Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Emotional Spectrum of LLMs: Leveraging Empathy and Emotion-Based Markers for Mental Health Support

Alessandro De Grandi, Federico Ravenda, Andrea Raballo, Fabio Crestani

TL;DR

The paper introduces RACLETTE, a fine-tuned LLM-based system that jointly performs emotion recognition and empathetic response generation for mental health support. It advances emotion embeddings as interpretable markers that summarize a user’s affective state over interactions, enabling comparison with disorder-specific emotional profiles for preliminary screening. By leveraging datasets like Empathetic Dialogues and Reddit-based mental health corpora, RACLETTE demonstrates superior emotional accuracy and empathetic reply quality relative to baselines, and extends this framework to deduce mental-state markers from emotional distributions. Additionally, the study explores Reddit-derived emotion embeddings to detect suicide risk via distance metrics (KL, JS, CS), achieving high recall, though at trade-offs in precision, underscoring both the potential and the ethical considerations of deploying such tools in real-world care settings.

Abstract

The increasing demand for mental health services has highlighted the need for innovative solutions, particularly in the realm of psychological conversational AI, where the availability of sensitive data is scarce. In this work, we explored the development of a system tailored for mental health support with a novel approach to psychological assessment based on explainable emotional profiles in combination with empathetic conversational models, offering a promising tool for augmenting traditional care, particularly where immediate expertise is unavailable. Our work can be divided into two main parts, intrinsecaly connected to each other. First, we present RACLETTE, a conversational system that demonstrates superior emotional accuracy compared to state-of-the-art benchmarks in both understanding users' emotional states and generating empathetic responses during conversations, while progressively building an emotional profile of the user through their interactions. Second, we show how the emotional profiles of a user can be used as interpretable markers for mental health assessment. These profiles can be compared with characteristic emotional patterns associated with different mental disorders, providing a novel approach to preliminary screening and support.

The Emotional Spectrum of LLMs: Leveraging Empathy and Emotion-Based Markers for Mental Health Support

TL;DR

The paper introduces RACLETTE, a fine-tuned LLM-based system that jointly performs emotion recognition and empathetic response generation for mental health support. It advances emotion embeddings as interpretable markers that summarize a user’s affective state over interactions, enabling comparison with disorder-specific emotional profiles for preliminary screening. By leveraging datasets like Empathetic Dialogues and Reddit-based mental health corpora, RACLETTE demonstrates superior emotional accuracy and empathetic reply quality relative to baselines, and extends this framework to deduce mental-state markers from emotional distributions. Additionally, the study explores Reddit-derived emotion embeddings to detect suicide risk via distance metrics (KL, JS, CS), achieving high recall, though at trade-offs in precision, underscoring both the potential and the ethical considerations of deploying such tools in real-world care settings.

Abstract

The increasing demand for mental health services has highlighted the need for innovative solutions, particularly in the realm of psychological conversational AI, where the availability of sensitive data is scarce. In this work, we explored the development of a system tailored for mental health support with a novel approach to psychological assessment based on explainable emotional profiles in combination with empathetic conversational models, offering a promising tool for augmenting traditional care, particularly where immediate expertise is unavailable. Our work can be divided into two main parts, intrinsecaly connected to each other. First, we present RACLETTE, a conversational system that demonstrates superior emotional accuracy compared to state-of-the-art benchmarks in both understanding users' emotional states and generating empathetic responses during conversations, while progressively building an emotional profile of the user through their interactions. Second, we show how the emotional profiles of a user can be used as interpretable markers for mental health assessment. These profiles can be compared with characteristic emotional patterns associated with different mental disorders, providing a novel approach to preliminary screening and support.
Paper Structure (21 sections, 4 equations, 6 figures, 7 tables)

This paper contains 21 sections, 4 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. A Psychological Rationale for our Approach
  5. Datasets
  6. Tailoring an LLM to React Empathetically
  7. Emotion Recognition and Empathetic Response
  8. Mental State as Mixtures of Emotions
  9. Reddit's Emotion embeddings applied to the Detection of Suicide Risk
  10. Results for Mental Health Classification
  11. Conclusions
  12. Limitations
  13. Ethical Considerations
  14. Finetuning Details
  15. Error Analysis
  16. ...and 6 more sections

Figures (6)

  • Figure 1: An example of how our 3-turns coversation structure has been implemented.
  • Figure 2: A visual explanation of how the emotional profile of a user is updated across a conversation and how to extract the final emotion embedding.
  • Figure 3: Overview of the main steps of RACLETTE pipeline.
  • Figure 4: (A) 2-Dimensional representation of mental disorders distribution after applying t-SNE dimensionality reduction. (B) Sorted emotion embedding of depression. (C.) Sorted emotion embedding of DailyDialog.
  • Figure 5: Emotional embeddings of subreddits related to high risk of suicide.
  • ...and 1 more figures