It is postulated that if correlation between the aective load of lyrics and the
audio features exists, not only the audio features, but also the lyrics can be used
to produce more precise playlist suggestions in a variety of applications such as
online radios and mobile music players. The proposed method of extraction
of aective load data from lyrics transformed into a bag-of-words model [3]
using the LSA [4] extended Aective Norms for English Words [5] is not
resource demanding and can be performed on a mobile device most likely faster
than in-depth audio analysis. Aective data can be used to facilitate playlist
suggestions, possibly extending systems such as PATS and PlaylistDJ discussed
in [6] by serving as an additional music describing parameter.