Several feature phrases including combinations of words are predetermined to indicate a feature of a song. TF-IDF values are computed based on appearances counts of the feature phrases. A song feature quantity is computed as a vector constituted by the TF-IDF values. This vector is an index indicating the feature of the song. Using the feature phrases results in proper exclusion of a comment not related to the feature of the song itself in computing the song feature quantity. This allows the song feature quantity to match, with a high accuracy, with a mood generated when the song is actually listened to.