Multi-dimensional Graph Neural Network for Sequential Recommendation.

• We design a multi-dimensional information integrated graph neural network, which integrates the category and time information into the graph embedding process, so as to obtain more fine-grained item representations. • Considering the lack of time sensitivity for the traditional self-attention mechanism, we integrate interaction time information into the self-attention mechanism, so as to capture the user's dynamic preferences more accurately. • We conduct extensive experiments on three real-world datasets. Our experiment results demonstrate the superior performance of our proposed method comparing with other state-of-the-art baselines. Graph neural networks (GNNs) technology has been widely used in recommendation systems because most information in recommendation systems has a graph structure in nature, and GNNs have advantages in graph representation learning. In sequential recommendation, the relationships between interacting items can be constructed as an isomorphic graph, and (GNNs) can capture high-order information between graph nodes. Many models have used graph-based methods for sequential recommendation, and achieved great success. However, the existing research only considers the number of interactions between items when constructing the item graph. As such, revisions are needed to capture the multi-dimensional transformation relationships between items. Hence, we emphasize the importance of multi-dimensional information, and we propose a C ategory and T ime information integrated G raph N eural N etwork (CTGNN), which combines the item category and interaction time information with a multi-layer graph convolution network to form multi-dimensional fine-grained item representations. In addition, we design a temporal self-attention network to model the dynamic user preference and make the next-item recommendation. Finally, we conduct extensive experiments on three real-world datasets, and the results demonstrate the excellent performance of the proposed model. [ABSTRACT FROM AUTHOR]