Introduction Matrix Factorization Methods Netflix Prize Competition Conclusion MATRIX FACTORIZATION TECHNIQUE FOR RECOMMENDER SYSTEMS Oluwashina Aladejubelo Universite Joseph Fourier, Grenoble, France June 6, 2015 Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
30
Embed
Matrix Factorization Technique for Recommender Systems
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
MATRIX FACTORIZATION TECHNIQUE FORRECOMMENDER SYSTEMS
Oluwashina Aladejubelo
Universite Joseph Fourier,Grenoble, France
June 6, 2015
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
About Me
Bachelor of Science, Ambrose Alli University, Nigeria(2004-2008)
IT Business Analyst, Virgin Nigeria Airlines (2009-2011)
Team Lead/Software Architect, Speckless InnovationsLimited (2011-2014)
Master of Informatics (M2 MOSIG), Universit JosephFourier, Grenoble (2014-2015)
Master Thesis on ”Distributed Large-Scale Learning” withPr. Massih-Reza Amini.
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Overview
1 Introduction
2 Matrix Factorization Methods
3 Netflix Prize Competition
4 Conclusion
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
1 IntroductionRecommender SystemsContent Filtering ApproachCollaborative Filtering ApproachContent vs Collaborative Filtering
2 Matrix Factorization MethodsMatrix Factorization Model (MFM)Stochastic Gradient DescentAlternating Least SquaresAdding BiasesAdditional Input SourceTemporal DynamicsVarying confidence levels
3 Netflix Prize Competition
4 Conclusion
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Recommender Systems
Recommender systems analyze patterns of user interest inproducts to provide personalized recommendations
They seek to predict the rating or preference that user wouldgive to an item
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Recommender Systems
Such systems are very useful for entertainment products suchas movies, music, and TV shows.
Many customers will view the same movie and each customeris likely to view numerous different movies.
Huge volume of data arise from customer feedbacks which canbe analyzed to provide recommendations
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Content Filtering Approach
creating profile for each user or product to characterize itsnature.programs associate users with matching products.
it requires gathering external information that may not beavailable
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Collaborative Filtering Approach
depends on past user behaviour, e.g. previous transactions orproduct rating
does not rely on creation of explicit profiles
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Collaborative Filtering Approach
the primary areas of collaborative filtering are neighborhoodmethods and latent factor models
neighborhood is based on computing the relationshipsbetween items or users
latent factor models tries to explain by characterizing bothitems and users on say, 20 to 100 factors inferred from theratings patterns
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Content vs Collaborative Filtering
Collaborative filtering address data aspects that are difficult toprofile.
it is generally more accurate
suffers from cold startup problem (new product / new user) inwhich case content filtering is better
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
1 IntroductionRecommender SystemsContent Filtering ApproachCollaborative Filtering ApproachContent vs Collaborative Filtering
2 Matrix Factorization MethodsMatrix Factorization Model (MFM)Stochastic Gradient DescentAlternating Least SquaresAdding BiasesAdditional Input SourceTemporal DynamicsVarying confidence levels
3 Netflix Prize Competition
4 Conclusion
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Matrix Factorization Model (MFM)
some of the most successful realizations of latent factormodels are based on matrix factorization
it characterizes both items and users by vectors of factorsinferred from item rating patterns
high correspondence between item and user factors leads to arecommendation
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Matrix Factorization Model (MFM)
MFM maps both users & items to a joint latent factor spaceof dimensionality f
the user-item interactions are modeled as inner products inspace f
each item i is associated with a vector qi ∈ Rf
each user u is associated with a vector pu ∈ Rf
Oluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Matrix Factorization Model (MFM)
the approximate user rating is given by
r̂ui = qTi Pu (1)
carelessly addressing only the relatively few known entries ishighly prone to overfitting
observed ratings can be modeled directly with regularizationas follows
minq∗,p∗∑
(u,i)∈κ
(rui − qTi pu)2 + λ(||qi ||2 + ||pu||2) (2)
κ is a set of (u, i) pairs for which rui is knownOluwashina Aladejubelo Matrix Factorization Techniques for Recommender Systems
IntroductionMatrix Factorization Methods
Netflix Prize CompetitionConclusion
Stochastic Gradient Descent (SGD) - Simon Funk; 2006
SGD approach can be used for solving the equation (2)
For each given training case, the system predicts ru i andcomputes the prediction error
eui = rui − qTi pu
it modifies the parameters by a magnitude proportional to γin the opposite direction of the gradient, yielding∈ Rf