Sentiment Analysis Approaches for Mobile Banking App Reviews Using Transfer Learning Method

Digital transformation of banking services, namely Mobile Banking or m-Banking, has succeeded in becoming a breakthrough in people's routines regarding banking transactional and non-transactional activities in Indonesia since 2021. This has resulted in customers' needs regarding the ease of using m-Banking process to be increasingly high. Thus, a process that can simplify the analysis of many user opinion reviews is important. This research intent to determine the best algorithm approach through comparative analysis of the final accuracy value with machine learning, deep learning, and focusing on transfer learning algorithms for the purpose of best sentiment analysis performance. The data was obtained from online reviews regarding user experience from a mobile banking application that has been widely used in Indonesia, XYZ Mobile. Knowledge Discovery in Database (KDD) method was conducted in this research with the stages of data selection, preprocessing, transformation, mining, and evaluation. The need to optimize accuracy values in machine learning and deep learning algorithms utilizes a swarm intelligence-based algorithm, namely Ant Colony Optimization (ACO) to improve the accuracy results that have been produced. Based on the research results, transfer learning algorithms show superiority in producing the best accuracy values among other modeling algorithms. Bidirectional Encoder Representations from Transformers (BERT) succeeded in producing the highest accuracy value of 97.02% followed by Fourier Neural Network (Fnet) at 96.56% and eXtra Long Neural Network (XLNet) at 96.37%.