Researchers at the University of Melbourne in Australia have developed a framework to evaluate whether an event reported on Twitter is a likely witness account by assessing the evidence of whether the poster is on the ground at the event.
The framework analyzes the textual and visual content of the tweet, seeking inferences that the poster was at the event they are tweeting about, and then testing that by looking for evidence of their presence. In the tweet's text, statements such as observations of the event, attached images, and the existence of geotags in the metadata build the user's credibility.
The framework surmounts various challenges by exploring different evidence sources within tweets, with procedures applied to exclude postings that cannot support inferences that the tweeter is at the event. Supervised machine-learning techniques then apply classification models to pull up evidence from the remaining tweets that support inferences that the tweeter is at the event.
From Pursuit (Australia)
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found