Using Regression to Combine Data Sources for Semantic Music Discovery

Mixture Hierarchical Model Instead of assuming that vt is normally distributed, assume that vt comes from a mixture of normal distributions Intuition: Suppose for “genre tags” clusters around 0.25 and for “acoustic tags” clusters around 0.05 then if = 0.20, then vt might have two peaks at 0.05 and -0.15. 10,870 songs 2 tag vocabularies 71 genre and subgenre tags (e.g., “rock”, “delta blues”, “trance”, “piano concerto”) 151 acoustic tags from Pandora’s Music Genome Project (e.g., “acoustic instrumentation”, “vocal harmonies”, “major key tonality”)