Reproduce “A Comprehensive Survey on Self-Supervised Learning for Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xubin Ren, Wei Wei, Lianghao Xia, Chao Huang
This survey from The University of Hong Kong synthesizes over 170 research papers on self-supervised learning (SSL) for recommender systems, proposing a refined taxonomy encompassing Contrastive, Reconstructive, and Adversarial Learning. It systematically details how these SSL paradigms address data sparsity and enhance representation learning across nine distinct recommendation scenarios, also providing an open-source library for implementation.
Key insights
- Self-supervised learning effectively mitigates data sparsity in recommender systems by leveraging intrinsic data structures to generate supervisory signals, reducing reliance on external labeled data.
- Contrastive Learning is the most versatile SSL paradigm, widely adopted across scenarios with diverse view creation techniques (data-based, feature-based, model-based) and InfoNCE-based objectives being prevalent.
- Reconstructive Learning, including Masked Autoencoding and Denoised Diffusion models, plays a crucial role in discovering underlying data patterns through reconstructing missing or corrupted information.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.