Sentiment Analysis of Tweets on the 2023 France Riots Using an Enhanced Stacking Classifier

Abstract

Social media platforms like Twitter increasingly accessed through immersive technologies and virtual environments like Metaverse, provide valuable insights into public opinions. This research analyzes tweets on the French riots of 2023 to gauge popular sentiment. A stacking classifier which can combine Naive Bayes, SVM, along with Random Forest is used. Tweets were scraped and labeled using Sentiment Intensity Analyzer (SIA). Traditional TF-IDF vectorization and tokenization were enhanced along with Word2Vec and n-gram tokenization. When utilizing SVM as the final estimator at thresholds −0.05 (negative) along with 0.35 (positive), hyperparameter adjustment produced an accuracy of 65%. Individual classifiers like Random Forest (57%), SVM (50%), along with Naive Bayes (40%), have been defeated by the model, demonstrating superior predictive performance.This study also highlights how sentiment analysis within the Metaverse context can provide early insights into collective digital behavior during real-world socio-political events.