Research map / Large language models / Language, safety and interpretability
UBAR: Towards Fully End-to-End Task-Oriented Dialog System with GPT-2
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 |
|---|---|---|---|
| Model whole dialog sessions with belief states, database results, acts, and responses in sequence | UBAR · MultiWOZ 2.0, generated belief and act | Combined: 95.7 -> 105.1 | Helped |
| Generate responses using generated dialog context rather than ground-truth response context (ablation) | UBAR · MultiWOZ 2.0, end-to-end | Inform: 82.6 -> 91.2 | Helped |
| Use generated responses in turn-level dialog history (ablation) | URUR · MultiWOZ 2.0, end-to-end | Combined: 94.8 -> 101.8 | Helped |
| Use only belief states, database results, and system acts instead of only user utterances and responses (ablation) | UBAR · MultiWOZ 2.0, end-to-end | Combined: 95.9 -> 101.9 | Helped |
| Use belief states and system acts in session history (ablation) | UBAR · MultiWOZ 2.0, end-to-end | Combined: 94.8 -> 105.1 | Helped |
| Use all previous turns rather than only the previous turn (ablation) | UBAR · MultiWOZ 2.0, generated belief and act | Combined: 101.6 -> 105.1 | Helped |
| Use ground-truth belief states in prior-turn context (ablation) | UBAR · MultiWOZ 2.0, all previous turns | Combined: 105.1 -> 106.1 | No clear effect |
| Use ground-truth acts and responses in context instead of the system's generated content (ablation) | UBAR · MultiWOZ 2.0, end-to-end | Combined: 105.1 -> 100.1 | Hurt |
| Use only user utterances and generated responses in context (ablation) | UBAR · MultiWOZ 2.0, end-to-end | Combined: 101.9 -> 95.9 | Hurt |
| Fine-tune the held-out-domain model on 100 sessions from the new domain (ablation) | UBAR base model · New domain, held-out-domain transfer | Combined | Helped |
| Fine-tune on 100 new-domain sessions (ablation) | UBAR base model · Original four domains after held-out-domain adaptation | Combined | Hurt |
| Use session-level UBAR with non-delexicalized responses for DST | DST-UBAR · MultiWOZ 2.0 | Joint Accuracy (%): 51.44 -> 52.59 | Helped |
Closest papers by meaning
- TA&AT: Enhancing Task-Oriented Dialog with Turn-Level Auxiliary Tasks and Action-Tree Based Scheduled Sampling (AAAI 2024 · similarity 0.88)
- From Retrieval to Generation: A Simple and Unified Generative Model for End-to-End Task-Oriented Dialogue (AAAI 2024 · similarity 0.86)
- StateTree: Enhancing Long-term Dialogue Reasoning via Reinforcement Learning (NeurIPS 2026 · similarity 0.85)
- ToolDial: Multi-turn Dialogue Generation Method for Tool-Augmented Language Models (ICLR 2025 · similarity 0.85)
- Winnie: Task-Oriented Dialog System with Structure-Aware Contrastive Learning and Enhanced Policy Planning (AAAI 2024 · similarity 0.85)
- An Efficient Self-Learning Framework For Interactive Spoken Dialog Systems (ICML 2024 · similarity 0.85)
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.