Event detection based on on-line news clustering

• Main Author: Shi, Ke
• Other Authors: Mao Kezhi
• Format: Theses and Dissertations
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/76044
• Institution: Nanyang Technological University

The terrorist attack directly affects personal safety, and it also has a lasting impact on international politics, civil liberties, and the economy. Internet produces massive amounts of terrorist attack news every day, o how to extract news of interest is time-consuming work. In order to provide organized information to readers, clustering technology is used to automatically arrange vast news. In this project, a document representation model is trained by CNN and LSTM to represent each news as a 48-dimensional vector. Meanwhile, a hierarchical structure is designed to do the K-means and Affinity Propagation clustering. The first step is to cluster samples by locations, and the second step is to cluster samples by content information. As a result, the overall model obtains a satisfactory performance as Purity at 85.19%, RI at 82.12% and NMI at 76.42%.