Topological data mapping of online hate speech, misinformation, and general mental health: A large language model based study (PLOS. Digital Health)

The emergence of social media has sparked grave worries about how it may contribute to discrimination and even social violence in the United States by disseminating false information and hate speech. It is unclear how these actions relate to users’ psychological wellbeing more broadly, despite some research connecting them to particular personality features. Analyzing enormous volumes of social media data to find hidden patterns is a major issue. Large language models and machine learning were used in this study to solve the problem. GPT-3 was used to integrate thousands of Reddit posts from specific communities, resulting in semantic-rich vectors. To investigate connections between speech patterns and community characteristics, these embeddings were examined using classification models. Lastly, relationships between hate speech, disinformation, mental illnesses, and general mental health were depicted using topological data analysis (TDA).

https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000935