EnsemJudge: Enhancing Reliability in Chinese LLM-Generated Text Detection through Diverse Model Ensembles

Zhuoshang Wang; Yubing Ren; Guoyu Zhao; Xiaowei Zhu; Hao Li; Yanan Cao

cs.CLMar 30, 2026

EnsemJudge: Enhancing Reliability in Chinese LLM-Generated Text Detection through Diverse Model Ensembles

Authors: Zhuoshang Wang, Yubing Ren, Guoyu Zhao, Xiaowei Zhu, Hao Li, Yanan Cao

Organizations: Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China · School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China

Abstract

Large Language Models (LLMs) are widely applied across various domains due to their powerful text generation capabilities. While LLM-generated texts often resemble human-written ones, their misuse can lead to significant societal risks. Detecting such texts is an essential technique for mitigating LLM misuse, and many detection methods have shown promising results across different datasets. However, real-world scenarios often involve out-of-domain inputs or adversarial samples, which can affect the performance of detection methods to varying degrees. Furthermore, most existing research has focused on English texts, with limited work addressing Chinese text detection. In this study, we propose EnsemJudge, a robust framework for detecting Chinese LLM-generated text by incorporating tailored strategies and ensemble voting mechanisms. We trained and evaluated our system on a carefully constructed Chinese dataset provided by NLPCC2025 Shared Task 1. Our approach outperformed all baseline methods and achieved first place in the task, demonstrating its effectiveness and reliability in Chinese LLM-generated text detection. Our code is available at https://github.com/johnsonwangzs/MGT-Mini.

Figures & tables

Figure 1: The EnsemJudge Detection Framework.

Split	Label Counts			Data Type
Split	LLM	Human	Total	Normal	Attack	Varying Length
train	24300	8100	32400	$\checkmark$	$\times$	$\times$
dev	1700	1100	2800	$\checkmark$	$\times$	$\times$
test	5500	5500	11000	$\checkmark$	$\checkmark$	$\checkmark$

Table 1: Details of the dataset provided by NLPCC2025 Shared Task 1.

text type	$w_{2}$	$w_{3}$	$w_{4}$	$w_{5}$	$w_{6}$	$w_{7}$	$w_{8}$	$w_{9}$	$w_{10}$	$w_{11}$	$w_{13}$	$w_{14}$	$w_{15}$	$w_{16}$	$w_{17}$	$\lambda$
ext. short	10	0	10	10	60	60	55	60	0	0	0	0	95	400	10	250
short	10	0	10	40	40	40	35	40	0	0	0	40	95	400	40	150
medium	10	0	10	40	40	40	35	40	0	0	0	100	80	90	40	0
general	10	10	10	40	70	70	70	75	50	60	85	400	40	60	80	0

Table 2: The hyper-parameters used in our system. The order of the weights

w

corresponds to the order of the base models listed in Table 3 .

Model Name	All	Normal	Attack			Varying Length (Short Text)
	All	Normal	Mixed	Paraphrase	Perturbation	64	128	256	512
	1-11000	1-4000	4001-5000	5001-6000	6001-7000	7001-8000	8001-9000	9001-10000	10001-11000
SpecialToken	0.3334	0.3333	0.3338	0.3333	0.3333	0.3333	0.3333	0.3333	0.3333
ConsecutivePunctuation	0.3377	0.3367	0.3395	0.3400	0.3400	0.3333	0.3356	0.3400	0.3400
CommonPhrase	0.5299	0.5926	0.5122	0.6306	0.5325	0.3813	0.4020	0.4324	0.4681
SentenceSegment	0.5899	0.5919	0.6225	0.6361	0.6387	0.4864	0.5373	0.5627	0.6243

Table 3: Main results on test set.

Panel of Figure 2 — Figure 2: Model performance (macro F1-score) for our system and different base models.

AI-Generated Text Detection

AI-generated text detection develops methods to distinguish human-written, machine-generated, and AI-edited text.

See all 170 papers

C-ReD: A Comprehensive Chinese Benchmark for AI-Generated Text Detection Derived from Real-World Prompts
Apr 13, 2026Chenxi Qing, Junxi Wu, Zheng Liu +5AI-Generated Text Detection Robustness of Machine-Generated Text Detection
Tsinghua University · Nankai University · Harbin Institute of Technology, Shenzhen +2
EVIL-Detect for NLPCC 2026 Shared Task 6: LLM-Generated Text Detection
Aug 11, 2026Hongrui Bao, Hangyu Rong, Zhuoshang Wang +2AI-Generated Text Detection Language Model Generation Evaluation
Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China · School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China
DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection
May 15, 2026Junchao Wu, Yefeng Liu, Chenyu Zhu +8Multilingual Language Model Evaluation AI-Generated Text Detection
NLP2CT Lab, Department of Computer and Information Science, University of Macau · Alibaba Group · Xiamen University

Apr 13, 2026J/K move · Enter open · S save

C-ReD: A Comprehensive Chinese Benchmark for AI-Generated Text Detection Derived from Real-World Prompts

Chenxi Qing, Junxi Wu, Zheng Liu, Yixiang Qiu+4

Tsinghua University · Nankai University · Harbin Institute of Technology, Shenzhen+2

AI-Generated Text Detection Robustness of Machine-Generated Text Detection

Open paper AskPDF arXiv

Recently, large language models (LLMs) are capable of generating highly fluent textual content. While they offer significant convenience to humans, they also introduce various risks, like phishing and academic dishonesty. Numerous research efforts have been dedicated to developing algorithms for detecting AI-generated text and constructing relevant datasets. However, in the domain of Chinese corpora, challenges remain, including limited model diversity and data homogeneity. To address these issues, we propose C-ReD: a comprehensive Chinese Real-prompt AI-generated Detection benchmark. Experiments demonstrate that C-ReD not only enables reliable in-domain detection but also supports strong generalization to unseen LLMs and external Chinese datasets-addressing critical gaps in model diversity, domain coverage, and prompt realism that have limited prior Chinese detection benchmarks. We release our resources at https://github.com/HeraldofLight/C-ReD.

Abstract

AI-Generated Text Detection

Explore similar work

C-ReD: A Comprehensive Chinese Benchmark for AI-Generated Text Detection Derived from Real-World Prompts

EVIL-Detect for NLPCC 2026 Shared Task 6: LLM-Generated Text Detection

DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection