Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Span-level Detection of AI-generated Scientific Text via Contrastive Learning and Structural Calibration

Zhen Yin, Shenghua Wang

TL;DR

Sci-SpanDet combines section-conditioned stylistic modeling with multi-level contrastive learning to capture nuanced human-AI differences while mitigating topic dependence, thereby enhancing cross-domain robustness and shows strong resilience under adversarial rewriting.

Abstract

The rapid adoption of large language models (LLMs) in scientific writing raises serious concerns regarding authorship integrity and the reliability of scholarly publications. Existing detection approaches mainly rely on document-level classification or surface-level statistical cues; however, they neglect fine-grained span localization, exhibit weak calibration, and often fail to generalize across disciplines and generators. To address these limitations, we present Sci-SpanDet, a structure-aware framework for detecting AI-generated scholarly texts. The proposed method combines section-conditioned stylistic modeling with multi-level contrastive learning to capture nuanced human-AI differences while mitigating topic dependence, thereby enhancing cross-domain robustness. In addition, it integrates BIO-CRF sequence labeling with pointer-based boundary decoding and confidence calibration to enable precise span-level detection and reliable probability estimates. Extensive experiments on a newly constructed cross-disciplinary dataset of 100,000 annotated samples generated by multiple LLM families (GPT, Qwen, DeepSeek, LLaMA) demonstrate that Sci-SpanDet achieves state-of-the-art performance, with F1(AI) of 80.17, AUROC of 92.63, and Span-F1 of 74.36. Furthermore, it shows strong resilience under adversarial rewriting and maintains balanced accuracy across IMRaD sections and diverse disciplines, substantially surpassing existing baselines. To ensure reproducibility and to foster further research on AI-generated text detection in scholarly documents, the curated dataset and source code will be publicly released upon publication.

Span-level Detection of AI-generated Scientific Text via Contrastive Learning and Structural Calibration

TL;DR

Sci-SpanDet combines section-conditioned stylistic modeling with multi-level contrastive learning to capture nuanced human-AI differences while mitigating topic dependence, thereby enhancing cross-domain robustness and shows strong resilience under adversarial rewriting.

Abstract

The rapid adoption of large language models (LLMs) in scientific writing raises serious concerns regarding authorship integrity and the reliability of scholarly publications. Existing detection approaches mainly rely on document-level classification or surface-level statistical cues; however, they neglect fine-grained span localization, exhibit weak calibration, and often fail to generalize across disciplines and generators. To address these limitations, we present Sci-SpanDet, a structure-aware framework for detecting AI-generated scholarly texts. The proposed method combines section-conditioned stylistic modeling with multi-level contrastive learning to capture nuanced human-AI differences while mitigating topic dependence, thereby enhancing cross-domain robustness. In addition, it integrates BIO-CRF sequence labeling with pointer-based boundary decoding and confidence calibration to enable precise span-level detection and reliable probability estimates. Extensive experiments on a newly constructed cross-disciplinary dataset of 100,000 annotated samples generated by multiple LLM families (GPT, Qwen, DeepSeek, LLaMA) demonstrate that Sci-SpanDet achieves state-of-the-art performance, with F1(AI) of 80.17, AUROC of 92.63, and Span-F1 of 74.36. Furthermore, it shows strong resilience under adversarial rewriting and maintains balanced accuracy across IMRaD sections and diverse disciplines, substantially surpassing existing baselines. To ensure reproducibility and to foster further research on AI-generated text detection in scholarly documents, the curated dataset and source code will be publicly released upon publication.

Paper Structure

This paper contains 14 sections, 10 equations, 8 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. AI-generated text detection
  4. Techniques for style and structure-aware modeling
  5. Proposed Method
  6. Overall
  7. Section-aware style discrimination
  8. Span-level localization
  9. Boundary confidence and calibration
  10. Experiment and Analysis
  11. Experimental setup
  12. Main results
  13. Ablation studies and analysis
  14. Conclusion

Figures (8)

  • Figure 1: Schematic illustration of the Sci-SpanDet framework for structure-aware detection of AI-generated content in scholarly text.
  • Figure 2: Overview of the proposed Sci-SpanDet framework for AIGC detection, comprising section-aware style discrimination, span-level localization, and boundary confidence modeling.
  • Figure 3: Effect of contextual weighting on F1(AI).
  • Figure 4: Effect of the balancing coefficient $\alpha$ in the style-aware loss.
  • Figure 5: Effect of rewriting intensity on detection performance.
  • ...and 3 more figures