Analisis Sentimen Pengguna Indodax Menggunakan FastText dan Convolutional Neural Network

Abstract

The rapid growth of cryptocurrency investors in Indonesia, surpassing stock market investors, highlights the importance of trading platforms like Indodax. User reviews are crucial for service evaluation, but their large volume necessitates automated analysis. This study aims to build and evaluate a deep learning model for sentiment analysis on Indodax application reviews from the Google Play Store. The method used is a combination of FastText for word embedding and a Convolutional neural network (CNN) for sentiment classification into three classes positive, negative, and neutral. The evaluation focused on three main aspects: the influence of CNN kernel size (3, 4, and 5), the impact of data preprocessing stages, and the overall model kinerjance. The results indicate that the model with a kernel size of 5 achieved the highest accuracy of 0,8079. Data preprocessing stages, particularly cleaning, proved crucial in enhancing model performance. Overall, the combined FastText and CNN model demonstrated solid performance, although it faced challenges in distinguishing between neutral and negative reviews. This research contributes to the application of sentiment analysis in the Indonesian digital finance domain.