Abstract
Social networks are important communication channel where individuals and emergency agencies can exchange information during disasters. The ability to detect disaster information or 'reporting' tweets would provide many advantages in disaster management during crowded events. This study explores Twitter behaviour during the Mina stampede tragedy in the 2015 Hajj by processing tweets posted over seven days during and after the incident (24-30 September 2015). Statistical features were derived from tweets, such as the number of hashtags, user mentions, and links, to provide an overview of the use of Twitter during this disaster. A classification model was built to filter reporting tweets using two Arabic natural language processing tools: Farasa and MADAMIRA. A support vector machine with a radial basis function kernel generated the best results in both tools (F-score: 88%-89%). The results will be useful to those who manage large, crowded events such as Hajj in Arabic-speaking regions.