WEAT_Slides

Transcript

1. Find Food for Us! Ming He weat
2. Where to Dine?
3. Where to Dine?
4. Where to Dine? weat Find Food for Us!
5. • 2K Users • 70K Ratings + Reviews • 3K Restaurants Yelp User Reviews
7. User’s X Ratings - - - -
8. Word 1 3 0 16 Word 2 0 12 9 ... Word n 31 1 2 Restaurants X Reviews User’s X Ratings - - - - Bag of words
9. Word 1 0.15 0 0.9 Word 2 0 0.8 0.3 ... Word n 0.6 0.1 0.2 Restaurants X Reviews User’s X Ratings - - - - TF-IDF
10. 1 0.9 0.2 0.9 1 0.15 0.2 0.15 1 Word 1 0.15 0 0.9 Word 2 0 0.8 0.3 ... Word n 0.6 0.1 0.2 Restaurants X Reviews + Latent Semantic Indexing User’s X Ratings - - - - Restaurant Similarity by Review TF-IDF
11. 1 0.9 0.2 0.9 1 0.15 0.2 0.15 1 User’s X Ratings - - - - Restaurant Similarity by Review
12. 1 0.9 0.2 0.9 1 0.15 0.2 0.15 1 User’s X Ratings - - - - 1 0.2 0.1 0.2 1 0.15 0.1 0.15 1 Restaurant Similarity by Rating Restaurant Similarity by Review
13. 1 0.9 0.2 0.9 1 0.15 0.2 0.15 1 User’s X Ratings - - - - 1 0.2 0.1 0.2 1 0.15 0.1 0.15 1 Restaurant Similarity by Rating Restaurant Similarity by Review (LSI) weight 1-weight
14. User’s X Ratings - - - - 1 0.8 0.4 0.8 1 0.2 0.4 0.2 1 Hybrid Restaurant Similarity
15. User’s X Ratings - - - - 1 0.8 0.4 0.8 1 0.2 0.4 0.2 1 Hybrid Restaurant Similarity 99% Sparse
16. User’s X Ratings - - - - 1 0.8 0.4 0.8 1 0.2 0.4 0.2 1 Hybrid Restaurant Similarity Item-based Collaborative Filtering 99% Sparse
17. User’s X Ratings 1 0.8 0.4 0.8 1 0.2 0.4 0.2 1 Hybrid Restaurant Similarity Item-based Collaborative Filtering
18. User’s X Ratings
19. User’s X Ratings 7 6
20. User’s X Ratings 7 6
21. Multi-Armed Bandit Validation Real-time learning to improve system Recommended 3 Restaurants Random Recommendation
22. Multi-Armed Bandit Validation Recommended 3 Restaurants Random Recommendation Collect user’s click-through to update the system in real time and display popular models more frequently
23. Multi-Armed Bandit Validation Recommended 3 Restaurants Random Recommendation Use group’s click-through to update the system in real time and display popular models more frequently
24. weat
25. Ming He Economics
26. Appendix
27. Why hybrid similarity matrix? Rating-based similarity matrix? Almost Uniform Ratings Rating Restaurant 1 Restaurant 2 ... Restaurant n User 1 4.334 4.333 4.333 4.334 User 2 2.564 2.563 2.564 2.563 Degree of Sparsity: 0.99
28. Why hybrid similarity matrix? Review-based similarity matrix Rating Restaurant 1 Restaurant 2 ... Restaurant n User 1 4.334 3.343 4.567 2.634 User 2 2.564 2.783 3.564 3.963 Parsing Reviews TF-IDF Latent Semantic Indexing