IARank: Ranking Users on Twitter in Near Real-Time, Based on Their Information Amplification Potential

This work introduces IARank, a novel, simple and accurate model to continuously rank influential Twitter users in real-time. Our model is based on the information amplification potential of a user, the capacity of the user to increase the audience of a tweet or another username that they find interesting, by retweets or mentions. We incorporate information amplification using two factors, the first of which indicates the tendency of a user to be retweeted or mentioned, and the second of which is proportional to the size of the audience of the retweets or mentions. We distinguish between cumulative influence acquired by a user over time, and an important tweet made by an otherwise not-important user, which deserves attention instantaneously, and devise our ranking scheme based on both notions of influence. We show that our methods produce rankings similar to PageRank, which is the basis for several other successful rankings of Twitter users. However, as opposed to PageRank-like algorithms, which take non-trivial time to converge, our method produces rankings in near-real time. We validate our results with a user-study, which shows that our method ranks top users similar to a manual ranking produced by the users themselves. Further, our ranking marginally outperformed PageRank, with 80% of the Top 5 most important users being classified as relevant to the event, whereas, PageRank had 60% of the Top 5 users marked as relevant. However, PageRank produces slightly better rankings, which correlates better with the user-produced rankings, when considering users beyond the top 5.