Can we repurpose multiple-choice question-answering models to rerank retrieved documents?

Jasper Kyle Catapang · 2025

Read Can we repurpose multiple-choice question-answering models to rerank retrieved documents? by Jasper Kyle Catapang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06276.txt.

Description

Jasper Kyle Catapang asks a practical question in a 2025 paper: can models built to answer multiple-choice questions also rank search results? Search systems first pull a set of documents, then rank them by how well each one fits the query. A good ranker matters for search and for chat tools that read retrieved text before they answer.

Catapang points to a math parallel between picking the right answer choice and judging how relevant a document is. On that basis he built a small proof-of-concept model that joins the two ideas. Tests showed the model improved ranking accuracy while staying lightweight. The paper is early work, offered as a sign that question-answering models hold untapped value for ranking tasks.

Catalog record

CREATORJasper Kyle Catapang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-06276

Browse the full library · About the Tower of Babel