BERT and Approximate String Matching for Automatic Recognition and Normalization of Professions in Spanish Medical Documents.

This publication presents the participation of the EdIE-KnowLab team in the MEDical DOcuments PROFessions recognition shared task from IberLeF 2021. The proposed system consists of a Spanish version of the BERT classification model, BETO, for the Named Entity Recognition tasks and an approximate string matching technique using Damerau–Levenshtein distance for the Normalization task. The NER systems reached 64.3% and 60.4% in Micro-Average F1 for Task 1 and Task 2, respectively. The approximate string matching approach obtained 17.8% in F1 for the Normalization task. Source code to reproduce the results is available under the MIT license at https://github.com/ vsuarezpaniagua/EdIE-MEDDOPROF.