The Global Terrorism Database (GTD) research team, in collaboration with researchers from the University of Maryland’s Computer Science (COMSCI) department, undertook a pilot project to evaluate the potential efficiency gains that could be achieved by relying more heavily on the use of artificial intelligence (AI) to compile the database. The primary motivation for this initiative was to reduce the time lag between the occurrence of real-time terrorist events and GTD data collection, as well as to reduce the costs associated with producing the data. A key area where it was hypothesized that the use of AI could have a positive impact is the time required for human analysts to identify and code events that meet the GTD’s inclusion criteria.
To evaluate the potential of using AI more extensively in the collection of the GTD, we conducted two experiments using different natural language processing (NLP) methodologies (i.e., automated and computational techniques for extracting information from text). The first experiment focused on automatically identifying individual terrorist attacks and clustering together all documents referring to the same events from a pool of potentially relevant news articles. The second experiment aimed to automatically extract detailed information about each attack. Our core research question was whether AI tools could correctly identify unique terrorism events from global news sources and then correctly map the relevant features of the events to the variables included in the GTD.
Publication Information
Colon, Carlos R., Benjamin Evans, and Margaret A. Hayden. 2024. "The Global Terrorism Database: Experiments in Machine-Assisted Data Collection." College Park, MD: START (December).