Experimental analysis on dependency about summarization strategy used by individual summarizer

摘要

The objective of this paper is to investigate how and to what extent summaries of one original document depend on individual human summarizer. We collect 140 students' summary and cluster them by Ward's method. The result shows that several sentences are likely to be adopted in the majority of summaries. We call these sentences as core sentences. Which sentences are adopted as core sentences does not depend on the degree of compression rate. However which sentences of core are selected depend on the cluster and degree of compression rate. Then, we analyze the relation between the distribution of the selected sentences from the core and automatic summarization method. The experimental data show that the actually selected sentences of each cluster from core correlate with one or two factors used in automatic summarization method, namely, weight of individual term, where the sentence occurs in the original document and rhetorical relation "but."