Causal Network Construction to Support Understanding of News

Fully understanding an event reported in a television news program, newspaper, or web site requires background knowledge of the event. Background knowledge is particularly important in understanding an event that is complicated or at a delicate stage. Without such knowledge, we can gain only a superficial or even false understanding of the event. To obtain the necessary background knowledge, users can search other articles or programs related to the event one by one. However, this can be a burden and users may still miss valuable components of knowledge. Out system that provides background knowledge of news events is needed to improve our understanding of news.


As one kind of background information, the causal relation, i.e. why the event happened, is crucial. In this paper, we propose a method to support news understanding based a novel causal relation model. Numerous technologies have been developed to extract causal relations from news articles and documents. However, for most, processing is performed per article and too much fragmented knowledge is extracted to facilitate understanding. Extracting useful background knowledge presents difficulties. Using causal relations as a basis for background knowledge requires that the whole cause relation related to the event, the root cause of the event, and other events affected by the event are provided. To achieve this aim, we propose a method of constructing a causal network by merging similar causal relations. Moreover, we propose a method of deleting unimportant causal relations.

project/en/ishii.txt · Last modified: 2011/11/25 04:52 by ylab