MOOD DETECTION BASED ON LAST SONG LISTENED ON SPOTIFY

Abstract

A Song Is One Medium Used To Express Someone's Emotion, Whether As A Performer Or Audience. With The Advancement Of Machine Learning And A Deeper Understanding Of Sentiment Analysis, We Decided To Study Mood Detection Based On The Last Song Listened To. One of the direct ways to measure someone's mood is by using a Four-dimensional Mood Scale (FDMS) device. This device categorized mood into four dimensions: low valence, high valence, low arousal, and high arousal. In this article, we used a variation of FDMS adapted to the Indonesian language called FDMS-55 to compare the result from our model. Our model is trained using song data collected from Spotify and Genius using their respective API (Application Programming Interface). We classified manually into a mood class and then processed further using Azure Cognitive Service Text Analytics API. Based on evaluation conducted on the model, the FastTreeOva algorithm produces the highest accuracy both on valence class with 0.8901 and arousal class with 0.9167. The comparison between the model result and respondent's FDMS-55 device result is made with cosine similarity and yields similarity value of 0.770 with 0.103 standard deviation. It is concluded that someone's mood is related to the song they listened to, and our model can precisely predict someone's mood based on the last song they listened to.