Comparative Analysis of Classical Machine Learning Models for Twitter Sentiment Classification

View at publisher Summarise with VIRAT
Authors: Manish kumar Jha, Gaurav Kumar Soni, Gavesh Jain, Satyam Tiwari, Khushal Gupta, Bhavesh Singhal
Cited by: 0
References: 13

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Sentiment analysis of social network sites such as Twitter is one of the main areas critical to the analysis of public opinion. The objective of this paper is to compare three different classical machine learning models for carrying out a sentiment classification task on Twitter, namely Logistic Regression (LR), Linear Support Vector Machine (SVM) and Bernoulli Naive Bayes (BNB). The data used is a large, well balanced dataset of$\mathbf{1. 6}$million tweets with the same number of samples in the positive and the negative sentiment classes. With 5000 features forming the vocabulary, the models are trained on a dataset containing 80 % of the total number of tweets, using TF-IDF vectorization on unigrams and bigrams for feature extraction. Different metrics of the models are analyzed on a test dataset containing 320,000 tweets, using metrics such as accuracy, ROC AUC, Average Precision (AP) and Matthews Correlation Coefficient (MCC). It was observed that Linear SVM performs considerably better than Bernoulli Naive Bayes while Logistic Regression also gives similarly better results than Bernoulli Naive Bayes. Logistic regression performed best with an accuracy of 79.61 %, ROC AUC of 0.8773, and MCC is 0.5923 while the linear SVM resulted in an accuracy of 79.53 %, ROC AUC of 0.8772 and MCC of 0.5909. The baseline BNB model gave overall the lowest results with a ROC AUC of 0.8404 and an accuracy of 76.65 %. It was found in the analysis that linear models such as SVMs and Logistic regression are very efficient in classifying large amounts of text and forms a good basis for sentiment analysis in Twitter.

Fields of Study & Subjects

Sentiment Analysis and Opinion Mining Text and Document Classification Technologies Hate Speech and Cyberbullying Detection

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1949–1954
Publication date
21 January 2026
Language
EN