Conspiracy Narratives on Voat: A Longitudinal Analysis of Cognitive Activation and Evolutionary Psychology Features
This repository contains an annotated Voat.co dataset and a script for prompting GPT-3.5-turbo related to establishing minimal sufficienct features for conspiracy narratives from evolutionary psychology. Binary features for each posting/comment are:
Secrecy
Pattern
Action
Actor
Threat
The corresponding text spans are:
Secrecy_span
Pattern_span
Action_span
Actor_span
Threat_span
Further, overall_ct
was annotated to to assess the conspiracy status irrespective of the five features.
The dataset is based on Voat.co a deplatformed online platform that resembles in structur Reddit and was known to host deplatformed Sub-Reddits (Mekacher & Papasavva, 2022). The dataset (N = 3,384) is a span-level multi-label conspiracy dataset [voat_annotation.csv]. It consists of postings and comments between 2014-06-20 and 2020-12-23 that each have been annotated on a text span level for the presence of five minimal sufficient evolutionary psychology features. The present subset builds on a dataset that has been originally collected by:
Mekacher, A., & Papasavva, A. (2022, May). " I Can’t Keep It Up." A Dataset from the Defunct Voat. co News Aggregator. In Proceedings of the International AAAI Conference on Web and Social Media (Vol. 16, pp. 1302-1311). https://doi.org/10.5281/zenodo.5841668
Different subverses have been sampled for this dataset, namely:
/v/anon
,/v/Conspiracy
,/v/GreatAwakening
,/v/pizzagate
,/v/theredpill
( each n = 1,880)/v/gaming
,/v/news
,/v/Science
,/v/Showerthoughts
(each n = 1,504)
You can find the dataset file in the data/
directory.
The prompting_script.txt
file in this repository provides a script for generating prompts for GPT-3 and the underlying annotation logic to discern the five features from each other.