A Similarity Grammatical Structures Based Method for Improving Open Information Systems

Erick Nilsen Pereira de Souza (Federal University of Bahia, Brazil)

Daniela Barreiro Claro (Federal University of Bahia, Brazil)

Rafael Glauber (Federal University of Bahia, Brazil)

Abstract: Open information extraction (Open IE) discovers facts as triples of relationships in texts. A major challenge to Open IE task is to reduce the proportion of invalid extractions. Current methods based on a set of specific features eliminate many inconsistent and incomplete facts. However, these solutions have the disadvantage of being highly language-dependent. This dependence arises from the difficulty in finding the most representative set of features, considering the peculiarities of each language. These solutions require extensive training sets, usually produced with the aid of a specialized linguistic knowledge. Furthermore, although linguistic knowledge resources are common in English, they are scarce in most other languages. Therefore, we propose a method for classifying extracted facts based on the similarity of grammatical structures, which builds models from morphological structures contained in the extraction through the application of algorithms for the detection of isomorphism in sub-graphs. In particular, Portuguese was chosen for the implementation and validation of the proposed approach as it is one of the languages that lack this type of resource.

Categories: L.1.0, M.0, M.7