Implementasi Metode Bidirectional Encoder Representations from Transformers (BERT) untuk Analisis Sentimen Komentar Pengguna Aplikasi Dana di Instagram

The rapid advancement of technology at present influences various aspects of life and provides convenience and efficiency in various areas. The implementation of technology is evident in the financial sector, with an increasing number of digital financial services facilitating financial transactions. One popular form of digital finance widely used in society is the Dana application. Dana offers services that users can utilize and often provides product information through its Instagram social media account. Feedback and comments about the application are gathered from users. By employing the Bidirectional Encoder Representations from Transformers (BERT) modeling of IndoBert in the sentiment analysis process of user comments on the DANA application on Instagram in this study, it is expected to provide information and facilitate understanding of user opinions. The goal is to detect issues and complaints, offering insights to users about the Dana application. Based on this background, research on sentiment analysis of user comments on the Dana application is conducted. The data used in the study consists of comments from the Dana Instagram account, totaling 2084. The data is divided into two categories: positive and negative, based on automatic labeling by the Transformers. The modeling results using the pre-trained Bidirectional Encoder Representations from Transformers (BERT) method from IndoBert show an accuracy of 98% on the training data and a validation accuracy of 93%. The hyperparameters used are a batch size of 32 and a training epoch of 10, with a 70:30 ratio for training and testing data. Subsequently, the modeling undergoes a deployment process using Streamlit to integrate it into a web-based system or application.

Keywords — Dana, pre-trained Bidirectional Encorder Representations From Transformers (BERT), IndoBert, Sentimen Analytic