Download presentation
Presentation is loading. Please wait.
Published byReginald Daniel Modified over 9 years ago
1
Collaborative Competitive Filtering: Learning recommender using context of user choices Shuang Hong Yang Bo Long, Alex Smola, Hongyuan Zha Zhaohui Zheng Georgia Tech Yahoo! Labs
2
2 Outline Motivation Collaborative filtering Collaborative competitive filtering Experiments Summary & Discussions
3
3 Too many options to choose from? Information filter (e.g., recommender system): = capture user's preference to: Identify what you truly like Avoid your disliked ones filter
4
Current information filters Learn preference by matrix(e.g., ratings) completion: X = missing The Netflix Prize data This paper: No more matrix completion! --- Information filters should encode the user-system interaction process!
5
5 User-system interaction A typical interaction at a recommender system: user visit the recommender system (e.g.visit a website) system offers a subset of options out of a huge inventory user chooses one from the offers and take actions context of user choice: user makes a choice (indicate preference) in the context of a offer set, rather than the whole inventory
6
6 Context of user choices Joe is a fan of action Joe likes DieHard over :Rocky and Terminator Joe is a fan of Bruce Willis User make choices (indicate preference) in the context of a small offer set rather than the whole inventory More accurate preference info from session data
7
7 Interaction session data Most nature data repository of a recommender system When degraded to rating matrix missing value in rating matrix user's implicite interactions with non-chosen items are taken as missing values if sessions are degraded to matrix cons: vulnerability to overfitting
8
8 Key insight No more matrix completion! --- Information filters should encode the user-system interaction process! 1.User make choices (indicate preference) in the context of a small offer set rather than the whole inventory 2. Matrix completion does not fully leverage the session data and is hence vulnerable to overfitting
9
9 Outline Motivation Collaborative filtering Collaborative competitive filtering Experiments Summary & Discussions
10
10 Current Recommender Systems Formulation Matrix completion (e.g., rating prediction): Methodology Collaborative filtering Content-based filtering user\itemHarry Potter God Farther Avatar Joe355 Harry45 George533 ?
11
11 Collaborative filtering Encoding the "Collaboration" effect Similar items receive similar responses from similar users John likes Terminator, David is similar to John → David likes Terminator John like Terminator, Rocky is similar to Terminator → John likes Rocky Pooling the evidences to remedy data sparseness Taxonomy Neighborhood approaches Latent factor approaches
12
12 Neighborhood approaches user, item, rating 1.Define similarity (e.g., between two users) 2.Estimate rating:
13
13 Latent factor approaches 1. Associate latent factor for each user, each item latent user profile, latent item profile 2. Multiplicative rating model 3. Estimate latent factors through optimization
14
14 CF: A motivating pitfall Over-optimistic prediction on binary responses Beyond matrix completion: new formulations are needed Beyond collaborative filtering: new mechanism needed
15
15 Outline Motivation Collaborative filtering Collaborative competitive filtering Experiments Summary & Discussions
16
16 An overview Directly modeling session data Encode collaboration effect: –Similar users have similar preference toward similar items Encode competitive effect: –Items compete for the attention of users
17
17 Interaction session At an interaction t : – User – Offer set – Decision set (assume one per choice)
18
18 Axiomatic view of user choice process Revenue (utility): user u 's gain from item i : r ui Opportunity cost: Given an offers set O, the opportunity cost is the potential loss of taking i which excluding u to take other offers. Profit: the net gain: Local optimality of user choices
19
19 Collaborative competitive filtering Directly modeling session data Encode collaboration effect: collaborative filtering –multiplicative utility function: : latent user profile, : latent item profile Encode competitive effect: local context-aware loss
20
20 Local context-aware loss Idea loss Cons: –NP-hard –The constraint restricts the utility only up to an arbitrary monotone transformation, thus cannot yield unique solution
21
21 Local context-aware loss Surrogate loss functions –Softmax (multinomial logit) –Hinge (bundle pairwise)
22
22 Collaborative competitive filtering Softmax model (multinomial logit model)
23
23 Collaborative competitive filtering Hinge model (max-score estimation)
24
24 Optimization Stochastic gradient descent
25
25 Optimization Large-scale implementation –distributed optimization with averaging (decomposing the objective/data w.r.t. users) on Hadoop –feature hashing
26
26 Extensions Encoding node features –X u : user features (age, gender, income, etc.) –X i : item features (description words,etc.)
27
27 Extensions Nonresponded sessions –user did not choose any of the offers –assume each user has a response threshold –user responds only if his satisfaction exceeds the threshold
28
28 Experiments Yahoo! Pulse social data –1.2M users, 400 items –29M interactions – simulated contexts Netflix prize data (5star ratings) –0.48M users, 18K items –100M interactions –simulated context Yahoo! front page –3.6M users, 2.5K items –110M interactions –real session context
29
29 Results Offline test (social & Netflix)
30
30 Results Offline test (Y! front page)
31
31 Results Online test (Y! front page)
32
32 Results Narrow-band filter (social)
33
33 Summary Information filtering beyond matrix completion --- modeling user-system interaction sessions –Placing user choice into context → more accurate preference modeling –Remedy data sparsity issue Local context-aware loss –softmax (multinomial logit) –hinge (bundled pariwise) Significant improvements –offline top-k ranking –online clickthrough rate
34
34 Thanks! Any comments would be appreciated!
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.