Research map / Large language models / Instruction and fine-tuning
A Generative Approach for Script Event Prediction via Contrastive Fine-Tuning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 event-centric pretraining (ablation) | BART base generative approach · Public dataset | accuracy: 61.08 -> 62.94 | Helped |
| Add task-specific contrastive fine-tuning (ablation) | BART base generative approach · Public dataset | accuracy: 40.00 -> 62.94 | Helped |
| Use generative contrastive fine-tuning instead of a linear classifier (ablation) | BART base · Public dataset | accuracy: 61.77 -> 62.94 | Helped |
| Use event-level blank infilling instead of random span masking (ablation) | BART base generative approach · Public dataset | accuracy: 61.44 -> 62.94 | Helped |
| Use the mean rather than the sum of log-probabilities (ablation) | BART base generative approach · Public dataset | accuracy: 60.84 -> 62.94 | Helped |
| Use ComplementEntropyLoss rather than Cross Entropy Loss (ablation) | BART base generative approach · Public dataset | accuracy: 62.71 -> 62.94 | Helped |
| Use ComplementEntropyLoss rather than Margin Ranking Loss (ablation) | BART base generative approach · Public dataset | accuracy: 61.18 -> 62.94 | Helped |
| Fine-tune BART with the proposed two-stage generative method instead of vanilla BART | BART base · Public dataset | accuracy: 60.00 -> 62.94 | Helped |
| Use the proposed method rather than GraphBERT | BART base · Public dataset | accuracy: 60.72 -> 62.94 | Helped |
| Train with the larger original-dataset training set | Ours (BART large) · Public dataset | accuracy: 64.82 -> 65.88 | Helped |
| Use the proposed method rather than MCPredictor | BART base · Original dataset | accuracy: 67.14 -> 67.21 | No clear effect |
Closest papers by meaning
- SAM-Net: Integrating Event-Level and Chain-Level Attentions to Predict What Happens Next (AAAI 2019 · similarity 0.90)
- Rich Event Modeling for Script Event Prediction (AAAI 2023 · similarity 0.90)
- FEEL: Featured Event Embedding Learning (AAAI 2018 · similarity 0.89)
- What Happens Next? Event Prediction Using a Compositional Neural Network Model (AAAI 2016 · similarity 0.88)
- Event Representations With Tensor-Based Compositions (AAAI 2018 · similarity 0.86)
- Realism VS Accuracy: Event Sequence Forecasting from a Generative Modeling Perspective (NeurIPS 2026 · similarity 0.83)
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.