Design And Implementation Of Relation Extraction System Based On Reinforcement Learning

With the rapid development of the Internet,information has explosively increased.Using information extraction technology to automatically extract structured information from text can greatly enhance the efficiency of information acquisition.Relation extraction aims at extracting the semantic relationship among entities from text.It is one of the main tasks of information extraction and has been widely used in the fields of construction of knowledge bases and automatic question answering.However,since the contextual information of some specific relationships is seriously insufficient in text,which leads to ambiguity in relation classification,and then affect the performance of the relation extraction.To solve the above problems,this paper proposes a novel relation extraction model based on a reinforcement learning,which tries to seek the relation evidence in the external information source,and integrates it into the relation extraction model.The main contributions are as follows:(1)Propose a novel model which can overcome the problem of lacking relation context information.The model is separated into two parts.One is the traditional relation extractor to determine the relation of a text between two entities,the other is the relation evidence retrieval component that can obtain the evidence to enrich the relation context.The two parts are successfully integrated into a reinforcement learning framework.(2)In order to verify the effectiveness of the proposed method,this thesis designs some experiments.In the first part of our experiments,we compare the proposed method with other existing relation extraction models.In the second part of experiments,we measure the effect of introduced manual rules and external relation evidence.The experimental results show that our model significantly outperforms other four models.On one hand,our method can effectively integrate deep learning with artificial knowledge.On the other hand,it can utilize external relation evidence to enhance the performance of relation extraction.(3)Design and implement a relation extraction system based on the proposed method in this paper.