Detection and Characterization of Influential Cross-lingual Information Diffusion on Social Networks

Social network services (SNSs) have become new global and multilingual information platforms due to their popularity. In SNSs with content-sharing functionality, such as "retweet" in Twitter and "share" in Facebook, posts are easily and quickly shared among users, and some of which can spread over different regions and languages. In this work, we first define the concept of cross-lingual information cascade on the basis of the main language of users and then try to characterize and detect those information cascades which can widely spread over different regions and languages on social networks. Understanding the cross-lingual characteristics of information cascades is not only valuable for sociological research, but also beneficial in the practical sense for those who want to know globally-influential events (e.g. ALS Ice Bucket Challenge and Terrorism in Europe) and estimate the impact of global advertisements on products (e.g. Samsung galaxy phone and a movie, Your Name). On the first attempt, we conducted statistical analysis of cascade growth and language distribution of information cascades with a large Twitter dataset. Based on the results, we propose a feature-based model, by which we successfully detected influential cross-lingual information cascades.