pith. machine review for the scientific record. sign in

arxiv: 2605.05950 · v1 · submitted 2026-05-07 · 💻 cs.CL

Recognition: unknown

Lightweight Stylistic Consistency Profiling: Robust Detection of LLM-Generated Textual Content for Multimedia Moderation

Siyuan Li , Aodu Wulianghai , Xi Lin , Xibin Yuan , Qinghua Mao , Guangyan Li , Xiang Chen , Jun Wu
show 1 more author
Jianhua Li
Authors on Pith no claims yet

Pith reviewed 2026-05-08 10:52 UTC · model grok-4.3

classification 💻 cs.CL
keywords LLM-generated text detectionstylistic consistencyadversarial robustnessmultimedia moderationparaphrasing attackscross-domain generalizationtext classification
0
0 comments X

The pith

A consistency profile of stylistic features across paraphrased variants detects LLM-generated text more reliably than prior methods.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces LiSCP as a lightweight way to detect LLM-generated text by building a profile from how discrete stylistic features and continuous semantic signals hold steady when the same text is paraphrased multiple times under multimodal guidance. A sympathetic reader would care because existing detectors rely on fragile statistical cues that break under common rewrites or mixed human-AI edits, which now appear routinely in news, movies, and online media. The method claims better accuracy inside the training domain and up to 11.79 percent gains when moved to new domains, plus resistance to attacks that try to hide the origin. This stability focus aims to give moderators a practical, interpretable signal without heavy model access.

Core claim

LiSCP constructs a consistency profile that combines discrete stylistic features with continuous semantic signals, leveraging stylistic stability across multimodal-guided paraphrased text variants. Experiments on real-world multimedia news and movie datasets plus conventional text domains show superior in-domain detection and up to 11.79 percent better cross-domain performance, along with notable robustness under adversarial attacks and hybrid human-AI settings.

What carries the argument

The LiSCP consistency profile, which measures stability of stylistic and semantic features across multiple paraphrased versions of a text.

If this is right

  • Superior detection accuracy holds on real-world multimedia datasets such as news articles and movie scripts.
  • Performance gains of up to 11.79 percent appear when the detector is applied to text domains not seen in training.
  • Robustness extends to adversarial paraphrasing attempts and to content that mixes human and LLM writing.
  • The same profile works across both multimedia-specific and conventional text domains without domain-specific retraining.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The emphasis on feature stability rather than model-specific cues implies the detector could continue working on future LLMs without immediate retraining.
  • Because the method is lightweight and does not require access to the original generator, it could be embedded directly into content pipelines for subtitles, comments, or article moderation.
  • Success against hybrid texts suggests the profile might also flag lightly edited AI content in collaborative writing platforms.

Load-bearing premise

The chosen stylistic features remain stable and discriminative for LLM text versus human text even after multimodal-guided paraphrasing.

What would settle it

Run the detector on outputs from an LLM that has been fine-tuned or prompted to produce paraphrases whose stylistic variability matches human writing at the same rate, then check whether accuracy falls to near-random levels.

Figures

Figures reproduced from arXiv: 2605.05950 by Aodu Wulianghai, Guangyan Li, Jianhua Li, Jun Wu, Qinghua Mao, Siyuan Li, Xiang Chen, Xibin Yuan, Xi Lin.

Figure 1
Figure 1. Figure 1: Overview of the LiSCP. (a) Multimodal-Guided view at source ↗
Figure 2
Figure 2. Figure 2: Evaluation of detection performance degradation view at source ↗
Figure 4
Figure 4. Figure 4: Explainable in-domain detection visualization: view at source ↗
Figure 3
Figure 3. Figure 3: Evaluation of detection performance degradation view at source ↗
read the original abstract

