Recommendation Engines - An Architectural GuidePresentation Transcript
RECOMMENDATIONENGINESAN ARCHITECTURAL GUIDEby Timothy Vogel; Sr. Data Architect; Sony Network Entertainment Inc(SNEI)email@example.com
the vagaries of aggregation!Presentation Science; start with ajoke...
...or two...the vagaries of personal taste!
...or three or four...the vagaries of human experiencerelative to human tasterelative to human expectations!
Semantic Transport Collaborative Filtering - tenets of "semantic transport" "Automated Collaborative Filtering and Semantic Transports" http://www.lucifer.com/~sasha/articles/ACF.html "Social Trust Bonds" all recommendations depend upon "trust" between the recommenderand target audience. behavioral similarity is the "online proxy" for social familiarity betweenhuman beings. trust is promoted and diminished by the efficacy of the recommendationitself via... the recommenders actual motivation the recommendees perception of that motivation the recomendees final judgment of the recommendations "quality"
Recommendation Engine Design Questionsand its architectural analog. Have you considered… How To Attract "Recommendees"? website design and targeting How To Proffer the Recommendation? interface How To Collect the Requisite Data? instrumentation/appropriate type and range of transaction-types How To "Learn" From The User? A-B-(A) testing How To "Learn" From Recommendation Performance? adding "clicked recommendation" and "ignored recommendation"transaction type for feedback learning mode How to Profit From Your "Service"? monetization
Conceptual ArchitectureYou need to decide… What is the context of the recommendation? overt (move, book, restaurant recommendation site) covert (website with pop-ups or directed navigation) Who/What is the mechanism of similarity? User transaction Item similarity (product-type, genre, meta-data, price, etc.) Item similarity via text-analysis How will the recommendation be made? pull – explicit user participation via ratings push – implicit user participation via clicks How will we collect the data required? apis that ask for it java-script that captures it deep-packet inspection
“How to Develop Online RecommendationSystemsThat Deliver Superior Business Performance”Cognizant Systemshttp://www.cognizant.com/InsightsWhitepapers/How-to-Develop-Online-Recommendation-Systems-that-Deliver-Superior-Business-Performance.pdf
In Reality, All Recommendation Engines AreGlorified Classification SystemsCognizant Systemshttp://www.cognizant.com/InsightsWhitepapers/How-to-Develop-Online-Recommendation-Systems-that-Deliver-Superior-Business-Performance.pdf
Components of Recommendationssocial construct : recommendation engines facsimile Familiarity; history of social interaction : transactional history Trust; similarity of past interactions : shared transactions Formality of suggestion; deep insight into a friends dilemma : "precision" mere familiarity with a friends needs : "recall" Efficacy; advice-request frequency/regularity: click-thru rate Learning; memory : A-B-(A) test system
End with a joke...Never forget;Recommendationis always apracticalapplication.
...or two...Never forget 2;Decide whiledesigning theRecommendation System justwhat constitutesthe risk ofmaking a "badrecommendation".