Mengenal Model BERT dan Implementasinya untuk Analisis Sentimen Ulasan Game

Abstract

This Systematic Literature Review (SLR) analyzes the use of the BERT (Bidirectional Encoder Representations from Transformers) model for sentiment analysis of game reviews. The research identifies methods of implementing BERT, evaluates its effectiveness compared to other models, and explores challenges and opportunities in this context. Results show that BERT consistently outperforms traditional models, with an average accuracy improvement of 3-7%. BERT architecture variations, adiptive fine-tuning techniques, and pre-processing specific to game languages proved effective. Practical applications include game sales prediction and identification of preferred features. Key challenges include high computational requirements and handling game-specific context. Future research directions include domain-specific model development, multi-modal integration, and improved interpretability. This SLR provides a foundation for further research in the use of advanced language models for sentiment analysis of game reviews.