Distinguishing Individuals from Organisations on Twitter

Sunghwan Mac Kim, Cécile Paris, Robert Power, Stephen Wan

Abstract: Twitter is a popular public platform for individuals to share information and express opinions. It is also widely used as an active communication channel by many organisations. Analysis results about public opinion could thus be misrepresented and distorted if the posts generated by non-individuals are appropriately handled motivating the need to identify the type of user accounts for social media analytics. In this paper, we demonstrate the utility of a joint text and network representation for classifying if a profile belongs to an individual or organisation. Crucially, combining the two feature types has not been shown before for this task. Our experimental results show that performance gains are possible but only when using our novel adaptation of the Text-Associated DeepWalk(TADW) method.