Uncovering Illegal Wildlife Trade on Social Media: Automatic Data Collection, Deep Learning Filters and Identification

Illegal wildlife trade is one of the biggest threats to biodiversity conservation, as many species, including iconic species such as rhinoceros and elephant taxa, are targeted for their meat, trophies and other body parts. Over the last years, the scale and nature of illegal wildlife trade has changed dramatically. The Internet is becoming a major market for wildlife products, as it provides cost-effective solutions, vast outreach and anonymity for illegal wildlife traders. A 2014 study by the International Fund for Animal Welfare found 33 000 items for sale on 280 online marketplaces. More recent findings suggest that the illegal market for wildlife has moved to social media and, to a lesser extent, to the dark web. So far, the use of social media data in conservation science has been limited [1]. There are survey efforts ongoing to determine the quantity, origins and destinations of illegal wildlife trade on social media, which are carried out in a manual, labour-intensive fashion. Machine-learning, and especially deep learning, has seen a surge in interest over the last decade. The amount of data produced by social media increased manifold and the involved stakeholders gained in interest in analysing the collected data on users, their actions, behaviour and networks. Deep learning describes a machine-learning approach in which input data from training sets is abstracted into a network tree of numerous abstract levels. It performs reasonably efficient on modern graphics processors and is able to process unstructured data. The use of deep learning in conservation science is still limited, but recently it has been applied for identifying wild animals in camera-trap images [2]. In a recent contribution [3] we demonstrated the usefulness of a deep learning approach in the search for illegal wildlife trade on social media. Here, we propose a framework using deep learning to filter and identify data pertaining illegal wildlife trade from social media platforms. We show how such an approach can be used to identify species and wildlife products – e.g. rhino horn and elephant ivory –, their origins, destinations, routes, and involved actors more efficiently from social media data. The proposed presentation is a methodological contribution. It will discuss, in a hands-on manner, the necessary steps to i) collect data from social media, ii) train a neural network, and iii) detect content which is relevant in the context of illegal wildlife trade. [1] Di Minin, E., Tenkanen, H. & Toivonen, T. (2015). Prospects and challenges for social media data in conservation science. Frontiers in Environmental Science. [2] Norouzzadeh, M.S., Nguyen, A., Kosmala, M., Swanson, A., Packer, C. & Clune, J. (forthcoming). Automatically identifying wild animals in camera-trap images with deep learning. [3] Di Minin, E., Fink, C., Tenkanen, H. & Hiippala, T. (2018). Machine learning for tracking illegal wildlife trade on social media. Nature Ecology & Evolution.peerReviewe