DOTS - Detection of Covid-19 contagion symptoms and self-diagnosis in social networks

Abstract

Social media present ways for people to share emotions, feelings, ideas, and even symptoms of disease, and is a great source of data for a variety of analyses. At the end of 2019, an alert was raised for a global pandemic of a virus that has a very high contamination rate and can cause respiratory complications in the contaminated people. To help identify those who may have the symptoms of this disease or to control who are already infected, this paper analyzed the performance of KNN, Naive Bayes, Decision Tree, Random Forest, SVM, simple Multilayer Perceptron, Convolutional Neural Networks and BERT algorithms to classify tweets that contained reports of Covid-19 symptoms or selfreports of infection. The dataset was labeled using a set of disease symptom keywords taken from a list provided by the World Health Organization. The tests on these models showed that the Random Forest algorithm performed best when classifying the tweets in a small dataset. This work demonstrated a superior performance of the Random Forest algorithm over other more robust algorithms for this type of classification and dataset.