Analyzing Major Indian Protests using Nature Inspired Algorithms

The rates at which strikes, riots and protests are increasing these days is alarming and a matter of grave concern. In order to understand the intensity of a protest and eventually take steps in order to minimize massive possible damage, it is critical to analyze the sentiments towards that particular event. Previous works in this field have majorly used traditional machine learning approaches. However, we believe that incorporating nature-inspired algorithms along with BERT for feature selection can greatly improve the accuracy. In this work, we have proposed a novel approach wherein we use an amalgamation of BERT and nature-inspired algorithms to perform sentiment analysis of major protests in India. For this, we combine the feature extraction capabilities of BERT with both swarm intelligence as well as evolutionary algorithms. For comparative study, we have taken a model which only uses BERT along with KNN and compared it with models which use nature inspired algorithms as an additional step to perform feature selection. We have achieved remarkable improvement in accuracy by incorporating nature inspired algorithms with the highest improvement of 17.78% in classifying sentiments by using the Grey-Wolf optimizer. Significant improvements in accuracy were also observed by using Particle Swarm Optimization (PSO), Harris Hawk Optimization (HHO) and Genetic Algorithm (GA). Snscrape has been used to extract tweets of major protests in India. Our results accurately correspond to the actual intensity of devastation that was observed during those protests.