The visual image has long been central to how war is seen, contested and legitimised, remembered and forgotten. Archives are pivotal to these ends as is their ownership and access, from state and other official repositories through to the countless photographs scattered and hidden from a collective understanding of what war looks like in individual collections and dusty attics. With the advent and rapid development of social media, however, the amateur and the professional, the illicit and the sanctioned, the personal and the official, and the past and the present, all seem to inhabit the same connected and chaotic space. However, to even begin to render intelligible the complexity, scale and volume of what war looks like in social media archives is a considerable task, given the limitations of any traditional human-based method of collection and analysis. We thus propose the production of a series of ‘snapshots’, using computer-aided extraction and identification techniques to try to offer an experimental way in to conceiving a new imaginary of war. We were particularly interested in testing to see if twentieth century wars, obviously initially captured via pre-digital means, had become more ‘settled’ over time in terms of their remediated presence today through their visual representations and connections on social media, compared with wars fought in digital media ecologies (i.e. those fought and initially represented amidst the volume and pervasiveness of social media images). To this end, we developed a framework for automatically extracting and analysing war images that appear in social media, using both the features of the images themselves, and the text and metadata associated with each image. The framework utilises a workflow comprising four core stages: (1) information retrieval, (2) data pre-processing, (3) feature extraction, and (4) machine learning. Our corpus was drawn from the social media platforms Facebook and Flickr.
What We Wore and Sketchfab
