E-VAds: A New Benchmark for Understanding E-Commerce Short Videos Using Multi-Modal LLMs
By
[Submitted on 9 Feb 2026 (v1), last revised 17 Jun 2026 (this version, v4)]
Summary
This paper introduces E-VAds, the first benchmark specifically designed for understanding e-commerce short videos. The authors propose a multi-modal information density assessment framework showing that e-commerce content has higher density across visual, audio, and textual modalities than mainstream datasets. They curated 3,961 high-quality videos from Taobao across various product categories and used a multi-agent system to generate 19,785 open-ended Q&A pairs across five tasks. They also developed E-VAds-R1, an RL-based reasoning model with a multi-grained reward design (MG-GRPO) that achieves a 109.2% performance gain in commercial intent reasoning with only a few hundred training samples.
Source
bskyE-VAds: A New Benchmark for Understanding E-Commerce Short Videos Using Multi-Modal LLMsarxiv.orgKey quotes
· 4 pulledE-commerce short videos represent a high-revenue segment of the online video industry characterized by a goal-driven format and dense multi-modal signals.
Current models often struggle with these videos because existing benchmarks focus primarily on general-purpose tasks and neglect the reasoning of commercial intent.
Our evaluation reveals that e-commerce content exhibits substantially higher density across visual, audio, and textual modalities compared to mainstream datasets, establishing a more challenging frontier for video understanding.
Experimental results demonstrate that E-VAds-R1 achieves a 109.2% performance gain in commercial intent reasoning with only a few hundred training samples.
You might also wanna read
Alibaba's Qwen3-VL AI Model Demonstrates Advanced Video Analysis Capabilities
Alibaba has released a technical report on its Qwen3-VL multimodal AI model, demonstrating exceptional capabilities in processing and analyz
the-decoder.com·6mo ago
Alibaba's Tongyi DeepResearch: Open-Source AI Research Agent Matches OpenAI Performance
Alibaba's Tongyi DeepResearch is presented as the first fully open-source web agent that achieves performance comparable to OpenAI's DeepRes
tongyi-agent.github.io·7mo ago
DatBench: A New Framework for More Faithful and Efficient Vision-Language Model Evaluation
The article introduces DatBench, a new evaluation framework for vision-language models (VLMs) that addresses critical issues in current eval
arxiv.org·5mo ago
GLM-5V-Turbo: A Native Multimodal Foundation Model for Agentic AI Tasks
GLM-5V-Turbo is a new multimodal foundation model developed by the GLM-V Team that integrates perception, reasoning, planning, tool use, and
arXiv.org·1mo ago
SkillsBench: A Benchmark for Evaluating AI Agent Skills Across Diverse Tasks
SkillsBench is a new benchmark for evaluating how well AI agent skills work across diverse tasks. The benchmark includes 86 tasks across 11
arxiv.org·4mo ago
StreamingVLM: Real-Time Vision-Language Model for Infinite Video Stream Processing
StreamingVLM is a new vision-language model designed for real-time understanding of infinite video streams, addressing the computational cha
arxiv.org·8mo ago