Rumour stance classification, the task that determines if each tweet in a discussing a rumour is supporting, denying, questioning or simply on the rumour, has been attracting substantial interest. Here we a novel approach that makes use of the sequence of transitions in tree-structured conversation threads in Twitter. The conversation are formed by harvesting users' replies to one another, which results a nested tree-like structure. Previous work addressing the stance task has treated each tweet as a separate unit. Here we analyse by virtue of their position in a sequence and test two sequential, Linear-Chain CRF and Tree CRF, each of which makes different about the conversational structure. We experiment with eight datasets, collected during breaking news, and show that exploiting the structure of Twitter conversations achieves significant improvements the non-sequential methods. Our work is the first to model Twitter as a tree structure in this manner, introducing a novel way of NLP tasks on Twitter conversations.
No takes yet. Share an insight, caveat, or question.
Zubiaga et al. (2016) studied this question.