Enhancing Mental Health Support with AI-Driven Emotion-Embedded Knowledge Graphs

Abstract

Background and Aim: Mental health conversation assistants (chatbots) have gained popularity with advancements in technology and AI; however, the level of empathy and emotion perception in these applications remains limited. To support more human-centered applications, we propose an Artificial Intelligence (AI) framework inspired by psychological models of emotions and empathy, designed to enhance chatbot interactions.   Method: The proposed AI framework integrates emotion-embedded knowledge graphs with Large Language Models (LLMs) to generate personalized responses. We employ knowledge graphs to enhance text analysis, enabling the identification of relationships between entities and structuring user (or patient) conversations into a coherent knowledge graph representation. Emotion detection techniques further enrich this representation, creating a unique knowledge graph for each individual based on their emotional state.   Results: We present the framework's validity by evaluating the proposed framework with a mental health conversation dataset comprising interviews with therapists in mental health. The key features of the framework allowed modelling these conversations in an emotion-embedded knowledge graph allowing the application to learn about the user. The generated responses were evaluated for similarity with therapist answers, coherence and empathetic expressions.   Discussion: The emotion perception and customisation based on users' emotional states provide a significant advantage in creating supportive and empathetic conversational agents, potentially transforming the landscape of mental health assistance. This AI-driven framework lays the foundation for developing innovative solutions to deliver more impactful and personalised mental health interventions through technology to enhance patient support and the overall quality of care in digital health services.