Research map / Label-efficient and robust learning / Recognition and long-tail
Mitigating Negative Style Transfer in Hybrid Dialogue System
Shimin Li, Qinyuan Cheng, Linyang Li, Xipeng Qiu
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 contrastive latent variables (ablation) | HiS-Dialog · FusedChat | Combined Score: 104.60 -> 106.25 (↑1.65) | Improved |
| Add style prefix (ablation) | HiS-Dialog · FusedChat | Combined Score: 104.60 -> 107.76 (↑3.16) | Improved |
| Add both contrastive latent variables and style prefix (ablation) | HiS-Dialog · FusedChat | Combined Score: 104.60 -> 108.39 (↑3.79) | Improved |
| Use contrastive latent variables and style prefix together rather than contrastive latent variables alone (ablation) | HiS-Dialog · FusedChat | Combined Score: 106.25 -> 108.39 | Improved |
| Use a prefix length of 70 rather than 50 (ablation) | HiS-Dialog · FusedChat | Combined Score: 108.39 -> 106.87 | Hurt |
| Use a prefix length of 100 rather than 50 (ablation) | HiS-Dialog · FusedChat | Combined Score: 108.39 -> 107.92 | Hurt |
| Use decoder-only style prefix rather than prefix at encoder, decoder, and cross-attention (ablation) | HiS-Dialog · FusedChat | Combined Score: 107.11 vs 108.39 | Hurt |
| Use encoder and decoder prefix without cross-attention prefix (ablation) | HiS-Dialog · FusedChat | Combined Score: 108.27 vs 108.39 | Hurt |
| Use triplet loss instead of InfoNCE (ablation) | HiS-Dialog · FusedChat and MultiWOZ | Task Score and BLEU | Improved |
| Set the triplet-loss margin to 0.3 (ablation) | HiS-Dialog · FusedChat and MultiWOZ | Task Score and BLEU | Improved |
Closest papers by meaning
- Towards Diverse, Relevant and Coherent Open-Domain Dialogue Generation via Hybrid Latent Variables (AAAI 2023 · similarity 0.88)
- Stylized Dialogue Response Generation Using Stylized Unpaired Texts (AAAI 2021 · similarity 0.87)
- Stylized Dialogue Generation with Multi-Pass Dual Learning (NeurIPS 2021 · similarity 0.87)
- From Natural Alignment to Conditional Controllability in Multimodal Dialogue (ICLR 2026 · similarity 0.86)
- Contrast and Generation Make BART a Good Dialogue Emotion Recognizer (AAAI 2022 · similarity 0.85)
- Improving the Robustness of Knowledge-Grounded Dialogue via Contrastive Learning (AAAI 2024 · 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.
