Twibo: comparing very large communities via massive social media datasets

Abstract

Online social media are becoming the standard infrastructure for social communication and dissemination of information. As social media platforms not only passively provide infrastructure but also actively perform algorithmic curation for their profit and user experience, an important concern, often called ``filter bubble'' arises: people are trapped in their own personalized bubble--being exposed only to the opinions that conform their beliefs and political positions, thus potentially creating social polarization and information ``islands''. Although the adoption of social media is an international phenomenon, language difference and policy barrier also create information islands. The goal of this paper is to develop methods/system to cross-link concepts and communities in different social media, and leverage them to study the extent and impact of filter bubbles. To accomplish this goal, the main objectives in this paper are to develop text/graph mining methods to connect concepts and entities in Twitter and Weibo through Wikipedia; and to compare two social media in the dimension of topics and networks to quantify the significance of language bubble.