Recommendation in social networks
This presentation is the property of its rightful owner.
Sponsored Links
1 / 58

Recommendation in Social Networks PowerPoint PPT Presentation


  • 91 Views
  • Uploaded on
  • Presentation posted in: General

Recommendation in Social Networks. Mohsen Jamali , Martin Ester Simon Fraser University Vancouver, Canada. UBC Data Mining Lab October 2010. Outline. Introduction Collaborative Filtering Social Recommendation Evaluating Recommenders TrustWalker SocialMF Conclusion. Outline.

Download Presentation

Recommendation in Social Networks

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Recommendation in social networks

Recommendation in Social Networks

Mohsen Jamali, Martin Ester

Simon Fraser University

Vancouver, Canada

UBC Data Mining Lab October 2010


Outline

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Outline1

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Introduction

Introduction

  • Need For Recommenders

    • Rapid Growth of Information

    • Lots of Options for Users

  • Input Data

    • A set of users U={u1, …, uN}

    • A set of items I={i1, …, iM}

    • The rating matrix R=[ru,i]NxM

Mohsen Jamali, Recommendation in Social Networks


Problem definitions in rss

Problem Definitions in RSs

  • Predicting the rating on a target item for a given user (i.e. Predicting John’s rating on Star Wars Movie).

  • Recommending a List of items to a given user (i.e. Recommending a list of movies to John for watching).

movie1 ??

Recommender

List of Top Movies ??

Recommender

Mohsen Jamali, Recommendation in Social Networks


Outline2

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Collaborative filtering

Collaborative Filtering

  • Most Used and Well Known Approach for Recommendation

  • Finds Users with Similar Interests to the target User

  • Aggregating their opinions to make a recommendation.

  • Often used for the prediction task

Mohsen Jamali, Recommendation in Social Networks


Collaborative filtering1

Target

Customer

Collaborative Filtering

Aggregator

Prediction

Mohsen Jamali, Recommendation in Social Networks


Item based collaborative filtering

Item based Collaborative Filtering

  • Normally, there are a lot more users than items

  • Collaborative Filtering doesn’t scale well with users

  • Item based Collaborative Filtering has been proposed in 2001

  • They showed that the quality of results are compatible in item based CF

Mohsen Jamali, Recommendation in Social Networks


Item based collaborative filtering1

Item-based Collaborative Filtering

Mohsen Jamali, Recommendation in Social Networks


Item item collaborative filtering

Item-Item Collaborative Filtering

Aggregator

Prediction

Mohsen Jamali, Recommendation in Social Networks


Outline3

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Recommendation in social networks1

Recommendation in Social Networks

  • Social Networks Emerged Recently

    • Independent source of information

  • Motivation of SN-based RS

    • Social Influence: users adopt the behavior of their friends

  • Social Rating Network

  • Social Network  Trust Network

Mohsen Jamali, Recommendation in Social Networks


Recommendation in social networks2

Recommendation in Social Networks

  • Cold Start users

    • Very few ratings

    • 50% of users

    • Main target of SN recommenders

A Sample Social Rating Network

Mohsen Jamali, Recommendation in Social Networks


Recommendation in social networks3

Recommendation in Social Networks

  • Classification of Recommenders

    • Memory based

    • Model based

  • Memory based approaches for recommendation in social networks

    • [Golbeck, 2005]

    • [Massa et.al. 2007]

    • [Jamali et.al. 2009]

    • [Ziegler, 2005]

Mohsen Jamali, Recommendation in Social Networks


Trust based recommendation

Trust-based Recommendation

  • Explores the trust network to find Raters.

  • Aggregate the ratings from raters for prediction.

  • Different weights for users

Mohsen Jamali, Recommendation in Social Networks


Outline4

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Evaluating recommenders

Evaluating Recommenders

  • Cross Validation

    • K-Fold

    • Leave-one-out

  • Root Mean Squared Error (RMSE)

  • Mean Absolute Error (MAE)

Mohsen Jamali, Recommendation in Social Networks


Data sets

Data Sets

  • Epinions – public domain

  • Flixster

    • Flixster.com is a social networking service for movie rating

    • The crawled data set includes data from Nov 2005 – Nov 2009

    • Available at http://www.cs.sfu.ca/~sja25/personal/datasets/

Mohsen Jamali, Recommendation in Social Networks


Data sets cont

Data Sets (cont.)

