Personalized Optimization of Elderly MusicTherapy Based on Emotion Analysis and Physiological Data

Abstract

With the accelerating pace of global population aging, mental health problems among older adults have become increasingly prominent. Although traditional music therapy has demonstrated benefits in improving mood and alleviating anxiety, its effectiveness remains limited by the lack of personalization and real-time feedback. To address this issue, this study proposes a personalized music therapy optimization framework based on multimodal emotion analysis and physiological data fusion. The framework integrates multiple sources-including heart rate variability (HRV), electrodermal activity (EDA), electroencephalography (EEG), speech, and facial expressions-and employs deep learning models for feature fusion, combined with a personalized recommendation mechanism that dynamically adjusts musical interventions to enhance therapeutic efficacy. Experimental results indicate that the proposed method improves emotion recognition accuracy by approximately 7.6 percentage points over conventional approaches and yields a notable increase in macro-F1. Moreover, the experimental group shows greater improvements in PANAS positive affect scores and in HRV indicators, including a larger reduction in the LF/HF ratio, compared with the control group, thereby validating the effectiveness of the framework for mental health interventions. The main contribution lies in introducing a novel optimization approach that fuses multimodal data with personalized recommendation, enriching the technical toolkit of music therapy and providing an innovative pathway for mental health interventions in older adults, while advancing the interdisciplinary integration of artificial intelligence and digital health.