Research On The Construction Method Of Medical Knowledge Graph And Application In Disease Diagnosis

As the amount of medical data increases,it is of great significance to discover new knowledge from medical entities such as diseases,drugs,treatments and genes,and to mine the hidden knowledge between medical data to assist the diagnosis of diseases.Knowledge graph technology has become an important technical support for knowledge questioning and domain knowledge discovery.Combining medical knowledge to construct medical knowledge map is the driving force for the development of intelligent medical treatment in the future.However,the current medical domain knowledge graph has the problem of poor interpretability and low efficiency.Therefore,according to the particularity and complexity of medical data,it is a problem worth studying to improve the representation ability and reasoning ability of medical knowledge graph to meet the practical medical application demand.In this paper,aiming at the problems of difficulty in automatic knowledge extraction and poor semantic reasoning ability in the construction of medical knowledge graph,medical knowledge extraction,medical knowledge reasoning method and its application in disease diagnosis were studied:1.Aiming at the problems of multiple examples of document-level medical entity relationships and the noise of distant supervision in medical relationship markers,a distant supervision relationship extraction model based on reinforcement learning was proposed,and the method of reinforcement learning was used to improve distant supervision.In order to solve the problem of poor interpretability of rules extracted from the network model of short and long term memory,this thesis add trigger words and position embedding,improve the loss function,and extract rules with strong interpretability.The experimental results show that the proposed model overcomes the problems of poor rule interpretability and noise and achieves an F value of 67.6%,which is better than the baseline method.2.An improved multi-path knowledge inference model is proposed to solve the problem that the existing medical knowledge reasoning model requires a large number of matrix operations and high complexity.Considering the particularity of medical data,the knowledge structure of knowledge graph is expanded by combining triplet relation with path feature.In order to solve the problem of low recall rate in knowledge reasoning,the logical reasoning path is used as the constraint condition,and the representation of entities and relationships is learned from the potential path features to complete the knowledge graph.The results showed that the MRR value reached 65.6% and 60.3%,which were higher than other classical models.3.On the basis of medical entity relationship extraction and knowledge reasoning,a disease diagnosis model based on markov logic network is constructed and applied to the disease diagnosis system.Aiming at the problem that the existing multiple linear regression method has insufficient expansion in disease diagnosis due to the large amount of medical knowledge graph,boltzmann machine was introduced into markov potential function,the expression of potential function was changed,and the conditional probability was deduced by boltzmann machine to calculate the disease probability.The experimental results showed that the DCG value reached 66%,which was higher than the other four diagnostic methods.