Invited Talk by Michal Lukasik, University of Sheffield on Modeling temporal dynamics of rumours in social media
The ability to model rumour dynamics helps with identifying those, which, if not debunked early, will likely spread very fast. In this talk we describe models of rumour prevalence based on point processes. In the first approach we developed a multi-task learning method parametrized by text from posts, allowing data statistics to be shared between groups of similar rumours (ACL 2015, EMNLP 2015). In the follow-up work, we developed a convolution kernel for comparing social media events and showed how it can be applied for rumour dynamics modeling (AAAI 2016). Evaluation on tweets from the 2014 Ferguson riots demonstrates that our models outperform several strong baseline methods for rumour popularity prediction.
Michal Lukasik did his PhD at the University of Sheffield. His PhD topic is on probabilistic modeling of rumours in social media. Michal spent a significant part of his PhD visiting the NLP group at the University of Melbourne. During his PhD he was an intern at Google Research in NYC and later at the Max Planck Institute in Germany.
Michal worked on a range of problems, including rumour stance classification, meme popularity prediction, visibility maximization in social networks, entity recommendation, and quality estimation for machine translation