Hugging Face Daily Papers · August 27, 2026 · 3 min read

RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Accepted at EMNLP 2026 </p>\n","updatedAt":"2026-08-27T07:47:01.526Z","author":{"_id":"6732fec1f3705c686e602ef8","avatarUrl":"/avatars/3a3b90f925397ac408e7594184ce1e29.svg","fullname":"Emre. Kuru","name":"emrekuruu","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.8770975470542908},"editors":["emrekuruu"],"editorAvatarUrls":["/avatars/3a3b90f925397ac408e7594184ce1e29.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.25625","authors":[{"_id":"6a8fea733bd48bb654ea69a2","user":{"_id":"6732fec1f3705c686e602ef8","avatarUrl":"/avatars/3a3b90f925397ac408e7594184ce1e29.svg","isPro":true,"fullname":"Emre. Kuru","user":"emrekuruu","type":"user","name":"emrekuruu"},"name":"Emre Kuru","status":"claimed_verified","statusLastChangedAt":"2026-08-27T08:45:05.010Z","hidden":false},{"_id":"6a8fea733bd48bb654ea69a3","name":"Mehmet Onur Keskin","hidden":false},{"_id":"6a8fea733bd48bb654ea69a4","name":"Reza Farahbakhsh","hidden":false},{"_id":"6a8fea733bd48bb654ea69a5","name":"Noel Crespi","hidden":false}],"publishedAt":"2026-08-26T00:00:00.000Z","submittedOnDailyAt":"2026-08-27T00:00:00.000Z","title":"RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval","submittedOnDailyBy":{"_id":"6732fec1f3705c686e602ef8","avatarUrl":"/avatars/3a3b90f925397ac408e7594184ce1e29.svg","isPro":true,"fullname":"Emre. Kuru","user":"emrekuruu","type":"user","name":"emrekuruu"},"summary":"Document retrieval increasingly supports high-stakes information access in finance, healthcare, and law. Modern retrieval pipelines vary both in modality (text or multimodal) and in retrieval architecture (dense or late-interaction). These choices impose a hard compromise: the most effective pipelines are too slow and expensive to run at scale, while the fastest fail to retrieve evidence from complex documents. Practitioners must therefore choose between missed evidence and unusable latency, with no principled basis for adapting that choice at the query level. We show that this compromise is unnecessary. Not every query requires the same pipeline. Across benchmarks spanning financial and scientific corpora, no static pipeline dominates. We introduce RetrievalRouter, a lightweight query-aware router that learns, from the query text alone, which retrieval pipeline best fits each query. A single tunable parameter exposes the full accuracy-latency frontier, and for every static baseline, RetrievalRouter offers an operating point that is simultaneously more accurate and faster. Against the best static baseline, RetrievalRouter is 2.5% more accurate and 12.4 times faster. Furthermore, compared with prior adaptive strategy selection methods, RetrievalRouter achieves significantly higher nDCG@5 across accuracy-oriented settings, while matching or numerically outperforming them on both nDCG@5 and latency in latency-oriented settings. Our code and data are available at https://github.com/emrekuruu/retrieval-router.","upvotes":1,"discussionId":"6a8fea733bd48bb654ea69a6","githubRepo":"https://github.com/emrekuruu/retrieval-router","githubRepoAddedBy":"user","ai_summary":"RetrievalRouter adaptively selects retrieval pipelines per query to improve both accuracy and speed across diverse document benchmarks.","ai_keywords":["dense retrieval","late-interaction","multimodal retrieval","RetrievalRouter","query-aware routing","nDCG@5","accuracy-latency frontier"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":0},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6732fec1f3705c686e602ef8","avatarUrl":"/avatars/3a3b90f925397ac408e7594184ce1e29.svg","isPro":true,"fullname":"Emre. Kuru","user":"emrekuruu","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.25625.md","query":{}}">

Papers

arxiv:2608.25625

RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval

Published on Aug 26

· Submitted by

Emre. Kuru on Aug 27

Upvote

Authors:

Emre Kuru ,

Abstract

RetrievalRouter adaptively selects retrieval pipelines per query to improve both accuracy and speed across diverse document benchmarks.

Generated by thinkingmachines/Inkling-Small

Document retrieval increasingly supports high-stakes information access in finance, healthcare, and law. Modern retrieval pipelines vary both in modality (text or multimodal) and in retrieval architecture (dense or late-interaction). These choices impose a hard compromise: the most effective pipelines are too slow and expensive to run at scale, while the fastest fail to retrieve evidence from complex documents. Practitioners must therefore choose between missed evidence and unusable latency, with no principled basis for adapting that choice at the query level. We show that this compromise is unnecessary. Not every query requires the same pipeline. Across benchmarks spanning financial and scientific corpora, no static pipeline dominates. We introduce RetrievalRouter, a lightweight query-aware router that learns, from the query text alone, which retrieval pipeline best fits each query. A single tunable parameter exposes the full accuracy-latency frontier, and for every static baseline, RetrievalRouter offers an operating point that is simultaneously more accurate and faster. Against the best static baseline, RetrievalRouter is 2.5% more accurate and 12.4 times faster. Furthermore, compared with prior adaptive strategy selection methods, RetrievalRouter achieves significantly higher nDCG@5 across accuracy-oriented settings, while matching or numerically outperforming them on both nDCG@5 and latency in latency-oriented settings. Our code and data are available at https://github.com/emrekuruu/retrieval-router.