Research map / Large language models / Language, safety and interpretability
DialogLM: Pre-trained Model for Long Dialogue Understanding and Summarization
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 |
|---|---|---|---|
| Use window-based dialogue denoising pre-training instead of the general-purpose objective | UniLM-base · AMI, ICSI, and QMSum | ROUGE-1: 51.92 -> 54.49; 46.75 -> 49.25; 29.14 -> 34.02 | Helped |
| Further pre-train using the original UniLM objective on dialogue data | UniLM-base · AMI, ICSI, and QMSum | ROUGE-1 | No clear effect |
| Add hybrid sparse attention to support longer inputs | DialogLM · AMI, ICSI, and QMSum | ROUGE-1: 54.49 -> 53.72; 49.25 -> 49.56; 34.02 -> 33.69 | Mixed |
| Add hybrid sparse attention | DialogLM · AMI and QMSum topic segmentation | Pk: 0.38 -> 0.36; 0.44 -> 0.38 | Helped |
| Add hybrid sparse attention | DialogLM · ForeverDreaming and TVMegaSite | ROUGE-1: 35.42 -> 35.75; 45.04 -> 45.58 | Helped |
| Add window-based dialogue pre-training (ablation) | DialogLM-sparse · QMSum | ROUGE-1: 29.14 -> 33.69 | Helped |
| Add window-based dialogue pre-training (ablation) | DialogLM-sparse · TVMegaSite | ROUGE-1: 43.42 -> 45.58 | Helped |
| Add Speaker Mask noise (ablation) | DialogLM-sparse · QMSum | ROUGE-1: 33.52 -> 33.69 | Helped |
| Add Speaker Mask noise (ablation) | DialogLM-sparse · TVMegaSite | ROUGE-1: 45.31 -> 45.58 | Helped |
| Add Turn Splitting and Turn Merging noise (ablation) | DialogLM-sparse · QMSum | ROUGE-1: 32.76 -> 33.69 | Helped |
| Add Turn Splitting and Turn Merging noise (ablation) | DialogLM-sparse · TVMegaSite | ROUGE-1: 44.23 -> 45.58 | Helped |
| Add Text Infilling noise (ablation) | DialogLM-sparse · QMSum | ROUGE-1: 33.27 -> 33.69 | Helped |
Closest papers by meaning
- Dialog Inpainting: Turning Documents into Dialogs (ICML 2022 · similarity 0.88)
- SD-Eval: A Benchmark Dataset for Spoken Dialogue Understanding Beyond Words (NeurIPS 2024 · similarity 0.87)
- F2RVLM: Boosting Fine-grained Fragment Retrieval for Multi-Modal Long-form Dialogue with Vision Language Model (AAAI 2026 · similarity 0.87)
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.87)
- StreamingDialogue: Prolonged Dialogue Learning via Long Context Compression with Minimal Losses (NeurIPS 2024 · similarity 0.87)
- Scaling Speech-Text Pre-training with Synthetic Interleaved Data (ICLR 2025 · similarity 0.87)
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.