Multi-criteria Dual-View Attention Network for Rating Prediction

There have always been recommender systems the majority of the researches predicting uses single ratings. Where on the other hand, Multi-criteria predictions are more accurate. In recommender systems, collaborative filtering is a well-known strategy. Deep learning has produced excellent results in various fields, including computer vision, voice, and text. Deep learning at recommender systems recently gained wide attention, and many recommendation models have been published. However, based on our knowledge, no research has been conducted to collect multi-criteria recommendations and collaborative filtering with a dual view attention network. This paper proposes a multi-criteria Dual-View Attention Network (MDVAN) to predict the multi-criteria ratings for items as they progress from coarse to multi-criteria fine attention level. We explore the dual-view model to fully use the User View and Item View advantages hidden in historical rating behaviors with the multi-criteria problem. Then create a ReLU universal pair-wise channel unit to model the relation of two multi-criteria domains. We implement an aggregation function to incorporate the multi-criteria items in the dual view model. It was shown through experimentation on a real-world dataset that we had exceeded all of the current approaches. And that proof dual-view attention network in multi-criteria recommendation systems achieved great success.