Text classification of UK smallholding communities through Twitter

Abstract

Within the UK, livestock holdings are registered so that livestock can be traced, and animal diseases be controlled. These regulations are enforced irrespective of farm size, however, tend to be better followed on traditional farms, whereas holdings new to keeping livestock are less likely to be aware of their obligations. These smallholdings thereby may evade registration and are less likely to participate in national disease surveillance and ultimately complicate national animal disease control. Less information is known about small-scale livestock keepers, in particular those without a traditional farming background. Smallholders have been known to play a vital role in zoonotic disease outbreaks and more action needs to be taken to improve surveillance systems by incorporating this demographic into current intelligence. Literature indicates that parts of these communities often utilise social media as a means of communication and information sharing. Twitter followers from a prominent smallholder user in the UK were extracted and manually categorized as a smallholder or not, based on profile descriptions. Manual coding of just under 1,000 Twitter profiles was conducted to build a robust training dataset. Text classification algorithms were applied on this annotated data, and the resulting classification algorithms produced accuracies of over 80%. Results indicate that classification can prove to be a highly successful tool, if a sufficient training dataset is curated, and there is enough textual information within the user profiles on social media.