Research And Implementation Of Knowledge Graph-oriented Attribute Value Extraction And Verification Technology

After the concept of Knowledge Graph was proposed by Google in 2012,this form of knowledge representation has attracted strong attention from academia and industry,and scholars regard it as the search engine of the future.The construction of a high-quality knowledge graph is a dynamic development process.It is necessary to add new knowledge in time to improve the knowledge graphs,and to verify the time-sensitive knowledge in the update graphs.New knowledge can be supplemented by information extraction technology oriented to network text,and relevant attribute value verification technology can be used to verify the timeliness knowledge in the updated graphs.However,traditional techniques are difficult to apply to most web texts for constructing knowledge graphs,which leads to a difficult problem in constructing high-quality knowledge graphs.In order to solve the problem,this paper studies the attribute value extraction and attribute value verification technology oriented to the knowledge graph.The main research work is as follows:1.Study oriented knowledge graph attribute value extracting technology,this paper puts forward the technique of extraction based on regular matching degree of attribute value,regular matching degree is refers to the regular expression and to extract text matching degree of attribute's value,the study technology can choose from regular expression set out the most suitable for a particular text extraction attribute value of regular expressions,to accurately extract the attribute value information.In addition,for the attribute value of "place name" or "institution name",the attribute value extraction technology based on named entity recognition tool is proposed,and the attribute value of this type can be quickly obtained from the text type by using named entity recognition function.Through comparative experimental analysis,it is proved that the accuracy of the two proposed techniques in attribute value extraction is better than that based on the regular expression set extraction technique.2.Aiming at the research of attribute value verification technology oriented to knowledge graph,this paper proposes an attribute value verification model based on hybrid manual verification,and uses the expectation maximization algorithm to select a small part of data with rich features for manual accurate verification.The verified data are used to assist the training of high-quality truth evaluators,and the effectiveness of the model to verify the authenticity of the whole data set is eventually improved.The experimental results show that,with the increase of data scale,the proposed verification technology is more accurate than the existing highquality verification technology and its simplified version,and the running time is between the existing high-quality verification technology and its simplified version.Although the running time of the research technology is slightly increased compared to the simplified version of the existing high-quality verification technology,the slight increase in the time cost is acceptable because the research technology improves the verification accuracy compared to the existing technology.3.Based on the attribute value extraction and verification technology studied in this paper,a high-quality knowledge graph construction and generation system is designed and implemented,and each module of the system is described and tested.Through the display of the interface of the search example,the expected goal of the system design is basically realized.