Research On Recommendation Algorithm Based On Graph Awareness

The advancement of Internet technology has resulted in a surge of available information,leading to the issue of information overload.Predictive models have emerged as a promising solution for filtering through vast amounts of data to locate relevant information.Personalized information filtering tools,such as recommendation systems,have shown the ability to effectively predict users’ information needs in a quick and efficient manner.The application of recommendation systems in addressing the issue of information overload presents both research opportunities and practical implications.The work in this paper is divided into three main parts:First,this study provides a comprehensive summary of recommendation algorithms based on graph neural networks and evaluates the effectiveness of embedding representation methods for learning user and item preferences from their historical behavior data.By treating user-item interactions as a bipartite graph,the recommendation problem is transformed into an edge prediction problem in the graph.Second,this study proposes a recommendation algorithm named NeighborhoodAware Graph Convolutional Network(NA-GCN).The NA-GCN algorithm utilizes graph convolutional networks to enhance the representation learning of the user-item interaction signals,resulting in improved performance.Furthermore,the algorithm incorporates an attention mechanism to dynamically assign weights to user-item interactions based on different semantic information contained in various network layers.This allows for a more accurate modeling of user preferences for different items,resulting in more efficient communication of information.Third,this study proposes a novel recommendation algorithm,which combines residual networks with graph neural networks,named Residual Graph Collaborative Filtering(R-GCF).Traditional graph neural network-based recommendation algorithms alleviate the data sparsity problem by increasing the number of layers.However,the increase in stacked layers may lead to an over-smoothing effect,reducing the uniqueness of users and items and the reliability of recommendations.Furthermore,the higher the number of layers,the more obvious the over-smoothing phenomenon becomes,making it difficult to distinguish the categories of information in higher-level nodes.To address this issue,this study fuses residual networks with graph neural networks to ensure effective information propagation in each network layer,while also ensuring that higherlevel networks have more node feature information.This approach enhances the reliability of recommendations.Experimental results show that the proposed R-GCF recommendation algorithm outperforms mainstream algorithms in terms of accuracy.