This paper presents a novel approach from a perspective of considering community structures to collaborative filtering. In our approach, multiple types of information are be explored and exploited, including item content, user demography, use-item ratings, use-item structure and user social information. Leveraging the types of information, we apply multiple techniques from data mining, including multi-relational data mining and graph data mining, to explicitly discovery user community structures, which in turn are used in collaborative filtering. Initial experimental results indicate that this community-based approach can significantly improve the effectiveness of a collaborative filtering system when sparsity and synonym are the issues.
展开▼
