@inproceedings{pan-etal-2026-navirank,
title = "{N}avi{R}ank: {LLM}-Informed Latent Space Navigation Enhances Reasoning-Intensive Information Retrieval",
author = "Pan, Jinwen and
Jiang, Haolin and
Liu, Hang",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.612/",
pages = "12880--12897",
ISBN = "979-8-89176-462-0",
abstract = "Retrieve-then-rerank pipelines restrict large language models (LLMs) to a fixed top-k candidate set, making their reasoning ability fundamentally limited by the recall ceiling of the first-stage dense retriever. On reasoning-intensive retrieval, we make two observations: first, relevant documents missed by dense top-k retrieval are frequently reachable within a few hops on the native approximate nearest neighbor (ANN) graph; second, LLM relevance judgments can identify high-yield anchors whose neighborhoods contain these missed documents. Based on these findings, we propose NaviRank, an LLM-guided navigation framework that performs greedy expansion over the HNSW graph: it uses highly parallel pointwise LLM judgments to steer anchor selection and local expansion, followed by listwise calibration for final ranking. Experiments on BRIGHT show that NaviRank achieves 27.0 NDCG@10, while establishing the strongest accuracy{--}efficiency Pareto frontier in both token usage and latency among all evaluated methods."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="pan-etal-2026-navirank">
<titleInfo>
<title>NaviRank: LLM-Informed Latent Space Navigation Enhances Reasoning-Intensive Information Retrieval</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jinwen</namePart>
<namePart type="family">Pan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haolin</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Retrieve-then-rerank pipelines restrict large language models (LLMs) to a fixed top-k candidate set, making their reasoning ability fundamentally limited by the recall ceiling of the first-stage dense retriever. On reasoning-intensive retrieval, we make two observations: first, relevant documents missed by dense top-k retrieval are frequently reachable within a few hops on the native approximate nearest neighbor (ANN) graph; second, LLM relevance judgments can identify high-yield anchors whose neighborhoods contain these missed documents. Based on these findings, we propose NaviRank, an LLM-guided navigation framework that performs greedy expansion over the HNSW graph: it uses highly parallel pointwise LLM judgments to steer anchor selection and local expansion, followed by listwise calibration for final ranking. Experiments on BRIGHT show that NaviRank achieves 27.0 NDCG@10, while establishing the strongest accuracy–efficiency Pareto frontier in both token usage and latency among all evaluated methods.</abstract>
<identifier type="citekey">pan-etal-2026-navirank</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.612/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>12880</start>
<end>12897</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T NaviRank: LLM-Informed Latent Space Navigation Enhances Reasoning-Intensive Information Retrieval
%A Pan, Jinwen
%A Jiang, Haolin
%A Liu, Hang
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F pan-etal-2026-navirank
%X Retrieve-then-rerank pipelines restrict large language models (LLMs) to a fixed top-k candidate set, making their reasoning ability fundamentally limited by the recall ceiling of the first-stage dense retriever. On reasoning-intensive retrieval, we make two observations: first, relevant documents missed by dense top-k retrieval are frequently reachable within a few hops on the native approximate nearest neighbor (ANN) graph; second, LLM relevance judgments can identify high-yield anchors whose neighborhoods contain these missed documents. Based on these findings, we propose NaviRank, an LLM-guided navigation framework that performs greedy expansion over the HNSW graph: it uses highly parallel pointwise LLM judgments to steer anchor selection and local expansion, followed by listwise calibration for final ranking. Experiments on BRIGHT show that NaviRank achieves 27.0 NDCG@10, while establishing the strongest accuracy–efficiency Pareto frontier in both token usage and latency among all evaluated methods.
%U https://aclanthology.org/2026.findings-emnlp.612/
%P 12880-12897
Markdown (Informal)
[NaviRank: LLM-Informed Latent Space Navigation Enhances Reasoning-Intensive Information Retrieval](https://aclanthology.org/2026.findings-emnlp.612/) (Pan et al., Findings 2026)
ACL