Classification and Visualization of Lyric Collections Using Guided LDA

Lyrics are one of the most important components of music, and they have a great impact on the appreciation of songs such as J-POP. Therefore, it is useful to classify and search songs based on lyrics. However, the impression of lyrics is subjective and may be influenced by musical elements other than lyrics, so the criteria for searching for lyrics required by users may vary from person to person. To address this issue, we are working on a research project to support active lyric search by visualizing the distribution of lyrics. Here, it is often difficult to appropriately calculate the distribution of the lyrics because lyrics have a higher degree of lexical freedom than articles and papers. In this study, we propose a method to visualize the distribution of lyrics calculated applying guided LDA (Latent Dirichlet Allocation) that interactively consumes guided words. This method facilitates the iterative visualization of lyric classification results based on the users' viewpoints. It also makes it possible to search for songs by focusing only on lyrics without taking other musical elements into account. Users can observe the differences in individuality and tendency of songs and artists, and the diversity of lyrics, by using the visualization results.