Building complete Collaborative Filtering Method System

Abstract

Collaborative filtering (CF) is a key technique in recommender system. Recently, general neighborhood problem existing in collaborative filtering is identified in our previous work, which could result into fatal wrong under multi-community or multi-interest case. In order to overcome it, collaborative filtering based on community (CFC) is presented. Unfortunately, CFC suffers from severer sparsity, which could result into worse performance. Various improved methods are proposed to enhance it. Based on a series of above methods, a complete and hierarchical Collaborative Filtering Method System (CFMS) is build. CFMS extend collaborative filtering, adapting to various different cases. Experiments are made to empirically valuate and compare various methods of CFMS.