
This article presents the automatic text summarizer PragmaSUM, which is independent from the language and knowledge domain of the source text, based on the Cassiopeia algorithm, which uses Luhn’s distribution and Zipf’s Law to select words in the text used for classifying sentences and generating the summary. A corpus is created for tests in Portuguese, composed of scientific articles from 10 different knowledge domains, for evaluating summaries generated by BLMSumm, GistSumm and PragmaSUM summarizers. Performance was observed using Recall, Precision and F-Measure metrics present in the assessment tool ROUGE. The end of the article presents the results of the summary assessment generated by the summarizers and PragmaSUM by employing two forms of summarization: with keywords for classifying sentences in the source text without using these words and by comparing summarizers. It was observed that using keywords in automatic text summarization allows for personalization of the summary according to the users’ needs by fetching sentences that really correspond to their interest domain.