ABSTRACT
Public health surveillance via social media can be a useful tool to identify and track potential cases of a disease. The aim of this research was to design a method for identifying tweets describing potential Covid-19 cases. The proposed method uses a Wide & Deep (W&D) architecture, which combines two learning branches fed from different features to improve classification effectiveness. The deep branch uses a BERT-type model, while the wide branch considers two different lexical-based features. It was evaluated on the data from Task 5 of the Social Media Mining For Health (#SMM4H) 2021 competition. Results show that the proposed W&D method performed better than the wide-only and deep-only models, achieving an F1-score of 0.79 which matches the results of the 1st place ensemble-model. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.