In this project, we aim to classify songs into genres using their lyrics. It is challenging for humans to perform this task, and there often is debate where a song fits since boundaries are not clearly defined and genres are overlapping. After preprocessing our data, we trained our own GloVe embeddings of the song lyrics and created different visualizations to better understand our data. As a baseline, we used our GloVe embeddings in two logistic regression models to classify them into genres. Then, we balanced our dataset so that there was a very similar number of lyrics for each of the genres. Finally, using our GloVe embeddings, we trained an LSTM model and bidirectional LSTM model. Our best LSTM model achieved an accuracy of 68%.
