Text Segmentation Using a Change of Keywords' Activation Levels

摘要

We propose, in this paper, a new method to segment a text into thematic units. The method enables us to find boundaries between segments by detecting changes of contextural information which approximates node activation process of a semantic network. An experiment to decide paragraph boundaries in 116 input texts, extracted from the column entitled 'Shunjyu' of 'Nihon Keizai Shimbun (Japanese business daily newspaper)', has shown that precision and recall rates are 67.8% and 64.8%, respectively.