Exploiting Tree-Structured Conversations for Rumour Stance Classification
A new paper led by the University of Warwick, in collaboration with the University of Sheffield, has explored the ability to classify rumour stance by leveraging the discursive nature of Twitter conversations. The paper is titled “Stance Classification in Rumours as a Sequential Task Exploiting the Tree Structure of Social Media Conversations“, published at the Natural Language Processing conference COLING, held in Osaka, Japan, in December 2016.
The paper exploits the tree structure of Twitter conversations, where a tweet discusses a rumour, with a nested set of replies responding to it. Tweets can respond by either supporting, denying, questioning or commenting on the rumour. To exploit the tree structure, the paper explores the use of two different settings of Conditional Random Fields (CRF), a sequential classifier: Linear CRF and Tree CRF. Comparing with a set of competitive baseline classifiers, the paper shows that the use of a sequential classifier can substantially boost performance, concluding that exploitation of discourse is helpful for rumour stance classification.