Totally agree, it's a little bit weak. Also the number of training and validating instances is somehow low... BUT anyhow interesting that it seems as if there a

Totally agree, it's a little bit weak. Also the number of training and validating instances is somehow low... BUT anyhow interesting that it seems as if there are enough features for prediction of mental illnesses in text. And it's a great approach to develop quantitative instruments for diagnosis of this "soft illnesses".