Temporal Pattern Attention Graph Convolutional Network for Recommendation

Recommendation system algorithms implemented based on graph convolutional networks have been widely applied. However, despite some researches having expanded the data used by models from the most basic user-item interaction data to other types of data, such as incorporating users' social friends and item knowledge graphs as side information, user historical behavior has not been taken into account. To further enhance the performance of recommendations, this paper proposes a graph convolutional network that incorporates temporal pattern attention for sequential recommendation, thereby capturing the temporal patterns in users' historical behaviors. The model's performance on three real-world public datasets outperforms other baselines, strongly demonstrating its effectiveness in recommendation.