The Design And Implementation Of Building Knowledge Graph System Based On Information Extraction

With the development of Internet Technology,the total amount of data generated in daily life is increasing rapidly,and people have higher demands on the speed of information acquisition and the efficiency of search.However,the original knowledge semantic network can not represent the relationship between nodes clearly,it is difficult to find the target content quickly.In 2012,Google proposed a representation model of Knowledge Graph,mapping subject and object entities in the information to nodes and the relationships to edges in the KG.It not only adapt to different fields,but also help people to find out the relevant information quickly,so building Knowledge Graph plays a positive role for the dissemination of knowledge.For the building of Knowledge Graph,the extraction of the triple is the core part,so this paper proposes the automatic extraction model of the triple and the implementation scheme of the system,based on the investigating of the existing triple extraction and the design of KG building system,the main contribution is as follows:1.In the triple extraction task of joint entity and relation,the extraction model based on the weighted pointer network is proposed,the dilated convolution kernel of different scales is used in the basic structure of the CNN,the duplicate entities are filtered by using weighted pointer network,the multiple groups of entities in different length are labeled from the overall point of view.The model is tested on SKE and CHIP datasets,the results show that the model is about 1.8%higher than the current mainstream BI-LSTM+Attention model.2.In the triple extraction task of event subject,the extraction model based on the multi-head attention mechanism is proposed,the multi-head attention mechanism is integrated into the Bi-LSTM network,to obtain the different semantic information of words,the core event words positioned by embedding the trigger word feature?The model of this paper is verified in CCKS and iFLYTEK datasets,and results show that the model gives more attention to the core entities,it is about 1.5%higher than the mainstream structure of CNN+CRF.3.This paper provides the overall structure of building Knowledge Graph system,designs and implements comprehensive solution such as uploading text,extracting triple,generating Knowledge Graph and managing Knowledge Graph,uses Flask as the back-end framework,takes Neo4j and MySQL as the database management system.After testing the system,it is confirmed that the service in the system has high feasibility and stability,it can be oriented to a variety of industries and users with different knowledge levels.