arXiv subjects

Kaichun Wang

Publications and source records attributed to Kaichun Wang.

3 recordsLinked to original sources

arXiv2026

Whitewashing Hate, Smearing Harmless Content: Annotator-Style Rebuttal Attacks on LLM-Based Moderation

Large language models (LLMs) are increasingly used for hate speech moderation, often within human--AI workflows in which reviewers provide feedback before a final decision. Such feedback introduces two manipulation directions: whitewashing hateful content as normal and smearing normal content as hateful. This study examines the susceptibility of initially correct model judgments to annotator-style rebuttals and analyzes whether attack effectiveness differs across manipulation directions. We introduce a rejudge protocol that extends direct contradiction with decision-boundary perturbations and adversarial rationales. Experiments with multiple LLMs on two hate speech datasets show that annotator-style rebuttals substantially degrade moderation performance, with stronger effects in multi-turn settings. The results further reveal stable, model-specific asymmetries between whitewashing and smearing across attack configurations, indicating distinct directional vulnerability patterns. Explicit reasoning prompts and defensive instructions reduce these effects but do not eliminate them. These findings highlight the need for direction-aware safeguards and dedicated feedback-robustness evaluation in human--AI moderation workflows.

cs.CL↗

arXiv2025

From Events to Trending: A Multi-Stage Hotspots Detection Method Based on Generative Query Indexing

LLM-based conversational systems have become a popular gateway for information access, yet most existing chatbots struggle to handle news-related trending queries effectively. To improve user experience, an effective trending query detection method is urgently needed to enable differentiated processing of such target traffic. However, current research on trending detection tailored to the dialogue system scenario remains largely unexplored, and methods designed for traditional search engines often underperform in conversational contexts due to radically distinct query distributions and expression patterns. To fill this gap, we propose a multi-stage framework for trending detection, which achieves systematic optimization from both offline generation and online identification perspectives. Specifically, our framework first exploits selected hot events to generate index queries, establishing a key bridge between static events and dynamic user queries. It then employs a retrieval matching mechanism for real-time online detection of trending queries, where we introduce a cascaded recall and ranking architecture to balance detection efficiency and accuracy. Furthermore, to better adapt to the practical application scenario, our framework adopts a single-recall module as a cold-start strategy to collect online data for fine-tuning the reranker. Extensive experiments demonstrate that our framework significantly outperforms baseline methods in both offline evaluations and online A/B tests, and user satisfaction is relatively improved by 27\% in terms of positive-negative feedback ratio.

cs.IR↗

arXiv2025

Is LLM an Overconfident Judge? Unveiling the Capabilities of LLMs in Detecting Offensive Language with Annotation Disagreement

Large Language Models (LLMs) have become essential for offensive language detection, yet their ability to handle annotation disagreement remains underexplored. Disagreement samples, which arise from subjective interpretations, pose a unique challenge due to their ambiguous nature. Understanding how LLMs process these cases, particularly their confidence levels, can offer insight into their alignment with human annotators. This study systematically evaluates the performance of multiple LLMs in detecting offensive language at varying levels of annotation agreement. We analyze binary classification accuracy, examine the relationship between model confidence and human disagreement, and explore how disagreement samples influence model decision-making during few-shot learning and instruction fine-tuning. Our findings reveal that LLMs struggle with low-agreement samples, often exhibiting overconfidence in these ambiguous cases. However, utilizing disagreement samples in training improves both detection accuracy and model alignment with human judgment. These insights provide a foundation for enhancing LLM-based offensive language detection in real-world moderation tasks.

cs.CL↗