Nowadays, Recommender System (RS) has become increasingly popular and essential in daily life. However, existing mainstream approaches consider only the user’s interests and the attributes of the item, ignoring the user’s social connections and the fact that social connections could influence the user’s choices. This can cause the recommended results to have problems called filter bubbles. We argue that Graph Neural Networks(GNNs) are highly suitable for recommender systems since most of the data in recommender systems can be represented as graph structures.
In this paper, we propose DREANRec(Deep Relation Enhanced Attention Networks for Social Recommendation), a novel graph neural network, which effectively incorporates social information among users and considers the heterogeneous strength of social relations and latent item-item relations through the attention mechanism.Extensive experiments were implemented to prove the effectiveness of our approach.