Relevance feedback in an information retrieval system based on Bayesian belief networks

People have been using information retrieval systems for decades, however users do not always find the retrieval results to be relevant. The discrepencies between what the systems consider relevant and what the users consider as relevant may not be totally avoided due to the uncertainty involved during the indexing process. Many IR systems adopt relevance feedback as a learning mechanism for understanding users' information needs. This paper discusses the issues of incorporating relevance feedback into an Tp system based on Bayesian networks. Methods for supporting the addition of evidence and for altering dependencies in the model are presented.