Word Sets, Keywords, and Text Contents: An Investigation of Text Topic on the Computer

Abstract

This study presents a methodology for the identification of coherent word sets. Eight sets were initially identified and further grouped into two main sets: a `company' set and a `non-company' set. These two sets shared very few collocates, and therefore they seemed to represent distinct topics. The positions of the words in the `company' and `non-company' sets across the text were computed. The results indicated that the `non-company' sets referred to `company' implicitly. Finally, the key words were compared to an automatic abridgment of the text which revealed that nearly all key words were present in the ahridgment. This was interpreted as suggesting that the key words may indeed represent the main contents of the text.