Automating Event Extraction for the Security Domain
This chapter presents on-going efforts at the Joint-Research Center of the European Commission for automating event extraction from news articles collected through the Internet with the Europe Media Monitor system. Event extraction builds on techniques developed over several years in the fields of information extraction, whose basic goal is to derive quantitative data from unstructured text. The motivation for automated event tracking is to provide objective incident data with broad coverage on terrorist incidents and violent conflicts from around the world. This quantitative data then forms the basis for populating incident databases and systems for trend analysis and risk assessment.
A discussion of the technical requirements for information extraction and the approach adopted by the authors is presented. In particular, we deploy lightweight methods for entity extraction and a machine-learning technique for pattern-based event extraction. A preliminary evaluation of the results shows that the accuracy is already acceptable. Future directions of improving the approach are also discussed.
Unable to display preview. Download preview PDF.