Combining content and temporal information for rumour stance classification
Work resulting from collaboration between the University of Sheffield, the University of Melbourne and the University of Warwick on classification of stance expressed in tweets towards rumours has been recently accepted for publication and presentation at the Annual Meeting of the Association for Computational Linguistics (ACL 2016).
This work makes progress on the state of the art on classifiers that determine the stance expressed in social media towards rumours. More specifically, the classifier is able to determine if a tweet is supporting, denying, querying or commenting on a rumour with over 65% accuracy, tested over four events collected from Twitter. We argue that the output of this classifier can be effectively used to automatically establish the likely veracity of a rumour, enabling among others the flagging of highly disputed rumours as being potentially false.
Our classifier, based on Hawkes Processes, is the first to combine textual and temporal information from tweet sequences to perform stance classification, which our experimentation proves useful over other competitive approaches. For more information on this work, please refer to our paper.
Pheme partners talk social media rumours in podcast Ontotext Work on PHEME Has Reached the Top 4 in the EC Competition for Innovation Radar Prize 2016
Comments are currently closed.