Searching and tracking events in news data is becoming a crucial issue for data analysts. This paper builds on a seminal work sponsored by DARPA in the 1990s. The objective is to provide a graphical tool, EventRiver, that allows the user to browse the flow of news and events in an efficient way. The paper is well documented, and describes clearly a number of previous studies that address the same issue. The authors propose a novel approach that is better than other approaches that qualify as state of the art. An information watcher can leverage the proposed platform for understanding the course of events.
The authors build their new system in three main steps. First, they use text mining and clustering techniques to regroup the news into events, and the events into stories. They adapt classic clustering algorithms to deal with texts as vectors of keywords. In the second step, they design a visualization framework based on “bubbles” positioned on a timeline. In the third step, the authors provide the system with an appropriate interface to help the user browse events and stories. The full platform is tested using qualitative and quantitative methods to convince the reader that EventRiver is better than the previous systems, IN-SPIRE and LensRiver. The evidence is mainly obtained by taking a close look at the results of the different approaches and the collection of user feedback.