Predicting Object Types in Linked Data by Text Classification
Xiang Zhang, Erjing Lin, Siyao Pi · 2017
Type information of objects is very valuable in linked data. However, many linked data are incomplete in type information. Traditional research of type inference is able to find missing types by means of reasoning, but it may become invalid in those data with incomplete or incorrect schema. In this paper, we propose a text-classification approach to type prediction in linked data. An Object Graph is proposed as the data model. A Virtual Document of Type Information is constructed for each object, and two strategies are proposed to characterize the inductiveness of different part of virtual document for type prediction. Two classifiers are trained to categorize each object into a set of types. Experiments validate that type prediction by text classification is feasible and well-performed.