LATIDIA · Investigación

TrustMod-SM: un punto de referencia multieje para evaluar la confiabilidad de los LLM en la moderación de contenido de redes sociales

arXiv: 2611.12970v1Tipo de anuncio: nuevo Resumen: los modelos de lenguaje grandes (LLM) muestran un gran potencial para la moderación de contenido de redes sociales, pero su confiabilidad más allá de la precisión general sigue siendo poco explorada. Presentamos

La noticia

arXiv:2611.12970v1 Announce Type: new Abstract: Large Language Models (LLMs) show strong potential for social media content moderation, but their trustworthiness beyond overall accuracy remains underexplored. We introduce TrustMod-SM, a benchmark of roughly 29,000 instances across eight datasets, to evaluate LLM moderators on five dimensions: trustfulness, fairness, safety, robustness, and context integrity. Evaluating 16 open-weight models (0.5B-284B parameters, spanning text-only and vision-language architectures) and one closed frontier model, we find persistent trustworthiness gaps. Higher detection accuracy often coincides with greater demographic disparity, and models frequently comply with concealment or exaggeration prompts, assign high confidence to wrong predictions, rarely escalate ambiguous cases, and struggle to separate hate speech from counter-speech or reclaimed