Cluster-Centric Approach to News Event Extraction
Jakub Piskorski, Hristo Tanev, Martin Atkinson, Van Der Goot Erik · Frontiers in artificial intelligence and applications · 2008
This paper presents a real-time and multilingual news event extraction system developed at the Joint Research Centre of the European Commission. It is capable of accurately and efficiently extracting violent and natural disaster events from online news. In particular, a linguistically relatively lightweight approach is deployed, in which clustered news are heavily exploited at all stages of processing. The paper focuses on the system's architecture, real-time news clustering, geolocating clusters, event extraction grammar development, adapting the system to the processing of new languages, cluster-level information fusion, visual event tracking and accuracy evaluation.