Research map / Graph learning / Knowledge and temporal graphs
Finding the Right Moment: Human-Assisted Trailer Creation via Task Composition
Pinelopi Papalampidi, Frank Keller, Mirella Lapata
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 |
|---|---|---|---|
| add graph structure to the transformer (ablation) | TP identification model · TP identification (test set) | PA@5: 22.22 -> 27.78 | Improved |
| add asynchronous auxiliary net with prediction consistency loss (ablation) | TP identification model · TP identification (test set) | PA@10: 35.56 -> 41.11 | Improved |
| add representation consistency loss to asynchronous distillation (ablation) | TP identification model · TP identification (test set) | PA@5: 28.89 -> 21.11 | Hurt |
| replace asynchronous with contrastive joint training (ablation) | TP identification model · TP identification (test set) | PA@5: 21.11 -> 33.33 | Improved |
| add pre-training of the auxiliary network (ablation) | TP identification model · TP identification (test set) | PA@10: 47.78 -> 50.00 | Improved |
| use only subtitles instead of all modalities (ablation) | TP identification model · TP identification (test set) | PA@5: 34.44 -> 21.11 | Hurt |
| use only video instead of all modalities (ablation) | TP identification model · TP identification (test set) | PA@10: 50.00 -> 46.67 | Hurt |
| add turning points to the random-walk criteria (ablation) | GraphTrailer · Trailer moment identification (dev) | accuracy (%): 24.28 -> 25.79 | Improved |
| add sentiment alone to the similarity criterion (ablation) | GraphTrailer · Trailer moment identification (dev) | accuracy (%): 24.28 -> 23.97 | Hurt |
| add sentiment on top of similarity plus turning points (ablation) | GraphTrailer · Trailer moment identification (dev) | accuracy (%): 25.79 -> 26.95 | Improved |
| add asynchronous auxiliary net training (ablation) | GraphTrailer · Trailer moment identification (dev) | accuracy (%): 22.63 -> 21.87 | No clear effect |
| add joint contrastive training of main and auxiliary networks (ablation) | GraphTrailer · Trailer moment identification (dev) | accuracy (%): 22.63 -> 25.44 | Improved |
Closest papers by meaning
- Learning Trailer Moments in Full-Length Movies with Co-Contrastive Attention (ECCV 2020 · similarity 0.90)
- Towards Automated Movie Trailer Generation (CVPR 2024 · similarity 0.87)
- Movie Summarization via Sparse Graph Construction (AAAI 2021 · similarity 0.85)
- CineTechBench: A Benchmark for Cinematographic Technique Understanding and Generation (NeurIPS 2025 · similarity 0.85)
- NarrativeBridge: Enhancing Video Captioning with Causal-Temporal Narrative (ICLR 2025 · similarity 0.84)
- MovieBench: A Hierarchical Movie Level Dataset for Long Video Generation (CVPR 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.