The increasing prevalence of Large Language Models (LLMs) in content creation has made distinguishing human-written textual content from LLM-generated counterparts a critical task for multimedia moderation. Existing detectors often rely on statistical cues or model-specific heuristics, making them vulnerable to paraphrasing and adversarial manipulations, and consequently limiting their robustness and interpretability. In this work, we proposeLiSCP , a novel lightweight stylistic consistency profiling method for robust detection of LLM-generated textual content, focusing on feature stability under adversarial manipulation. Our approach constructs a consistency profile that combines discrete stylistic features with continuous semantic signals, leveraging stylistic stability across multimodal-guided paraphrased text variants. Experiments spanning real-world multimedia news and movie datasets and conventional text domains demonstrate that LiSCP achieves superior performance on in-domain detection and outperforms existing approaches by up to 11.79% in cross-domain settings. Additionally,it demonstrates notable robustness under adversarial scenarios, including adversarial attacks and hybrid human-AI settings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes LiSCP, a lightweight stylistic consistency profiling method for detecting LLM-generated text. It constructs a profile by combining discrete stylistic features with continuous semantic signals and claims this profile remains stable under multimodal-guided paraphrasing. Experiments on multimedia news/movie datasets and standard text domains are reported to show superior in-domain detection, up to 11.79% gains over baselines in cross-domain settings, and robustness to adversarial attacks plus hybrid human-AI content.

Significance. If the central claims are substantiated with proper controls, the work could contribute a practical, interpretable, and computationally light detector for AI-generated content in multimedia moderation pipelines. The emphasis on stylistic stability across paraphrased variants addresses a known weakness in existing statistical or model-specific detectors.

major comments (3)
  1. [Experimental evaluation] Experimental evaluation section: no baselines are enumerated, no error bars or confidence intervals are provided for the performance metrics, and no data exclusion rules or run counts are stated. This directly undermines verification of the headline 11.79% cross-domain improvement and the robustness claims under adversarial/hybrid settings.
  2. [Methods] Methods section: the selection criteria for the discrete stylistic features are not described, nor is it stated whether the feature set is fixed a priori or tuned on the reported datasets. Because the central robustness claim rests on the stability of precisely these features after paraphrasing, the absence of this information makes the outperformance non-reproducible and potentially circular.
  3. [Results] Results section: no ablation is presented that isolates the contribution of the stylistic features versus the semantic signals alone. Without this, it cannot be determined whether the reported cross-domain gains arise from the proposed consistency profile or from the semantic component that many prior detectors already exploit.
minor comments (2)
  1. [Abstract] Abstract: 'proposeLiSCP' is missing a space; 'it demonstrates' should be capitalized after the period.
  2. [Abstract] Abstract: the phrase 'notable robustness' is qualitative; the paper should state the exact metrics (e.g., accuracy drop under attack) used to support this claim.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and will revise the manuscript to improve reproducibility and clarity.

read point-by-point responses

  1. Referee: [Experimental evaluation] Experimental evaluation section: no baselines are enumerated, no error bars or confidence intervals are provided for the performance metrics, and no data exclusion rules or run counts are stated. This directly undermines verification of the headline 11.79% cross-domain improvement and the robustness claims under adversarial/hybrid settings.

    Authors: We agree that additional experimental details are needed for verification. In the revised manuscript, we will enumerate all baselines with references and implementations, report error bars or confidence intervals from multiple runs, and specify data exclusion rules along with the number of runs. This will substantiate the 11.79% cross-domain gains and robustness results. revision: yes

  2. Referee: [Methods] Methods section: the selection criteria for the discrete stylistic features are not described, nor is it stated whether the feature set is fixed a priori or tuned on the reported datasets. Because the central robustness claim rests on the stability of precisely these features after paraphrasing, the absence of this information makes the outperformance non-reproducible and potentially circular.

    Authors: The discrete stylistic features are drawn from established stylometric literature and fixed a priori without any tuning on the evaluation datasets, to preserve interpretability and generalizability. We will revise the Methods section to list the features explicitly, detail their selection criteria from prior work, and confirm the fixed (non-tuned) nature of the set. This resolves the reproducibility concern. revision: yes

  3. Referee: [Results] Results section: no ablation is presented that isolates the contribution of the stylistic features versus the semantic signals alone. Without this, it cannot be determined whether the reported cross-domain gains arise from the proposed consistency profile or from the semantic component that many prior detectors already exploit.

