JRC terrorism and extremism database : data quality optimisation for machine learning

The JRC Terrorism and Extremism Database is a tool created by the Joint Research Centre to gather data on terrorism and extremism-related events in the European Union, as well as on the global level. By automatically categorising and grouping events using machine-learning approaches, the database provides insights for short term event monitoring, developing trends and can support related threat assessments. Building on previous work on the subject, this report focuses on the preliminary research conducted to further improve data quality using machine-learning models. The methodology section of this report is crucial, given its technical nature. Initially, the report will provide a comprehensive overview of Machine Learning in the context of text classification. It will underscore the significance of preprocessing and vectorisation, and provide related examples. Furthermore, it will expound the logic behind the selection, evaluation, and deployment of models. Subsequently, the report will offer an overview of various considerations pertinent to the application of Machine Learning techniques in text classification tasks. It is important to note that these models can be prone to biases and require ongoing monitoring to ensure their effective operation. Moving closer to the subject of classification, the report will briefly review the literature on using Machine Learning methods in Terrorism studies, outline the rationale for the new categorisation, and introduce the categories created for automated text classification. The next section will outline the outcomes of an initial application of a Machine Learning model on a corpus of manually labelled news articles. These results reveal a bias within the model, the potential causes of which will be discussed. Finally, the report will explore the current limitations inherent in the existing data and propose a framework to further enhance the quality of the Database, pending the development of a viable Machine Learning model.