KUSUMO, VALENTINO PUTRA BUDI KUSUMO (2024) SENTIMENT ANALYSIS REVIEW COMMENT USING TRANSFORMER MODEL. S1 thesis, UNIVERSITAS KATOLIK SOEGIJAPRANATA.

Preview	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_COVER.pdf Download (436kB) | Preview
	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_BAB I.pdf Restricted to Registered users only Download (413kB)
	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_BAB II.pdf Restricted to Registered users only Download (787kB)
	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_BAB III.pdf Restricted to Registered users only Download (918kB)
	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_BAB IV.pdf Restricted to Registered users only Download (1MB)
	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_BAB V.pdf Restricted to Registered users only Download (533kB)
Preview	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_DAPUS.pdf Download (530kB) | Preview
	Text 21.K1.0005_VALENTINO PUTRA BUDI KUSUMO_LAMP.pdf Restricted to Registered users only Download (735kB)

Abstract

This research investigates the application of transformer-based models for sentiment analysis of movie reviews, emphasizing the role of preprocessing techniques. Traditional machine learning approaches have shown limited capability in capturing the emotional complexity of textual data. The study explores preprocessing methods such as case folding, symbol cleaning, tokenization, stopword removal, and stemming to evaluate their impact on the performance of the T5-small model integrated with a Multilayer perceptron (MLP). Using kaggle-sourced dataset, the research identifies as the optimal preprocessing combinations to maximize accuracy, precision, recall and F1-score. Experiments demonstrate that preprocessing enhances model;s generalization, reducing overfitting and improving performance metrics compared to unprocessed data. Grid search optimization identifies the best hyperparameters, achieving a peak validation accuracy of 70.56% with balanced precision and recall. The results. The results underline the necessity of tailored preprocessing for effective sentiment analysis while highlighting gaps for future research, such as dataset diversity and advanced hyperparameter tuning.

Actions (login required)

View Item