A Virtual Music-Therapist Dialogue System Based on Natural Language Processing
Year: 2025
Type: conference-paper
Abstract: This study proposes an intelligent and personalized virtual music-therapy system that integrates multimodal emotion recognition, emotion-driven dialogue management, and music recommendation for university students. The system extracts emotional cues from speech, text, and facial expressions and aligns them via a cross-modal attention mechanism. Emotion tracking and empathic-response strategies are embedded in the dialogue manager to adapt dynamically to the user's affective trajectory. The music-recommendation module combines emotion matching with collaborative filtering to improve both relevance and diversity. Experiments on the IEMOCAP and MELD datasets demonstrate that the proposed system significantly outperforms unimodal approaches and the current state-of-the-art baseline (M2FNet) in weighted accuracy (WA) and F1-score. Improvements are also observed in HitRate@10 and intra-list diversity (ILD). Subjective Mean Opinion Score (MOS) evaluations confirm enhanced user experience and perceived emotional-regulation efficacy. The research offers a technical pathway toward intelligent and personalized music therapy and provides a reference for integrating online mental-health interventions with music-recommendation services.
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7122772129