    Authors: We acknowledge that an ablation would strengthen the claims. The revised Results section will include an ablation comparing the full consistency profile (stylistic + semantic) against semantic-only and stylistic-only variants, to isolate the contribution to cross-domain gains and robustness. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical method with independent experimental validation

full rationale

The paper introduces LiSCP as a profile combining discrete stylistic features with continuous semantic signals and validates it through experiments on multimedia news, movie, and text datasets, reporting in-domain and cross-domain gains plus robustness to attacks. No equations, parameter-fitting steps, or self-citations appear in the abstract or described derivation chain that would reduce any claimed result to its inputs by construction. Performance metrics are presented as outcomes of held-out testing rather than tautological re-statements of fitted values, satisfying the criteria for a self-contained empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no free parameters, axioms, or invented entities are specified in the provided text.

pith-pipeline@v0.9.0 · 5485 in / 1043 out tokens · 31265 ms · 2026-05-08T10:52:51.441197+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

55 extracted references · 17 canonical work pages · 5 internal anchors

  1. [1]

    The llama 3 herd of models.arXiv e-prints, pages arXiv–2407, 2024

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models.arXiv e-prints, pages arXiv–2407, 2024

    2024

  2. [2]

    Agentchain: Blockchain-empowered multi-agent coordination for trustworthy llm question-answering systems.IEEE Transactions on Dependable and Secure Computing, 2026

    Bei Chen, Gaolei Li, Jun Wu, Jianhua Li, Mingzhe Chen, and Jiacheng Wang. Agentchain: Blockchain-empowered multi-agent coordination for trustworthy llm question-answering systems.IEEE Transactions on Dependable and Secure Computing, 2026

    2026

  3. [3]

    Large language models driven neural architecture search for universal and lightweight disease diagnosis on histopathology slide images.npj Digital Medicine, 8(1):682, 2025

    Xiu Su, Qinghua Mao, Zhongze Wu, Xi Lin, Shan You, Yue Liao, and Chang Xu. Large language models driven neural architecture search for universal and lightweight disease diagnosis on histopathology slide images.npj Digital Medicine, 8(1):682, 2025

    2025

  4. [4]

    GPT-4o System Card

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    work page internal anchor Pith review arXiv 2024

  5. [5]

    Trustworthy ai-generative con- tent for intelligent network service: Robustness, security, and fairness,

    Siyuan Li, Xi Lin, Yaju Liu, and Jianhua Li. Trustworthy ai-generative content in intelligent 6g network: Adversarial, privacy, and fairness.arXiv preprint arXiv:2405.05930, 2024

    work page arXiv 2024

  6. [6]

    Advance- ments in ai-generated content forensics: A systematic literature review.ACM Computing Surveys, 58(3):1–36, 2025

    Qiang Xu, Wenpeng Mu, Jianing Li, Tanfeng Sun, and Xinghao Jiang. Advance- ments in ai-generated content forensics: A systematic literature review.ACM Computing Surveys, 58(3):1–36, 2025

    2025

  7. [7]

    Can ai-generated text be reliably detected? stress testing ai text detectors under various attacks.Transactions on Machine Learning Research, 2025

    Vinu Sankar Sadasivan, Aounon Kumar, Sriram Balasubramanian, Wenxiao Wang, and Soheil Feizi. Can ai-generated text be reliably detected? stress testing ai text detectors under various attacks.Transactions on Machine Learning Research, 2025

    2025

  8. [8]

    A survey on llm-generated text detection: Necessity, methods, and future directions.Computational Linguistics, 51(1):275–338, 2025

    Junchao Wu, Shu Yang, Runzhe Zhan, Yulin Yuan, Lidia Sam Chao, and Derek Fai Wong. A survey on llm-generated text detection: Necessity, methods, and future directions.Computational Linguistics, 51(1):275–338, 2025

    2025

  9. [9]

    Evobench: Towards real-world llm-generated text detection bench- marking for evolving large language models

