Increasing Diversity Through Furthest  Neighbor-Based Recommendation   Alan Said, Benjamin Kille, Brijnesh J. Jain, Sahin ...
Agenda   Problem   Approach: k Furthest Neighbor   Experimental settings   Results   Conclusions   Discussion27.02.2...
Problem: Missing diversity               • Accurate recommendations               • However, all items appear             ...
Problem: Missing diversity                                                              Intersection:   Intersection:     ...
Desired features of Recommendations   Reflect a user‘s preferences   Correct ranking   Novelty   Serendipity          ...
k Furthest Neighbor                                               dislike                                                 ...
k Furthest Neighbor27.02.2012      CC IRML   Folie 7
Experimental settingsData set: randomly sampled 1 million ratings out ofMovieLens (1M100k)     excluded: 100 most popular...
Results I Precision @ NN                    5          10             25               50        100       200Pearson Sim...
Results II Recall @ NN                    5          10             25               50        100       200Pearson Simil...
Results III Overlap 27.02.2012    Information Retrieval & Machine   11                           Learning
Conclusion „The enemy of my enemy is my friend“ seems to hold in    the context of recommender systems kFN achieved wors...
Thanks for your attention!!! http://recsyswiki.com27.02.2012          CC IRML    Folie 13
Contact             Benjamin Kille             Researcher of Competence Center    +49 (0) 30 / 314 – 74 128             In...
Discussion How to optimize the approach? Are there other ways to introcude more diverse    recommendations? How to eval...
Upcoming SlideShare
Loading in...5
×

Increasing Diversity Through Furthest Neighbor-Based Recommendation

585

Published on

Published in: Technology, Education
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
585
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
6
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Increasing Diversity Through Furthest Neighbor-Based Recommendation

  1. 1. Increasing Diversity Through Furthest Neighbor-Based Recommendation Alan Said, Benjamin Kille, Brijnesh J. Jain, Sahin Albayrak 1
  2. 2. Agenda Problem Approach: k Furthest Neighbor Experimental settings Results Conclusions Discussion27.02.2012 Information Retrieval & Machine 2 Learning
  3. 3. Problem: Missing diversity • Accurate recommendations • However, all items appear similar27.02.2012 Information Retrieval & Machine 3 Learning
  4. 4. Problem: Missing diversity Intersection: Intersection: Actors Plot Harry Potter and the Chamber of Secrets 49,0% 32,3% Harry Potter and the Prisoner of Azkaban 52,9% 26,7% Harry Potter and the Goblet of Fire 39,2% 29,2% Harry Potter and the Order of the Phoenix 49,0% 16,8% Harry Potter and the Half-Blood Prince 41,2% 15,5% Harry Potter and the Deathly Hallows: Part 1 43,1% 16,8% Harry Potter and the Deathly Hallows: Part 2 49,0% 22,4% Source: imdb.com27.02.2012 Information Retrieval & Machine 4 Learning
  5. 5. Desired features of Recommendations Reflect a user‘s preferences Correct ranking Novelty Serendipity Idea: Combining orthogonal recommendation Diversity27.02.2012 CC IRML Folie 5
  6. 6. k Furthest Neighbor dislike like27.02.2012 Information Retrieval & Machine 6 Learning
  7. 7. k Furthest Neighbor27.02.2012 CC IRML Folie 7
  8. 8. Experimental settingsData set: randomly sampled 1 million ratings out ofMovieLens (1M100k)  excluded: 100 most popular movies (rating frequency)  excluded: users with < 40 ratings  44,214 users; 9,432 moviesApproaches: Evaluation: – kNN Pearson  precision @N – kNN cosine  recall @N – kFN Pearson  overlap – kFN cosine N ϵ {5; 10; 25; 50; 100; 200} 27.02.2012 Information Retrieval & Machine 8 Learning
  9. 9. Results I Precision @ NN 5 10 25 50 100 200Pearson Similarity 0,0007 0,0110 0,0170 0,0280 0,0410 0,0900Cosine Similarity 0,0050 0,0070 0,0160 0,0270 0,0570 0,0000 27.02.2012 Information Retrieval & Machine 9 Learning
  10. 10. Results II Recall @ NN 5 10 25 50 100 200Pearson Similarity 0,0080 0,0130 0,0210 0,2300 0,0140 0,0100Cosine Similarity 0,0020 0,0060 0,0070 0,0060 0,0050 0,0040 27.02.2012 Information Retrieval & Machine 10 Learning
  11. 11. Results III Overlap 27.02.2012 Information Retrieval & Machine 11 Learning
  12. 12. Conclusion „The enemy of my enemy is my friend“ seems to hold in the context of recommender systems kFN achieved worse precision kFN provided higher recall with N > 50 kFN did provide orthogonal recommendations27.02.2012 Information Retrieval & Machine 12 Learning
  13. 13. Thanks for your attention!!! http://recsyswiki.com27.02.2012 CC IRML Folie 13
  14. 14. Contact Benjamin Kille Researcher of Competence Center +49 (0) 30 / 314 – 74 128 Information Retrieval +49 (0) 30 / 314 – 74 003 & Machine Learning benjamin.kille@dai-labor.de27.02.2012 CC IRML Folie 14
  15. 15. Discussion How to optimize the approach? Are there other ways to introcude more diverse recommendations? How to evaluate diversity in the context of recommender system?27.02.2012 CC IRML Folie 15
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×