Salmaa, Dannes Luthfiyah (2025) Implementasi Metode Deep Learning Dan Machine Learning Untuk Analisis Sentimen Peringatan Darurat Pada Media Sosial Youtube. Other thesis, Universitas Islam Riau.

Text 213510021.pdf - Submitted Version Restricted to Registered users only Download (8MB) | Request a copy

Abstract

Sentiment analysis has become an increasingly important method for understanding public opinion on social and political issues. This study aims to evaluate the performance of deep learning methods (LSTM and CNN) and machine learning methods (SVM and Naïve Bayes) in conducting sentiment analysis on YouTube comments regarding the Peringatan Darurat issue. The data was classified using various labeling methods, including GPT-4o Mini Zero-Shot, Few-Shot, Lexicon-Based, and TextBlob, with the application of the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance in the dataset. The results indicate that the dataset labeling method and the application of SMOTE significantly affect model performance. The best accuracy for deep learning was achieved by the LSTM model using the GPT-4o Mini Zero-Shot labeling method without SMOTE (0.71), while the best accuracy for machine learning was obtained by the SVM model using the GPT-4o Mini Zero-Shot labeling method with SMOTE (0.78). Overall, machine learning models (SVM and Naïve Bayes) outperformed deep learning models (LSTM and CNN), with SVM achieving the highest accuracy. Future research is expected to use larger and more diverse datasets, explore feature representation techniques such as Word2Vec or BERT, and develop a Flask-based application for real-time sentiment analysis.

Actions (login required)

View Item