Loading...

The Emotion Recognition of Social Media Users’ Comments during Covid-19 Outbreak

Bahari Ghale’ Roudkhani, Zhalerokh | 2022

279 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55667 (31)
  4. University: Sharif University of Technology
  5. Department: Languages and Linguistics Center
  6. Advisor(s): Rezaei, Saeed; Bahrani, Mohammad
  7. Abstract:
  8. In the last three years, the lives of many people around the world have changed with the spread of the Corona virus. In order to better manage the consequences related to the spread of this disease, extensive research has been done on this virus, and researchers in data science and artificial intelligence have devoted a part of their research to studying the effects of this virus on the people in one or different societies.On the other hand, the study of social networks about a specific issue or trend topic, allows us to examine more closely the atmosphere that governs the society and analyze the emotions, feelings and the level of concern of the members of the society about that issue.The present research is also in the same direction. In this research, the sentiments of Persian users of the Twitter regarding the spread of the Covid-19 virus have been investigated. In the first stage, about 7000 tweets of Persian language users with the keywords "Corona", "Covid-19", "Mask", "Vaccine" etc.were collected and labeled with 8 different labels, in which 5 categories of Ekman's six categories (happiness, sadness, fear, anger, surprise) and three other categories (joking, wonder, neutral) are used. These data will be available to the public after the work is finished. It has been tried to make the amount of data in different categories almost equal. The results of the analysis of the dataset showed that the highest number of tweets, aside from the neutral class, is related to the two classes of sadness and anger. Then the main topics discussed in the tweets are identified and compared in terms of frequency. In the next step, a classifier has been created using the ParsBERT algorithm, which can be used to place new tweets in one of the 8 introduced categories. The accuracy of this classification is above 76%, which is more accurate than the previous researches for the classification of Persian tweets, while in the current research, the data was classified into 8 groups, which brought its own challenges
  9. Keywords:
  10. COVID-19 ; Twitter Social Network ; Emotion Detection ; Deep Learning ; Bidirectional Encoder Representations from Transformers (BERT)Model ; Attention Mechanism

 Digital Object List

 Bookmark

No TOC