The Extended Belief Network Retrieval Model Based On Reference Relationship Of Scientific Literatures

The belief network model is a retrieval model based on Bayesian network whichprovides a framework for combining different evidence information, through combine helpfulevidences the retrieval effectiveness of the model will be improved.In the library of scientific literature, each literature affect each other, it mainly shows asreference relationship between literatures. The value of reference relationship in informationretrieval is it can represents the important degree of literature, which can use attention andvalue of the two quantitative indicators to represent. The influencing factors of attention arecited counts, the importance of literature which is cited by, the topic closeness betweencitation and cited document. The influencing factors of value are citation counts, theimportance of citation document, the topic closeness between citation and cited document.In this paper, we use attention and value as new evidences to extend the belief networkmodel. Considering document content, attention and value have different contributions toquery result, we introduced two parameters α and β to adjust the three evidences. Wegive two combing method of disjunctive (or) and conjunctive (and) to extend the basicmodel. In our experiments, we determined the optimal value of α and β throughinduction method, the disjunctive (or) method has better retrieval performance thanconjunctive (and) method and basic model under the reasonable parameters.