Research Of News Recommendation Technology Based On Knowledge Graph

Nowadays,the recommendation system can help people find the information they are interested in from the vast amount of information,so that the pressure of users to find information is alleviated.Therefore,many experts and scholars have carried out research on the recommendation system and proposed many solutions for the recommendation system.However,the existing recommendation systems generally recommend results by using user's history click record or the user's rating of the item,ignoring a lot of useful information for the recommendation,such as user item interaction information,common sense and so on.The structural knowledge in the knowledge map can provide additional information for the recommendation system and assist with the knowledge level in the recommendation.Therefore,the introduction of the knowledge map into the recommendation system has certain research value.Based on the two methods which introduce knowledge graph into the recommendation system:sequential learning and joint learning,A method for dynamically updating the feature vector is proposed.Based on the feature vector of entities obtained by the knowledge graph,this method takes into account the user's history browsing data in the recommendation,simulates the spread of user interest on the knowledge graph,and adjusts the feature vector of entities appropriately,which makes it better meet the requirement of the recommendation.According to the characteristics of the news,a recommendation method is proposed to fully exploit the semantic level and knowledge level information in the news headlines.On this basis,the news recommendation model is constructed.In this paper,through the experiment,it verifies the effect of the knowledge map,the validity of the dynamic update of the entity feature vector and the influence of the LSTM on the semantic mining.The influence of eigenvector on recommendation model under different knowledge representation models is also listed.Finally,the results compared with other recommendation models show that the news recommendation system based on knowledge graph has certain practical value.