  • General Statistics of Flixster and Epinions

  • Flixster: 1M users, 47K items

    • 150K users with at least one rating

    • Items: movies

    • 53% cold start

  • Epinions: 71K users, 108K items

    • Items: DVD Players, Printers, Books, Cameras,…

    • 51% cold start

Mohsen Jamali, Recommendation in Social Networks


Outline5

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Trustwalker motivation

TrustWalker - Motivation

  • Issues in Trust-based Recommendation

    • Noisy data in far distances

    • Low probability of Finding rater at close distances

Mohsen Jamali, Recommendation in Social Networks


Trustwalker motivation1

TrustWalker - Motivation

  • How Far to Go into Network?

    • Tradeoff between Precision and Recall

  • Trusted friends on similar items

  • Far neighbors on the exact target item

Mohsen Jamali, Recommendation in Social Networks


Trustwalker

TrustWalker

  • TrustWalker

    • Random Walk Model

    • Combines Item-based Recommendation and Trust-based Recommendation

  • Random Walk

    • To find a rating on the exact target item or a similar item

    • Prediction = returned rating

Mohsen Jamali, Recommendation in Social Networks


Single random walk

Single Random Walk

  • Starts from Source user u0.

  • At step k, at node u:

    • If u has rated I, return ru,i

    • With Φu,i,k, the random walk stops

      • Randomly select item j rated by u and return ru,j .

    • With 1- Φu,i,k, continue the random walk to a direct neighbor of u.

Mohsen Jamali, Recommendation in Social Networks


Stopping probability in trustwalker

Stopping Probability in TrustWalker

  • Item Similarities

  • Φu,i,k

    • Similarity of items rated by u and target item i.

    • The step of random walk

Mohsen Jamali, Recommendation in Social Networks


Recommendation in trustwalker

Recommendation in TrustWalker

  • Prediction = Expected value of rating returned by random walk.

Mohsen Jamali, Recommendation in Social Networks


Properties of trustwalker

Properties of TrustWalker

  • Special Cases of TrustWalker

    • Φu,i,k = 1

      • Random Walk Never Starts.

      • Item-based Recommendation.

    • Φu,i,k = 0

      • Pure Trust-based Recommendation.

      • Continues until finding the exact target item.

      • Aggregates the ratings weighted by probability of reaching them.

      • Existing methods approximate this.

  • Confidence

    • How confident is the prediction

Mohsen Jamali, Recommendation in Social Networks


Experimental setups

Experimental Setups

  • Evaluation method

    • Leave-one-out

  • Evaluation Metrics

    • RMSE

    • Coverage

    • Precision = 1- RMSE/4

Mohsen Jamali, Recommendation in Social Networks


Comparison partners

Comparison Partners

  • Tidal Trust [Golbeck, 2005]

  • Mole Trust [Massa, 2007]

  • CF Pearson

  • Random Walk 6,1

  • Item-based CF

  • TrustWalker0 [-pure]

  • TrustWalker [-pure]

Mohsen Jamali, Recommendation in Social Networks


Experiments cold start users

Experiments – Cold Start Users

Mohsen Jamali, Recommendation in Social Networks


Experiment all users

Experiment- All users

Mohsen Jamali, Recommendation in Social Networks


Experiments confidence

Experiments - Confidence

  • More confident Predictions have lower error

Mohsen Jamali, Recommendation in Social Networks


Outline6

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Matrix factorization

Matrix Factorization

  • Model based approach

  • Latent features for users

  • Latent features for items

  • Ratings are scaled to [0,1]

  • g is logistic function

U and V have normal priors

Mohsen Jamali, Recommendation in Social Networks


Social trust ensemble 2009

Social Trust Ensemble [2009]

Mohsen Jamali, Recommendation in Social Networks


Social trust ensemble cont

Social Trust Ensemble (cont.)

