음악가사와 오디오 특성 변수의 통합 분석 모델 구축에 관한 연구: 텍스트 마이닝과 기계학습을 중심으로 = A Study on the Integrated Analysis Model of Music Lyrics and Audio Characteristic : Focusing on Text Mining and Machine Learning

다국어 초록 (Multilingual Abstract)

The objective of this study is to build a model that could predict the characteristics of music by extracting and integrating the lyrics and audio characteristics. The study presents an integrated model between two variables by implementing lyrics analysis using text mining and audio features provided by Spotify. In Study 1, variables were derived from lyrics using cluster analysis and sentiment analysis. In Study 2, correlation analysis was used to explore the relationship between lyrics and audio features. And finally, in Study 3, musical genres are predicted using the input variables derived from Studies 1 and 2 as well as machine learning methodology. We confirmed that lyrical themes and sentiment variables could increase the accuracy of the estimation model in comparison to genre prediction that relied only on audio characteristics. Furthermore, we found that lyrical characteristics contribute more to genre estimation than audio characteristics when a specific genre has differentiation in its lyrical content. We believe that this study can be extended to predict music needed for specific applications, including in advertising. To increase the necessary practical utilization, efforts should be made in parallel to collect various music type and derive appropriate input variables.