Multi-Agent VLMs Guided Self-Training with PNU Loss for Low-Resource Offensive Content Detection

2025-11-14

AI Summary

To address the scarcity of labeled offensive content in low-resource social media scenarios, this paper proposes a multi-agent vision-language model (MA-VLM) framework for collaborative pseudo-labeling. Methodologically, it employs dual-perspective MA-VLMs—user and moderator—to generate high-confidence pseudo-labels; introduces a Positive-Negative-Unlabeled (PNU) loss to jointly optimize positive, negative, and unlabeled samples; and integrates pseudo-label consistency filtering with iterative self-training using a lightweight classifier. The key contributions are multi-perspective pseudo-label augmentation and PNU-driven noise-robust learning. Evaluated on multiple benchmark datasets, the method achieves substantial performance gains over state-of-the-art approaches using only a small number of annotated examples, approaching the accuracy of fully supervised large models. This demonstrates its effectiveness and generalizability under low-resource settings.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Adversarial Agents

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Accurate detection of offensive content on social media demands high-quality labeled data; however, such data is often scarce due to the low prevalence of offensive instances and the high cost of manual annotation. To address this low-resource challenge, we propose a self-training framework that leverages abundant unlabeled data through collaborative pseudo-labeling. Starting with a lightweight classifier trained on limited labeled data, our method iteratively assigns pseudo-labels to unlabeled instances with the support of Multi-Agent Vision-Language Models (MA-VLMs). Un-labeled data on which the classifier and MA-VLMs agree are designated as the Agreed-Unknown set, while conflicting samples form the Disagreed-Unknown set. To enhance label reliability, MA-VLMs simulate dual perspectives, moderator and user, capturing both regulatory and subjective viewpoints. The classifier is optimized using a novel Positive-Negative-Unlabeled (PNU) loss, which jointly exploits labeled, Agreed-Unknown, and Disagreed-Unknown data while mitigating pseudo-label noise. Experiments on benchmark datasets demonstrate that our framework substantially outperforms baselines under limited supervision and approaches the performance of large-scale models

Problem

• Detecting offensive content with scarce labeled data
• Leveraging unlabeled data through collaborative pseudo-labeling
• Mitigating pseudo-label noise using multi-agent VLMs

Innovation

• Self-training framework leverages unlabeled data
• Multi-Agent VLMs simulate dual perspectives for labeling
• PNU loss jointly optimizes labeled and pseudo-labeled data