Recognizing Emotions in Music

Music is a universal phenomenon among humans and is a powerful medium for expressing and evoking emotions. The interdisciplinary field of Music Emotion Recognition (MER) leverages this intrinsic connection by developing algorithms and systems to automatically recognize these emotions in music. A major challenge in MER is that the datasets available for training machine learning models are often limited in size and may not fully represent the diversity of genres and emotions in music. To address this gap, we created a comprehensive MER dataset by combining publicly available data from different platforms containing music metadata. A new string-matching algorithm was developed, along with a domain-specific matching process, to combine data from these platforms. The resulting dataset contains metadata for 508,872 songs and 546 emotion labels. To evaluate the dataset and to provide a foundation for future research, a baseline for machine learning models was established for the 100 most frequent emotion labels in the dataset.