    Xiao Yu, Yi Yu, Dongrui Liu, Kejiang Chen, Weiming Zhang, Nenghai Yu, and Jing Shao. Evobench: Towards real-world llm-generated text detection bench- marking for evolving large language models. InFindings of the Association for Computational Linguistics: ACL 2025, pages 14605–14620, 2025

    2025

  10. [10]

    DSIPA: Detecting LLM-Generated Texts via Sentiment-Invariant Patterns Divergence Analysis

    Siyuan Li, Aodu Wulianghai, Guangyan Li, Xi Lin, Qinghua Mao, Yuliang Chen, Jun Wu, and Jianhua Li. Dsipa: Detecting llm-generated texts via sentiment- invariant patterns divergence analysis.arXiv preprint arXiv:2604.26328, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  11. [11]

    Gated Multimodal Units for Information Fusion

    John Arevalo, Thamar Solorio, Manuel Montes-y Gómez, and Fabio A González. Gated multimodal units for information fusion.arXiv preprint arXiv:1702.01992, 2017

    work page Pith review arXiv 2017

  12. [12]

    Mtikguard system: A transformer-based multimodal system for child-safe content moderation on tiktok.arXiv preprint arXiv:2511.17955, 2025

    Dat Thanh Nguyen, Nguyen Hung Lam, Anh Hoang-Thi Nguyen, and Trong-Hop Do. Mtikguard system: A transformer-based multimodal system for child-safe content moderation on tiktok.arXiv preprint arXiv:2511.17955, 2025

    work page arXiv 2025

  13. [13]

    Prdetect: Perturbation-robust llm-generated text detection based on syntax tree

    Xiang Li, Zhiyi Yin, Hexiang Tan, Shaoling Jing, Du Su, Yi Cheng, Huawei Shen, and Fei Sun. Prdetect: Perturbation-robust llm-generated text detection based on syntax tree. InFindings of the Association for Computational Linguistics: NAACL 2025, pages 8290–8301, 2025

    2025

  14. [14]

    A practical synthesis of detecting ai-generated textual, visual, and audio content.arXiv preprint arXiv:2504.02898, 2025

    Lele Cao. A practical synthesis of detecting ai-generated textual, visual, and audio content.arXiv preprint arXiv:2504.02898, 2025

    work page arXiv 2025

  15. [15]

    Spot- ting llms with binoculars: Zero-shot detection of machine-generated text

    Abhimanyu Hans, Avi Schwarzschild, Valeriia Cherepanova, Hamid Kazemi, Aniruddha Saha, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Spot- ting llms with binoculars: Zero-shot detection of machine-generated text. In International Conference on Machine Learning, pages 17519–17537. PMLR, 2024

    2024

  16. [16]

    Gltr: Statistical detection and visualization of generated text

    Sebastian Gehrmann, Hendrik Strobelt, and Alexander M Rush. Gltr: Statistical detection and visualization of generated text. InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 111–116, 2019

    2019

  17. [17]

    Model-agnostic sentiment distribution stability analysis for robust llm-generated texts detection

    Siyuan Li, Xi Lin, Guangyan Li, Zehao Liu, Aodu Wulianghai, Li Ding, Jun Wu, and Jianhua Li. Model-agnostic sentiment distribution stability analysis for robust llm-generated texts detection. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 35608–35616, 2026

    2026

  18. [18]

    ExaGPT: Example-Based Machine-Generated Text Detection for Human Interpretability

    Ryuto Koike, Masahiro Kaneko, Ayana Niwa, Preslav Nakov, and Naoaki Okazaki. Exagpt: Example-based machine-generated text detection for human inter- pretability.arXiv preprint arXiv:2502.11336, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    Adversarial watermarking transformer: To- wards tracing text provenance with data hiding

    Sahar Abdelnabi and Mario Fritz. Adversarial watermarking transformer: To- wards tracing text provenance with data hiding. In2021 IEEE Symposium on Security and Privacy (SP), pages 121–140. IEEE, 2021

    2021

  20. [20]

    A watermark for large language models

    John Kirchenbauer, Jonas Geiping, Yuxin Wen, Jonathan Katz, Ian Miers, and Tom Goldstein. A watermark for large language models. InInternational Conference on Machine Learning, pages 17061–17084. PMLR, 2023

