U E D R , A S I H C RSS

Project Prometheus/Collaborative Filtering

일단은 본격적인 CF로 가는 것보다 아마존의 "Customers who bought this book also bought"식으로 좀 더 간단한 것을 하는 것이 좋을 듯 하다. 이것은 꼭 Clustering이 필요없다 -- Clustering이 효과를 발휘하려면 상당량의 데이타(NoSmok:CriticalMass )가 쌓여야 하는데, 쉬운 일이 아닐 것이다. 다음은 JuNe이 생각한 간단한 알고리즘. 일종의 Item-to-Item Correlation Recommendation.
  • userPref is a set of book preferences of a user.
  • For every meaningful action(searching, viewing, writing a review, rating without writing reviews) there is a pref coefficient. (prefCoef)
  • bookPref is a set of book preferences of a book. (similar concept to "also bought")
  • When a user does a specific well-defined action, bookPref is updated as "prefCoef*userPref+bookPref" and resorted. (some books at the end of the list might be eliminated for performance reasons)
  • Any top Ns are shown from the bookPref


Iteration 2 에서 만든 Prototype & 알고리즘 (추후 작성 예정)



Valid XHTML 1.0! Valid CSS! powered by MoniWiki
last modified 2021-02-07 05:24:05
Processing time 0.0147 sec