Filtering Tweets for Social Unrest

There has been substantial interest in building technologies that can use social media postings to help forecast civil unrest [1]-[3]. The Arab Spring of 2011 compellingly illustrates how social media can both reflect and influence political (in)stability [4]. Since social media data is generated on such a large and rapid scale, computational tools are potentially extremely useful in helping to render meaning from that data. While previous work has focused on forecasting specific near-term unrest events [2], in this current paper we are interested in filtering social media content for postings that are relevant to social unrest, with the idea that downstream systems or human experts would use this filtered content for further analysis. In particular, we experiment with filtering tweets written in Arabic for relevance to social unrest.