Semantic contours in tracks based on emotional tags

Andrius Butkus

pp. 45-66

in: Kristoffer Jensen (ed), Computer music modeling and retrieval. genesis of meaning in sound and music, Berlin, Springer, 2009

Abstract

Outlining a high level cognitive approach to how we select media based on affective user preferences, we model the latent semantics of lyrics as patterns of emotional components. Using a selection of affective last.fm tags as top-down emotional buoys, we apply LSA latent semantic analysis to bottom-up represent the correlation of terms and song lyrics in a vector space that reflects the emotional context. Analyzing the resulting patterns of affective components, by comparing them against last.fm tag clouds describing the corresponding songs, we propose that it might be feasible to automatically generate affective user preferences based on song lyrics.