EviRerank: Adaptive Evidence Construction for Long-Document LLM Reranking

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Minghan, Gaussier, Eric, Li, Juntao, Zhou, Guodong
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915710308974592
author Li, Minghan
Gaussier, Eric
Li, Juntao
Zhou, Guodong
author_facet Li, Minghan
Gaussier, Eric
Li, Juntao
Zhou, Guodong
contents Decoder-only LLM rerankers struggle with long documents: inference is costly and relevance signals can be diluted by irrelevant context. Motivated by an attention analysis indicating a consistent degradation trend when non-relevant text is appended, we propose EviRerank, an evidence-based long-document reranking framework for decoder-only LLMs. EviRerank (i) scores document blocks with a lightweight selector (BM25, bi-encoder, or cross-encoder), (ii) constructs a compact reranking context under a hard token cap by dynamically budgeting evidence blocks with Adaptive Evidence Budgeting (AEB) and adding a global summary cue via Summary Augmentation (SA), and (iii) reranks with a decoder-only LLM. Across TREC DL'19, DL'23, and MLDR-zh, EviRerank consistently outperforms full-document LLM reranking and strong block-selection baselines while substantially reducing the required input length. On TREC DL'19, EviRerank achieves 0.743 nDCG@10 and 0.307 MAP, establishing a new best result and improving over RankLLaMA (0.701/0.288) by +0.042 nDCG@10 (+6.0%) and +0.019 MAP (+6.6%).
format Preprint
id arxiv_https___arxiv_org_abs_2411_06254
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle EviRerank: Adaptive Evidence Construction for Long-Document LLM Reranking
Li, Minghan
Gaussier, Eric
Li, Juntao
Zhou, Guodong
Information Retrieval
Decoder-only LLM rerankers struggle with long documents: inference is costly and relevance signals can be diluted by irrelevant context. Motivated by an attention analysis indicating a consistent degradation trend when non-relevant text is appended, we propose EviRerank, an evidence-based long-document reranking framework for decoder-only LLMs. EviRerank (i) scores document blocks with a lightweight selector (BM25, bi-encoder, or cross-encoder), (ii) constructs a compact reranking context under a hard token cap by dynamically budgeting evidence blocks with Adaptive Evidence Budgeting (AEB) and adding a global summary cue via Summary Augmentation (SA), and (iii) reranks with a decoder-only LLM. Across TREC DL'19, DL'23, and MLDR-zh, EviRerank consistently outperforms full-document LLM reranking and strong block-selection baselines while substantially reducing the required input length. On TREC DL'19, EviRerank achieves 0.743 nDCG@10 and 0.307 MAP, establishing a new best result and improving over RankLLaMA (0.701/0.288) by +0.042 nDCG@10 (+6.0%) and +0.019 MAP (+6.6%).
title EviRerank: Adaptive Evidence Construction for Long-Document LLM Reranking
topic Information Retrieval
url https://arxiv.org/abs/2411.06254