Reproduce “Interpretable Feature Discovery from Unstructured Data Using LLMs and Statistical Choice Models” 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
Baptiste Arnaudo, Amine Bennouna, Sébastien Martin
The authors turn pairwise LLM rationales into shared features and item-level scores. A second LLM groups feature descriptions, while a Bradley–Terry-family model uses mentions and omissions to estimate scores and account for position bias. On synthetic cover letters, the method recovered all 16 relevant signals, versus 13 for direct LLM extraction; this controlled result does not establish recovery in real hiring.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.