Um motor de inferência para relações de identidade em grafos de conhecimento

Abstract

The growing demand for realtime information access requires high cost – financial and computational – for data integration due to lack of standardization, resulting in problems during modeling and display data. The Knowledge Graphs were used to deal these problems. By providing a structured, scalable and understandable machine model, the creation and maintenance are vulnerable to errors due to automatic reasoning difficulties in large data from different domains – which can produce inaccurate, erroneous or incomplete results – mainly related with ambiguity. The problems are normally caused by ambiguous relationships and by inaccuracy in determining Identity Relations (IR) in a domain. Recent studies compare all attributes without considering that some of them can be more relevant. This work applied an automatic IR detection mechanism which execute an automatic selection of relevant attributes for a domain from entropy analysis and statistical correlation between the attributes. The proposed solution was applied in 12 real datasets that include software development activities. The characters which were automatically selected obtained better IR detection accuracy than the criteria recommended by a domain expert.