Application of Extended Belief Network Model for Scientific Document Retrieval

This work presents an extended belief network model to deal with the problem of searching for scientific document. The original belief network model is extended by combine reference evidence. An algorithm is introduced to compute the two metrics: worth and concern which are used to measure the important degree of scientific document based on the reference evidence. We discuss a strategy of the extended belief network model, and experiment with it using a Chinese document collection from the CNKI. The results show that this combination can improve the retrieval performance without requiring any extra information from the user query. In our experiment, the improvements reach up to 11% in terms of average precision.