Research map / Large language models / Agents, code and math
AI-Salesman: Towards Reliable Large Language Model Driven Telemarketing
Qingyu Zhang, Chunlei Xin, Xuanang Chen, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, Qing Ye, Qianlong Xie, Xingxing Wang
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 |
|---|---|---|---|
| train directly with GRPO and the Bayesian reward (no SFT) plus DOGA inference | Qwen2.5-7B-Instruct · TeleSales Syn-Data (mean) | LLM-judge mean score: 5.46 -> 6.49 | Improved |
| add SFT before GRPO (ablation) | GRPO · TeleSales Syn-Data (mean) | LLM-judge mean score: 6.49 -> 5.77 | Hurt |
| add GRPO on top of SFT model | SFT-only · TeleSales Syn-Data (mean) | LLM-judge mean score: 5.69 -> 5.77 | No clear effect |
| add domain SFT to baseline | Qwen2.5-7B-Instruct · TeleSales Syn-Data (Role-playing) | LLM-judge mean score: 5.54 -> 5.66 | No clear effect |
| add domain SFT to baseline | Qwen2.5-7B-Instruct · TeleSales Syn-Data (Business Analysis) | LLM-judge mean score: 6.49 -> 6.78 | Improved |
| add Bayesian-supervised reasoning reward (ablation) | GRPO · TeleSales Syn-Data (mean) | LLM-judge mean score: 6.15 -> 6.49 | Improved |
| add semantic similarity reward (ablation) | GRPO · TeleSales Syn-Data (mean) | LLM-judge mean score: 6.39 -> 6.49 | Improved |
| use DOGA dynamic outline instead of static prompt (ablation) | Ours · Objection Handling / Operational Guidance / Business Analysis | LLM-judge score: +11.9% / +14.8% / +5.1% | Improved |
| use DOGA instead of static prompt on simple tasks (ablation) | Ours · Role-playing and Idle-chat Rejection | LLM-judge score | Hurt |
| scale base model from 7B to 14B (ablation) | Ours · Real-Data | performance score: 6.62 -> 6.67 | No clear effect |
| scale base model from 32B to 72B (ablation) | Ours · Real-Data | performance score: 7.17 -> 7.01 | Hurt |
Closest papers by meaning
- DialogXpert: Driving Intelligent and Emotion-Aware Conversations Through Online Value-Based Reinforcement Learning with LLM Priors (AAAI 2026 · similarity 0.84)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · similarity 0.83)
- Plug-and-Play Policy Planner for Large Language Model Powered Dialogue Agents (ICLR 2024 · similarity 0.83)
- AGILE: A Novel Reinforcement Learning Framework of LLM Agents (NeurIPS 2024 · similarity 0.83)
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.83)
- iMAD: Intelligent Multi-Agent Debate for Efficient and Accurate LLM Inference (AAAI 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.
