Applying machine learning to identify musical taste

Discovering the musical taste of a person has an obvious application in recommendation mechanisms used by music service providers. We are interested in a less obvious application, related to the work environment of a software developer. In this work we compare two algorithms used in data mining as classifiers. The goal is to compare Support Vector Machine (SVM) and k-Nearest Neighbor (k-NN) as predictors of the musical taste of a user. We use a database of songs previously classified with a label indicating whether the user likes or dislikes each song. The database includes features of the song; each classifier uses the same combinations of features in the learning process and then classifies new instances of songs according to the user's predicted taste. This initial study indicated SVM as a better predictor than k-NN for this particular context. Future investigations intend to evaluate the user in a synchronous environment, our hypothesis is that it might be possible to understand more than the like / dislike scenario and expand to what the user wants to hear at a particular moment, capturing her mood. Eventually correlate the mood of a software developer to the fault proneness of the code she has written.