[3G5-ES-1] Knowledge engineering (2)

[3G5-ES-1-05] A Consideration on Complementing Subclass-of Property on Wikidata with Outputting Reliable Confidence Score

〇YUXI JIN¹, Shun Shiramatsu¹, Ikue Kawamura¹ (1. Nagoya Institute of Technology)

Keywords:Wikidata, superclass estimation

Wikidata, growing linked data which is full potential to provide reliable and powerful service, bases on the knowledge of Wikipedia. Although multiple projects required food information on Wikidata, the subclass-of property in food entities are not as sufficient as Wikipedia. To automatically complement the subclass-of properties, we need a confidence score of the automatic analysis because non-reliable output of automatic analysis should not be used for the complementing. Therefore, in this paper, we try to improve the confidence scores as output of classifiers, e.g., SVM and LSTM. Concretely, we apply the logistic regression to generate reliable the confidence score of the classifiers. Our empirical experiments show that LSTM with logistic regression is provided with better credibility for subclass-of property estimation with outputting reliable confidence score.