2 files

VTKEL: 30K documents dataset for Visual-Textual-Knowledge Entity Linking

posted on 09.09.2020, 14:15 by Shahi DostShahi Dost, Luciano Serafini, Marco Rospocher, Lamberto Ballan, Alessandro Sperduti

(Updated version, after fixed some bugs)

VTKL dataset, contains documents composed of pictures with five corresponding textual captions for each image. The VTKL dataset is obtained by extending the Flikr30k dataset, designed for visual-textual mention alignment, with links to YAGO ontolgy, one of the largest web knowledge base. These links are obtained automatically by processing each image caption with PIKES, an NLP tool for entity recognition and linking.


Fondazione Bruno Kessler, Italy