Analisis Sentimen Ulasan Aplikasi Shopee Pada Google Play Store Menggunakan Metode Bidirectional Long Short-Term Memory (BiLSTM)

Sentiment analysis of application reviews on the Google Play Store faces challenges in capturing the complex patterns of natural language, especially in Indonesian reviews that are often informal and unstructured. Deep learning models such as BiLSTM require effective feature extraction techniques to improve sentiment prediction accuracy. This study employs a Bidirectional Long Short- Term Memory (BiLSTM) model with Word2Vec as a feature extraction technique, utilizing review data of the Shopee application collected from the Google Play Store. The data undergoes preprocessing stages, including cleaning, tokenizing, normalization, stopword removal, and stemming, followed by model training with various parameter combinations such as batch size and learning rate. The experimental results show that using a batch size of 32 with a learning rate of 0.001 yields the highest accuracy of 75%, while a batch size of 64 with the same learning rate also results in an accuracy of 75%. Reducing the learning rate to 0.00001 for both batch sizes significantly decreased the accuracy, with the lowest accuracy recorded at 68% to 63% for batch size 32 and 64, respectively. Overall, the use of Word2Vec as a feature extraction technique proved effective in enhancing the model's ability to capture positive, negative, and neutral sentiments from Shopee app reviews. Our finding contributes to the development of sentiment analysis models based on deep learning for e-commerce applications, particularly in the context of Shopee app reviews.