LuxVeri at GenAI Detection Task 3: Cross-Domain Detection of AI-Generated Text Using Inverse Perplexity-Weighted Ensemble of Fine-Tuned Transformer Models

TL;DR

This paper introduces an ensemble of fine-tuned transformer models with inverse perplexity weighting to improve cross-domain detection of AI-generated text, achieving competitive results in the COLING-2025 workshop.

Contribution

It proposes a novel inverse perplexity-weighted ensemble method for transformer models to enhance cross-domain AI-generated text detection accuracy.

Findings

Achieved TPR of 0.826 in non-adversarial detection

Achieved TPR of 0.801 in adversarial detection

Demonstrated effectiveness of inverse perplexity weighting

Abstract

This paper presents our approach for Task 3 of the GenAI content detection workshop at COLING-2025, focusing on Cross-Domain Machine-Generated Text (MGT) Detection. We propose an ensemble of fine-tuned transformer models, enhanced by inverse perplexity weighting, to improve classification accuracy across diverse text domains. For Subtask A (Non-Adversarial MGT Detection), we combined a fine-tuned RoBERTa-base model with an OpenAI detector-integrated RoBERTa-base model, achieving an aggregate TPR score of 0.826, ranking 10th out of 23 detectors. In Subtask B (Adversarial MGT Detection), our fine-tuned RoBERTa-base model achieved a TPR score of 0.801, securing 8th out of 22 detectors. Our results demonstrate the effectiveness of inverse perplexity-based weighting for enhancing generalization and performance in both non-adversarial and adversarial MGT detection, highlighting the potential for transformer models in cross-domain AI-generated content detection.