Research map / Video / Video-language understanding
Video Timeline Modeling For News Story Understanding
Meng Liu, Mingda Zhang, Jialu Liu, Hanjun Dai, Ming-Hsuan Yang, Shuiwang Ji, Zheyun Feng, Boqing Gong
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 |
|---|---|---|---|
| Model nodes as learnable embeddings with three Transformer encoders (Tri-Transformer vs V-Transformer) | V-Transformer · YouTube-News-Timeline test | micro precision: 64.4% -> 66.6% | Improved |
| Tri-Transformer vs V-Transformer | V-Transformer · YouTube-News-Timeline test | macro Hamming distance: 0.477 -> 0.453 | Improved |
| Tri-Transformer vs V-Transformer | V-Transformer · YouTube-News-Timeline test | micro Euclidean distance: 0.916 -> 0.837 | Improved |
| Add cross-modal distillation from text-aware teacher | Tri-Transformer · YouTube-News-Timeline test | macro precision: 71.4% -> 72.5% | Improved |
| Add cross-modal distillation | Tri-Transformer · YouTube-News-Timeline test | micro Hamming distance: 0.490 -> 0.470 | Improved |
| Add cross-modal distillation | Tri-Transformer · YouTube-News-Timeline test | micro Euclidean distance: 0.837 -> 0.804 | Improved |
| Add cross-modal distillation | Tri-Transformer · YouTube-News-Timeline test | macro pairwise agreement accuracy: 85.4% -> 85.2% | Hurt |
| Add cross-modal distillation | Tri-Transformer · YouTube-News-Timeline test | macro recall: 71.2% -> 71.1% | No clear effect |
| Give the model text node descriptions as input (reference optimum teacher) | Tri-Transformer student · YouTube-News-Timeline test | macro Hamming distance: 0.436 -> 0.308 | Improved |
| Remove video release-time positional encoding (ablation) | Tri-Transformer · YouTube-News-Timeline test | macro Hamming distance: 0.453 -> 0.621 | Improved |
| Remove video release-time positional encoding (ablation) | Tri-Transformer · YouTube-News-Timeline test | micro pairwise agreement accuracy: 94.3% -> 72.1% | Improved |
| Remove video PE (ablation) | Tri-Transformer · YouTube-News-Timeline test | macro Euclidean distance: 0.716 -> 1.523 | Improved |
Closest papers by meaning
- NewsNet: A Novel Dataset for Hierarchical Temporal Segmentation (CVPR 2023 · similarity 0.85)
- Video Summarization Using Fully Convolutional Sequence Networks (ECCV 2018 · similarity 0.84)
- Video Summarization with Large Language Models (CVPR 2025 · similarity 0.83)
- Seq2Time: Sequential Knowledge Transfer for Video LLM Temporal Grounding (CVPR 2025 · similarity 0.82)
- SeriesBench: A Benchmark for Narrative-Driven Drama Series Understanding (CVPR 2025 · similarity 0.82)
- Scaling Up Video Summarization Pretraining with Large Language Models (CVPR 2024 · 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.
