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

2026-08-11

AI Summary

This work addresses the challenges of detecting large language model (LLM)-generated text in real-world Chinese scenarios, where distribution shifts and conflicting signals significantly degrade detector performance. To tackle these issues, the authors propose EVIL-Detect, a novel framework that integrates edit-level regression, zero-shot likelihood contrast, lexical statistical features, and conservative textual rules. The method further introduces a conflict-aware multi-signal ensemble mechanism and a calibrated decision boundary to enhance robustness and stability under severe out-of-distribution shifts. Evaluated on the NLPCC 2026 Shared Task 6 official benchmark, EVIL-Detect achieves state-of-the-art performance with a macro-F1 score of 0.8888, ranking first among all participating systems.

Abstract

The rapid development of large language models (LLMs) has increased the need for reliable detection of LLM-generated text, especially in realistic Chinese scenarios involving human-written text (HWT), LLM-generated text (LGT), and LLM-refined text (HLT). This paper presents EVIL-Detect, a multi-signal ensemble framework with conflict-aware fusion for NLPCC 2026 Shared Task 6. The system integrates edit-extent regression, zero-shot likelihood-contrast signals, lexical statistics, and conservative text rules. With calibrated decision boundaries and conflict-aware integration, our system improves robustness under strong out-of-distribution shifts, achieving a macro-F1 score of 0.8888 and ranking first in the official evaluation. Our code is available at https://github.com/bbbbhrrrr/evildetect.

Problem

• LLM-generated text detection
• Chinese text
• human-written text
• LLM-refined text
• out-of-distribution robustness

Innovation

• multi-signal ensemble
• conflict-aware fusion
• LLM-generated text detection
• out-of-distribution robustness
• zero-shot likelihood contrast