Recommender systemsRecommender systems

Drew CulbertDrew CulbertIST 497IST 497

12/12/0212/12/02

OverviewOverview

• DefinitionDefinition

• Ways its usedWays its used

• ProblemsProblems

• MaintenanceMaintenance

• The futureThe future

What is it?What is it?

• Recommender systems are a technological Recommender systems are a technological proxy for a social process.proxy for a social process.

• Recommender systems are a way of Recommender systems are a way of suggesting like or similar items and ideas suggesting like or similar items and ideas to a users specific way of thinking.to a users specific way of thinking.

• Recommender systems try to automate Recommender systems try to automate aspects of a completely different aspects of a completely different information discovery model where people information discovery model where people try to find other people with similar tastes try to find other people with similar tastes and then ask them to suggest new things.and then ask them to suggest new things.

ExampleExample

• Customer ACustomer A– Buys Metalica CDBuys Metalica CD– Buys Megadeth CDBuys Megadeth CD

• Customer BCustomer B– Does search on MetalicaDoes search on Metalica– Recommender system Recommender system

suggests Megadeth from suggests Megadeth from data collected from data collected from customer Acustomer A

Motivation for Recommender Motivation for Recommender SystemsSystems

• Automates quotes like:Automates quotes like:– "I like this book; you might be interested "I like this book; you might be interested

in it" in it" – "I saw this movie, you’ll like it“"I saw this movie, you’ll like it“– "Don’t go see that movie!" "Don’t go see that movie!"

Further MotivationFurther Motivation

• Many of the top commerce sites use Many of the top commerce sites use recommender systems to improve recommender systems to improve sales.sales.

• Users may find new books, music, or Users may find new books, music, or movies that was previously unknown movies that was previously unknown to them.to them.

• Also can find the opposite for e.g.: Also can find the opposite for e.g.: movies or music that will definitely movies or music that will definitely not be enjoyed.not be enjoyed.

Where is it used?Where is it used?

• Massive E-commerce sites use this Massive E-commerce sites use this tool to suggest other items a tool to suggest other items a consumer may want to purchaseconsumer may want to purchase

• Web personalizationWeb personalization

Ways its usedWays its used

• Survey’s filled out by past users for Survey’s filled out by past users for the use of new usersthe use of new users

• Search-style AlgorithmsSearch-style Algorithms

• Genre matchingGenre matching

• Past purchase queryingPast purchase querying

Recommender System Recommender System TypesTypes• Collaborative/Social-filtering systemCollaborative/Social-filtering system – aggregation of – aggregation of

consumers’ preferences and recommendations to consumers’ preferences and recommendations to other users based on similarity in behavioral patternsother users based on similarity in behavioral patterns

• Content-based systemContent-based system – supervised machine learning – supervised machine learning used to induce a classifier to discriminate between used to induce a classifier to discriminate between interesting and uninteresting items for the userinteresting and uninteresting items for the user

• Knowledge-based systemKnowledge-based system – knowledge about users – knowledge about users and products used to reason what meets the user’s and products used to reason what meets the user’s requirements, using discrimination tree, decision requirements, using discrimination tree, decision support tools, case-based reasoning (CBR)support tools, case-based reasoning (CBR)

Content-based Collaborative Content-based Collaborative Information FilteringInformation Filtering

• Relevance feedbackRelevance feedback – positive/negative – positive/negative prototypes.prototypes.

• Feature selectionFeature selection – removal of non- – removal of non-informative terms.informative terms.

• Learning to recommendLearning to recommend – agent counts – agent counts with 2 matrices; user vs. category with 2 matrices; user vs. category matrix (for successful classification) matrix (for successful classification) and user’s recommendation factor (1 and user’s recommendation factor (1 to 5) or binary.to 5) or binary.

products

Cu

stom

ers

ExamplesExamplesAmazon.comAmazon.com Books, movies, musicBooks, movies, music

CDNOW.comCDNOW.com MusicMusic

Ebay.comEbay.com (feedback (feedback forms)forms)

AnythingAnything

Reel.comReel.com MoviesMovies

Barnes & NobleBarnes & Noble BooksBooks

ProblemsProblems

• Inconclusive user feedback formsInconclusive user feedback forms

• Finding users to take the feedback surveysFinding users to take the feedback surveys

• Weak AlgorithmsWeak Algorithms

• Poor resultsPoor results

• Poor DataPoor Data

• Lack of DataLack of Data

• Privacy Control (May NOT explicitly Privacy Control (May NOT explicitly collaborate with recipients)collaborate with recipients)

MaintenanceMaintenance

• CostlyCostly

• Information becomes outdatedInformation becomes outdated

• Information quantity (large, disk Information quantity (large, disk space expansion)space expansion)

The Future of Recommender The Future of Recommender SystemsSystems

• Extract implicit negative ratings Extract implicit negative ratings through the analysis of returned item.through the analysis of returned item.

•How to integrate community with recommendations

•Recommender systems will be used in the future to predict demand for products, enabling earlier communication back the supply chain.

ResourcesResources

• http://www.acm.org/cacm/MAR97/http://www.acm.org/cacm/MAR97/resnick.htmlresnick.html

• http://www.ercim.org/publication/ws-http://www.ercim.org/publication/ws-proceedings/DelNoe02/proceedings/DelNoe02/CliffordLynchAbstract.pdfCliffordLynchAbstract.pdf

• http://people.cs.vt.edu/~ramakris/papers/http://people.cs.vt.edu/~ramakris/papers/ppp.pdfppp.pdf

• http://www.sims.berkeley.edu/~sinha/http://www.sims.berkeley.edu/~sinha/talks/UMD-Rashmi.pdftalks/UMD-Rashmi.pdf

Resources continuedResources continued

• http://www.cs.umn.edu/Research/http://www.cs.umn.edu/Research/GroupLens/papers/pdf/ec-99.pdfGroupLens/papers/pdf/ec-99.pdf

• http://www.rashmisinha.com/talks/http://www.rashmisinha.com/talks/Recommenders-SIGIR.pdfRecommenders-SIGIR.pdf

• http://www.grouplens.org/papers/http://www.grouplens.org/papers/pdf/slides-1.pdfpdf/slides-1.pdf

THANK YOUTHANK YOUANY ANY

QUESTIONS?QUESTIONS?