FGR-ColBERT: Identifying Fine-Grained Relevance Tokens During Retrieval

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jarolím, Antonín, Fajčík, Martin
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918421446262784
author Jarolím, Antonín
Fajčík, Martin
author_facet Jarolím, Antonín
Fajčík, Martin
contents Document retrieval identifies relevant documents but does not provide fine-grained evidence cues, such as specific relevant spans. A possible solution is to apply an LLM after retrieval; however, this introduces significant computational overhead and limits practical deployment. We propose FGR-ColBERT, a modification of ColBERT retrieval model that integrates fine-grained relevance signals distilled from an LLM directly into the retrieval function. Experiments on MS MARCO show that FGR-ColBERT (110M) achieves a token-level F1 of 64.5, exceeding the 62.8 of Gemma 2 (27B), despite being approximately 245 times smaller. At the same time, it preserves retrieval effectiveness (99% relative Recall@50) and remains efficient, incurring only a ~1.12x latency overhead compared to the original ColBERT.
format Preprint
id arxiv_https___arxiv_org_abs_2604_00242
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FGR-ColBERT: Identifying Fine-Grained Relevance Tokens During Retrieval
Jarolím, Antonín
Fajčík, Martin
Information Retrieval
Computation and Language
Document retrieval identifies relevant documents but does not provide fine-grained evidence cues, such as specific relevant spans. A possible solution is to apply an LLM after retrieval; however, this introduces significant computational overhead and limits practical deployment. We propose FGR-ColBERT, a modification of ColBERT retrieval model that integrates fine-grained relevance signals distilled from an LLM directly into the retrieval function. Experiments on MS MARCO show that FGR-ColBERT (110M) achieves a token-level F1 of 64.5, exceeding the 62.8 of Gemma 2 (27B), despite being approximately 245 times smaller. At the same time, it preserves retrieval effectiveness (99% relative Recall@50) and remains efficient, incurring only a ~1.12x latency overhead compared to the original ColBERT.
title FGR-ColBERT: Identifying Fine-Grained Relevance Tokens During Retrieval
topic Information Retrieval
Computation and Language
url https://arxiv.org/abs/2604.00242