Important Sentence Extraction Using Contextual Semantic Network

In this paper, we propose a method for calculating important scores of sentences for text summarization. In this method, Contextual Semantic Network is used to calculate scores of importance for sentences included in input documents. The Contextual Semantic Network is constructed by using the Associative Concept Dictionary which includes semantic relations and distance information among the words in the documents. The concept dictionary was built using the results of association experiments which adopted basic nouns as stimulus words in Japanese elementary school textbooks. For evaluating the method, we compared the quality of the important score ranking obtained from our proposed method with that obtained from human subjects and that obtained from a conventional method using term frequency (tfidf). We used eight documents from the Japanese textbooks for the evaluation and carried out an experiment where 40 human subjects chose the five most important sentences from each of the eight documents. The results show that summarization accuracy can be improved by applying our method.