Data for Automatic Detection of Influence Actors in Disinformation Networks

This repository contains additional data used for this paper:
S.T. Smith, E.K. Kao, E.D. Mackin, D.C. Shah, O. Simek, and D.B. Rubin, "Automatic detection of influential actors in disinformation networks," Proc. Natl. Acad. Sci. U.S.A., to appear, doi:10.1073/pnas.2011216118.

The data represent narrative networks for both the English (en) and French (fr) language narratives analyzed in the paper. The comma-separated value (.csv) files en_influence_network.csv and en_tweet_time_weight.csv represent the English narrative network, and the correspondingly named files represent the French network.

The files *_influence_network.csv contain a directed graph whose vertices are defined by the Twitter user id along with edge weights determined by the number of times a Twitter user (to column) retweets another user (from column) within the corresponding narrative, so that edge direction corresponds to the direction of influence. This graph is used as the prior Poisson distribution of the influence network, described in the main paper, Section Methodology: Network Discovery.

The files *_tweet_time_weight.csv contain a list of tweets on the narrative by specific Twitter users (uid column), tweet id (tweet_id), tweet or retweet time (tweet_time) in coordinated universal time (UTC), and weight of the tweet (narrative_weight) within the narrative represented in Figure 2 of the main paper. These tweets are the observed outcomes used for impact estimation of a specific narrative, described in the main paper, Section Methodology: Impact Estimation.