Extended SBN Retrieval Model Based On Ontology Terms Relationship

The Bayesian Network Retrieval Model is one of the probability models in information retrieval. By extending this retrieval model with reasonable relations of terms, the retrieval function may be enhanced effectively. The ontology refers to the formalized specification of a shared conceptual model which possesses both the conceptual layered structure and the logical reasoning function. Based on ontology, the relationship among terms can be accurately obtained.This thesis first extends the double-layer structure of SBN (Sample Bayesian Network) model, comprising 1 single term layer and 1 document layer, into a three-layer structure comprising 2 term layers and 1 document layer; and then obtains the relationship between 2 term layers through the ontology and calculates the relative degree between the terms of both layers by means of ontology-based method; and finally gives the probability estimate of all the layer nodes of the extended model as well as the reasoning mechanism of the retrieval model. In the experiment, this thesis first of all builds up 5 instances of ontology with different themes by skeletal methodology, each containing 10 to 20 terms; and then obtains the relative degree among all the terms by ontology-based calculating method for term relative degree. After that it takes the small Chinese test collections as the testing data and picks 5 searching subjects for the original SNB retrieval model and the expanded model as well; and finally gets recall and precision of both retrieval results by interpolation method and analyzes the data of each step of the expanded model. The experiment result shows that in comparison with the original SBN retrieval model, extended SBN retrieval model based on ontology term relationships boasts better retrieval function.