Empowering Multi-Criteria Recommender Systems Using Advanced Graph Learners

Multi-criteria (MC) recommender systems enhance recommendation accuracy with an aid of multiple criteria beyond single-criterion ratings. In this study, we survey and analyze recent advancements in MC recommender systems, particularly focusing on representative deep neural network (DNN) models such as AEMC, DMCF, and CPA-LGC. These models improve recommendation accuracy by capturing nonlinear relationships between criteria and overall ratings. However, earlier DNN-based methods like AEMC and DMCF have limitations in fully exploiting complex relations among MC ratings. To address these challenges, CPA-LGC was recently proposed by adopting a graph neural network (GNN)-based approach. In comparison with two earlier methods, we evaluate CPA-LGC's ability to better capture intricate relationships in MC ratings, demonstrating its potential to significantly enhance MC recommendation performance.