動態協同式過濾推薦之系統實做

Abstract

在現今廣受應用的資料探勘領域裡，推薦系統廣泛被應用於建議相關商品給使用者。在眾多新興電子商務中，線上電影推薦系統如"IMDb(Internet Movie Database)"， "MovieFinder.com"，"MovieLens"等，皆已廣為大眾喜愛和接受。這些電影推薦系統都是使用「協同過濾」(Collaborative Filtering) 技術來推薦相關影片給使用者。在這篇研究中，我們嘗試根據使用者線上瀏覽歷史和影片評價紀錄，運用「使用者相關性」(UserCorrelation)、「商品相關性」(ItemCorrelation)，以及「斜率性預測」(SlopeOne Predictor)，來建立一個「動態協同式過濾推薦系統」。此外所實做之系統運用「點閱流樹」(ClickStream Tree)技術，來預測使用者下一步將瀏覽的網頁。研究方法發現「商品相關性」(ItemCorrelation)是協同過濾演算法中較有效率的，而「斜率性預測」(SlopeOne Predictor)則是較精確的方法。此動態推薦系統以「服務性架構」(Service Oriented Architecture)及「商業流程執行語言」(Business Process Execution Language)為實做基礎，能有效地根據資料的特性進行相關商品推薦。

As a popular application of data mining, recommender systems attempt to predict items that a user may be interested in, given some information about the user's profile. With the gradually increasing use of IMDb (Internet Movie Database), MovieFinder.com, MovieLens and the likes, recommendation systems are gaining more popularity and acceptance by the public. Such film recommendation systems make use of collaborative filtering technology to recommend films to users. In this research, we propose a dynamic collaborative filtering system that makes use of UserCorrelation, ItemCorrelation, SlopeOne Predictor, and Clickstream collaborative filtering model to make use of both user navigation patterns along side historical purchased items for users with similar buying behavior. With the proposed dynamic model, we predicted the potential next page (movie title) of interest with higher confidence via the help of clickstream tree. We observed that ItemCorrelation is the fastest recommendation scheme, and SlopOne Predictor is the most accurate scheme. Our dynamic recommendation system based on SOA, orchestrated by BPEL dynamically switches among the schemes to generate most accurate recommendation within a timely fashion in a scalable manner.