Research And Implementation Of Content Recommendation Algorithm Based On Knowledge Graph

Knowledge graph is often used in intelligent search engine,recommendation system and other fields to improve its accuracy because of its strong organizational ability and relationship processing ability.At present,the recommendation system based on knowledge graph is mainly divided into feature-based recommendation and path-based recommendation.The feature-based recommendation algorithm use knowledge graph to directly knowledge representation,but fails to introduce multi-hop relation,it is difficult to use semantic network information of knowledge.The path-based recommendation utilize multi-hop knowledge of knowledge graph to effectively utilize semantic network information but usually relies on prior knowledge.Therefore,it's of great research value to carry out the research of application of knowledge graph recommendation system aiming at the problem of how to effectively utilize the semantic correlation information of knowledge by recommendation algorithm.Aiming at the problem of combining knowledge feature representation and recommendation algorithm in the application of knowledge graph in recommendation system,this thesis studies the recommendation system based on knowledge graph,and designing a knowledge representation learning model based on self-attention,which use the entity relation to learn the entity relation semantic information,so as to achieve the high quality representation of knowledge characteristics and bring more and more useful information for recommendation.A content recommendation model with behavioral knowledge characteristics embedded in it is constructed,and the dynamic learning of knowledge characteristics based on users' historical preferences and knowledge graph can bring more accurate and diverse recommendations.The main contributions of this thesis are as follows:1.Proposing a learning knowledge representation model based on the attention to solve the knowledge representation problems,around the triple to determine entities to the importance of the semantic differences,using the attention mechanism to learn from the triple semantics,in order to improve the knowledge characteristics of quality,providing more high quality information for recommendation system.The performance is demonstrated by link prediction and triad classification experiments,and the feasibility of the method is proved.2.A content recommendation algorithm with unified embedding of behavioral knowledge features is proposed,which use historical preferences and semantic relationship structure of knowledge graphs to deeply explore users' interests and hobbies.By clicking the prediction experiment,the ability of the model to dynamically learn semantic relevance information and dig preferences in depth is demonstrated,and the effectiveness of the algorithm is proved.3.Built a recommendation system based on knowledge graph,designed and implemented a news recommendation system based on the application of news recommendation,and realized the function of recommending news preferences to users based on the previous two contributions,and demonstrated the effect of the system.