Research map / Large language models / Language, safety and interpretability
NaturalConv: A Chinese Dialogue Dataset Towards Multi-turn Topic-driven Conversation
Xiaoyang Wang, Chen Li, Jianqiao Zhao, Dong Yu
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 |
|---|---|---|---|
| rerank BM25 responses with BERT | Retrieval on NaturalConv · NaturalConv test | BLEU-1/2: 23.30 / 13.12 -> 24.96 / 13.82 | Improved |
| rerank BM25 responses with BERT | Retrieval on NaturalConv · NaturalConv test | DISTINCT-1/2: 8.48 / 43.21 -> 8.27 / 42.31 | Hurt |
| add retrieved document sentence to GRU input (ablation) | GRU · NaturalConv test | BLEU-1/2: 27.89 / 14.23 -> 27.86 / 14.24 | No clear effect |
| add retrieved document sentence to GRU input (ablation) | GRU · NaturalConv test | Manual (1-3): 1.97 -> 2.03 | Improved |
| add retrieved document sentence to LSTM input (ablation) | LSTM · NaturalConv test | BLEU-1/2: 26.09 / 13.35 -> 26.79 / 14.54 | Improved |
| add retrieved document sentence to LSTM input (ablation) | LSTM · NaturalConv test | DISTINCT-1/2: 0.98 / 4.30 -> 2.13 / 9.49 | Improved |
| add retrieved document sentence to LSTM input (ablation) | LSTM · NaturalConv test | Manual (1-3): 2.10 -> 2.16 | Improved |
| add retrieved document sentence to Transformer input (ablation) | Transformer · NaturalConv test | BLEU-1/2: 25.17 / 12.39 -> 24.47 / 13.12 | Mixed |
| add retrieved document sentence to Transformer input (ablation) | Transformer · NaturalConv test | F1: 25.73 -> 27.01 | Improved |
| add retrieved document sentence to Transformer input (ablation) | Transformer · NaturalConv test | Manual (1-3): 2.04 -> 2.09 | Improved |
| use retrieval rather than generation | NaturalConv dialogue systems · NaturalConv test | DISTINCT-1/2: GRU: 1.80 / 8.17; Retrieval: 8.48 / 43.21 | Improved |
| use Transformer rather than GRU or LSTM | Generation models trained on NaturalConv · NaturalConv test | DISTINCT-1/2: Transformer: 2.91 / 15.32; GRU: 1.80 / 8.17; LSTM: 0.98 / 4.30 | Improved |
Closest papers by meaning
- CORECODE: A Common Sense Annotated Dialogue Dataset with Benchmark Tasks for Chinese Large Language Models (AAAI 2024 · similarity 0.86)
- MultiTalk: Scaling Full-Duplex Speech Models to Long, Multi-Party, Bilingual Conversation (NeurIPS 2026 · similarity 0.86)
- Aligning Spoken Dialogue Models from User Interactions (ICML 2025 · similarity 0.85)
- Zhongjing: Enhancing the Chinese Medical Capabilities of Large Language Model through Expert Feedback and Real-World Multi-Turn Dialogue (AAAI 2024 · similarity 0.85)
- LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset (ICLR 2024 · similarity 0.85)
- Improving Open-Domain Dialogue Response Generation with Multi-Source Multilingual Commonsense Knowledge (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.
