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

2026-05-07

AI Summary

This work addresses the challenge of detecting large language model–generated text in multimedia content moderation, particularly the limited robustness of existing methods against paraphrasing and adversarial attacks. The authors propose a lightweight style consistency modeling approach that, for the first time, introduces style consistency profiling into AI-generated text detection. By integrating discrete stylistic features with continuous semantic signals under multimodal guidance, the method effectively captures the stability of textual style. The resulting model achieves an optimal balance among lightweight design, cross-domain generalization, and adversarial robustness. Experimental results demonstrate that it outperforms current state-of-the-art methods by up to 11.79% in cross-domain detection performance on real-world news and movie datasets, while also exhibiting strong effectiveness in human-AI mixed and adversarial scenarios.

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.

Problem

• LLM-generated text detection
• multimedia moderation
• adversarial robustness
• stylistic consistency
• cross-domain generalization

Innovation

• stylistic consistency
• LLM-generated text detection
• adversarial robustness
• lightweight profiling
• cross-domain generalization