ITAINNOVA at SocialDisNER: A Transformers cocktail for disease identification in social media in Spanish

AbstractITAINNOVA participates in SocialDisNER with a hybrid system which combines Transformer-based Language Models (LMs) with a custom-built gazetteer for Approximate String Matching (ASM) and dedicated text processing techniques for the social media domain. Additionally, zero-shot classification capabilities have been explored in order to support different parts of the system. An extensive analysis on the interactions of these components has been accomplished, making the system stand out above the mean performance of all the participating teams.