Exploiting Structured News Information to Improve Event Detection via Dual-Level Clustering

With massive amount of news articles come to fore every seconds, helping people navigate through correlated news on a given topic of interest is a pressing need for news portals. This paper tackles Event Detection task, which refers to aggregate news articles report on the same event into tightlyfocused, topic-centered news sets. We propose a novel Dual-level Clustering Model based on News Representation with Time2vec. The advantage of our model lies in taking structured news information into consideration by constructing news vector with context vector and time vector. Besides, the effect of key entities has been taken into account by Dual-level Clustering framework. Experiments conducted on real world news corpus illustrate that our model achieves considerable success compared with several 'state-of-the-art' methods.