Can Twitter Help to Predict Outcome of 2019 Indian General Election: A Deep Learning Based Study

Abstract

Traditionally, elections polls have been widely used to analyse trend and to predict likely election results. However, these methods are very expensive and labour intensive. With the widespread development of several social media platforms, a large amount of unstructured data become easily available, which in turn could be processed and analysed to extract meaningful information about several topics and events such as election, sports, natural hazards etc. Hence, in this study, we utilise twitter data to analyse the 2019 Indian general election and to predict possible outcomes. We have collected 41 million election related tweets during the months of April and May 2019. The proposed approach works by initially performing hashtags based tweets segregation to generate data for training the classification model. Subsequently, we augment different word embeddings with deep learning classification model to support improved classification accuracy. The model achieves 87.30% classification accuracy in amalgamation with fasttext word embedding. Finally, the opinion analysis is performed on the classified tweets to determine possible election outcomes. To evaluate our proposed model, the output results are validated with the available ground truth and have shown a close correlation to the actual results. Furthermore, we have also performed an in-depth analysis of the selected political parties tweets, to provide an insight into top trending hashtags and mentions used by people during election campaigns.