cse 258
play

CSE 258 Web Mining and Recommender Systems Advanced Recommender - PowerPoint PPT Presentation

Nov 14, 2023 •471 likes •1.74k views

CSE 258 Web Mining and Recommender Systems Advanced Recommender Systems This week Methodological papers Bayesian Personalized Ranking Factorizing Personalized Markov Chains Personalized Ranking Metric Embedding Translation-based

  1. Translation-based Recommendation bias • Benefit from using metric embeddings • Model (u, i, j) with a single component • Recommendations can be made by a simple NN search

  2. Translation-based Recommendation

  3. Translation-based Recommendation ● Automotives ● Office Products ● Toys & Games ● Video Games ● Cell Phones & Accessories ● Clothing, Shoes, and Jewelry ● Electronics May 1996 - July 2014

  4. Translation-based Recommendation check-ins at different venues movie ratings Dec. 2011 - Apr. 2012 Nov. 2005 - Nov. 2009 (all available online) user reviews Jan. 2001 - Nov. 2013

  5. Translation-based Recommendation 11.4M reviews & ratings of 4.5M users on 3.1M local businesses restaurants, hotels, parks, shopping malls, movie theaters, schools, military recruiting offices, bird control, mediation services ... Characteristics : vast vocabulary of items, variability, and sparsity http://cseweb.ucsd.edu/~jmcauley/

  6. Translation-based Recommendation

  7. Translation-based Recommendation varying sparsity

  8. Translation-based Recommendation Unified

  9. Translation-based Recommendation TransRec

  10. Translation-based Recommendation Works well with… Doesn’t work well with…

  11. Overview Morals of the story: • Today we looked at two main ideas that extend the recommender systems we saw in class: 1. Sequential Recommendation: Most of the dynamics due to time can be captured purely by knowing the sequence of items 2. Metric Recommendation: In some settings, using inner products may not be the correct assumption

  12. Assignment 1

  13. Assignment 1

  14. CSE 258 Web Mining and Recommender Systems Real-world applications of recommender systems

  15. Recommending product sizes to customers

  16. Recommending product sizes to customers Goal: Build a recommender system that predicts whether an item will “fit”:

  17. Recommending product sizes to customers Challenges: • Data sparsity: people have very few purchases from which to estimate size • Cold-start: How to handle new customers and products with no past purchases? • Multiple personas: Several customers may use the same account

  18. Recommending product sizes to customers Data: • Shoe transactions from Amazon.com • For each shoe j , we have a reported size c_j (from the manufacturer), but this may not be correct! • Need to estimate the customer’s size (s_i ), as well as the product’s true size (t_j)

  19. Recommending product sizes to customers Loss function:

  20. Recommending product sizes to customers Loss function:

  21. Recommending product sizes to customers Loss function:

  22. Recommending product sizes to customers

  23. Recommending product sizes to customers Loss function:

  24. Recommending product sizes to customers Model fitting:

  25. Recommending product sizes to customers Extensions: • Multi-dimensional sizes • Customer and product features • User personas

  26. Recommending product sizes to customers Experiments:

  27. Recommending product sizes to customers Experiments: Online A/B test

  28. Playlist prediction via Metric Embedding

  29. Playlist prediction via Metric Embedding Goal: Build a recommender system that recommends sequences of songs Idea: Might also use a metric embedding (consecutive songs should be “nearby” in some space)

  30. Playlist prediction via Metric Embedding Basic model: (compare with metric model from last lecture)

  31. Playlist prediction via Metric Embedding Basic model (“single point”):

  32. Playlist prediction via Metric Embedding “Dual - point” model

  33. Playlist prediction via Metric Embedding Extensions: • Popularity biases

  34. Playlist prediction via Metric Embedding Extensions: • Personalization

  35. Playlist prediction via Metric Embedding Extensions: • Semantic Tags

  36. Playlist prediction via Metric Embedding Extensions: • Observable Features

  37. Playlist prediction via Metric Embedding Experiments: Yes.com playlists • Dec 2010 – May 2011 “Small” dataset: • 3,168 songs • 134,431 + 1,191,279 transitions “Large” dataset • 9,775 songs • 172,510 transitions + 1,602,079 transitions

  38. Playlist prediction via Metric Embedding Experiments:

  39. Playlist prediction via Metric Embedding Experiments: Small Big

  40. Efficient Natural Language Response Suggestion for Smart Reply

Recommend

More recommend