Verifying baselines for crisis event information classification on Twitter

Social media are rich information sources during and in the aftermath of crisis events such as earthquakes and terrorist attacks. Despite myriad challenges, with the right tools, significant insight can be gained which can assist emergency responders and related applications. However, most extant approaches are incomparable, using bespoke definitions, models, datasets and even evaluation metrics. Furthermore, it is rare that code, trained models, or exhaustive parametrisation details are made openly available. Thus, even confirmation of self-reported performance is problematic; authoritatively determining the state of the art (SOTA) is essentially impossible. Consequently, to begin addressing such endemic ambiguity, this paper seeks to make 3 contributions: 1) the replication and results confirmation of a leading (and generalisable) technique; 2) testing straightforward modifications of the technique likely to improve performance; and 3) the extension of the technique to a novel and complimentary type of crisis-relevant information to demonstrate it’s generalisability.