Towards credibility of micro-blogs: characterising witness accounts

Information about events can be opportunistically harvested from social media, however, a major challenge is assessing the credibility of the information derived, and the credibility of the micro-bloggers who are the source of the information. Witnesses to events are intrinsically linked with credibility for many disciplines including journalism and the criminal justice system. This research seeks to determine whether likely witness accounts of an event can be differentiated from social media feeds. A conceptual model of a witness account, and related impact accounts and relayed accounts is developed. Additionally, influence regions defining a relationship between witnesses and events are inferred, from different categories of witness accounts. This model is explored and tested using a bushfire event as a case study. In depth manual analysis of Twitter data related to this event and its effects, confirms the expected revelations of characteristics of direct observations of a bushfire that witnesses report, and the impacts and actions potential witnesses report. A visualisation of influence regions for smoke and traffic congestion observations is provided. Additionally, for the case study event, it is observed that witness accounts contain fewer place name references, but more personal place descriptions such as ‘my home’. These findings suggest implications for automatic data mining from place descriptions that will enable an assessment of the credibility of extracted event information.