by users Find items which users would have found anyway 30.03.2012 3
- 4. Popularity Bias What is popularity bias? Some things are more popular than others Blockbuster movies1: Pulp Fiction, Inception, etc. Best selling books2: Steve Jobs Bio, A Song of Ice and Fire, etc. Apps3: Angry Birds, Skype, Kindle 1 : IMDb most popular 2 : Amazon 2011 best sellers 3 : Most downloaded Android apps 30.03.2012 4
- 6. Popularity Bias ed rat hly = hig ms ite lar pu Po 30.03.2012 6
- 7. Collaborative Filtering Looks for users who share rating patterns Use ratings from like-minded users to calculate a prediction for the user Boils down to: The most similar users create a neighborhood. Those items which are most popular in the neighborhood will be recommended. 30.03.2012 7
- 8. Collaborative Filtering: Similarities Standard CF approaches do not consider the popularity of items when creating neighbor- hoods of similar users. i.e. not considering the popularity bias. Percentage of ratings given to different popularity classes of movies in the Movielens 10 Million ratings dataset 30.03.2012 8
- 9. Collaborative Filtering: Similarities Standard CF approaches do not consider the popularity of items when creating neighbor- hoods of similar users. i.e. not considering the popularity bias. Disitribution of ratings given to the three most popular movies in the Movielens 10 Million dataset 30.03.2012 9
- 10. Weighting Schemes 30.03.2012 10
- 11. Experiments Approach: Test two similarity weighting strat- egies in different scenarios on two different movie rating datasets. Weighting: Linear Inverse & Inverse User frequency Datasets: Movielens10M & Moviepilot Scenarios: Cold Start, Post Cold Start, Power Users 30.03.2012 11
- 12. Results 30.03.2012 12
- 13. Results When is it good to use popularity weighting? Movielens 10M >20% improvement in Precision Ratings: 1-5 stars ← 30 - 100 items each → 13 30.03.2012
- 14. Results When is it not good to use popularity weighting? Moviepilot No significant improvement in Precision Ratings: 0-10 stars 30.03.2012 14
- 15. Conclusion Popular items create a problem for recom- mender systems due to favorable bias. Similarity weighting can lessen the effects of the bias when the rating scale is “compact” when the users have “more than few” and “less than many” ratings 30.03.2012 15
- 16. Ongoing Work What if lower precision does not mean poorer quality? Lower precision can be an indicator of new, novel, serendipitous recommendations – these will produce lower precision values in offline evaluation Currently evaluating the quality of recommender algorithms based on user feedback, not only precision/recall/etc. Values. Users and Noise: The Magic Barrier of Recommender Systems – UMAP'12 User satisfaction survey: www.dai-lab.de/~alan/survey 30.03.2012 16
- 17. Questions? Thank You! 30.03.2012 17