Metraipolis › archive › dc_64e3bbc0c35e

Interpretable Feature Discovery from Unstructured Data Using LLMs and Statistical Choice Models — arXiv:2610.38082

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.38082
License (as archived)
CC0-1.0
Archived
2026-10-31T06:11:17.735Z
SHA-256 of the archived text
b327a219ec0158a50da69537707f809312cddb08e987a01712eae1cbcec21f67
Stable URL of this page
https://metraipolis.com/documento/dc_64e3bbc0c35e

How to cite

Interpretable Feature Discovery from Unstructured Data Using LLMs and Statistical Choice Models — arXiv:2610.38082 — arXiv (en), https://arxiv.org/abs/2610.38082 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_64e3bbc0c35e (Metraipolis archive dc_64e3bbc0c35e, sha256 b327a219ec0158a50da69537707f809312cddb08e987a01712eae1cbcec21f67)
BibTeX
@misc{dc64e3bbc0c35e,
  title = {Interpretable Feature Discovery from Unstructured Data Using LLMs and Statistical Choice Models — arXiv:2610.38082},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.38082},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_64e3bbc0c35e; sha256 b327a219ec0158a50da69537707f809312cddb08e987a01712eae1cbcec21f67},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.38082. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Feature extraction from unstructured data typically relies on opaque embeddings unsuited to interpretable decision-making. Interpretable features, on the other hand, typically require costly, and often infeasible, hand-engineering. Large language models (LLMs) offer a natural bridge, as they can read raw data and describe it in interpretable terms. One could simply ask an LLM to extract features from a large sample of the dataset, but context limitations and diluted attention can lead to unsatisfactory results. We introduce a method that automatically discovers interpretable, decision-relevant features from unstructured data. Our key idea is to perform pairwise comparisons between items, asking an LLM to articulate, in interpretable terms, the features distinguishing each pair. We treat these explanations, or rationales, as data: we develop a statistical choice model that combines these rationales across the comparisons to identify relevant features and estimate per-item, per-feature scores. We first test our method on a synthetic hiring dataset of cover letters with known ground-truth features and show that it significantly outperforms direct LLM feature extraction in feature recovery. We then apply it to the ChangeMyView (Reddit) debate dataset, where it discovers several interpretable features associated with persuasion while reaching predictive accuracy comparable to established baselines. We also show how the same rationales can be used to audit LLM decisions, yielding a holistic score whose decomposition over interpretable features reveals what the LLM rewards.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (stat.ML) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.38082 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-31T06:07:08.621Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).