Design And Implementation Of Knowledge Credibility Evaluation Tool For Domain Knowledge Graph

With the concept of " Knowledge Graph" put forward by Google in 2012,more and more people put into the research and application of knowledge graph.Nowadays,people rely on natural language processing and machine learning technology to construct knowledge graph automatically.Compared with manual construction,although automation technology can save a lot of manpower and time,it can also introduce wrong knowledge.There are potential risks in the construction of the knowledge graph,if the wrong knowledge is left unaddressed.Therefore,it is necessary to evaluate the credibility of knowledge in the knowledge graph..At present,there are few studies on the measurement of knowledge credibility in knowledge graph,but the concept of "credibility" has been applied in the related tasks of knowledge graph to some extent.In the knowledge graph completion task,confidence is used to evaluate the accuracy of triples.In the knowledge graph error detection task,the correctness of triples is detected.In order to evaluate the credibility of knowledge in the knowledge graph,a knowledge credibility evaluation scheme for domain knowledge graph is proposed in this paper,and a knowledge credibility evaluation tool is designed and implemented based on the scheme.The evaluation scheme in this paper starts from three dimensions of relationship,attribute value and problem solving respectively.In the evaluation of relationship credibility,the knowledge representation learning algorithm is firstly used to vectorize the entities and relationships in the knowledge graph,then the energy function is used to quantify the degree of relationship credibility,and finally the Sigmoid function is used to calculate the relationship credibility.In the evaluation attribute value credibility,the attribute value of the entity in the knowledge graph is evaluated according to the attribute standard set constructed manually.The attribute value of numerical type is directly compared to get the credibility,while the attribute value of string type is calculated by combining lexical forest similarity and editing distance.In the evaluation of problem solving,a question-answering system based on knowledge graph is constructed firstly,then the question-answering standard set is constructed manually,and finally the question-answering system and the question-answering standard set are used to calculate the reliability of problem solving of knowledge in the knowledge graph.The validity of the aboved method is verified by using the military domain knowledge graph.The experimental results show that the relational credibility evaluation method,attribute value credibility evaluation method and problem solving credibility evaluation method all have high accuracy,precision and recall,which proves that the credibility evaluation method of the three dimensions in this paper can effectively distinguish trustworthy objects from untrustworthy objects.