Temporal Gate-Attention Network for Meta-path based Explainable Recommendation

Heterogeneous Information Network (HIN) has garnered significant attention in the field of explainable recommendation based on paths due to its rich semantic information. However, many existing explainable recommendation systems solely focus on the connection information between nodes in a HIN while neglecting the temporal information between nodes. This leads to reduced accuracy and explainability of recommendation. For example, in a HIN of users and merchandises, many existing methods only consider the purchased merchandises by users, whereas our approach takes into account the temporal order of users' purchases, resulting in the creation of a new HIN. Additionally, we consider the possibility of unrelatedness between the previous and next merchandise in the sequence of users' purchases, which is a factor overlooked by current methods. Therefore, we propose the Temporal Gate-Attention Network for Meta-path based Explainable Recommendation(TGAMER). Our model utilizes gate recurrent unit(GRU) to capture the temporal information between items. Furthermore, it introduce a novel gate-attention mechanism to enhance feature aggregation among similar items while eliminating it among dissimilar ones. Extensive experiments conducted on three real-world datasets provide compelling evidence of the effectiveness of our model in terms of recommendation performance.