Suppressing Deceitful Influence in Cites-or-Writes Social Networks

This paper presents a method for analysing influence in a class of social networks that can be viewed as consisting of messengers and messages. A messenger may either write a message or cite (e.g., re-tweet, repost, grade, or like) a message originally written by another messenger. We term such social networks as cites-or-writes networks. The free citability of messages may lead to collusion among a small set of messengers to gain unwarranted influence. We propose a mathematically rigorous method for inferring the influence of a messenger and the impact of her message. As a key feature, our method does not require any configurable parameters. It only uses a matrix of messenger-writes-message data and a matrix of messenger-cites-message data. Another key feature of our technique is the suppression of collusion; i.e., messengers who deliberately promote each other by preferential citations are automatically suppressed in their influence. We discuss several examples of social networks that follow the cites-or-writes model. We show that our technique is fast enough to do large-scale social network analysis, and present a sample analysis for a large network.