Research On Knowledge Reasoning Method In Chinese Knowledge Graph Construction

With the development of Internet and artificial intelligence technology,more attention has been paid to knowledge and the relationships among them.The Internet has transformed from a hyperlinked document Web to data World Wide Web,which contains a large number of relationships between entities,and it also can be a semantic network which holds the rich relationship between entities,called knowledge graph,which is regarded as the kernel of search engine of the future.Many companies i.e.Google,Baidu and Sogou et al.,all have launched research of the knowledge graph to advance the quality of their search.The current mainstream knowledge graphs,i.e.Freebase,DBpedia and NELL et al.,have contained hundreds of millions of factual data,but they also have deficiencies.There are many potentially unknown relationships between entities in the process of automatically constructing the knowledge graph,but not fully identified.Therefore,based on the knowledge inference algorithm to automatically deduce the relationship between the facts to enrich the knowledge graph,construct a relatively completed knowledge graph,which is significant in semantic search,Artificial intelligence and decision-making in business.In this paper,aiming at the problem of path connectivity problem and computational cost in the present knowledge-based inference algorithm,an optimization method has been proposed.The main contributions of this paper are:(1)based on the analysis of the present mainstream situation of knowledge inference at home and abroad,comprehensively introduce the key technology involved in the construction of knowledge graph and knowledge inference,and also point out the existing problems and challenges in the current inference methods.(2)an optimization method based on the feature word is presented for path completion.In the process of Random Walk,the paths between entities are not missing due to the lack of relationships in the knowledge graph,so that the effective path cannot be obtained.This paper makes full use of the description text of the entity and TF-IDF algorithm to extract the feature word set of the entity to supplement the path missing information,which solves the problem of path connectivity in high-efficiency.The experimental results show that this method improves the accuracy of inference model.(3)an method of using the hypernyms to divide the path constraint based random walk pruning path in entity domain is proposed.Considering the number of paths generated by the edge combination is very large,which causes the number of paths to grow exponentially in each step.To reduce the order of the path search and obtain the effective path,the path discovery strategy is improved.The method constructs hypernym table in the hierarchical structure which is based on the encyclopedia classification system,and trains the word vector through Word2 vec tool to calculate the similarity of the string and the the semantic to obtain,which can divide the entities in the knowledge graph by domains.Through only the relevant areas of the entity nodes to walk,the number of path types under conditional constraints is greatly reduced.According to the experimental results,this strategy improves the computation efficiency.