基於共分群模型整合內容式與協同式之即時推薦系統

摘要

Recommender systems have become an essential research field because of a high interest from academia and industries. Collaborative filtering (CF), a branch of recommender system, is frequently confronted with the sparsity issue (resulted in fewer records (rating / clicking) against the unknowns that need to be predicted) and “cold start” problem (hard to make prediction for new user and new item), while Content-based (CB) approaches are limited by recommending similar items without user-item click information. Empirically, CF is better than CB, but is helpful to solve cold-start problem. Therefore, many hybrid approaches have been proposed to integrate collaborative filtering and content-based approach.In this thesis, we propose a hybrid approach that combines content-based approach with collaborative filtering under a unified model called co-clustering with augmented matrices (CCAM). CCAM is based on information theoretic co-clustering but further considers augmented matrices like user profile and item description. We then build a collaborative filtering model based on content-based information and co-clustering result to reduce the sparsity problem and solve cold-start problem. Finally, a parallel approach is proposed to solve the scalability problem of large data set.