To retrieve music by mood tags in a social network, we introduce a mood vector, which allows moods of music pieces and mood tags to be represented Pinternally by numeric values. A mood vector consists of 12 arousal-valence pairs each represents a mood of Thayer's two-dimensional mood model. To determine the mood vector of a music piece, a Support Vector regressor is created for arousal and valence each using features of a music piece. Then, the regressors predict a mood vector. To map a mood tags to its mood vector, we investigate the relationship between them based on tagging data retrieved from Last.fm. To show the benefits of the proposed method, in this paper, we create a test set by using last.fm tags and their synonyms, and measure its retrieval performance over the keyword-based approach using the test set. The results illustrate that the proposed method can be useful in many respects including solving the problem caused by synonyms.
Available online: https://ieeexplore.ieee.org/document/8436833