Exploration of Sentiment Prediction Models For Social Media Posts

Abstract

Sentiment analysis is a text analysis technique that can be used to understand the opinion, feeling, or sentiment of a text. This research aims to explore and compare sentiment prediction models on social media data with three algorithms, namely GaussianNB, Logistic Regression, and Support Vector Machine (SVM). The dataset used is taken from www.kaggle.com, which consists of social media posts from the Twitter, Facebook, and Instagram platforms with positive, negative, and neutral sentiment categories. The analysis process involves text data preprocessing, data labeling, feature extraction with Bag of Words (BoW) and TF-IDF, and handling data imbalance with SMOTE. The results showed that the SVM model with TF-IDF and SMOTE performed best, with 93.25% accuracy on training data and 92.50% on test data. This research contributes to determining the best model for sentiment analysis of social media data and can be a reference in developing better sentiment prediction systems in the future.