Silvia,
The beta value for the modulating variable is not percent signal change. Instead it is a kind of slope of the modulating effect relative to the percent signal change of BOLD response.
Whether you need to normalize the modulating effect depends on whether the emotional valence is comparable across subjects. For example, does a valence of 5 for one subject mean the same thing for another subject? If not, normalization within each subject might be desirable before you take those slopes for group analysis.
Gang