Sentiment Analysis of Russia-Ukraine Conflict: A Hybrid Approach Using VADER, GloVe-embedding and LSTM

Soumen Sinha, Saketh Innani, Pawan Chinnari, Mehek Khan

Abstract: Russian-Ukraine conflict is going on for quite a long period and it has become very crucial to understand the public opinion on this issue for various reasons like policy implication, peace building efforts, humanitarian reasons and many more. This paper aims to study the public sentiment on Russia - Ukraine conflict from twitter data using a hybrid approach. It presents a hybrid approach to perform sentiment analysis on twitter data using VADER, GloVe-embedding and LSTM. Our research analyses and classifies sentiments using deep learning techniques. A lot of work done till now using twitter data requires a lot of time to label the tweets. Twitter possesses huge amount of data and it becomes difficult to label each one of them manually. For training our model tweets were scraped from twitter using Tweepy library in python. We scraped 1 million tweets from first week of January 2023 till first week of February 2023. Tweets were scraped using various queries which are there in Tweepy library. In this paper we have shown the application of VADER which can be used to label unlabelled dataset. After labelling the data using VADER we performed word and GloVe embedding on our dataset. After performing GloVe embedding we performed sentiment analysis using Bi-directional LSTM. Our model achieved an overall accuracy of 97.09%. We also compare the accuracy achieved by other models with our hybrid approach.