Research On Relevance Feedback Based-Conceptual Graphs And Its Implementation

Relevance Feedback (RF) is one of hot technology of information retrieval (PR) . Nowadays, most RF technologies are still designed by methods of keyword matching. In the course of retrieval information, semantic information was seldom included , most RF technologies were mainly based on the system's evaluation of document relevance feedback to the system, nor user's evaluation. So the Recall and the Precision is not better and users are not satisfied with the answer about their query from the IR Model.In this thesis, a solution about Relevance Feedback technology based on Conceptual Graphs (CGSRF) about Chinese was discussed. HowNet (version 2002) was used as the Knowledge Base(KB) and Conceptual Graphs(CGs) was used as Knowledge Representation(KR) about Chinese semantic information. Based on the Search Engineer(SE), A CGSRF system was designed about relevance feedback Model from user's evaluation to back to system and took it as an independency system that is using on the other IR model.The main works in this thesis are as the following.Firstly, extraction the concepts and relations from the information or query, creation the node of CGs and implementation the interface of between CGs and HowNet. Secondly Similarity Computation. Thirdly Semantic Analysis. In the end implementation of the system of Relevance Feedback base on CGs.In this thesis, methods about knowledge representation on Semantic Network were made a comprehensive comparison. A method was discussed for create the concept node and relation node of CGs, implemented for the interface of between Concept Graphs and HowNet. Algorithms were discussed in extracting concepts and relations from searched information or query and built for CGs Base.In this thesis, the information from the IR Model was processed including word segmentation, simplify syntactic parsing and semantic analysis and hand to CGs process.In this thesis, rules were researched about translation-interaction between Chineseand CGS, especially the algorithms in aspect of Chinese tense and voice.In this thesis, methods were discussed about similarity computation. In the Conceptual Graphs, the CGSRF system could easily compute the similarity of concepts, relations and CGs.On the basis of CGs and HowNet, the relevance feedback based on CGs was achieved. In the thesis the modules about the CGSRF system were designed and main technologies and algorithms were described.Testing by two corpuses, the result from the CGSRF could make users satisfied.