Heterogeneous Trust-based Social Recommendation via Reliable and Informative Motif-based Attention
Supriyo Mandal, Abyayananda Maiti · 2022 International Joint Conference on Neural Networks (IJCNN) · 2022
Recently, social recommender systems are promising to enhance the quality of recommendations by integrating user-user social networks and user-item bipartite networks. However, there are very little success in that direction. Pessimistic findings are ascribed mainly to three factors. (1) Very few works focus on the identification of implicit neighbors to overcome the sparsity problem of explicit links. Furthermore, these works do not con-sider higher-order and complex patterns of interactivity among users. (2) Very less number of trust-based social recommender systems integrate heterogeneous trust relationships, and this het-erogeneity is considered for explicit social links only. Moreover, these works ignore user-user heterogeneous trust relationships of higher-order network structure and user-item heterogeneous interactivity. (3) Existing works overlook the reliability (or lack of that) problem of links in higher-order and complex patterns of interactivity. To address the above mentioned challenges, we develop, a Graph$\bar{C}$onvolutional$\bar{N}$etworks via$\bar{R}$eliable and$\bar{I}$nformative$\bar{M}$otif-based Attention Model (CNRIM). To the best of our knowledge, it is the first work that investigates user-user heterogeneous trust relationships and user-item heterogeneous interactivity via reliable, informative motif-based attention mech-anisms. Varying reliability and informative motifs introduce the heterogeneity. The experiments on publicly available real-world datasets, and empirical analyses present the superiority of our model over popular baselines.