Research On Global Entity Relation Extraction In Music

With the rapid growth of information of the Internet, Information Extraction is payed more and more attention, while the Relation Extraction is the most important subtask in Information Extraction. Our study found that traditional Entity Relation Extraction just handle the Entities in the same sentence, which losts a lot of relation. This paper proposes a concept of Global Entity Relation, which means extracting the relation between every two entities in a text and claasify the relation.We start our research in music domain, after detailed statistics and analysis, we found that the relation between two entities can be affected by the other relations among the entities, like the equivalence and the parallel relation. So we can get global entity relation by merge the different relation among entities and can make some simple inference. We first recognize all the Mentions of the Entity in music domain by using the some methods that based on rules, using Dependency Parser, and the integration of the both; Then do Coreference Resolution by using the methods that based on rules, binary classification and the integration of the both; After thatwe apply the kernel methods in Entity Relation Extraction; At last we weil achieve the global relation by merge all the realtions among the entities and so some simple inference. Evaluation results show that our system is 13.8% higher that the traditional system, what's more, our system can be applied to any field. At the end of this paper, we also designed and implemented Corefrence Resolution Platform and Text Mining Platform, they provide better support for research and application in NLP.