Abstract:Recommender systems based on graph neural networks receive increasing research interest due to their excellent ability to learn a variety of side information including social networks. However, previous works usually focus on modeling users, not much attention is paid to items. Moreover, the possible changes in the attraction of items over time, which is like the dynamic interest of users are rarely considered, and neither do the correlations among items. To overcome these limitations, this paper proposes graph neural networks with dynamic and static representations for social recommendation (GNN-DSR), which considers both dynamic and static representations of users and items and incorporates their relational influence. GNN-DSR models the short-term dynamic and long-term static interactional representations of the user's interest and the item's attraction, respectively. Furthermore, the attention mechanism is used to aggregate the social influence of users on the target user and the correlative items' influence on a given item. The final latent factors of user and item are combined to make a prediction. Experiments on three real-world recommender system datasets validate the effectiveness of GNN-DSR.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the limitations of existing graph - neural - network - based recommendation systems in dealing with users and items. Specifically, most previous studies mainly focused on modeling user behavior and paid less attention to item characteristics. In addition, there was also insufficient consideration for the time - varying attractiveness of items and the correlations between items. These problems led to the lack of accuracy and timeliness in the recommendation system. To solve the above problems, the author proposed a graph neural network model combining dynamic and static representations for social recommendation (GNN - DSR). This model not only considered the dynamic and static features of users and items but also introduced the influence of relationships between them, including the social influence of users and the correlation influence between items. ### Summary of the core problems in the paper: 1. **Imbalance between user behavior and item characteristics**: Most previous studies focused on user - behavior modeling and paid less attention to item modeling. 2. **Temporal change in item attractiveness**: The change in item attractiveness over time was ignored, which is similar to the dynamic interests of users. 3. **Correlation between items**: The inter - item correlations were not fully considered, which may lead to inaccurate recommendation results. 4. **Social influence and correlation influence**: Existing methods failed to fully integrate the influence of user social networks and item - association networks. ### Overview of the solution: GNN - DSR solves these problems in the following ways: - **Dynamic and static representations**: Recurrent neural networks (RNNs) are used to capture short - term dynamic representations, and the graph attention mechanism is utilized to model long - term static representations. - **Relational graph aggregation**: The relationship influences from user - user graphs and item - item graphs are aggregated through the graph attention mechanism. - **Multilayer perceptron (MLP)**: The interaction representations and relationship influences are combined for score prediction. ### Formula presentation: 1. **Similarity calculation**: \[ \text{sim}(j, k)=\frac{R(·, j)^T R(·, k)}{\|R(·, j)\|_2 \|R(·, k)\|_2} \] 2. **LSTM dynamic representation**: \[ h^S_i = \text{LSTM}(X(i)), \quad h^S_j=\text{LSTM}(Y(j)) \] 3. **Attention mechanism static representation**: \[ h^L_i=\sigma(W^{uv}_0 \sum_{j \in N^{uv}_i} \alpha_{ij} x_{i \leftarrow j}+b^{uv}_0) \] \[ h^L_j=\sigma(W^{vu}_0 \sum_{i \in N^{vu}_j} \alpha_{ji} y_{j \leftarrow i}+b^{vu}_0) \] 4. **Interaction representation**: \[ h^I_i = h^S_i\odot h^L_i, \quad h^A_j = h^S_j\odot h^L_j \] 5. **Output layer prediction**: \[ \hat{r}_{ij}=g_{\text{output}}([h^u_i, h^v_j]) \] Through these improvements, GNN - DSR can more comprehensively capture the dynamic and static features of users and items and their relationship influences, thereby improving the performance of the recommendation system.

Graph Neural Networks with Dynamic and Static Representations for Social Recommendation

Self-Propagation Graph Neural Network for Recommendation

A Graph Neural Network Framework for Social Recommendations

Dynamic and Static Features-Aware Recommendation with Graph Neural Networks

Exploiting Structural and Temporal Influence for Dynamic Social-Aware Recommendation

Graph-based Dynamic Preference Modeling for Personalized Recommendation

Graph neural networks for preference social recommendation

Graph Attention Network with High-Order Neighbor Information Propagation for Social Recommendation

Multi-dimensional Graph Neural Network for Sequential Recommendation

Using Graph Neural Networks for Social Recommendations

Graph Neural Networks in Modern Recommender Systems

GDSRec: Graph-Based Decentralized Collaborative Filtering for Social Recommendation

Dual Variational Graph Reconstruction Learning for Social Recommendation

GDSRec: Graph-Based DecentralizedCollaborative Filtering for SocialRecommendation

Enhancing Matrix Factorization-based Recommender Systems Via Graph Neural Networks

Adaptive Feedback Graph-Enhanced Network ForSocial Recommendation

Social Recommendation through Heterogeneous Graph Modeling of the Long-term and Short-term Preference Defined by Dynamic Time Spans

Incorporating A Triple Graph Neural Network with Multiple Implicit Feedback for Social Recommendation

A Graph-Neural-Network-Based Social Network Recommendation Algorithm Using High-Order Neighbor Information

Federated social recommendation with graph neural network

A Deep Graph Neural Network-Based Mechanism for Social Recommendations