Abstract:This paper addresses key challenges in enhancing recommendation systems by leveraging Graph Neural Networks (GNNs) and addressing inherent limitations such as over-smoothing, which reduces model effectiveness as network hierarchy deepens. The proposed approach introduces three GNN-based recommendation models, specifically designed to mitigate over-smoothing through innovative mechanisms like residual connections and identity mapping within the aggregation propagation process. These modifications enable more effective information flow across layers, preserving essential user-item interaction details to improve recommendation accuracy. Additionally, the study emphasizes the critical need for interpretability in recommendation systems, aiming to provide transparent and justifiable suggestions tailored to dynamic user preferences. By integrating collaborative filtering with GNN architectures, the proposed models not only enhance predictive accuracy but also align recommendations more closely with individual behaviors, adapting to nuanced shifts in user interests. This work advances the field by tackling both technical and user-centric challenges, contributing to the development of robust and explainable recommendation systems capable of managing the complexity and scale of modern online environments.

What problem does this paper attempt to address?

This paper aims to address two key challenges in recommendation systems: enhancing the performance of recommendation systems using graph neural networks (GNNs) and solving the over - smoothing problem that occurs as the network depth increases. Specifically: 1. **Over - smoothing problem**: As the number of GNN layers increases, node features gradually become similar, leading to a decrease in the model's effectiveness. The paper proposes to alleviate this problem by introducing residual connections and identity mapping, ensuring the effective flow of information between different layers, preserving the key details of user - item interactions, and thus improving the accuracy of recommendations. 2. **Interpretability**: Modern recommendation systems need not only to provide accurate recommendation results but also to be able to explain the reasons for the recommendations. The paper emphasizes the importance of interpretability in recommendation systems, aiming to provide transparent and reasonable suggestions to adapt to users' dynamic preferences. To address these challenges, the paper proposes three GNN - based recommendation models. These models reduce the over - smoothing phenomenon through innovative mechanisms such as residual connections and identity mapping, and at the same time combine collaborative filtering techniques to enhance the accuracy and personalization of predictions. ### Key Technologies and Methods 1. **Embedding Layer**: - Initialize the embedding vectors of each user and item, and construct a normalized adjacency matrix to capture user - item relationships. 2. **Aggregation Propagation Layer**: - Introduce initial residual connections and identity mapping to reduce the over - smoothing phenomenon and aggregate more historical user - item interaction information. - Use nonlinear activation functions and parameter adjustments to enhance the flexibility and expressiveness of the network. 3. **Rating Prediction Layer**: - Aggregate user - item matrices of different layers to generate the final predicted rating, and calculate the overall similarity between users and items through inner product. ### Experimental Verification The paper uses three datasets (Gowalla, Yelp - 2018, Amazon - Book) for experiments and compares the performance of the proposed algorithm with five mainstream recommendation algorithms (BPRMF, GCMC, NGCF, LR - GCCF, LGC). The experimental results show that the proposed algorithm outperforms other algorithms in both recall and normalized discounted cumulative gain (NDCG) metrics, especially when dealing with large - scale datasets. ### Conclusion The paper proposes an innovative deep graph neural network - based recommendation algorithm. By introducing initial residual connections and identity mapping, it effectively solves the over - smoothing problem and improves the accuracy and interpretability of the recommendation system. This algorithm not only technically enhances the robustness of the recommendation system but also meets users' needs for transparent and dynamic recommendations, providing a valuable framework and tools for the development of future personalized recommendation systems.

Enhancing Recommendation Systems with GNNs and Addressing Over-Smoothing

Self-Propagation Graph Neural Network for Recommendation

DGR: A General Graph Desmoothing Framework for Recommendation via Global and Local Perspectives

A Dynamic Adaptive Multi-view Fusion Graph Convolutional Network Recommendation Model with Dilated Mask Convolution Mechanism

Beyond smoothness: A general optimization framework for graph neural networks with negative Laplacian regularization

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

Fully Utilizing Neighbors for Session-Based Recommendation with Graph Neural Networks

Multi-Aspect enhanced Graph Neural Networks for recommendation

A Scalable Social Recommendation Framework with Decoupled Graph Neural Network

ConsisRec: Enhancing GNN for Social Recommendation via Consistent Neighbor Aggregation

ApeGNN: Node-Wise Adaptive Aggregation in GNNs for Recommendation

HybridGNN: Learning Hybrid Representation for Recommendation in Multiplex Heterogeneous Networks

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

Neural Multi-network Diffusion towards Social Recommendation

A Graph Neural Network Framework for Social Recommendations

A feature-enhanced knowledge graph neural network for machine learning method recommendation

GNNCL: A Graph Neural Network Recommendation Model Based on Contrastive Learning

Matrix Completion of Adaptive Jumping Graph Neural Networks for Recommendation Systems

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

Hybrid-Order Gated Graph Neural Network for Session-Based Recommendation

Knowledge-aware fine-grained attention networks with refined knowledge graph embedding for personalized recommendation