Widaug. Data augmentation for named entity recognition using Wikidata

Resumen

The current state of the art of Natural Language Processing models are based on the use of a big amount of data to be trained. The more, the better. However, this is quite a limitation in the creation of datasets for specific natural language processing tasks such as Named Entity Recognition, which involves one or more annotators to read, understand and annotate those required named entities along a corpus. Currently, there are many good general domain corpora for the English language. However, particular domains or scenarios and other non-English languages are still not so represented in the research community. Thus, data augmentation techniques are explored to create synthetic data similar to the originals to enrich the training process of the models. On the other hand, knowledge graphs contain a lot of valuable information that is not being used to help in the data augmentation process. This work proposes a data augmentation method based on the Wikidata knowledge graph which is tested in a Spanish corpus for a Named Entity Recognition challenge.