    2023

  21. [21]

    Online detection of llm-generated texts via sequential hypothesis testing by betting

    Can Chen and Jun-Kun Wang. Online detection of llm-generated texts via sequential hypothesis testing by betting. InInternational Conference on Machine Learning, pages 9231–9276. PMLR, 2025

    2025

  22. [22]

    Beyond binary: Towards fine-grained llm-generated text detection via role recognition and involvement measurement

    Zihao Cheng, Li Zhou, Feng Jiang, Benyou Wang, and Haizhou Li. Beyond binary: Towards fine-grained llm-generated text detection via role recognition and involvement measurement. InProceedings of the ACM on Web Conference 2025, pages 2677–2688, 2025

    2025

  23. [23]

    Adadetectgpt: Adaptive detection of llm-generated text with statistical guarantees.arXiv preprint arXiv:2510.01268, 2025

    Hongyi Zhou, Jin Zhu, Pingfan Su, Kai Ye, Ying Yang, Shakeel AOB Gavioli- Akilagun, and Chengchun Shi. Adadetectgpt: Adaptive detection of llm-generated text with statistical guarantees.arXiv preprint arXiv:2510.01268, 2025

    work page arXiv 2025

  24. [24]

    A survey on image-text multimodal models.arXiv preprint arXiv:2309.15857, 2023

    Ruifeng Guo, Jingxuan Wei, Linzhuang Sun, Bihui Yu, Guiyong Chang, Dawei Liu, Sibo Zhang, Zhengbing Yao, Mingjun Xu, and Liping Bu. A survey on image-text multimodal models.arXiv preprint arXiv:2309.15857, 2023

    work page arXiv 2023

  25. [25]

    Paraphrasing evades detectors of ai-generated text, but retrieval is an effective defense.Advances in Neural Information Processing Systems, 36, 2024

    Kalpesh Krishna, Yixiao Song, Marzena Karpinska, John Wieting, and Mohit Iyyer. Paraphrasing evades detectors of ai-generated text, but retrieval is an effective defense.Advances in Neural Information Processing Systems, 36, 2024

    2024

  26. [26]

    Pald: Detection of text partially written by large language models

    Eric Lei, Hsiang Hsu, and Chun-Fu Chen. Pald: Detection of text partially written by large language models. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  27. [27]

    Glimpse: Enabling white-box methods to use proprietary models for zero-shot llm-generated text detection

    Guangsheng Bao, Yanbin Zhao, Juncai He, and Yue Zhang. Glimpse: Enabling white-box methods to use proprietary models for zero-shot llm-generated text detection. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  28. [28]

    Llm-as-a- coauthor: Can mixed human-written and machine-generated text be detected? pages 409–436

    Qihui Zhang, Chujie Gao, Dongping Chen, Yue Huang, Yixin Huang, Zhenyang Sun, Shilin Zhang, Weiye Li, Zhengyan Fu, Yao Wan, and Lichao Sun. Llm-as-a- coauthor: Can mixed human-written and machine-generated text be detected? pages 409–436. Association for Computational Linguistics, June 2024

    2024

  29. [29]

    Detecting fake content with relative entropy scoring

    Thomas Lavergne, Tanguy Urvoy, and François Yvon. Detecting fake content with relative entropy scoring. InProceedings of the 2008 International Conference on Uncovering Plagiarism, Authorship and Social Software Misuse-Volume 377, pages 27–31, 2008

    2008

  30. [30]

    Unifying human and statistical evaluation for natural language generation.arXiv preprint arXiv:1904.02792, 2019

    Tatsunori B Hashimoto, Hugh Zhang, and Percy Liang. Unifying human and statistical evaluation for natural language generation.arXiv preprint arXiv:1904.02792, 2019

    work page arXiv 1904

  31. [31]

    Auto- matic detection of machine generated text: A critical survey

    Ganesh Jawahar, Muhammad Abdul-Mageed, and VS Laks Lakshmanan. Auto- matic detection of machine generated text: A critical survey. InProceedings of the 28th International Conference on Computational Linguistics, pages 2296–2309, 2020

