DptOIE: a portuguese Open Information Extraction system based on dependency analysis

Abstract

It is estimated that more than 80% of the information on the Web is stored in textual form. For humans, the task of extracting useful information from data that comes up daily is difficult. In order to automate the process, techniques of Open Information Extraction (OIE) methods, which are capable of extracting facts from large textual bases, have been proposed. At first, most OIE methods were developed for the English language. However, other languages, such as Portuguese, have tackled special attention, since it covers approximately $2.5\%$ of all content available on websites. For English languages, methods based on hand-crafted rules and dependency analysis have gained good results. Nevertheless, methods based on similar approaches, in Portuguese, have not presented equivalent performance. We believe that the rules defined are generic and do not cover specific aspects of the language. For this reason, our DptOIE method defined a new set of hand-craft rules and explore sentences through a dependency analysis by a depth-first search (DFS) approach. DptOIE was compared against two other OIE methods which extract facts in Portuguese: PragmaticOIE and ArgOE. DptOIE outstands the other works, obtaining a greater area under the precision-yield curve. Precision was superior as well as the number of coherent facts extracts. As far as we know, this is the most outperforming method to extract fact on OIE for the Portuguese language.