MIA

MIA is a side information-enhanced recommender system named after its most innovative component, the Mutual Information-based Alignment metric, as shown in the system architecture. It is designed to better digest side information provided to a recommender system along with the user interaction history.

A thorough comparison to a broad selection of baselines and a careful complexity and convergence rate analysis on MIA and the best performing baseline, KGAT, have shown that MIA is able to achieve state-of-the-art performance with a notable speed improvement.

More details about the algorithm and experiment results can be found in the full paper.