Abstract:Social recommendation aims to fuse social links with user-item interactions to alleviate the cold-start problem for rating prediction. Recent developments of Graph Neural Networks (GNNs) motivate endeavors to design GNN-based social recommendation frameworks to aggregate both social and user-item interaction information simultaneously. However, most existing methods neglect the social inconsistency problem, which intuitively suggests that social links are not necessarily consistent with the rating prediction process. Social inconsistency can be observed from both context-level and relation-level. Therefore, we intend to empower the GNN model with the ability to tackle the social inconsistency problem. We propose to sample consistent neighbors by relating sampling probability with consistency scores between neighbors. Besides, we employ the relation attention mechanism to assign consistent relations with high importance factors for aggregation. Experiments on two real-world datasets verify the model effectiveness.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the **Social Inconsistency Problem** in social recommendation. Specifically, social recommendation aims to integrate social links and user - item interactions to alleviate the cold - start problem in rating prediction. However, most of the existing social recommendation methods based on Graph Neural Networks (GNN) overlook the social inconsistency problem, that is, social links are not necessarily consistent with the rating prediction process. This inconsistency can be divided into two levels: 1. **Context - level**: It means that users connected in the social graph may have different item contexts. For example, if all the items rated by user \(u_3\) are books, while all the items rated by user \(u_2\) are sports goods, then although they are connected in the social graph, their item contexts are very different, resulting in \(u_3\) becoming an inconsistent neighbor of \(u_2\). 2. **Relation - level**: When simultaneously modeling the social graph and the user - item graph, there are multiple relationships. For example, in addition to social relationships, user - item relationships can also be distinguished by users' rating values. If both user \(u_1\) and user \(u_2\) are connected to item \(t_1\), but \(u_1\) gives \(t_1\) a high score (5 points), while \(u_2\) gives \(t_1\) a low score (1 point), this will lead to an inconsistency at the relation - level, because although they are socially connected, they are inconsistent in item preferences. To solve these problems, the paper proposes a new framework named **ConsisRec** to enhance the GNN model through the following methods: - **Query Layer**: Generate query embeddings for selecting consistent neighbors. - **Neighbor Sampling**: Dynamically sample consistent neighbors according to the similarity score between the query embedding and the neighbor embedding. - **Relation Attention**: Assign a higher importance factor to consistent relationships, thus paying more attention to these consistent relationships in the aggregation process. Through these methods, ConsisRec can aggregate node embeddings from consistent contexts and relationships, thereby improving the accuracy of rating prediction. Experimental results show that ConsisRec has shown significant performance on two real - world datasets.

ConsisRec: Enhancing GNN for Social Recommendation via Consistent Neighbor Aggregation

Self-Propagation Graph Neural Network for Recommendation

NC-GNN: Consistent Neighbors of Nodes Help More in Graph Neural Networks

A Graph Neural Network Framework for Social Recommendations

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

Revisiting Graph Based Social Recommendation: A Distillation Enhanced Social Graph Network

Neural Multi-network Diffusion towards Social Recommendation

Diversifying Collaborative Filtering via Graph Spreading Network and Selective Sampling

GDSRec: Graph-Based DecentralizedCollaborative Filtering for SocialRecommendation

GDSRec: Graph-Based Decentralized Collaborative Filtering for Social Recommendation

Dual Variational Graph Reconstruction Learning for Social Recommendation

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

ConsRec: Learning Consensus Behind Interactions for Group Recommendation

A Scalable Social Recommendation Framework with Decoupled Graph Neural Network

Matrix Completion of Adaptive Jumping Graph Neural Networks for Recommendation Systems

Enhancing Recommendation Systems with GNNs and Addressing Over-Smoothing

SGNNRec: A Scalable Double-Layer Attention-Based Graph Neural Network Recommendation Model

Graph Learning Augmented Heterogeneous Graph Neural Network for Social Recommendation

Disentangled Graph Social Recommendation

NGAT4Rec: Neighbor-Aware Graph Attention Network For Recommendation

Robust Preference-Guided Denoising for Graph based Social Recommendation