Multi Head Text Mining Applying Deep Learning and Bangla NLP Using COVID-19 Feedback From Students

Abstract

The COVID-19 epidemic bound administration all around the globe to lock down their frontier, their companies, their schools, and citizens from leaving their houses unless absolutely required. The mental health of both individuals and society as a whole can be seriously harmed by being so imprisoned. This negatively affects students' social and emotional well-being. Having schools, colleges, and universities closed had a significant negative influence on students' academic lives. We physically examined Bangla text that was physically obtained from students using a google form. It was all written in Bangla. With two label classes—positive and negative—through text categorization, we looked at how covid affected social relationships, mental health, and academic success. We tested several ML algorithms like logistic regression, decision trees, random forests, multi-naive Bayes, KNN, SGD, linear SVM, and RBF SVM, and discovered that SGD had the highest accuracy for academic impact. We get the greatest accuracy for the influence on social life column using KNN, and the best score for the impact on mental health with both multi-naive Bayes and SGD. Additionally, we used the CNN, LSTM, CNN-LSTM, BiLSTM, and CNN-BiLSTM as deep learning models in the aforementioned three columns, and we achieved an academic impact accuracy of 80%, mental health impact accuracy of 98.75% and Social life impact accuracy 83.75% using LSTM. The accuracy is 92.50%, 85%, and 92.50% using BiLSTM for academic, mental, and social impact columns. while its accuracy in terms of using CNN, CNN-LSTM, and CNN-BiLSTM is 82.50%, 70%, 92.50%; 85%, 90%, 92.50%; and 85%, 85%, 90% for academic, mental, and social impact columns, respectively.