Personae Entity Relations Extraction And Analysis In Chinese Microblog Text

As the development of internet, many convenient social network platforms have appeared. For example, Twitter, Face Book, Sina microblog and tencent microblog. This reason makes social network services like microblog very popular, and microblog text showing a trend of explosive growth. It is significance for discovering relations and behaviors between personae. At present, the researches of person features in social network and social relations have become hot topics. At present, many relation extraction methods can't work well at microblog corpus because of short text in microblog. For solving this problem, the research of this thesis shown as follows:(1) According to microblog text features, we improved dependency trigram kernel. The improvements include two parts, one is words semantic similarity and another one is words syntax similarity. In this thesis use How Net which is a words semantic similarity tool to increase dependency trigram sets words semantic similarity. And then the thesis proposes “POS-GR(Part of Speech-Grammatical role)” two-tuples to improve dependency trigram sets words syntax similarity. The improved dependency trigram kernel used in basic level for extracting personae interactive relations(IR).(2) After obtain IR, the thesis find that relations are too complicated to apply IR for constructing knowledge graph. For solve this problem, this thesis consider the IR as input sets of meta-learning level. According to IR, the thesis proposes seven relations describe words extraction rules and relation key words classification algorithm(RKWCA). The IR is classified as 4 types of social relations, such as friend, business, family and enemy.(3) This thesis evaluated the validity of above methods, and extracted IR and social relations in microblog. The thesis also constructing visualization knowledge graph by using person entities as nodes, relations as edges and relation key words as attributes.In this thesis, used syntax dependency tree for solving short text problem, and utilize dependency trigram kernel to find person relations features. The thesis utilizes meta-learning strategy to construct visualization knowledge graph. Experiment proved the thesis possess practical and theoretical significance.