Advanced Summer Research Scholar
Major: Math and Computer Science
Faculty mentor: Filippo Menczer
This research provides a tool to measure the trustworthiness of any account in a retweet network based on data about a few of the accounts. The main finding of this project is we can Identify accounts that spread links to low-credibility news sources on Twitter by applying PageRank-like propagation algorithms to retweet networks. We tested five algorithms: Weighted PageRank, Weighted Personalized PageRank, Reputation Scaling with Weighted PageRank, Reputation Scaling with Weighted Personalized PageRank, and TrustRank. We built a directed, weighted retweet network from tweets about Covid-19 where the links pointed from the retweeted account to the retweeting account and the weight corresponded to the number of times the user was retweeted by another user. The accounts in this network were labelled as either “trustworthy” or “untrustworthy” based on if they had retweeted a link to a low-credibility news source (gathered from Hoaxy and Newsguard), and we gave the algorithm approximately half of the low-credibility accounts as seeds to see if the algorithm could identify the other half of the low-credibility accounts. We used the AUC score, or the area under the ROC curve, to measure the performance of these algorithms. For AUC score greater is better and a score of 0.5 indicates randomness. All the methods showed some predictive power with AUC scores above 0.5 and TrustRank performed the best on our dataset with an AUC score of 0.644. This shows that PageRank and PageRank-like propagation algorithms can be applied to retweet networks to gain knowledge of low-credibility users, and with more fine tuning we hope to improve accuracy and perhaps build a tool based on this knowledge.