Date of Award
May 2021
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
School of Computing
Committee Member
Nina NH Hubig
Committee Member
Bart BK Knijnenburg
Committee Member
Taufiquar TK Khan
Abstract
Music and emotions are inherently intertwined. Humans leave hints of their personality everywhere, and particularly their music listening behavior shows conscious and unconscious diametric tendencies and influences. So, what could be more elegant than finding the underlying character given the attributes of a certain music piece and, as such, identifying the likelihood that music preference is also imprinted or at least resonating with its listener? This thesis focuses on the music audio attributes or the latent song features to determine human personality. Based on unsupervised learning, we cluster several large music datasets using multiple clustering techniques known to us. This analysis led us to classify song genres based on audio attributes, which can be deemed a novel contribution in the intersection of Music Information Retrieval (MIR) and human psychology studies. Existing research found a relationship between Myers-Briggs personality models and music genres. Our goal was to correlate audio attributes with the music genre, which will ultimately help us to determine user personality based on their music listening behavior from online music platforms. This target has been achieved as we showed the users’ spectral personality traits from the audio feature values of the songs they listen to online and verified our decision process with the help of a customized Music Recommendation System (MRS). Our model performs genre classification and personality detection with 78% and 74% accuracy, respectively. The results are promising compared to competitor approaches as they are explainable via statistics and visualizations. Furthermore, the RS completes and validates our pursuit through 81.3% accurate song suggestions. We believe the outcome of this thesis will work as an inspiration and assistance for fellow researchers in this arena to come up with more personalized song suggestions. As music preferences will shape specific user personality parameters, it is expected that more such elements will surface that would portray the daily activities of individuals and their underlying mentality.
Recommended Citation
Hridi, Anurata Prabha, "Mining User Personality from Music Listening Behavior in Online Platforms Using Audio Attributes" (2021). All Theses. 3508.
https://tigerprints.clemson.edu/all_theses/3508