Content filtering of decentralized P2P search system based on heterogeneous neural networks ensemble

Abstract

A Peer-to-Peer (P2P) based decentralized personalized information access system called PeerBridge for edge nodes of the Internet network is proposed to provide user-centered, content-sensitive, and high quality information search and discovery service from Web and P2P network timely. The general system architecture, user modeling and content filtering mechanism of Peer-Bridge are discussed in detail. Moreover in order to only find information which users are interested in, a new heterogeneous neural network ensemble (HNNE) classifier is presented for filtering irrelevant information, which combines several component neural networks to accomplish the same filtering task, and improves the generalization performance of a classification system. Performance evaluation in the experiments showed that PeerBridge is effective to search relevant information for individual users, and the filtering effect of the HNNE classifier is better than that of support vector machine, Naïve Bayes, and individual neural network. © Springer-Verlag Berlin Heidelberg 2005.