Discovering Complex Networks of Events and Relations in News Surveillance

When faced with the need for analyzing vast streams of on-line text data, we require methods that go well beyond keyword-based queries. Large-scale surveillance of on-line news streams requires an understanding of the text on a deeper level than is afforded by names and keywords alone, it becomes essential to understand complex interactions among the entities relationships and events. We will discuss the interplay between two aspects of this kind of deep analysis: a. how to extract knowledge from text "upstream" and b. how that knowledge may be utilized in downstream applications. We will use as live examples several systems in different application domains: cross-border crime and security, epidemiological surveillance, and business intelligence. We will present the experiences from the development of such systems and from interaction with real-world users, who are experts in their respective domains.