The Development of Music Recommendation Systems Based on Song Characteristics

This research constructed four music recommendation systems based on the combination of sentiments in a specific pieces of lyrics, the semantic difference of words within specific pieces of lyrics (with two more specific types, one using Word Mover's Distance and one simply utilizing the Doc2Vector method), and a generalized approach finding similar vectors between various pieces of music containing the musician, genre, and lyric content. The last method has been applied to multiple participants and five of which (that filled out the survey) gave an average rating of 93.8 out of 100 for the general quality of recommendation.