A new benchmark of 2,833 evasive text samples and 13,961 images shows current LLMs and VLMs frequently miss veiled policy violations in Chinese e-commerce ads.
Qwen3.https://qwenlm.github.io/blog/qwen3//, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
EVADE-Bench: Multimodal Benchmark for Evaluating and Enhancing Evasive Content Detection
A new benchmark of 2,833 evasive text samples and 13,961 images shows current LLMs and VLMs frequently miss veiled policy violations in Chinese e-commerce ads.