Developing an AI-based System for Personalized Music Mood Analysis

Thesis Abstract

In an era where music consumption increasingly relies on personalized experiences, understanding and accurately analyzing individual emotional responses to music remains a significant challenge. The proliferation of digital music platforms and the advent of artificial intelligence (AI) offer new opportunities to develop systems that can tailor music recommendations based on users’ mood states. However, existing approaches often lack precision and fail to adapt dynamically to the nuanced emotional variations in listeners. This study aims to develop an intelligent, AI-driven system capable of analyzing and predicting individual music-induced moods with high accuracy, thereby enhancing personalized music recommendation services. The specific objectives include (1) to design and implement a machine learning model that classifies music-induced moods based on physiological and contextual data; (2) to evaluate the effectiveness of multimodal data—such as facial recognition, voice tone analysis, and user interaction logs—in mood detection; (3) to explore the integration of the Circumplex Model of Affect as a theoretical framework to inform the classification process; and (4) to propose a user-centered interface for real-time mood analysis and music recommendation. The study adopts a mixed-methods research design, combining quantitative data collection through experimental trials and qualitative insights obtained via user surveys. The population comprises 200 voluntary participants aged 18 to 45 who regularly listen to digital music platforms in urban settings. Stratified random sampling ensures balanced representation across gender, age groups, and musical preferences. Data collection instruments include physiological sensors (heart rate monitors, galvanic skin response devices), facial expression recognition software, voice tone analyzers, and self-report mood questionnaires administered before and after musical exposure. The instrument validity and reliability are established through pilot testing and Cronbach’s alpha assessments, respectively. Data analysis employs supervised machine learning techniques such as Random Forest and Support Vector Machines (SVM) to classify moods, with model performance evaluated using metrics like accuracy, precision, recall, and F1-score. Additionally, statistical analyses such as multiple regression and ANOVA assess the relationship between physiological signals and self-reported moods, while thematic analysis interprets qualitative feedback. Expected findings suggest that multimodal data significantly improves the accuracy of mood detection compared to unimodal approaches, with anticipated classification accuracy exceeding 85%. The integration of physiological signals and facial and vocal cues explains a substantial proportion of variance in mood states, validating the Suitability of the Circumplex Model of Affect as a theoretical basis. The developed AI system is projected to demonstrate real-time responsiveness, with a user interface facilitating dynamic music recommendations aligned with detected moods. These insights will contribute to academic understanding by advancing knowledge on multimodal emotion recognition and its practical application in personalized music services. The study also aims to provide a framework for future innovations in affective computing and user-centered multimedia systems. The primary conclusion underscores the potential of AI-based multimodal analysis to revolutionize personalized music experiences, fostering emotional well-being and user engagement. Recommendations include integrating such systems into commercial music platforms, further refining algorithms for diverse demographic groups, and exploring neurological data for enhanced accuracy. Future research should examine longitudinal impacts on user mood and mental health, as well as cross-cultural validations. Overall, this study offers a significant contribution to affective computing, music psychology, and human-computer interaction, providing a scalable model for personalized emotion-aware systems that can adapt to the complex, dynamic nature of human emotional experiences in digital environments.