Abstract:Emotional Support Conversation (ESC) systems are pivotal in providing empathetic interactions, aiding users through negative emotional states by understanding and addressing their unique experiences. In this paper, we tackle two key challenges in ESC: enhancing contextually relevant and empathetic response generation through dynamic demonstration retrieval, and advancing cognitive understanding to grasp implicit mental states comprehensively. We introduce Dynamic Demonstration Retrieval and Cognitive-Aspect Situation Understanding (\ourwork), a novel approach that synergizes these elements to improve the quality of support provided in ESCs. By leveraging in-context learning and persona information, we introduce an innovative retrieval mechanism that selects informative and personalized demonstration pairs. We also propose a cognitive understanding module that utilizes four cognitive relationships from the ATOMIC knowledge source to deepen situational awareness of help-seekers' mental states. Our supportive decoder integrates information from diverse knowledge sources, underpinning response generation that is both empathetic and cognitively aware. The effectiveness of \ourwork is demonstrated through extensive automatic and human evaluations, revealing substantial improvements over numerous state-of-the-art models, with up to 13.79\% enhancement in overall performance of ten metrics. Our codes are available for public access to facilitate further research and development.

What problem does this paper attempt to address?

The paper attempts to address two key challenges: 1. **How can an Emotional Support Conversation (ESC) system efficiently retrieve and integrate context - relevant examples to generate relevant and empathetic responses?** - Traditional end - to - end generation models often struggle to produce context - appropriate responses in open - ended conversations. Although previous research has improved response consistency and maintained conversation goals by enhancing retrieval mechanisms, customizing these retrieval methods according to users' personalized needs in the specific context of ESC remains an under - explored area. 2. **How can the cognitive - aspect context - understanding ability of ESC systems be improved to fully grasp users' implicit mental states?** - Emotional conversation is a two - sided task, including emotional recognition and cognitive understanding. Although recent research has mainly focused on the emotional dimension, the equally important cognitive component has often been overlooked. Some studies have attempted to incorporate common - sense knowledge into empathetic conversation systems to strengthen cognitive empathy, but no one has yet used cognitive understanding to provide profound empathetic support. To address these challenges, the authors propose a new method named **Dynamic Demonstration Retrieval and Cognitive - Aspect Situation Understanding (D2RCU)**. This method, through the Dynamic Demonstration Selector and the Cognitive - Aspect Situation Understanding module, combined with the Multi - Knowledge Fusion Decoder, aims to improve the quality of emotional support conversations. ### Specific Methods 1. **Dynamic Demonstration Selector** - This module dynamically retrieves query - paragraph pairs semantically aligned with the current help - seeker's situation and personal attributes from the training set through Dense Passage Retrieval (DPR). This helps generate more personalized and context - relevant responses. 2. **Cognitive - Aspect Situation Understanding Module** - This module utilizes four cognitive relationships (Effect, Intent, Need, Want) in the ATOMIC knowledge source to deepen the understanding of the help - seeker's situation and implicit mental state. By encoding and refining these cognitive states, the system can more comprehensively understand the user's needs and intentions. 3. **Multi - Knowledge Fusion Decoder** - This module combines the encoded example pairs and cognitive states with the conversation history through a dual - cross - attention mechanism to generate relevant and cognitively consistent supportive responses. Through a weighted aggregation strategy, these features are balanced and integrated to form the final hidden state vector, which serves as the decoder's input. ### Experimental Results The paper verifies the effectiveness of D2RCU through extensive automatic and human evaluations. The results show that D2RCU achieves the highest scores on ten indicators and improves the performance by an average of 13.79% compared to the strongest baseline model. Ablation studies indicate that context learning significantly enhances the model's performance. ### Main Contributions 1. **Propose a new method for dynamic example selection using personalized information, generating more informative and personalized candidate pairs.** 2. **Introduce a dedicated cognitive understanding module, emphasize the importance of cognitive awareness, and design a response - generation method based on retrieval enhancement and cognitive understanding.** 3. **Through extensive experimental verification, prove that D2RCU significantly improves the performance of the existing best methods in ESC tasks.** These contributions enable D2RCU to better understand and respond to users' needs in emotional support conversation systems and provide more effective emotional support.

Dynamic Demonstration Retrieval and Cognitive Understanding for Emotional Support Conversation