    2020

  32. [32]

    Eve: Towards end-to-end video subtitle extraction with vision-language models.arXiv preprint arXiv:2503.04058, 2025

    Haiyang Yu, Mengyang Zhao, Jinghui Lu, Ke Niu, Yanjie Wang, Weijie Yin, Weitao Jia, Teng Fu, Yang Liu, Jun Liu, et al. Eve: Towards end-to-end video subtitle extraction with vision-language models.arXiv preprint arXiv:2503.04058, 2025

    work page arXiv 2025

  33. [33]

    Asap: Advancing semantic alignment promotes multi-modal ma- nipulation detecting and grounding

    Zhenxing Zhang, Yaxiong Wang, Lechao Cheng, Zhun Zhong, Dan Guo, and Meng Wang. Asap: Advancing semantic alignment promotes multi-modal ma- nipulation detecting and grounding. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 4005–4014, 2025

    2025

  34. [34]

    PSF-Med: Measuring and Explaining Paraphrase Sensitivity in Medical Vision Language Models

    Binesh Sadanandan and Vahid Behzadan. Psf-med: Measuring and explain- ing paraphrase sensitivity in medical vision language models.arXiv preprint arXiv:2602.21428, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  35. [35]

    A survey on detection of llms-generated content

    Xianjun Yang, Liangming Pan, Xuandong Zhao, Haifeng Chen, Linda Petzold, William Yang Wang, and Wei Cheng. A survey on detection of llms-generated content. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 9786–9805, 2024

    2024

  36. [36]

    Gptzero: An ai text detector„ 2023

    Edward Tian. Gptzero: An ai text detector„ 2023. URL https://gptzero.me/

    2023

  37. [37]

    Emma Strubell, Ananya Ganesh, and Andrew McCallum

    Irene Solaiman, Miles Brundage, Jack Clark, Amanda Askell, Ariel Herbert-Voss, Jeff Wu, Alec Radford, Gretchen Krueger, Jong Wook Kim, Sarah Kreps, et al. Release strategies and the social impacts of language models.arXiv preprint arXiv:1908.09203, 2019

    work page arXiv 1908

  38. [38]

    Radar: Robust ai-text detection via adversarial learning.Advances in Neural Information Processing Systems, 36, 2024

    Xiaomeng Hu, Pin-Yu Chen, and Tsung-Yi Ho. Radar: Robust ai-text detection via adversarial learning.Advances in Neural Information Processing Systems, 36, 2024

    2024

  39. [39]

    Detecting gen- erated text via rewriting

    Chengzhi Mao, Carl Vondrick, Hao Wang, and Junfeng Yang. Detecting gen- erated text via rewriting. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  40. [40]

    Hlpd: Aligning llms to human language preference for machine-revised text detection

    Fangqi Dai, Xingjian Jiang, and Zizhuang Deng. Hlpd: Aligning llms to human language preference for machine-revised text detection. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 30440–30448, 2026

    2026

  41. [41]

    Detectgpt: Zero-shot machine-generated text detection using probability curvature

    Eric Mitchell, Yoonho Lee, Alexander Khazatsky, Christopher D Manning, and Chelsea Finn. Detectgpt: Zero-shot machine-generated text detection using probability curvature. InInternational Conference on Machine Learning, pages 24950–24962. PMLR, 2023

    2023

  42. [42]

    Fast- detectgpt: Efficient zero-shot detection of machine-generated text via conditional probability curvature

    Guangsheng Bao, Yanbin Zhao, Zhiyang Teng, Linyi Yang, and Yue Zhang. Fast- detectgpt: Efficient zero-shot detection of machine-generated text via conditional probability curvature. InThe Twelfth International Conference on Learning Repre- sentations, 2024

    2024

  43. [43]

    Biscope: Ai-generated text detection by checking memorization of preceding tokens.Advances in Neural Information Processing Systems, 37:104065–104090, 2024

