While social networks can provide an ideal platform for up-to-dateinformation from individuals across the world, it has also proved to be a placewhere rumours fester and accidental or deliberate misinformation often emerges.In this article, we aim to support the task of making sense from social mediadata, and specifically, seek to build an autonomous message-classifier thatfilters relevant and trustworthy information from Twitter. For our work, wecollected about 100 million public tweets, including users' past tweets, fromwhich we identified 72 rumours (41 true, 31 false). We considered over 80trustworthiness measures including the authors' profile and past behaviour, thesocial network connections (graphs), and the content of tweets themselves. Weran modern machine-learning classifiers over those measures to producetrustworthiness scores at various time windows from the outbreak of the rumour.Such time-windows were key as they allowed useful insight into the progressionof the rumours. From our findings, we identified that our model wassignificantly more accurate than similar studies in the literature. We alsoidentified critical attributes of the data that give rise to thetrustworthiness scores assigned. Finally we developed a software demonstrationthat provides a visual user interface to allow the user to examine theanalysis.
展开▼
