Research map / Video / Video-language understanding
E-VAds: An E-commerce Short Videos Understanding Benchmark for MLLMs
Xianjie Liu, Yiman Hu, Liang Wu, Ping Hu, Yixiong Zou, Jian Xu, Bo Zheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train with RL using multi-grained reward G on top of SFT | Qwen3-VL-8B · E-VAds test (ALL tasks) | Strict score: .153 -> .320 | Improved |
| use SFT with thinking flow instead of direct answer (ablation) | Qwen2.5-VL-7B · E-VAds test (ALL tasks) | Strict score: .136 -> .131 | No clear effect |
| use more lenient single-granularity RL reward (R5 instead of S) (ablation) | Qwen2.5-VL-7B · E-VAds test (ALL tasks) | Strict score: .136 -> .152 | Improved |
| use multi-grained reward G instead of R5 (ablation) | E-VAds-R1 (Qwen2.5-VL-7B) · E-VAds test (ALL tasks) | Strict score: .163 -> .180 | Improved |
| add explicit reasoning-trace reward (ablation) | E-VAds-R1 (Qwen2.5-VL-7B) · E-VAds test (ALL tasks) | Strict score: .163 -> .169 (R5 rewards) | Improved |
| use G for think and answer rewards instead of R5 (ablation) | E-VAds-R1 (Qwen2.5-VL-7B) · E-VAds test (ALL tasks) | Strict score: .169 -> .192 | Improved |
| weight answer reward 0.8 and thinking reward 0.2 instead of equal weights (ablation) | E-VAds-R1 (Qwen2.5-VL-7B) · E-VAds test (ALL tasks) | Strict score: .192 -> .193 | Improved |
| add ASR transcript to video input (ablation) | E-VAds-R1 · E-VAds test (ALL tasks) | Strict score: .257 -> .320 | Improved |
| add video to ASR-only input (ablation) | E-VAds-R1 · E-VAds test (ALL tasks) | Strict score: .162 -> .257 | Improved |
| scale model from 7B to 32B | Qwen2.5-VL-Instruct · E-VAds test (ALL tasks) | Strict score: .111 -> .142 | No clear effect |
| use thinking variant instead of instruct | Qwen3-VL-8B · E-VAds test (ALL tasks) | Strict score: 0.153 -> 0.186 | Improved |
| use thinking variant instead of instruct | Qwen3-Omni-30B-A3B · E-VAds test (ALL tasks) | Strict score: .205 -> .207 | No clear effect |
Closest papers by meaning
- VideoWebArena: Evaluating Long Context Multimodal Agents with Video Understanding Web Tasks (ICLR 2025 · similarity 0.83)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · similarity 0.83)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.82)
- Needle In A Video Haystack: A Scalable Synthetic Evaluator for Video MLLMs (ICLR 2025 · similarity 0.82)
- OVO-Bench: How Far is Your Video-LLMs from Real-World Online Video Understanding? (CVPR 2025 · similarity 0.82)
- VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning? (ICLR 2026 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
