MutualRec: Joint friend and item recommendations with mutualistic attentional graph neural networks

摘要

Many social studies and practical cases suggest that people's consumption behaviors and social behaviors are not isolated but interrelated. However, most existing research either predicts users' consumption preference or recommends friends to users without dealing with them simultaneously. In this paper, we propose a novel framework called MutualRec that jointly accomplishes the two tasks based on graph neural networks, attention mechanisms, and mutualistic model. MutualRec first uses a spatial attention layer and a spectral attention layer to learn latent embeddings from observed data, and then merges them via a mutualistic attention layer. The first two layers can relieve data sparsity without violating users' preference sequence, while the last captures the relationship between user' consumption and social behaviors. We demonstrate the effectiveness of MutualRec in both social recommendation and link prediction via extensive experiments.