Toward a Scoring Schema to Rank Candidate Instances of Ontological Classes: Extracting Brazilian Portuguese Texts from the Web

Abstract

With the emergence of Information Extraction Systems driven by ontologies, boosted by the Semantic Web, there is a need for the development of scoring schemas that enable the automatic classification of information. These schemas, even so little explored in the Portuguese language, provide measures used in the stage of classification of relevant instances to ontological classes. In this way, this paper presents: (i) a brief discussion about existing scoring measures based on PMI (Pointwise Mutual Information); (ii) new scoring measures based on PMI and Standard Deviation Calculation; and (iii) an evaluation of all discussed measures in the context of Brazilian Portuguese texts from the web.