Automatic extraction of document topics

Abstract

A keyword or topic for a document is a word or multi-word (sequence of 2 or more words) that summarizes in itself part of that document content. In this paper we compare several statistics-based language independent methodologies to automatically extract keywords. We rank words, multi-words, and word prefixes (with fixed length: 5 characters), by using several similarity measures (some widely known and some newly coined) and evaluate the results obtained as well as the agreement between evaluators. Portuguese, English and Czech were the languages experimented. © 2011 IFIP International Federation for Information Processing.