@inproceedings{upadhyay-etal-2026-certified,
title = "Certified Cost-Optimal Triage for Suggestion Detection in Reviews",
author = "Upadhyay, Aniket and
Trivedi, Aakash and
Kumar, Dhruv and
Narang, Pratik",
editor = "Atanasova, Pepa and
Haffari, Gholamreza and
Wang, Minghan",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-industry.248/",
pages = "4196--4211",
ISBN = "979-8-89176-460-6",
abstract = "Detecting suggestions in reviews forces a trade- off: an LLM on every review is slow and costly; a cheap classifier is fast but its error rate is uncontrolled, measurable after the fact, not settable in advance. We present a certified triage layer: a calibration step computes ex- actly where a classifier may act alone so its error there stays below an operator-chosen tar- get {\ensuremath{\alpha}}; the rest is escalated. On a 14,995-review gold set over 68 industry categories, natural- weighted to a {\ensuremath{\sim}}9.5{\%} deployment positive rate, it auto-handles 74{--}93{\%} of the stream at cer- tified error for {\ensuremath{\alpha}} {\ensuremath{\in}} {2, 4, 5, 10}{\%}. Heuris- tic thresholding violates the {\ensuremath{\alpha}}=5{\%} target on 36{--}49{\%} of draws and an in-expectation con- formal gate on 25{--}28{\%}; only a betting certifi- cate bounds the violation rate by {\ensuremath{\delta}}, making a strict 2{\%} SLA rigorous at 74{\%} order-free cover- age. Cost-sensitive training, not threshold tun- ing, lowers the certifiable error floor to 0.39{\%}, and {\ensuremath{\sim}}200 target labels restore it under cross- industry shift. Of 13 open-weight LLMs (20B{--} 675B), 11 are worse detectors than our fine- tuned 435M encoder, supervision, not scale, is why (within-model control, {\textsection}5.6), and the cer- tificate rides on either backbone. Escalation targets a human: on the deferred slice the best LLM oracle still errs at 21.5{\%}."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="upadhyay-etal-2026-certified">
<titleInfo>
<title>Certified Cost-Optimal Triage for Suggestion Detection in Reviews</title>
</titleInfo>
<name type="personal">
<namePart type="given">Aniket</namePart>
<namePart type="family">Upadhyay</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aakash</namePart>
<namePart type="family">Trivedi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dhruv</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pratik</namePart>
<namePart type="family">Narang</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pepa</namePart>
<namePart type="family">Atanasova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gholamreza</namePart>
<namePart type="family">Haffari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minghan</namePart>
<namePart type="family">Wang</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">979-8-89176-460-6</identifier>
</relatedItem>
<abstract>Detecting suggestions in reviews forces a trade- off: an LLM on every review is slow and costly; a cheap classifier is fast but its error rate is uncontrolled, measurable after the fact, not settable in advance. We present a certified triage layer: a calibration step computes ex- actly where a classifier may act alone so its error there stays below an operator-chosen tar- get \ensuremathα; the rest is escalated. On a 14,995-review gold set over 68 industry categories, natural- weighted to a \ensuremath\sim9.5% deployment positive rate, it auto-handles 74–93% of the stream at cer- tified error for \ensuremathα \ensuremathın 2, 4, 5, 10%. Heuris- tic thresholding violates the \ensuremathα=5% target on 36–49% of draws and an in-expectation con- formal gate on 25–28%; only a betting certifi- cate bounds the violation rate by \ensuremathδ, making a strict 2% SLA rigorous at 74% order-free cover- age. Cost-sensitive training, not threshold tun- ing, lowers the certifiable error floor to 0.39%, and \ensuremath\sim200 target labels restore it under cross- industry shift. Of 13 open-weight LLMs (20B– 675B), 11 are worse detectors than our fine- tuned 435M encoder, supervision, not scale, is why (within-model control, §5.6), and the cer- tificate rides on either backbone. Escalation targets a human: on the deferred slice the best LLM oracle still errs at 21.5%.</abstract>
<identifier type="citekey">upadhyay-etal-2026-certified</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-industry.248/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>4196</start>
<end>4211</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Certified Cost-Optimal Triage for Suggestion Detection in Reviews
%A Upadhyay, Aniket
%A Trivedi, Aakash
%A Kumar, Dhruv
%A Narang, Pratik
%Y Atanasova, Pepa
%Y Haffari, Gholamreza
%Y Wang, Minghan
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-460-6
%F upadhyay-etal-2026-certified
%X Detecting suggestions in reviews forces a trade- off: an LLM on every review is slow and costly; a cheap classifier is fast but its error rate is uncontrolled, measurable after the fact, not settable in advance. We present a certified triage layer: a calibration step computes ex- actly where a classifier may act alone so its error there stays below an operator-chosen tar- get \ensuremathα; the rest is escalated. On a 14,995-review gold set over 68 industry categories, natural- weighted to a \ensuremath\sim9.5% deployment positive rate, it auto-handles 74–93% of the stream at cer- tified error for \ensuremathα \ensuremathın 2, 4, 5, 10%. Heuris- tic thresholding violates the \ensuremathα=5% target on 36–49% of draws and an in-expectation con- formal gate on 25–28%; only a betting certifi- cate bounds the violation rate by \ensuremathδ, making a strict 2% SLA rigorous at 74% order-free cover- age. Cost-sensitive training, not threshold tun- ing, lowers the certifiable error floor to 0.39%, and \ensuremath\sim200 target labels restore it under cross- industry shift. Of 13 open-weight LLMs (20B– 675B), 11 are worse detectors than our fine- tuned 435M encoder, supervision, not scale, is why (within-model control, §5.6), and the cer- tificate rides on either backbone. Escalation targets a human: on the deferred slice the best LLM oracle still errs at 21.5%.
%U https://aclanthology.org/2026.emnlp-industry.248/
%P 4196-4211
Markdown (Informal)
[Certified Cost-Optimal Triage for Suggestion Detection in Reviews](https://aclanthology.org/2026.emnlp-industry.248/) (Upadhyay et al., EMNLP 2026)
ACL