Sentiment Analysis in Domain-Specific X-Tweets Using Support Vector Machine and Long Short-Term Memory with TF-IDF Vectorizer

Abstract

Interest in sentiment analysis has grown as user comments on social media have proliferated on platforms. The hybridized machine learning framework used in this paper focuses on the classification of user feedback on tweet post. Other researchers have applied machine learning models to address this issue, but their models struggle with the complex sentiment nuance and achieve low prediction accuracy. To close this disparity, we used the powerful classification ability of support vector machine (SVM) in conjunction with LSTM (Long short-term-memory) to classify tweet sentiment and Term Frequency-Inverse Document (TF-IDF) Vectorizer for converting tweets data into numeric values. This experiment used the 1.5 million tweet sentiment 140 dataset available on Kaggle. The SVM-LSTM hybrid model developed in this study demonstrated 99% accuracy on both the training and test data. We used another state-of-the-art metric, ROC-AUC. This model achieved superhuman performance and outperformed many baseline models in NLP. This paper’s novelty is achieving a high-performing model above many baseline models and specifically targeting nuances in sentiment. This demonstrates the model’s effectiveness at classifying tweet sentiment, especially for domain specific applications such as social media sentiment analysis.