Technology

our results indicate that hate speech communities on Reddit share speech patterns with Reddit communities for Schizoid Personality Disorder, three Cluster B personality disorders (Borderline, Narcissistic, and Antisocial), and Complex Post-Traumatic Stress Disorder. While the Cluster B disorders would be expected due to prior studies into the Dark Triad and hate speech, this nonetheless offers confirmation that they are acting similarly to their Dark Triad counterparts when it comes to hate speech. Furthermore, the two non-Cluster B disorders have not been discussed in hate speech or misinformation literature to the best of our knowledge, and offer new routes of investigation.

The association in speech patterns between certain psychiatric disorders and hate speech (and misinformation to a lesser extent) suggests that, despite hate speech and misinformation being a social phenomenon, framing and approaching hate speech/misinformation as though they were psychiatric disorders may prove beneficial in combating them. For example, designing counter-messaging against these issues could utilize elements of therapy used for the psychiatric disorders that are most similar to hate speech and misinformation. Although hate speech embeddings being classified most frequently as Cluster B personality disorders could be seen as problematic due to those disorders’ historical resistance to treatment, the successes in treating Borderline Personality Disorder through Dialectical Behavior Therapy indicates that treatment of these disorders is not impossible and could be applied here [59,60]. Furthermore, although hate speech shares similarities with Cluster B personality disorders, that does not mean that it actually is a Cluster B personality disorder; it is most likely less resistant to change than a personality disorder is.

Misinformation’s relations to psychiatric disorder communities proved to remain elusive; a clear connection can be seen in the TDA mapping, and zero-shot classification had over 25% of the misinformation embeddings assigned to psychiatric disorders, but it was difficult to elucidate a clear pattern with the dataset at hand. A more robust dataset may be able to identify clearer trends.