DIFFSTRACT: distinguishing the content of texts

Abstract

Nowadays, it is almost a standard issue to generate summaries of texts automatically. In contrast, it is still a problem to identify the differences in the statements of the two publications. For the most part, this still requires a human being to read and evaluate at least excerpts of the relevant passages. Finding a so-called text differentiation with appropriate tools is becoming an increasingly interesting and important task to effectively cope with the daily flood of information on the WWW. For years, co-occurrence graphs have been a proven means of deriving statements of various kinds from texts. So-called text- representing centroids (TRC's) has often been an effective tool for identifying, comparing and categorizing texts or sections. The present article examines how a different form of co-occurrence graphs can take place and be helpful. First, different co-occurrence graphs are built from a larger corpus and various individual texts or text groups. Subsequently, the calculated difference graphs can be used to create summaries that precisely characterize the differences between texts. Experimental results show that this new method works well.