Regresar

Identification of a suitable NLP model for the detection of symptoms mentioned in textual conversations of Covid-19 infected persons

Abstract:

In March 2020, the World Health Organization (WHO) declared Covid-19 disease as a global pan-demic. Therefore, the need for reliable information arose, so several Virtual Health Assistants emerged to provide information to the public to teach the population how to prevent or cope with the Covid-19 Alpha variant infection, but progressively emerged the Beta, Delta, Omicron variants with different symptomatology which triggered new waves of infections and deaths in the world. For this reason, the present study promoted the creation of a NLP (Natural Language Processing) model to analyze the experiences of infected people in Guayaquil and to detect the predominant symptoms mentioned in their textual conversations. For this purpose, the Quantitative Methodology was followed by performing surveys through Google form, reaching 2873 people in the city of Guayaquil who had and overcame the Covid-19. Thus, the corpus of textual conversations was generated. On the other hand, the Qualitative Methodology was used by executing interviews to NLP experts, which allowed corroborating the classifiers suggested in this type of software. Finally, twenty-two different NLP models were built based on classifier algorithms like K-Nearest Neighbors, Random Forest, and Long-Short Term Memory (LSTM), and their quality was evaluated using metrics such as Accuracy, Precision, Recall, and AUC, finding that the model based on LSTM version 3 obtained the highest performance.