@inproceedings{dai-etal-2026-think,
title = "When Not to Think: Mitigating the Overthinking Effect in Content Moderation",
author = "Dai, Yu and
Zhang, Jielei and
Ou, Yang and
Yang, Zian and
Yang, Run and
Li, Xiang and
Zhuang, Qingyao and
Li, Peiyi and
Gao, Longwen",
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.110/",
pages = "1888--1899",
ISBN = "979-8-89176-460-6",
abstract = "Chain-of-thought (CoT) reasoning has become a default paradigm across reasoning-intensive tasks and is widely assumed to be, at worst, harmless. Recent efficient-reasoning work treats unnecessary CoT purely as a cost to be reduced. We show that in boundary-dense classification tasks such as content moderation and toxicity detection, CoT is instead actively harmful. We identify the overthinking effect: extended reasoning amplifies weak surface signals into false evidence of violation, inflating false positive rates on ambiguous non-violating samples while its gains on ambiguous violating samples mask the harm in aggregate metrics. To address this, we propose Oreki, an adaptive CoT framework that uses K-fold cross-probing to label per-sample reasoning necessity and fine-tunes on a mixture of full-reasoning and reasoning-skipped formats. Across four base models on video moderation and multi-label toxicity detection, Oreki preserves CoT{'}s recall gains while suppressing its false positive inflation, outperforming both Always-CoT and Never-CoT. On toxicity detection, it cuts low-agreement false positive rate from 0.467 to 0.257 while raising overall F1 from 0.792 to 0.851."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dai-etal-2026-think">
<titleInfo>
<title>When Not to Think: Mitigating the Overthinking Effect in Content Moderation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jielei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Ou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zian</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Run</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiang</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qingyao</namePart>
<namePart type="family">Zhuang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peiyi</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Longwen</namePart>
<namePart type="family">Gao</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>Chain-of-thought (CoT) reasoning has become a default paradigm across reasoning-intensive tasks and is widely assumed to be, at worst, harmless. Recent efficient-reasoning work treats unnecessary CoT purely as a cost to be reduced. We show that in boundary-dense classification tasks such as content moderation and toxicity detection, CoT is instead actively harmful. We identify the overthinking effect: extended reasoning amplifies weak surface signals into false evidence of violation, inflating false positive rates on ambiguous non-violating samples while its gains on ambiguous violating samples mask the harm in aggregate metrics. To address this, we propose Oreki, an adaptive CoT framework that uses K-fold cross-probing to label per-sample reasoning necessity and fine-tunes on a mixture of full-reasoning and reasoning-skipped formats. Across four base models on video moderation and multi-label toxicity detection, Oreki preserves CoT’s recall gains while suppressing its false positive inflation, outperforming both Always-CoT and Never-CoT. On toxicity detection, it cuts low-agreement false positive rate from 0.467 to 0.257 while raising overall F1 from 0.792 to 0.851.</abstract>
<identifier type="citekey">dai-etal-2026-think</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-industry.110/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1888</start>
<end>1899</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T When Not to Think: Mitigating the Overthinking Effect in Content Moderation
%A Dai, Yu
%A Zhang, Jielei
%A Ou, Yang
%A Yang, Zian
%A Yang, Run
%A Li, Xiang
%A Zhuang, Qingyao
%A Li, Peiyi
%A Gao, Longwen
%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 dai-etal-2026-think
%X Chain-of-thought (CoT) reasoning has become a default paradigm across reasoning-intensive tasks and is widely assumed to be, at worst, harmless. Recent efficient-reasoning work treats unnecessary CoT purely as a cost to be reduced. We show that in boundary-dense classification tasks such as content moderation and toxicity detection, CoT is instead actively harmful. We identify the overthinking effect: extended reasoning amplifies weak surface signals into false evidence of violation, inflating false positive rates on ambiguous non-violating samples while its gains on ambiguous violating samples mask the harm in aggregate metrics. To address this, we propose Oreki, an adaptive CoT framework that uses K-fold cross-probing to label per-sample reasoning necessity and fine-tunes on a mixture of full-reasoning and reasoning-skipped formats. Across four base models on video moderation and multi-label toxicity detection, Oreki preserves CoT’s recall gains while suppressing its false positive inflation, outperforming both Always-CoT and Never-CoT. On toxicity detection, it cuts low-agreement false positive rate from 0.467 to 0.257 while raising overall F1 from 0.792 to 0.851.
%U https://aclanthology.org/2026.emnlp-industry.110/
%P 1888-1899
Markdown (Informal)
[When Not to Think: Mitigating the Overthinking Effect in Content Moderation](https://aclanthology.org/2026.emnlp-industry.110/) (Dai et al., EMNLP 2026)
ACL
- Yu Dai, Jielei Zhang, Yang Ou, Zian Yang, Run Yang, Xiang Li, Qingyao Zhuang, Peiyi Li, and Longwen Gao. 2026. When Not to Think: Mitigating the Overthinking Effect in Content Moderation. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 1888–1899, Budapest, Hungary. Association for Computational Linguistics.