Research On The Solution Of Data Sparsity Problem Of Recommendation System Based On Knowledge Graph

With the rise of online content platforms and e-commerce,the data on the Internet has begun to explode,and Internet users are suffering from serious information overload problems.As a tool to solve the problem of information overload,recommendation system is one of the most important technology.Recommendation systems also suffer from serious data sparsity problems.Various types of side information have been used to solve the data sparsity problem of recommendation systems.As a semantic network with a heterogeneous graph structure with rich information,knowledge graphs have been widely used in recommendation systems in recent years.The thesis firstly introduces the related technologies of the sparsity of recommendation systems and knowledge graph-based recommendation systems based on knowledge graphs,including why traditional methods suffer from data sparsity,how previous work uses side information to solve data sparsity,and how to use the knowledge graph and how to solve the data sparsity problem of group recommendation problem in previous work.Then,the problems existing in previous work are analyzed combined with the characteristics of group recommendation problem.Then,this thesis proposes an end-to-end knowledge graph-based attention group recommendation model.The model first uses the structure of the graph convolutional network to spread the information of entities in the knowledge graph and learns the representation of users and items.Then the user's influence within the group is learned adaptively by comprehensively considering the relationship between the target item and the users in the group through the attention mechanism to obtain the representation of the group.Finally,this thesis uses a margin loss function and additional user-item interaction information to make the model learn better.Finally,this thesis validates the proposed method by conducting performance comparison and ablation experiments.Then,this thesis studies the impact of hyperparameters of the proposed method.Finally,an explanation is given by show a real case.