1) The document proposes LightGCN, a simplified graph convolutional network (GCN) model for recommender systems. LightGCN removes feature transformation and nonlinear activation layers that are commonly used in GCNs but provide no benefit for collaborative filtering. 2) LightGCN is compared to NGCF, an existing GCN-based recommender system model. LightGCN demonstrates improved performance over NGCF while being less complex. 3) Key contributions of LightGCN include empirically showing that common GCN designs like feature transformation and nonlinear activation are unnecessary for recommendations, proposing a simplified GCN model, and outperforming NGCF on benchmark datasets.