Exploring Event Evolution Patterns at the Atomic Level.

Lei Deng,Zhaoyun Ding,Bingying Xu,Bin Zhou,Yan Jia,Peng Zou

DOI: https://doi.org/10.1109/cyberc.2011.16

2011-01-01

Abstract:The event evolution mining for news corpus is beneficial for people who are less interested in the set of documents related by a topic rather than the underlying stories. Most of state-of-the-art approaches which derived from the TDT field considered events at the document level, which made different granularity for each event in evolution graph. In this paper, we consider events from a unified perspective by introducing the IE techniques into this task. We propose an unsupervised approach to explore event evolution patterns through extracting atomic event from documents, identifying their co-reference, and measuring their relationship automatically. And then we construct the event evolution graph. Meanwhile, we propose two policies to find a newsworthy subset from a mass of atomic events in a news corpus to simplify the event evolution graph. Finally, we show experimentally that our method which works on a Chinese news corpus can construct the atomic event evolution graph, whose vertexes are standing for important events of the topic, and edges are reasonable relationship between their adjacent events.