Context representation using word sequences extracted from a news corpus

Word meaning changes dynamically depending on context. We need to specify the context to identify this meaning. However, context varies depending on specificity of the topic and the viewpoint of the writer. In this paper, we propose that a word sequence can be used to identify context. Both contexts identified by word sequences and word sets related to the contexts are shown concretely. We used 800,000 Reuters news articles, and extracted the word sets using the confabulation model and five statistical measures as relations. We compared the measures and found that cogency and mutual information were the most effective. We demonstrate the usefulness of the word sequence to identify the context.