Abstrakt
It is estimated that more than 80% of the information on the Web is stored in textual form. As such, it has become increasingly difficult for humans to sort and extract useful information from the daily influx of data.
In order to automate this process, open information extraction (OIE) methods have been proposed, which can extract facts from large textual bases. While most OIE methods were initially developed for the English language, the importance of developing methods for other languages, such as Portuguese, has been increasingly recognized in recent literature.
OIE methods based on hand-crafted rules and shallow syntactic analysis have achieved good performances for the English language. Nevertheless, methods based on similar approaches in the Portuguese language have not achieved equivalent success.
We believe that the shallow syntactic patterns previously explored in the literature do not cover important aspects of the Portuguese language syntax. For this reason, we propose the DptOIE method based on a new set of syntax-based rules using dependency parsers and a depth-first search (DFS) algorithm for OIE and a set of grammar-based rules to cover specific syntactic phenomena of the language.
DptOIE was compared against the state-of-the-art OIE for the Portuguese language, obtaining favorable results both in our empirical evaluation and at the IberLEF evaluation track of OIE systems for the Portuguese language. Furthermore, we believe our method can be easily adapted to other Romance languages related to Portuguese.