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

2026-04-29

AI Summary

Current methods for detecting text generated by large language models exhibit vulnerability to adversarial perturbations, paraphrasing attacks, and cross-domain transfer, often relying on model parameters or extensive labeled data. This work proposes a training-free, black-box detection framework that leverages, for the first time, the differential stability of sentiment distributions under controlled stylistic perturbations as a discriminative signal. By introducing two unsupervised metrics—sentiment distribution consistency and retention—the method enables robust, zero-shot detection without requiring access to internal model information or annotated data. Experimental results demonstrate that the approach achieves up to a 49.89% improvement in F1 score across five text domains and exhibits strong generalization and resilience against attacks on state-of-the-art models, including GPT-5.2 and Gemini-1.5-pro.

Abstract

The rapid advancement of large language models (LLMs) presents new security challenges, particularly in detecting machine-generated text used for misinformation, impersonation, and content forgery. Most existing detection approaches struggle with robustness against adversarial perturbation, paraphrasing attacks, and domain shifts, often requiring restrictive access to model parameters or large labeled datasets. To address this, we propose DSIPA, a novel training-free framework that detects LLM-generated content by quantifying sentiment distributional stability under controlled stylistic variation. It is based on the observation that LLMs typically exhibit more emotionally consistent outputs, while human-written texts display greater affective variation. Our framework operates in a zero-shot, black-box manner, leveraging two unsupervised metrics, sentiment distribution consistency and sentiment distribution preservation, to capture these intrinsic behavioral asymmetries without the need for parameter updates or probability access. Extensive experiments are conducted on state-of-the-art proprietary and open-source models, including GPT-5.2, Gemini-1.5-pro, Claude-3, and LLaMa-3.3. Evaluations on five domains, such as news articles, programming code, student essays, academic papers, and community comments, demonstrate that DSIPA improves F1 detection scores by up to 49.89% over baseline methods. The framework exhibits superior generalizability across domains and strong resilience to adversarial conditions, providing a robust and interpretable behavioral signal for secure content identification in the evolving LLM landscape.

Problem

• LLM-generated text detection
• adversarial robustness
• domain shift
• misinformation
• content forgery

Innovation

• sentiment-invariant patterns
• zero-shot detection
• black-box LLM detection
• distributional stability
• unsupervised metrics