Automated Self-Supervised Learning for Recommendation

Abstract

Graph neural networks (GNNs) have emerged as the state-of-the-art paradigm\nfor collaborative filtering (CF). To improve the representation quality over\nlimited labeled data, contrastive learning has attracted attention in\nrecommendation and benefited graph-based CF model recently. However, the\nsuccess of most contrastive methods heavily relies on manually generating\neffective contrastive views for heuristic-based data augmentation. This does\nnot generalize across different datasets and downstream recommendation tasks,\nwhich is difficult to be adaptive for data augmentation and robust to noise\nperturbation. To fill this crucial gap, this work proposes a unified Automated\nCollaborative Filtering (AutoCF) to automatically perform data augmentation for\nrecommendation. Specifically, we focus on the generative self-supervised\nlearning framework with a learnable augmentation paradigm that benefits the\nautomated distillation of important self-supervised signals. To enhance the\nrepresentation discrimination ability, our masked graph autoencoder is designed\nto aggregate global information during the augmentation via reconstructing the\nmasked subgraph structures. Experiments and ablation studies are performed on\nseveral public datasets for recommending products, venues, and locations.\nResults demonstrate the superiority of AutoCF against various baseline methods.\nWe release the model implementation at https://github.com/HKUDS/AutoCF.\n