Lyrics Features for Song Classification: Impact of Language
For song classification tasks (e.g., genre detection, hit song prediction), a large number of different types of features are available. One of those feature types are lyrics features, i.e., textual features based on the lyrics of a song. Examples of such features include n-grams, tf-idf vectors, part-of-speech features, features based on parse trees etc.
The goal of this thesis is to investigate how the expressiveness (i.e., predictive power) of such lyrics features changes for lyrics in different languages. Further, the thesis should try to construct features that work well independently of the language.