MUSIC AND EMOTION: A MACHINE LEARNING APPROACH TO MOOD CLASSIFICATION FOR PERSONALIZED PLAYLIST CURATION

Abstract

A vital component of human existence, music elicits a range of feelings and moods. In recent years, research on understanding and classifying music based on its emotional content—also referred to as music mood classification—has grown significantly. Applications like mood-based playlist creation, emotion-aware music therapy, and tailored music recommendations depend heavily on the analysis of music's emotional content. Traditionally, to classify music moods, music specialists would listen to each track and manually tag each one with a mood (happy, sad, quiet, energetic, etc.). Because of human biases, this procedure was quite subjective and prone to discrepancies. Afterwards, the annotated data would be utilized to create manually constructed rule-based systems or basic statistical models that would categorize music into various moods. These methods have limitations in terms of accuracy, scalability, and generality even if they offered some new insights. Furthermore, hand annotation is subjective, costly, and time-consuming. Furthermore, differences in how a given piece of music is interpreted emotionally among human listeners may result in inconsistent labeled data. There is a need for automated and data-driven methods to get over these obstacles and enable large-scale mood analysis of music collections. By using computational models to extract patterns and correlations from data, machine learning techniques provide a viable answer to this issue and allow music to be automatically classified according to its emotional content. In light of this, this project creates an emotion recognition-based music recommendation system that analyzes user moods before recommending songs based on those moods. The suggested mood categorization system may be incorporated with any music recommendation engine, as confirmed by trials conducted on real data.