Abstract:Empathetic response generation necessitates the integration of emotional and intentional dynamics to foster meaningful interactions. Existing research either neglects the intricate interplay between emotion and intent, leading to suboptimal controllability of empathy, or resorts to large language models (LLMs), which incur significant computational overhead. In this paper, we introduce ReflectDiffu, a lightweight and comprehensive framework for empathetic response generation. This framework incorporates emotion contagion to augment emotional expressiveness and employs an emotion-reasoning mask to pinpoint critical emotional elements. Additionally, it integrates intent mimicry within reinforcement learning for refinement during diffusion. By harnessing an intent twice reflect the mechanism of Exploring-Sampling-Correcting, ReflectDiffu adeptly translates emotional decision-making into precise intent actions, thereby addressing empathetic response misalignments stemming from emotional misrecognition. Through reflection, the framework maps emotional states to intents, markedly enhancing both response empathy and flexibility. Comprehensive experiments reveal that ReflectDiffu outperforms existing models regarding relevance, controllability, and informativeness, achieving state-of-the-art results in both automatic and human evaluations.

What problem does this paper attempt to address?

### The Problem Addressed by the Paper This paper aims to address several key issues in lightweight empathetic response generation models: 1. **Interaction Mechanism between Emotion and Intent**: Existing research often overlooks the complex interplay between emotions and intents, leading to poor controllability of empathy. 2. **Computational Overhead**: Methods relying on large-scale language models (LLMs) are effective but come with high computational costs. 3. **Depth and Flexibility of Emotional Understanding**: Lightweight models mainly depend on external knowledge signals rather than underlying psychological mechanisms, which limits their empathetic capability and flexibility. 4. **Lack of Multi-task Datasets**: Existing lightweight models lack unified multi-task dataset support for emotional reasoning masking, intent prediction, and empathetic dialogue generation. To address these issues, the paper proposes a framework named **ReflectDiffu**, which combines emotional contagion and intent imitation mechanisms and refines intents during the diffusion process through reinforcement learning. Specifically, ReflectDiffu improves empathetic response generation in the following ways: - Introducing an Emotional Reasoning Annotator (ERA) to enhance emotional understanding. - Proposing an Intent Twice mechanism, i.e., Exploring-Sampling-Correcting mechanism, to improve the consistency between emotions and intents. - Utilizing large-scale language models to extend annotations on the EmpatheticDialogue dataset and create a multi-task dataset. Experimental results show that ReflectDiffu outperforms existing empathetic dialogue generation models in both automatic and human evaluations, excelling in relevance, controllability, and informativeness.

ReflectDiffu:Reflect between Emotion-intent Contagion and Mimicry for Empathetic Response Generation via a RL-Diffusion Framework

DiffusEmp: A Diffusion Model-Based Framework with Multi-Grained Control for Empathetic Response Generation

MFDR: Multiple-stage Fusion and Dynamically Refined Network for Multimodal Emotion Recognition

Human-Robot Emotional Interaction Model Based on Reinforcement Learning

EmpCRL: Controllable Empathetic Response Generation via In-Context Commonsense Reasoning and Reinforcement Learning

Improving Empathetic Response Generation by Emotion Recognition and Information Filtration

Empathy Level Alignment via Reinforcement Learning for Empathetic Response Generation

Enhancing Empathetic Response Generation by Augmenting LLMs with Small-scale Empathetic Models

EmotionIC: emotional inertia and contagion-driven dependency modeling for emotion recognition in conversation

EmpHi: Generating Empathetic Responses with Human-like Intents

CIRG-SL: Commonsense Inductive Relation Graph framework with Soft Labels for Empathetic Response Generation

Wish I Can Feel What You Feel: A Neural Approach for Empathetic Response Generation

Robust Facial Reactions Generation: An Emotion-Aware Framework with Modality Compensation

Facilitating Multi-turn Emotional Support Conversation with Positive Emotion Elicitation: A Reinforcement Learning Approach

An Iterative Associative Memory Model for Empathetic Response Generation

Mimicking the Thinking Process for Emotion Recognition in Conversation with Prompts and Paraphrasing

Chinese Emotional Dialogue Response Generation via Reinforcement Learning

Towards Empathetic Conversational Recommender Systems

Empathetic Response Generation with State Management

AlignDiff: Aligning Diverse Human Preferences via Behavior-Customisable Diffusion Model

Multi-dimensional Evaluation of Empathetic Dialog Responses