Research On Semantic Inference Based Knowledge Similarity And Contradiction Detection

Development of large knowledge base becomes trend nowadays. However, knowledge base faces knowledge redundancy and contradiction problem when inputting new knowledge. This thesis proposes a semantic inference based knowledge redundancy and contradiction detection method. Recognizing Textual Entailment (RTE) is one of the major problems in semantic inference. The thesis designs a word alignment based RTE method for the reason that word alignment is the key step in human judging process. This method uses several knowledge bases as semantic resource. It combines machine learning with rule-based judgment to apply inference between two sentences. Then the thesis focuses on how to incorporate RTE with knowledge redundancy and contradiction detection. As a result of large amount of data in knowledge base, inference between sentences will cost too much time. Therefore, this thesis applies information retrieval technology to reduce the number of sentences in order to reduce time. By analyzing the characteristics of similar knowledge, it proposes concept of textual locality, which is used for redundant knowledge detection. For the common contradict knowledge, it use rule-based judgment method.