Subjective Evaluation of Vector Representation of Emotion Flow for Music Retrieval

Because it simply consists of an initial point and a terminal point in a two dimensional emotion plane, vector representation of music emotion provides an intuitive and instant visualization of the dynamics of music emotion. In this paper, we investigate the performance of this representation for music information retrieval by conducting a series of subjective tests. A music retrieval system is created, and the user experience data are evaluated by seven metrics: learnability, ease of use, affordance, usefulness, joyfulness, novelty, and overall satisfaction. Compared with the point representation, the vector representation performs relatively better in affordance, novelty, and joyfulness but slightly worse in learnability and ease of use. The overall satisfaction score is 5.19 for the point representation and 5.43 for the vector representation. The results suggest that each representation has its own strengths, and the choice between the two representations depends on which metrics carry more weight in an application at hand.