Classifying Various Types of Symptoms of COVID-۱۹ (CTSC) in Twitter Using Deep Learning Algorithms

Data mining has many uses in the field of health, including diagnosis of diseases, classification ofpatients in disease management, finding patterns for faster diagnosis, and preventing complications.In this field Research for extracting public health data in social networks such as Twitter hasgrown exponentially. Many researchers have decided to use machine learning and deep learningalgorithms for such analysis.Background and aims: whether Social networks can enable the early detection of symptoms ofdiseases such as COVID-۱۹, whose treatment is not known for sure, and our knowledge about itis expanding. The aim of this study prediction COVID-۱۹ of symptoms.Method: In the CTSC method, the meaningless information is eliminated by collecting tweetsfrom ۲۶ June, ۲۰۲۱, to ۴ July, ۲۰۲۱, concerning the symptoms of COVID-۱۹. The root of thewords is found, and then they are classified based on the dictionary of five categories: respiratory,digestive, muscular, smell-taste, and sinusitis. Various deep learning algorithms such as convolutionalneural network, recurrent neural network, and gated recurrent unit have been used toclassify these tweets. To evaluate the proposed method, we have considered the accuracy andpercentage of error (Loss).Results: The results show that users diagnosed with covid۱۹ show respiratory symptoms, includingsneezing, pneumonia, sore throat, coughing, fever, gasping breathing, and heart problems are۴۱% likelier than others. We also obtained the best performance for evaluating the CTSC methodby deep algorithms with ۹۷% accuracy.Conclusion: In this research, we provide a method based on different types of symptoms of COVID-۱۹ using three deep learning algorithms CNN, LSTM, and GRU.In this study, the Twitter data has been pulled out from Twitter social media for ۹ days, and tweetsare extracted based on positive COVID-۱۹ hashtags. Tweets are then cleaned and matched againsta symptoms lexicon and then labeled based on various types of symptoms of COVID-۱۹ usingmentioned algorithms. The evaluation results show that the presented method with an accuracy of۹۷ percent and a Loss of ۰.۰۶ has the best performance. In future work, we can apply this methodto Persian tweets and include more symptoms in each category to obtain more accurate results.

کلیدواژه ها:

COVID-۱۹ ، Respiratory ، Twitter ، Deep Learning ، CTSC