Automatic Song Title Generation: A Summarization Approach

Thesis Type Master
Thesis Status
Student Simon Hachmeier
Thesis Supervisor
Research Field

More and more research efforts are being made towards automatic music composition and generation. We can already foresee that this will be an important topic for both research and industry. However, what is still missing, are methods to automatically generate song titles that match the generated music.
Since a song title is the short textual representation of a musical piece, we formalize the song title generation problem as an automatic text summarization problem. Two paradigms are distinguished in automatic text summarization: extractive methods and abstractive methods. The output of extractive methods are key-words or key-phrases from the source, in this case the lyrics of the song. Abstractive methods, however, additionally generate a new textual sequence for content representation of the source. In this thesis, both approaches are aimed to be implemented. The research results then allow for observation of strengths and weaknesses of both of the approaches.