Sentiment Analysis of Twitter Data Using Machine Learning and Deep Learning

Abstract

In recent years, sentiment analysis has become a popular topic of discussion because of the rapid growth of various social media sites and e-commerce sites. At its core sentiment analysis analyze people’s opinion and try to determine the polarity of that opinion. It is very handy to determine customer’s review and also to determine various social trends that is going on. Its aim is to identify if a text contains a positive or negative meaning. Nowadays people can post anything on social sites like facebook, tweeter, Instagram. In our thesis we used a dataset that contains more than 29530 tweets of various users. The aim of this thesis is to determine if these tweets contain any hatred content or they are not hatred. Different machine learning technique such as Naïve Bayes, Support Vector Machine, Logistic Regression, Random Forest are implemented in this thesis work. Using these algorithms, we have performed the classification task and the performance evaluation using various parameters such as precision, recall, f1-score, and accuracy. Furthermore, we analyzed their results and compared their outcome with one another. All our model gave almost the similar result varying within 1-2%. Among the machine learning algorithms, we got high accuracy (96.24%) using the Random Forest algorithm. Our work does not finish here as we also tried to implement deep learning to our project. Our aim was to compare the result between machine learning and deep learning and see which model provides the best outcome. From deep learning we used Bidirectional-Long Short Term Memory for this thesis work. However, the result we got from it was a tiny bit less accurate that our machine learning model. Therefore, our final decision is machine learning performs better for our small dataset than deep learning. Finally, in this thesis we also discussed the challenges and limitations of our research.