Deep Sentiment Classification using Topic Modeling for Covid-19

Abstract

Objectives: To extract various topics related to Covid-19 from Twitter API using LDA topic modelling technique and to analyse the sentiment of the people about the extracted topics. An interactive Q/A system with both voice and text interface is also proposed to guide COVID-19 related decision-making. And also to summarize the tweets containing a query and to suggest suitable solutions. Method: The proposed extracts Covid-19 related tweets from twitter API and uses Natural Language Process (NLP) method based on topic modeling to uncover various issues related to COVID-19 from public opinions. The training dataset consists of 3,38,666 COVID 19 related comments and the testing dataset consists of 1,12,888 comments. LSTM recurrent neural network is used for sentiment analysis of the extracted tweets and to produce summary for each topic identified through topic modelling. Findings: The accuracy comparison has been done for the existing system against the proposed model with respect various machine learning classifiers. The findings are-LSTM gives an accuracy of 79.5%, the Naïve Bayes classifier gives the accuracy of 74%, the Multinomial Naïve Bayes gives an accuracy of 74.5%, whereas the linear regression classifier achieves an accuracy of 76%, KNN classifier achieves an accuracy of 74.5% and the random forest with an accuracy of 75.5%. Novelty: The proposal of interactive Question Answering system is first of its kind. This work sheds light on the importance of using public opinions and suitable computational techniques to understand issues surrounding Covid 19 and to guide related decision-making.