A Comprehensive Classification of Sentiment Reviews of Twitter Data in the Domain of Climatology using Machine Learning Techniques

Purpose: This study aims at classification of sentiment reviews of Twitter data in the domain of climatology using machine learning techniques. It focuses on the text classification in order to determine the people’s intension about the climatic issues i.e., climate change, climate variability, environmental aspects etc. This paper portrays a comparison of results obtained by applying different classification algorithms like Naive Bayes (NB), Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbour (KNN), Decision Tree Classifier, Neural Network classifier etc. These algorithms are used to classify a sentimental review and people’s emotions associated with climate. Design/Methodology/Approach: Total 2265 climate reviews data have been taken from Twitter’s developers’ account. After that, we pre-processed the total dataset by removing various symbols, HTTP tags, punctuation, etc. The pre-processed text were analysed and represented through Topic modelling, Multi Dimensional Scaling (MDS) and also Visualization of Heatmap. Next, bag of words are evaluated through various algorithms such as Naive Bayes (NB), Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbour (KNN), Decision Tree Classifier and Neural Network classifier. After applying above mentioned classifier, datasets are tested and scores are noted. For the experiment, 70 % of total reviews (i.e.1586) are used for model training and 30% of total reviews (i.e. 680) are used for testing the models. Findings: By performing different algorithms, it shows that Random Forest classifier algorithm works well than other mentioned classifiers and most of the people have positive sentiment towards climate according to Valence Aware Dictionary for Sentiment Reasoning (VADER).