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). Unlabeled 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.

Hateful videos have become prevalent on online platforms, highlighting an urgent need for effective detection. However, existing studies primarily focus on binary classification and fail to provide contextual rationales that reveal the implicit meanings behind these judgments, significantly undermining model explainability. To fill this gap, we aim to achieve explainable hateful video detection, enabling models to provide contextual rationales that integrate relevant evidence and logical reasoning alongside decisions. This approach can comprehensively enhance the understanding of video content and the explainability of the decision-making process. We first introduce two datasets, Ex-HateMM and Ex-ImpliHateVid, for explainable hateful video detection. Each dataset provides fine-grained annotations of multimodal harmful elements, along with contextual rationales. We then propose an Information Augmentation and Reasoning Enhancement (IARE) framework designed for explainable detection. The framework employs an information augmentation phase that leverages the multimodal chain-of-thought to integrate harmful elements, thereby enriching rationale evidence. Additionally, IARE incorporates a reasoning enhancement phase, in which Direct Preference Optimization guides the model toward correct reasoning paths and away from incorrect ones, thereby improving the logical coherence of its justifications. We conduct extensive experiments on the two datasets, comparing multiple baselines with our proposed IARE framework. The results demonstrate that IARE achieves state-of-the-art performance while also generating accurate rationales.

Patronizing and Condescending Language (PCL) is a form of discriminatory toxic speech targeting vulnerable groups, threatening both online and offline safety. While toxic speech research has mainly focused on overt toxicity, such as hate speech, microaggressions in the form of PCL remain underexplored. Additionally, dominant groups’ discriminatory facial expressions and attitudes toward vulnerable communities can be more impactful than verbal cues, yet these frame features are often overlooked. In this paper, we introduce the PCLMM dataset, the first Chinese multimodal dataset for PCL, consisting of 715 annotated videos from Bilibili, with high-quality PCL facial frame spans. We also propose the MultiPCL detector, featuring a facial expression detection module for PCL recognition, demonstrating the effectiveness of modality complementarity in this challenging task. Our work makes an important contribution to advancing microaggression detection within the domain of toxic speech.

Haohao Zhu, Junyu Lu, Zeyuan Zeng, Zewen Bai, Xiaokun Zhang, Liang Yang, Hongfei Lin. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Hateful memes are prevalent on the Internet, raising the urgent need for effective detection.Given their implicit nature, incorporating rationales with background knowledge is crucial for enhancing model understanding.However, existing methods often suffer from limited quality of external rationales and misalignment with original meme information.These challenges hinder model comprehension, leading to reduced accuracy and explainability.To address these challenges, we propose a Multimodal Multi-agent Knowledge Enhanced (M2KE) framework for hateful meme detection.M2KE introduces a multi-agent rationale discovery mechanism to extract high-quality rationales relevant to meme content and an adaptive knowledge interaction mechanism to ensure alignment between original meme information and external rationales.Specifically, multi-agent rationale discovery mechanism improves the reliability of rationales by collaboratively verifying and refining them with multiple agents, supported by large language models (LLMs) due to their extensive knowledge.And adaptive knowledge interaction mechanism uses information entropy to dynamically balance the model's attention between original meme information and external rationales, preventing over-reliance on rationales and enabling a more comprehensive understanding.Experimental results on three datasets demonstrate that M2KE significantly outperforms existing models.Further analysis underscores the importance of effectively integrating accurate rationales to enhance model performance.Disclaimer: Samples in this paper may be considered offensive.

Researchers have attempted to mitigate lexical bias in toxic language detection (TLD). However, existing methods fail to disentangle the “useful” and “misleading” impact of lexical bias on model decisions. Therefore, they do not effectively exploit the positive effects of the bias and lead to a degradation in the detection performance of the debiased model. In this paper, we propose a Counterfactual Causal Debiasing Framework (CCDF) to mitigate lexical bias in TLD. It preserves the “useful impact” of lexical bias and eliminates the “misleading impact”. Specifically, we first represent the total effect of the original sentence and biased tokens on decisions from a causal view. We then conduct counterfactual inference to exclude the direct causal effect of lexical bias from the total effect. Empirical evaluations demonstrate that the debiased TLD model incorporating CCDF achieves state-of-the-art performance in both accuracy and fairness compared to competitive baselines applied on several vanilla models. The generalization capability of our model outperforms current debiased models for out-of-distribution data.

Harmful memes have proliferated on the Chinese Internet, while research on detecting Chinese harmful memes significantly lags behind due to the absence of reliable datasets and effective detectors.To this end, we present the comprehensive detection of Chinese harmful memes.We introduce ToxiCN MM, the first Chinese harmful meme dataset, which consists of 12,000 samples with fine-grained annotations for meme types. Additionally, we propose a baseline detector, Multimodal Knowledge Enhancement (MKE), designed to incorporate contextual information from meme content, thereby enhancing the model's understanding of Chinese memes.In the evaluation phase, we conduct extensive quantitative experiments and qualitative analyses on multiple baselines, including LLMs and our MKE. Experimental results indicate that detecting Chinese harmful memes is challenging for existing models, while demonstrating the effectiveness of MKE.

The samples presented by this paper may be considered offensive or vulgar.The widespread dissemination of toxic online posts is increasingly damaging to society.However, research on detecting toxic language in Chinese has lagged significantly.Existing datasets lack fine-grained annotation of toxic types and expressions, and ignore the samples with indirect toxicity.In addition, it is crucial to introduce lexical knowledge to detect the toxicity of posts, which has been a challenge for researchers.In this paper, we facilitate the finegrained detection of Chinese toxic language.First, we build MONITOR TOXIC FRAME, a hierarchical taxonomy to analyze toxic types and expressions.Then, a fine-grained dataset TOXICN is presented, including both direct and indirect toxic samples.We also build an insult lexicon containing implicit profanity and propose Toxic Knowledge Enhancement (TKE) as a benchmark, incorporating the lexical feature to detect toxic language.In the experimental stage, we demonstrate the effectiveness of TKE.After that, a systematic quantitative and qualitative analysis of the findings is given. 1

The fast spread of hate speech on social media impacts the Internet environment and our society by increasing prejudice and hurting people. Detecting hate speech has aroused broad attention in the field of natural language processing. Although hate speech detection has been addressed in recent work, this task still faces two inherent unsolved challenges. The first challenge lies in the complex semantic information conveyed in hate speech, particularly the interference of insulting words in hate speech detection. The second challenge is the imbalanced distribution of hate speech and non-hate speech, which may significantly deteriorate the performance of models. To tackle these challenges, we propose a novel dual contrastive learning (DCL) framework for hate speech detection. Our framework jointly optimizes the self-supervised and the supervised contrastive learning loss for capturing span-level information beyond the token-level emotional semantics used in existing models, particularly detecting speech containing abusive and insulting words. Moreover, we integrate the focal loss into the dual contrastive learning framework to alleviate the problem of data imbalance. We conduct experiments on two publicly available English datasets, and experimental results show that the proposed model outperforms the state-of-the-art models and precisely detects hate speeches.