Vicomtech's Language Technologies team obtains the best score for the second consecutive year in the main task of the IberLEF eHealth Knowledge Discovery Challenge competition

Vicomtech's Natural Language Processing team has participated in the eHealth-KD competition held within the 3rd edition of IberLEF, obtaining excellent results. For the second year, the team has won the main task and has achieved second place in the others.

The competition consists of 3 tasks about entity recognition and relation extraction in texts of the health domain. The Vicomtech team has participated in all the tasks implementing systems based on deep neural networks that jointly encode entities and relationships.

Natural language processing (NLP) methods are increasingly used to extract knowledge from unstructured health texts. Recent advances in text processing techniques make it possible to go beyond shallow reading of texts and facilitate the extraction of relevant data for decision-making.

The eHealth-KD 2021 proposes a knowledge structure to model human language with the aim of encouraging the development of software technologies to automatically extract useful data, mainly from eHealth documents written in Spanish.

Although this challenge is oriented to the domain of health, the structure of the knowledge proposed is general-purpose. In fact, although most of the phrases used to train and evaluate the systems are in Spanish and belong to the domain of health, the challenge has also included a small selection of sentences from different domains and languages, to promote multidomain and multilingual approaches.