CateReR: A Graph Neural Network-Based Model for Category-Wise Reliability-Aware Recommendation

In recent years of Web, it has been an interesting research topic to recommend items/objects based on the user's choice. Due to the exchange of opinions about items over the social network, trust plays a crucial role in liking or disliking an item. Although the Graph Neural Networks (GNNs), with their natural ability to integrate node information and topological structure, have shown enormous potential in the trust-aware social recommendation, these do not implicitly deal with external factors, such as 'item category', that may have a remarkable impact on user-trust. In this paper, we present a novel approach that project trust as dependent on the category of product. Subsequently, we augment GNN-based social recommendation by defining a concept of category-based user-reliability value. Our proposed graph neural network-based model for category-wise reliability-aware recommendation (CateReR) finds user-embedding and item-embedding with consideration to the variation of user's reliability over different product categories. CateReR is also capable of dealing with trust propagation and trust composition, which are often ignored by existing GNN-based models. We have experimented with CateReR on two real-life datasets to show the usefulness of the model.