MGAT: Multimodal Graph Attention Network for Recommendation

Highlights：

• We develop a new method MGAT, which incorporates attention mechanism into the graph neural network framework, to disentangle user preferences on different modalities.

• Technically, the model introduces the gated attention mechanism to control and weight the information flow in multimodal interaction graphs, which facilitates the understanding of user behaviors.

• We perform extensive experiments on two datasets to verify the rationality and effectiveness of MGAT. Moreover, because of user privacy, only user IDs are considered in this work. We will release the code and parameter settings upon acceptance.