Where does Information come from? Corpus Analysis for Automatic Abstracting

We report on our study of a corpus of abstracts and parent documents to determinate which structural parts of the parent document are used to extract useful information for an abstract. The results give us a sound basis for automatic abstracting of research articles. Our method for automatic abstracting, called selective analysis, is intended to produce user-oriented abstracts which are indicative in the essential content of the document and informative in the user's interest.