MyMoods: Personalized Mood Classification

TitleMyMoods: Personalized Mood Classification
Publication TypeMaster Thesis
Year of Publication2009
AuthorsCabrera, J. J.
AbstractAutomatic music classification is one of the most relevant topics in the field of Music Technology. Nowadays, due to the large amount of music that a listener can choose from, different approaches to Music Information Retrieval are very active among researchers. Mood classification is one of the most popular topics, but its high degree of subjectivity combined with the need to construct a ground truth are two important difficulties that researchers have to face. In this Master Thesis, we propose a content-based mood classification system in which the main contribution is the use of social tags as ground truth. That way, we are avoiding the need to build a database through manual annotation, as this process is usually a source of error and a tedious task. Information provided by music social networks will be extracted automatically, so that any emotion category selected by the user can be learned by the system by observing the global knowledge generated by social tags. Aside from this knowledge, the user will provide the system with feedback that will be used to characterize their preferences, leading to a personalized mood recommendation system.