Yudana, Rizky Aditya (2024) Analisis Sentimen Twitter Dalam Bahasa Minang Menggunakan Model Bert Dan Cnn: Perbandingan Hasil Dengan Dan Tanpa Pemrosesan Kamus. Other thesis, Universitas Islam Riau.

Text Rizky Aditya Yudana.pdf - Submitted Version Restricted to Registered users only Download (6MB) | Request a copy

Abstract

This research focuses on sentiment analysis of Twitter in the Minangkabau language using BERT and CNN models to compare results with and without dictionary processing. Data was collected from Twitter using Minangkabau-specific keywords and hashtags. The preprocessing steps included text cleaning, case folding, tokenization, stopword removal, and stemming. The use of a manual dictionary for data processing enhanced the accuracy of sentiment analysis. The research findings show that the BERT model with dictionary processing achieved an F1-Score, precision, recall, and accuracy of 0.91, while the BERT model without dictionary processing reached an F1-Score of 0.86, precision of 0.83, recall of 0.89, and accuracy of 0.85. For the CNN model, the results with dictionary processing showed an F1-Score of 0.53, precision of 0.62, recall of 0.57, and accuracy of 0.57. On the other hand, the CNN model without dictionary processing had an F1-Score of 0.34, precision of 0.25, recall of 0.50, and accuracy of 0.50. Based on the Accuracy, Precision, Recall, and F1-score obtained from each model, BERT produced the best results. This confirms that dictionary processing contributes significantly to improving model performance, particularly for regional languages.

Actions (login required)

View Item