Conference article

This paper describes a method for extracting attractive phrases of lyric focusing on linguistic expressions. Not only chorus but also linguistic expressions seem to be a cause of attractive phrases. We conducted impressive evaluation experiments to clarify the important factors of attraction of phrase. As the result; it was confirmed that “uniqueness of co-occurred terms” and “repetition” especially influenced attraction. Therefore; we modeled the uniqueness of co-occurred terms and repetition as seven mathematical features. And the proposed method detected attractive phrases using support vector machine with the modeled features; which is known as a high performance pattern recognition method. Through the attractive phrase detection experiments; we confirmed availability of the proposed method: the accuracy level and the precision was each 69% and 86%; respectively. Moreover; we discussed about the correctly detected attractive phrases comparing key sentences detected by the existing summarization methods. As the result of the discussions; the proposed method correctly detected the phrases that were ranked in low by the conventional methods though human evaluated the phrases as attractive. From these facts; it was suggested that lyrical linguistic expressions were well modeled in the proposed method; and the proposed method detected the attractive phrases better than the existing summarization method.

Music; Lyric; Attractive Phrase; Natural Language Processing