Research On The Method Of Domain Knowledge Graph Updating And Knowledge Recommendation

When the insured want to know social insurance-related information in real time,such as the latest policies and regulations,news,cases,etc.,traditional information retrieval cannot satisfy people's need for fast and accurate knowledge acquisition.Therefore,research on social insurance knowledge recommendation methods can make each user able to get targeted recommendations.However,the traditional recommendation method will produce cold start problem and information sparse problem.Hence,this paper introduces social insurance knowledge graph as auxiliary information to help knowledge recommendation.In order to maintain the freshness and completeness of knowledge recommendation,it is necessary to improve the freshness of knowledge graph.Thus,it is important to study the automatic updating method of knowledge graph before recommending knowledge in social insurance catagory.Based on the above background,this paper studies the automatic updating method of knowledge graph in social insurance catagory firstly,and continues to study the knowledge recommendation method based on the knowledge graph.In the automatic updating method of the social insurance knowledge graph,the missing information is complemented by the existing information of the knowledge graph to achieve internal update,and the seed entity is discovered according to the popular articles in the web and the entity is expanded to achieve external update.For the internal update,this paper proposes a Bi-GRU-based social insurance domain representation method,which uses the loss function of the translation model and the entity negative sampling method to train the completion model,and finally completes the extracted information into the knowledge graph through the entity link prediction.For the external updating,this paper explores the popular social insurance entity as the seed entity from the web and finds the expand entity,and proposes that the entity update frequency predictor sets the priority of the expand entity,and updates the expand entity according to the priority.In the part of knowledge recommendation,this paper proposes the knowledge graph embedding method firstly,linking the entities in the articles that the user used to interest to the knowledge graph,and using the context of the entity in the knowledge graph as the extended information to rich representation.The sentence embedding method that mixes domainknowledge was proposed to represent the title of the user history interest article and the candidate article,and uses the attention mechanism to give different weights to the user's historical interest.Finally,computing and sorting the probability that the user is interested in the candidate article.Recommending candidate articles to users by the probability.Finally,this paper experiments and analyzes the methods and models proposed or adopted above to verify the validity and accuracy of the proposed algorithm and model,and illustrates the effectiveness and application value of the automatic updating method of knowledge graph and the knowledge recommendation method in social insurance domain.