Initial Progress of Identification of the Appropriate NLP Technique for Content Evaluation in Textual Conversations of People Infected by Sars-Cov-2

Abstract

When Covid-19 became a pandemic on March 2020, an urgent need arose for reliable info and advice, so Virtual Assistants were created to help teach the public how to avoid the Alpha variant. But when new variants like Beta, Delta, and Omicron appeared with different symptoms, they caused new waves of infections and deaths. To tackle this, a Natural Language Processing prototype was created to analyze experiences of 4422 people, who had been infected in Ecuador, and to detect which symptoms were most common in their conversations. For this purpose, Python language was used, Google Collab platform, and several combinations of text processing techniques with various classifiers were tested. Finally, the results were measured using quality metrics, accuracy, precision, Recall, F1, to identify the most appropriate model, finding that the combination of Stop Word, Tokenization, stemming techniques together with the LSTM classifier reached high effectiveness among the options tested for a classifier model with multi-label output.