Reproduce “Debiasing the Cloze Task in Sequential Recommendation with Bidirectional Transformers” 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
Khalil Damak, Sami Khenissi, Olfa Nasraoui
Researchers from the University of Louisville developed Inverse Temporal Propensity Scoring (ITPS) to debias the Cloze task in sequential recommender systems, which consistently outperformed naive and traditional IPS-debiased models by accurately estimating true item relevance and mitigating popularity bias.
Key insights
- Exposure bias is inherent in sequential recommendation models using the Cloze task, as user interactions reflect both relevance and exposure, not just true preferences.
- Traditional Inverse Propensity Scoring (IPS) methods are insufficient for sequential recommendation due to their static view of exposure and their inability to model the single-choice nature of sequential interactions.
- Accounting for the temporal dynamics of item exposure through temporal propensity scoring is crucial for effective debiasing in sequential recommendation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.