Sentiment Analysis of Coretax Tax Application Users Using IndoBERT and Web Scraping on the X (Twitter) Platform

Case Study on Indonesian Taxpayer Digital Service Feedback

Abstract

This study analyzes public sentiment toward the Coretax tax system based on user opinions posted on the X (Twitter) platform. The objective is to assess how the public perceives the system’s stability, accessibility, and performance during periods of high usage. A quantitative text-based approach was applied using Natural Language Processing (NLP) techniques. Data were collected through web scraping of tweets containing Coretax-related keywords and processed through six preprocessing stages: case folding, cleaning, tokenizing, normalization, stopword removal, and stemming. Sentiment classification was conducted using the IndoBERT model mdhugol/indonesia-bert-sentiment-classification, which categorized tweets into positive, negative, and neutral classes. The results show that 181 tweets expressed positive sentiment, 171 negative sentiment, and 29 neutral sentiment. Negative sentiment predominantly relates to system errors and login difficulties, whereas positive sentiment commonly appears when the system functions normally. These findings demonstrate that system instability remains the primary factor influencing negative perceptions of Coretax and provide useful insights for improving the reliability of digital tax services.