The study of political violence is commonly concerned with understanding mass violence like civil war onset or rebellion. Such events are relatively easy to identify as common sources of textual data reliably report on their occurrence. Other types of conflict, however, may not rise to such a destructive level in order to be reported in these sources. Using machine learning, this article develops a new method to identify the occurrence of violent events by mining social media text. Using a new method of natural language processing known as word embeddings in concert with a convolutional neural network, we demonstrate an increased discovery rate of politically violent events of over thirty percent compared to previously proposed methods of estimating violent events in unstructured text. Moreover, by comparing violent events discovered by our machine learning platform directly to prominent event-based datasets, we demonstrate that data coded exclusively from news reports undercounts the number of politically violent events during time periods under study. We conclude with some suggestions about how to increase reporting accuracy in event data using neural networks.

David Muchlinski is a postdoctoral research associate at the University of New South Wales, with the Atrocity Forecasting Project. The Atrocity Forecasting Project aims to develop accurate quantitative forecasts of targeted mass killings, genocides, and other atrocities by state and non-state actors. Dr. Muchlinski's research agenda centers on explaining and predicting the onset of violent conflict including civil wars, electoral violence, and genocides and other targeted mass killings. During various postdoctoral fellowships, his work has begun to focus more closely on computational social science, especially for the purposes of prediction and forecasting onsets of political violence, as well as the generation, coding, and measurement of data. The work related to this research has been published in Political Analysis, Terrorism and Political Violence, Democratization, Politics and Religion, and by several presses, including Oxford University Press.