Research On Personal Relation Extraction Using Chinese Online Resource

Extracting personal relation from massive text data become an important research in the field of relation extraction. As relationship can be used in Searching System, Business Promotion, Intelligence Analysis, etc, this research has been developed widely. Current research methods in personal relation extraction can be categorized as manual definition of relationship type or automatic generation of relationship description. The former is difficult to uniform granularity of type in different methods and to define types comprehensively because of manual intervention, while the latter has the problems of rough generated relationship description and low precision. In order to define relationship type automatically, accurately and comprehensively, this paper used personal relation lists on Chinese online encyclopedia and proposed the method of personal relation extraction using Chinese online resource. The contents researched in this thesis are as follows:1. proposed the method of personal relation extraction using Chinese online resource. This method builds knowledge base with large-scale personal relation information on encyclopedia. Then we can get the relationship type set that are suitable for extracting text and labeled corpus without manual intervention. The experimental results show the method is feasible and has the advantages of automation and applicability.2. proposed the method of personal relation extraction based on combining pattern with similar context matching. This method follows "divide and rule" and can extracts more relation pairs that couldn’t been matched with patterns to improve the recall rate. To compare different methods in the experimental, our method got the best extraction results.3. proposed the method of personal relation extraction based on credible co-occurrence. This paper put forward the calculating formula of confidence to co-occurrence, and filter co-occurrence with lower confidence to improve the precision rate. Experimental results show the improved method is more effective in relation extraction.