The Digital Detective's Discourse - A toolset for forensically sound collaborative dark web content annotation and collection

Abstract

In the last decade, the proliferation of machine learning (ML) algorithms and their application on big data sets have benefited many researchers and practitioners in different scientific areas. Consequently, the research in cybercrime and digital forensics has relied on ML techniques and methods for analyzing large quantities of data such as text, graphics, images, videos, and network traffic scans to support criminal investigations. Complete and accurate training data sets are indispensable for efficient and effective machine learning models. An essential part of creating complete and accurate data sets is annotating or labelling data. We present a method for law enforcement agency investigators to annotate and store specific dark web content. Using a design science strategy, we design and develop tools to enable and extend web content annotation. The annotation tool was implemented as a plugin for the Tor browser. It can store web content, thus automatically creating a dataset of dark web data pertinent to criminal investigations. Combined with a central storage management server, enabling annotation sharing and collaboration, and a web scraping program, the dataset becomes multifold, dynamic, and extensive while maintaining the forensic soundness of the data saved and transmitted. To manifest our toolset's fitness of purpose, we used our dataset as training data for ML based classification models. A five cross-fold validation technique was used to evaluate the classifiers, which reported an accuracy score of 85 - 96%. In the concluding sections, we discuss the possible use-cases of the proposed method in real-life cybercrime investigations, along with ethical concerns and future extensions.