Multi-class approach for user behavior prediction using deep learning framework on twitter election dataset

Abstract

Among the broad assortment of Machine Learning approaches, deep learning has recently attracted attention particularly in the domain of user behavior analysis. The notion to study user behavior from the unstructured tweets shared on social media is an interesting yet challenging task. A social platform such as Twitter yield access to the unprompted views of the wide-ranging users on particular events like election. These views cater government and corporates to remold strategies, assess the areas where better measures need to be put forward and monitor common opinion. With the advent of the general election in India (largest democracy) people tend to articulate their views or issues. Tweets related to general elections 2019 of India is used as data corpus for the study. Multi-class classification fabricated with novel deep learning approach is implemented to analyses the user opinion. Here, we have used nine different classes, which is representing larger issues in the nation for election agenda. Moreover, comparative analysis between tradition approaches such as Naive Bayes, SVM, decision tree, logistic regression and employed approach with deep learning method is presented. Experimental results revels that the proposed model can reach up to 98.70% accuracy on multiclass based prediction in machine learning. The results assist the government and businesses to know about grave issue offering a shot to revise strategic policy and make welfare scheme program.