Research On Recommendation Based On Knowledge Graph And Graph Neural Network

With the rapid development of Internet technology and industries,the global digital economy has expanded rapidly,and the amount of data has also grown exponentially.With the rapid development of the Internet,a large amount of information is presented to us at the same time.As an important means of information filtering,personalized recommendation system is one of the most effective methods to solve the problem of information overload.However,most of the traditional recommendation methods use explicit feedback or implicit feedback as input,causing serious data sparsity problems.On the other hand,traditional recommendation methods treat each user interaction with an item as an independent action,ignoring the higher-order relationships between items.In order to solve the above problems,it is an effective solution to use the knowledge graph as an auxiliary input of the recommender system.The process of using knowledge graph for recommendation is to connect users,items,or users and items,and further improve the performance of the recommendation system by enhancing the semantic information of the data.Aiming at the problems existing in the existing recommendation algorithms based on knowledge graphs in various scenarios,this paper designs a more efficient and accurate recommendation algorithm to optimize the user experience.The main work and innovations of this paper include the following three aspects:Firstly,in view of the fact that the existing recommendation algorithms cannot accurately describe the importance of users to items from various relationships,this paper proposes a recommendation algorithm based on knowledge graph and graph attention mechanism.In this paper,the representation of the relationship is firstly introduced to model the semantic information of the knowledge graph,and the inner product of the user representation and the relationship representation is used as the attention weight to describe the user's attention to the different relationships of the items.Further,this paper utilizes a graph neural network to receive attention weights and iteratively update item representations.In addition,this paper constructs a subgraph containing limited domain nodes at a specific node,and iteratively aggregates entity representations on the subgraph instead of the global graph.The proposed method not only captures the user's attention weights for various relations of items,but also significantly relieves the computational burden.Secondly,in view of the over-smoothing problem of graph convolutional networks and the homogeneity of user preferences when applied to recommender systems,this paper proposes a recommendation model that combines graph neural networks and label propagation.The label propagation algorithm assists model training to alleviate graph Nodes are over-smoothed,thereby alleviating preference homogeneity in recommender systems and improving recommendation performance.This section first introduces the label propagation algorithm and the graph convolutional network,then studies the theoretical relationship between the label propagation algorithm and the graph convolutional network,and proves the necessity of introducing the graph convolutional network and label propagation in the recommendation algorithm,and finally uses the label propagation algorithm.LPA assists the graph convolutional network GCN to train the edge weights of the knowledge graph.In addition,this paper introduces an attention network to capture the attention weights of user-item pairs.Experimental results show that the proposed GCNLP model outperforms existing state-of-the-art models.Thirdly,existing recommendation methods based on graph convolutional networks face the following problems: most aggregators ignore the strong synergistic signals in knowledge graphs and fail to extract valuable features;existing Recommendation methods based on graph convolutional networks do not consider the importance of different neighborhood layers,which negatively affects recommendation performance.To address the above problems,this paper proposes a novel recommendation model named FMA-GCN.In this paper,a fixed-size receptive field is first introduced into the knowledge graph to control the amount of computation.Next,this paper elaborates a simplified version of neighbor aggregator to capture strong cooperative signals in knowledge graphs.Finally,this paper designs a layer-oriented message measurement mechanism to measure the information of each network layer in a finer granularity to form the final representation.This paper conducts a large number of comparative experiments and ablation experiments on three open source datasets to demonstrate the effectiveness of the FMA-GCN algorithm.Experimental results show that the algorithm outperforms the state-of-the-art algorithms.