What this kind of analysis needs is a blind test. Ask three groups to rate text independently in the same metrics - 1) General public. 2) Psychologists 3) Algorithm.

The results should be interesting. However, if the authors claim subliminal effects, then I do not how that could be tested.