Ontology based Distributed Personalized Searching for Peer-to-peer architecture using hierarchical neural networks evidence combination

In this paper, we propose a novel ontology based distributed personalized searching method. The user's information is distributed among the nodes in a peer-to- peer network to reduce the computational costs of the system. The user profile is modeled as a weighted concept hierarchy. We use weighing methods based on the user's surfing pattern to weigh the concepts in the reference ontology. The ontology is partitioned and each node learns a dimension of the user's interest using one partition, all partitions together form the user's profile. The system adapts itself to the changing interests of the user by means of aging. We use hierarchical neural networks to classify the documents into concepts in reference ontology. To overcome the problem of training in cases where insufficient documents are available for a particular concept and to increase the scalability we propose to use two different hierarchical neural network classifiers, each using a different learning function. Their beliefs are combined using Dempster-Shafer theory to eliminate any weaknesses in classification into concepts in the ontology. Our system has an overall performance improvement of around 20%.