Recommendation Model of Graph Neural Network Based on Neighbor Aware Interaction

The personalized recommendation system plays a major role in various online services. In recent years, the graph learning based emerging research about recommendation systems has developed rapidly. Some scholars use graph neural network to formulate high-level connection between users and items to improve embedding. The relationship information between neighbor nodes implies the expression differences of diverse neighbors, and it can reflect the signal strength of item characteristics. However, most studies have not considered the difference in implicit relationships in diverse neighbors. In order to distinguish the importance of different neighbors, this paper proposes a neighbor aware interaction graph attention network (NAGAT) for recommending tasks. It uses a novel neighbor aware attention layer to calculate the similarity between each pair of neighbors, the contribution of each pair of neighbors in neighbor interaction is distinguished by allocating distinct neighbor aware attention weight for each neighbor. Then, interaction information of neighbor pairs is combined with the node representation aggregated through the graph attention network to generate a novel embedding representation. Several experimental researches were conducted on two public datasets in Yelp2018 and Amazon-books, and the results show that the proposed is 1.8%-2.0% higher compared to currently advanced methods.