  • Issues with STE

    • Feature vectors of neighbors should influence the feature vector of u not his ratings

    • STE does not handle trust propagation

    • Learning is based on observed ratings only.

Mohsen Jamali, Recommendation in Social Networks


The socialmf model

The SocialMF Model

  • Social Influence  behavior of a user u is affected by his direct neighbors Nu.

  • Latent characteristics of a user depend on his neighbors.

  • Tu,v is the normalized trust value.

Mohsen Jamali, Recommendation in Social Networks


The socialmf model cont

The SocialMF Model (cont.)

Mohsen Jamali, Recommendation in Social Networks


The socialmf model cont1

The SocialMF Model (cont.)

Mohsen Jamali, Recommendation in Social Networks


The socialmf model cont2

The SocialMF Model (cont.)

Mohsen Jamali, Recommendation in Social Networks


The socialmf model cont3

The SocialMF Model (cont.)

Mohsen Jamali, Recommendation in Social Networks


The socialmf model cont4

The SocialMF Model (cont.)

Mohsen Jamali, Recommendation in Social Networks


The socialmf model cont5

The SocialMF Model (cont.)

  • Properties of SocialMF

    • Trust Propagation

    • User latent feature learning possible with existence of the social network

      • No need to fully observed rating for learning

      • Appropriate for cold start users

Mohsen Jamali, Recommendation in Social Networks


Experimental setups1

Experimental Setups

  • 5-fold cross validation

  • Using RMSE for evaluation

  • Comparison Partners

    • Basic MF

    • STE

    • CF

  • Model parameters

    • SocialMF:

    • STE:

Mohsen Jamali, Recommendation in Social Networks


Results for epinions

Results for Epinions

  • Gain over STE: 6.2%. for K=5 and 5.7% for K=10

Mohsen Jamali, Recommendation in Social Networks


Results for flixster

Results for Flixster

  • SocialMF gain over STE (5%) is 3 times the STE gain over BasicMF (1.5%)

Mohsen Jamali, Recommendation in Social Networks


Results cont

Results (cont.)

  • Lower error for Flixster

Epinions

Flixster

Mohsen Jamali, Recommendation in Social Networks


Sensitivity analysis on t

Sensitivity Analysis on λT

Sensitivity Analysis for Epinions

Mohsen Jamali, Recommendation in Social Networks


Sensitivity analysis on t1

Sensitivity Analysis on λT

Sensitivity Analysis for Flixster

Mohsen Jamali, Recommendation in Social Networks


Experiments on cold start users

Experiments on Cold Start Users

RMSE values on cold start users (K=5)

Mohsen Jamali, Recommendation in Social Networks


Experiments on cold start users1

Experiments on Cold Start Users

RMSE values on cold start users (K=5)

Mohsen Jamali, Recommendation in Social Networks


Experiments on cold start users2

Experiments on Cold Start Users

Mohsen Jamali, Recommendation in Social Networks


Analysis of learning runtime

Analysis of Learning Runtime

  • SocialMF:

  • STE:

  • SocialMF is faster by factor

Mohsen Jamali, Recommendation in Social Networks


Outline7

Outline

  • Introduction

    • Collaborative Filtering

    • Social Recommendation

    • Evaluating Recommenders

  • TrustWalker

  • SocialMF

  • Conclusion

Mohsen Jamali, Recommendation in Social Networks


Conclusion

Conclusion

  • TrustWalker [KDD 2009]

    • Memory-based

    • Random walk approach

  • SocialMF [RecSys 2010]

    • Model based

    • Matrix Factorization approach

  • Other work

    • Top-N Recommendation (RecSys 2009)

    • Link Prediction (ACM TIST 2010)

Mohsen Jamali, Recommendation in Social Networks


Conclusion1

Conclusion

  • Future Work

    • Framework for Clustering, Rating and Link Prediction

    • Explaining the recommendations

    • Constructing the social network from observed data.

Mohsen Jamali, Recommendation in Social Networks


Recommendation in social networks

Thank you!

Mohsen Jamali, Recommendation in Social Networks


  • Login