Graph Neural Network Enhancing Recommendation System based on Social Relationship and Attention Mechanism

Traditional recommendation methods generally only take into account the information of entity objects, such as less consideration of social relationships, and generates the problem of insufficient capacity. Because the actual user-object interaction is non-Euclidean data, traditional recommendation algorithms are difficult to deal with this structural type of data, and it is difficult to achieve excellent recommendation results, so the traditional recommendation methods generally take into account the information of the user and the object itself, which makes the topological structure of the composition between the user-object information is lost, and the recommendation effect is limited. In this paper, on the basis of the variogram self-encoder that integrates the attention mechanism, we construct the user-item graph neural network and the user social graph neural network respectively, and improve the recommendation effect through the auxiliary social relationship network when making product recommendation, and propose a variogram self-encoder recommendation model that integrates the social relationship and the attention mechanism, which can fill in information of the target node with the auxiliary information of the user-item graph and the user-item graph neural network when the data is sparse. The model can fill in the information of the target node with assistance when the data is sparse, avoiding the problem that the data size has an excessive impact on the prediction effect, and improving the recommendation effect by introducing social relations and using different mapping relations between user-user and user-item at the same time.