User-Aware Active Knowledge Acquisition for Emotional Support Dialogue

Emotional support plays an important role in dialogue systems, and its success depends on adapting to a user’s evolving and implicit needs across multi-turn interactions while leveraging the strong reasoning capacity of large language models. However, since signals about user needs are often weak, indirect, and can only be disambiguated through multi-turn interaction, existing emotional support methods often struggle to acquire and generalize relevant conversational knowledge efficiently. To bridge this gap, we introduce User-Aware Active Knowledge Acquisition (UKA), a gradient-free active dialogue learning framework that explicitly represents uncertainty about user needs and incorporates active learning into both knowledge acquisition and response selection. We propose a Theory-of-Mind uncertainty estimation mechanism that allows the model to prioritize responses, thereby eliciting more informative user feedback. UKA is capable of efficiently exploring user-aligned conversational knowledge during training while maintaining robustness at test time. Experiments across multiple dialogue benchmarks and model architectures demonstrate that our approach consistently outperforms strong baselines in dialogue quality and user alignment.

Large language models are becoming increasingly capable conversational assistants, but providing emotional support remains difficult because people often do not clearly express what they need. A user who sounds anxious may want empathy rather than advice, while another may prefer practical suggestions instead of emotional reassurance. Existing systems often fail to distinguish between these situations. We introduce UKA (User-aware active Knowledge Acquisition), a framework that helps AI systems better understand users during emotional support conversations. Instead of passively responding, the system actively selects replies that both support the user and reveal more information about the user’s underlying needs. It also stores successful interaction experiences as reusable emotional-support knowledge, allowing the system to gradually improve without retraining the model. Experiments across multiple dialogue benchmarks and large language models show that UKA consistently improves conversation quality and user alignment compared to existing approaches. Our results suggest that future conversational AI systems can become more helpful by actively learning about users throughout an interaction.