MD-ProTector: Positioning Multiple Data-Driven Prototypes for LLM-Generated Text Detection

2026-08-11

AI Summary

This work addresses the challenge that existing detection methods struggle to effectively distinguish between human- and large language model–generated text in large-scale, multi-style, multilingual, and multi-model settings, while conventional binary classification approaches overlook intra-class diversity. To this end, the paper proposes MD-ProTector, which uniquely integrates a multi-prototype mechanism with explicit modeling of intra-class variation. Specifically, multiple trainable prototypes are introduced for each class within the encoder embedding space, accompanied by a novel Prototype Positioning loss that automatically disentangles inter-class structure from intra-class variability, thereby yielding finer decision boundaries. Evaluated across three benchmarks and five diverse settings—spanning domains, generators, languages, and adversarial perturbations—the method achieves state-of-the-art performance in terms of average recall (AvgRec) and AUROC, while attaining the lowest FPR95, demonstrating significantly enhanced generalization capability.

Abstract

As LLM-generated content becomes more sophisticated, detection systems for distinguishing those texts from human-written text must operate at scale while handling diverse writing styles, domains, languages, and generator models. Input-only encoder detectors are suitable for practical deployment setting, but standard binary classification supplies only the class label and does not explicitly organize the substantial variation within either class. We propose MD-ProTector, which represents each class with multiple trainable reference vectors in the encoder embedding space, referred to as prototypes. These prototypes provide separate decision boundaries for different groups of texts within the same class. However, adding multiple prototypes alone does not determine which variation each prototype should represent. MD-ProTector addresses this problem with Prototype Positioning loss, which separates class-level structure from the within-class variation that differentiates individual prototypes. Evaluated across five settings from three large-scale benchmarks covering domain, generator, language, and adversarial variation, MD-ProTector achieves the highest AvgRec on MAGE CDCM and RAID and the highest AUROC and lowest FPR95 on RAID among the compared encoder-based methods.

Problem

• LLM-generated text detection
• prototype-based classification
• within-class variation
• encoder-based detection
• text authenticity

Innovation

• prototype-based detection
• LLM-generated text detection
• within-class variation modeling
• Prototype Positioning loss
• encoder-based detector