Liila,
We don't publish the specifications for attributes. Valence takes into account pitch, timbre, mode, and other acoustic information driven, like all the attributes, by a human-trained machine learning model applied to every track we analyze.
Thanks, David