Research map / Large language models / Language, safety and interpretability
User-Aware Active Knowledge Acquisition for Emotional Support Dialogue
Mufan Xu, Kehai Chen, Jiahao Hu, Xinchao Xu, Muyun Yang, Tiejun Zhao, Min Zhang
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 knowledge acquisition and reuse (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 70.5 -> 82.9 | Improved |
| add ToM uncertainty (belief tracking) (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 79.2 -> 82.9 | Improved |
| add active candidate response selection (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 78.1 -> 82.9 | Improved |
| use ToM uncertainty instead of LM generation probability (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 82.5 -> 82.9 | No clear effect |
| use belief-aware retrieval instead of random knowledge (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 76.3 -> 82.9 | Improved |
| increase candidate set size N from 1 to 4 (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 78.1 -> 81.1 | Improved |
| increase hypothesis budget M from 1 to 3 at N=4 (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) | average emotion score: 80.9 -> 82.9 | Improved |
| increase candidate set size N from 1 to 4 (ablation) | UKA · Sentient Eval (Qwen3-235B-A22B) Avoidant | Avo. score: 93.1 -> 86.7 | Hurt |
| grow knowledge base from 0 to 200+ entries (ablation) | UKA · Sentient Eval | average emotion score: 70.5 -> 82.9 | Improved |
| add UKA on top of prompting baseline | Qwen3-32B · Sentient Eval | average emotion score: 28.5 -> 33.9 | Improved |
| add MetaMind-style ToM prompting | Qwen3-32B · Sentient Eval | average emotion score: 28.5 -> 21.3 | Hurt |
| use UKA instead of PRINCIPLES | Seed-1.6-36B · ESConv | success rate: 0.450 -> 0.525 | Improved |
Closest papers by meaning
- DialogXpert: Driving Intelligent and Emotion-Aware Conversations Through Online Value-Based Reinforcement Learning with LLM Priors (AAAI 2026 · similarity 0.89)
- Simulation-Free Hierarchical Latent Policy Planning for Proactive Dialogues (AAAI 2025 · similarity 0.87)
- SayNext-Bench: Why Do LLMs Struggle with Next-Utterance Anticipation? (NeurIPS 2026 · similarity 0.87)
- RLVER: Reinforcement Learning with Verifiable Emotion Rewards for Empathetic Agents (ICLR 2026 · similarity 0.87)
- Enhancing Personalized Multi-Turn Dialogue with Curiosity Reward (NeurIPS 2025 · similarity 0.87)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · similarity 0.86)
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.
