Exploring The Impact of Feature Engineering and Data Organization on Sentiment Analysis of Twitter Data Using Machine Learning Algorithms

Abstract

As one of the most popular micro-blogging platforms, Twitter generates millions of tweets daily, making manual sentiment analysis of such large volumes impractical. Consequently, leveraging machine learning algorithms for efficient sentiment analysis has become a critical challenge. This paper explores the performance of four machine learning algorithms—Logistic Regression (LR), Gaussian Naive Bayes (GNB), Decision Tree (DT), and Gradient Boosting Machine (GBM)—across four datasets of varying sizes. The models were trained using four feature extractors which include unigrams, bigrams, a combination of unigrams and bigrams, and the pre-trained Global Vectors (GloVe) word embedding model, with feature dimensions of 100, 200, and 300. The study reveals the impact of dataset size and feature combinations on the performance of these algorithms, identifying the most effective feature extraction methods. These findings provide valuable insights into the relationship between data scale, feature representation, and algorithmic performance, offering innovative perspectives for future research in sentiment analysis based on tweets.