Consp2VecD: A Dataset Based on Emotional Dynamics Expressed by Reddit Conspiracy Groups for Information Disorder Analysis

This paper introduces Consp2VecD, a publicly available dataset designed to support the analysis of Information Disorder through emotional representations extracted from Reddit communities. The dataset includes posts and comments collected from conspiracy and non-conspiracy subreddits within a common temporal window (January–October 2025), enriched with metadata and 28-dimensional emotional vectors derived using a fine-grained emotion detection model. The resource enables the joint analysis of discourse, interaction patterns, and affective dynamics at the community level. Preliminary experiments using Logistic Regression and a Multilayer Perceptron show that emotional vectors provide discriminative information between the two classes, with Logistic Regression achieving a mean accuracy of 0.874 across multiple 80/20 train–test splits. Consp2VecD aims to provide a reproducible and reusable benchmark for studying emotional dynamics in online communities and their relation to Information Disorder phenomena.