[4M2-GS-5] Web intelligence: Web intelligence

[4M2-GS-5-02] Extraction of Linguistic Expression for resolving incompleteness between Linked Data and documents

〇Yuki Okumura¹, Sachio Hirokawa², Kazuhiro Takeuchi¹ (1. Osaka Electro-Communication University, 2. Kyusyu University)

Keywords:Linked Data, Content creation support, Ontology, Information extraction

Named Entities (NEs) and their semantic roles play a central role in sentence generation and understanding.
The knowledge registered in Wikidata shows the semantic roles of NEs in Wikipedia articles as linked-data.
Their combination is an insightful research resource for understanding the correspondence between structured knowledge and the structural linguistic expressions used to explain it.
In this paper, we investigate linguistic expressions on Wikipedia to describe property-value relation in Wikidata.
Precisely, We extract linguistic expressions that represent the characteristics of the property by applying a feature selection method using SVM (Support Vector Machine).
As a result, we confirmed that these characteristic linguistic expressions appeared with low frequency, and can clearly distinguish with them from the informative articles that correspond to leaf-nodes of Wikidata from others.
This finding contributes to fill slots of property-value in Wikipdata and to resolve the inconsistency between it's linked knowledge and description in Wikipedia.