New benchmark VARM-Bench evaluates AI reasoning for Chinese abusive speech moderation

Researchers have introduced VARM-Bench, a new benchmark designed to evaluate the verifiable structured reasoning capabilities of AI models in moderating Chinese abusive speech. Unlike previous benchmarks that focused on classification or categorization, VARM-Bench provides explicit anchors for six key moderation decisions within each instance, including target, target type, author stance, and harmfulness label. The benchmark employs a deterministic protocol to assess the correctness and completeness of these moderation records, aiming to ensure auditable and reproducible evaluations that go beyond simple label accuracy. AI

IMPACT This benchmark aims to improve the reliability and auditability of AI models used for content moderation, particularly for Chinese abusive speech.

RANK_REASON The cluster describes a new academic paper introducing a benchmark for AI research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Standard Chinese
• VARM-Bench

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →