Mapping (Dis-)Information Flow about the MH17 Plane Crash

Digital media enables not only fast sharingof information, but also disinformation. Oneprominent case of an event leading to circu-lation of disinformation on social media isthe MH17 plane crash. Studies analysing thespread of information about this event on Twit-ter have focused on small, manually anno-tated datasets, or used proxys for data anno-tation. In this work, we examine to what ex-tent text classifiers can be used to label datafor subsequent content analysis, in particularwe focus on predicting pro-Russian and pro-Ukrainian Twitter content related to the MH17plane crash. Even though we find that a neuralclassifier improves over a hashtag based base-line, labeling pro-Russian and pro-Ukrainiancontent with high precision remains a chal-lenging problem. We provide an error analysisunderlining the difficulty of the task and iden-tify factors that might help improve classifica-tion in future work. Finally, we show how theclassifier can facilitate the annotation task forhuman annotators