Research map / Video / Video-language understanding
Video2Reaction: Mapping Video to Audience Reaction Distribution in the Wild
Trang Nguyen, Sidong Zhang, Shiv Shankar, Gauri Jagatap, Deepak Chandran, Andrea Fanelli, Madalina Fiterau
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 |
|---|---|---|---|
| LoRA finetuning on Video2Reaction | LLaVA-Next-Video-7B · Video2Reaction | Chebyshev: 0.4110 -> 0.1882 | Improved |
| LoRA finetuning on Video2Reaction | LLaVA-Next-Video-7B · Video2Reaction | Top-3 F1: 0.1374 -> 0.7672 | Improved |
| LoRA finetuning on Video2Reaction | LLaVA-Next-Video-7B · Video2Reaction | KL divergence: 1.5547 -> 3.1765 | Hurt |
| LoRA finetuning on Video2Reaction | Qwen2.5-VL · Video2Reaction | cosine: 0.4401 -> 0.8427 | Improved |
| LoRA finetuning on Video2Reaction | Qwen2.5-VL · Video2Reaction | KL divergence: 1.5216 -> 3.4431 | Hurt |
| add text (clip description) modality (ablation) | SA-BFGS · Video2Reaction | MRR: 0.552 -> 0.709 | Improved |
| add audio on top of visual + text (ablation) | SA-BFGS · Video2Reaction | MRR: 0.709 -> 0.716 | No clear effect |
| add text to visual input (ablation) | LLaVA-Next · Video2Reaction | MRR: 0.731 -> 0.783 | Improved |
| two-stage (rephrase then label) annotation instead of single-stage (ablation) | LLM annotation pipeline · human-LLM correlation | mean Spearman correlation: 0.34 -> 0.40 | Improved |
| fine-tune on Video2Reaction plus 1% VCE vs 10% in-domain VCE only | Qwen2.5-VL · VCE benchmark | Top-3 accuracy: 0.35 -> 0.46 | Improved |
| specialized LDL (SA-BFGS) vs finetuned VLM on dominant reactions | SA-BFGS · Video2Reaction | Top-1 F1: 0.6521 (LLaVA FT) vs 0.5283 | Hurt |
Closest papers by meaning
- Infer Induced Sentiment of Comment Response to Video: A New Task, Dataset and Baseline (NeurIPS 2024 · similarity 0.87)
- ViMU: Benchmarking Video Metaphorical Understanding (NeurIPS 2026 · similarity 0.84)
- VidProM: A Million-scale Real Prompt-Gallery Dataset for Text-to-Video Diffusion Models (NeurIPS 2024 · similarity 0.84)
- OpenS2V-Nexus: A Detailed Benchmark and Million-Scale Dataset for Subject-to-Video Generation (NeurIPS 2025 · similarity 0.84)
- Exploiting Fine-Grained Skip Behaviors for Micro-Video Recommendation (AAAI 2025 · similarity 0.84)
- VideoUFO: A Million-Scale User-Focused Dataset for Text-to-Video Generation (NeurIPS 2025 · similarity 0.84)
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.
