LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation LightGCN is a simplified graph convolution network (GCN) designed for recommendation tasks. Unlike traditional GCNs, which include complex operations like feature transformation and nonlinear activation, LightGCN only uses neighborhood aggregation. This approach simplifies the model, making it easier to train and more effective for collaborative filtering. LightGCN learns user and item embeddings by linearly propagating them on the user-item interaction graph and uses the weighted sum of embeddings from all layers as the final embedding. This simple, linear model outperforms Neural Graph Collaborative Filtering (NGCF) by about 16% on average. The paper provides empirical evidence that the two most common designs in GCNs—feature transformation and nonlinear activation—do not contribute to the effectiveness of collaborative filtering and may even hinder model performance. LightGCN's design is rational and effective, as shown through extensive experiments and analyses. The model is implemented in both TensorFlow and PyTorch. The paper also discusses the relationship between LightGCN and other graph neural network models, showing that LightGCN can achieve similar performance to other models like APPNP. The results show that LightGCN outperforms other state-of-the-art methods in recommendation tasks, demonstrating its effectiveness and simplicity. The paper concludes that LightGCN is a promising model for recommendation systems due to its simplicity and effectiveness.LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation LightGCN is a simplified graph convolution network (GCN) designed for recommendation tasks. Unlike traditional GCNs, which include complex operations like feature transformation and nonlinear activation, LightGCN only uses neighborhood aggregation. This approach simplifies the model, making it easier to train and more effective for collaborative filtering. LightGCN learns user and item embeddings by linearly propagating them on the user-item interaction graph and uses the weighted sum of embeddings from all layers as the final embedding. This simple, linear model outperforms Neural Graph Collaborative Filtering (NGCF) by about 16% on average. The paper provides empirical evidence that the two most common designs in GCNs—feature transformation and nonlinear activation—do not contribute to the effectiveness of collaborative filtering and may even hinder model performance. LightGCN's design is rational and effective, as shown through extensive experiments and analyses. The model is implemented in both TensorFlow and PyTorch. The paper also discusses the relationship between LightGCN and other graph neural network models, showing that LightGCN can achieve similar performance to other models like APPNP. The results show that LightGCN outperforms other state-of-the-art methods in recommendation tasks, demonstrating its effectiveness and simplicity. The paper concludes that LightGCN is a promising model for recommendation systems due to its simplicity and effectiveness.