Sentiment and Statistical Analysis on Custom Twitter Dataset for 2022 Russo-Ukrainian Conflict

For a better understanding of the situation during a war or crisis, it is helpful to comprehend the public opinion. There are various social media networks which provide a platform for people across the world to express their opinions on globally significant issues. Recently, Russo-Ukrainian war began and people have shared their opinions on the same on various social media platforms including Twitter. In this paper, we have extracted tweets about the war using some of the popular hashtags, spanning over 7 months. The dataset will be open sourced so that people can contribute to it and use it for their own research. The study consisting of tweets extraction, preprocessing, assigning sentiments using polarity and subjectivity, fine tuning BERT based model on English language tweets, using M-BERT on 13 different languages supported by Text Blob and performing sentiment analysis. The study accumulated 1.5 million tweets out of which 7.5 lakh tweets, in English language, were used to train Naive Bayes Classifier and Random Forest classifier to get a F1 score of 0.79 and 0.911 respectively and fine-tuned BERT to get a F1 Score of 0.99. M-BERT was fine-tuned on a corpus of 2.5 lakh tweets having other 13 different languages to get a F1 Score of 0.98. The scope of this dataset can further be extended to various NLP downstream applications including hate speech detection, fake tweet detection and many more.