Addressing Interest Diversity in P2P Based Collaborative Spam Filtering

Collaborative information filtering tends to be a promising technology in the fight again Internet spam, and the peer-to-peer framework is believed to be more suitable to implement this collaboration compared to centralized ones. However, the assumption of uniform information interests across peers in collaborative filtering limits the improvement in filtering accuracy and increases network traffic unnecessarily. To address this problem, we propose to construct multiple interest groups for a peer, each corresponding to one message category; and to use a percolation search algorithm to retrieve message feedbacks in scale-free email networks. Experiments show the proposed model can get remarkable improvement in false positive reduction and good performance in spam feedback retrieval