Reproduce “Adversarial Text Rewriting for Text-aware Recommender Systems” 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
Sejoon Oh, Gaurav Verma, Srijan Kumar
Researchers at Georgia Institute of Technology demonstrate how text-aware recommender systems are susceptible to adversarial text rewriting, where item descriptions are subtly altered to artificially boost product rankings. Their Adversarial Text Rewriting (ATR) framework, using either two-phase fine-tuning or in-context learning, significantly improves item ranks while maintaining text quality, exposing a practical vulnerability for such systems.
Key insights
- Text-aware recommender systems are vulnerable to adversarial text rewriting, a previously underexplored attack vector that does not require user profile manipulation or fake interactions.
- The proposed ATR-ICL method leverages in-context learning with large language models to generate high-quality, fluent, and semantically similar adversarial text while still significantly boosting item rankings.
- Different recommender systems exhibit varying levels of robustness, with AMR and FDSA showing higher resistance to text rewriting attacks compared to HybridMF and UniSRec.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.