    Hanxi Guo, Siyuan Cheng, Xiaolong Jin, Zhuo Zhang, Kaiyuan Zhang, Guanhong Tao, Guangyu Shen, and Xiangyu Zhang. Biscope: Ai-generated text detection by checking memorization of preceding tokens.Advances in Neural Information Processing Systems, 37:104065–104090, 2024

    2024

  44. [44]

    Paraphrasing attack resilience of various machine-generated text detection methods

    Andrii Shportko and Inessa Verbitsky. Paraphrasing attack resilience of various machine-generated text detection methods. InProceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 4: Student Research Workshop), pages 450–

    2025

  45. [45]

    URL https://aclanthology

    Association for Computational Linguistics, 2025. URL https://aclanthology. org/2025.naacl-srw.46/

    2025

  46. [46]

    Learning to rewrite: Generalized llm-generated text detection

    Chengzhi Mao et al. Learning to rewrite: Generalized llm-generated text detection. InProceedings of the 63rd Annual Meeting of the Association for Computational Lin- guistics (Volume 1: Long Papers), pages 5897–5912. Association for Computational Linguistics, 2025. URL https://aclanthology.org/2025.acl-long.322/

    2025

  47. [47]

    Styledecipher: Robust and explainable detection of llm-generated texts with stylistic analysis,

    Siyuan Li, Aodu Wulianghai, Xi Lin, Guangyan Li, Xiang Chen, Jun Wu, and Jianhua Li. Styledecipher: Robust and explainable detection of llm-generated texts with stylistic analysis.arXiv preprint arXiv:2510.12608, 2025

    work page arXiv 2025

  48. [48]

    Deep kernel relative test for machine-generated text detection

    Yiliao Song, Zhenqiao Yuan, Shuhai Zhang, Zhen Fang, Jun Yu, and Feng Liu. Deep kernel relative test for machine-generated text detection. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  49. [49]

    Ghostbuster: Detecting text ghostwritten by large language models

    Vivek Verma, Eve Fleisig, Nicholas Tomlin, and Dan Klein. Ghostbuster: Detecting text ghostwritten by large language models. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 1702–1717, 2024

    2024

  50. [50]

    N-gram feature selection for author- ship identification

    John Houvardas and Efstathios Stamatatos. N-gram feature selection for author- ship identification. InInternational conference on artificial intelligence: Methodol- ogy, systems, and applications, pages 77–86. Springer, 2006

    2006

  51. [51]

    Ivypanda essays dataset

    IvyPanda. Ivypanda essays dataset. Hugging Face, 2022. URL https://huggingface. co/datasets/qwedsacf/ivypanda-essays

    2022

  52. [52]

    Character-level convolutional networks for text classification.Advances in neural information processing systems, 28, 2015

    Xiang Zhang, Junbo Zhao, and Yann LeCun. Character-level convolutional networks for text classification.Advances in neural information processing systems, 28, 2015

    2015

  53. [53]

    Visual news: Benchmark and challenges in news image captioning

    Fuxiao Liu, Yinghan Wang, Tianlu Wang, and Vicente Ordonez. Visual news: Benchmark and challenges in news image captioning. In Marie-Francine Moens, Xuanjing Huang, Lucia Specia, and Scott Wen-tau Yih, editors,Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6761– 6771, Online and Punta Cana, Dominican Republic...

    work page doi:10.18653/v1/2021.emnlp-main.542 2021

  54. [54]

    Raid: A shared bench- mark for robust evaluation of machine-generated text detectors.arXiv preprint arXiv:2405.07940, 2024

    Liam Dugan, Alyssa Hwang, Filip Trhlik, Josh Magnus Ludan, Andrew Zhu, Hainiu Xu, Daphne Ippolito, and Chris Callison-Burch. Raid: A shared bench- mark for robust evaluation of machine-generated text detectors.arXiv preprint arXiv:2405.07940, 2024

    work page arXiv 2024

  55. [55]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    Leland McInnes, John Healy, and James Melville. Umap: Uniform manifold approximation and projection for dimension reduction, 2020. URL https://arxiv. org/abs/1802.03426

    work page internal anchor Pith review arXiv 2020