Research On Knowledge Graph Reasoning And Medical Application

With powerful semantic representation and data organization ability,knowledge graph represents knowledge in the form which is easy for computer to understand,and provides a new feasible scheme for information retrieval.Keyword retrieval technique needs to manually select useful information from massive web pages,while knowledge question-answering relies on knowledge graph as the data foundation to quickly respond to people’s desired answers..Knowledge reasoning can discover the hidden knowledge in knowledge graph,and applying knowledge reasoning to knowledge question-answering can make the feedback answer more accurate and complete.However,materialized reasoning is difficult to satisfy the expanding knowledge base.Therefore,how to improve efficiency while ensuring the completeness of reasoning has become a hot topic in the industry.The thesis focuses on two key problems,including modular reasoning of knowledge graph and its application in knowledge question-answering.Firstly,the thesis studies the modular reasoning of knowledge graph.Standard inference engines load and calculate the knowledge graph as a whole.The scale of knowledge graph is increasing now while the computing and storage resources are limited,the existing reasoning systems cannot meet the performance requirements of reasoning.The thesis presents a modular knowledge graph reasoning method.The facts of knowledge graph are divided by predicate type and entity.According to the background knowledge involved in fact module,ontology module satisfying reasoning requirement is extracted from the pattern constraint of knowledge graph.On the basis of the reasoning requirement,ontology module is partially loaded for knowledge graph reasoning.Experiments show that the method proposed in the thesis can improve the time and memory consumption of reasoning significantly on the condition of ensuring the completeness and reliability of reasoning.Secondly,the thesis studies the application of modular reasoning in the medical field.The medical domain is a vertical domain of the knowledge graph,and the rich knowledge structures in the medical domain knowledge graph can provide a base for knowledge question-answering.The lack of openness in the existing Chinese medical knowledge graph limits the research and development of knowledge graph questionanswering in the Chinese medical field.Moreover,the existing Chinese medical knowledge graph lacks fine-grained medical knowledge graph for specific diseases.The thesis focuses on cardiovascular diseases,and the bottom up approach is used to construct a Chinese cardiovascular diseases knowledge graph,which includes 352 diseases,793 drugs and 42863 triplets.Then the knowledge graph is used as the underlying data,the knowledge graph modular reasoning method is integrated to build a Chinese cardiovascular disease knowledge graph question-answering framework supported by reasoning.Experiments show that applying reasoning to knowledge question-answering can make the answers more comprehensive.The research of the thesis provides a feasible idea and method for the modular reasoning of knowledge graph and its application in the medical field.It has important theoretical and practical significance for the further research of modular reasoning of knowledge graph and the application of knowledge reasoning in the medical field.