A Multi-Layer Business Analytics Framework for Evaluating DJP Online's Digital Tax Policy

This dataset was compiled to support the analysis of public sentiment and engagement toward the Directorate General of Taxes’ (DJP) digital tax platform, DJP Online and its mobile application M-Pajak. Data were collected through web scraping of user reviews from the Google Play Store (Application ID: id.go.pajak.djp) and include anonymized attributes such as username, review content, rating score (1–5 stars), and posting date. The final dataset contains 1,500 entries reflecting public perceptions of DJP’s digital services.

All data were preprocessed for text cleaning, normalization, and anonymization to remove personal identifiers. The dataset was further used for sentiment classification (positive, neutral, negative) using machine learning models including Multinomial Naïve Bayes, SVM, and IndoBERT, and for ARIMAX-based predictive modeling to evaluate the relationship between sentiment, engagement, and funnel conversion.

This dataset supports quantitative research on e-government service evaluation, digital engagement behavior, and policy effectiveness in Indonesia’s tax administration context.

本数据集旨在支撑针对印度尼西亚税务总局（Directorate General of Taxes，简称DJP）的电子税务平台“DJP Online”及其移动应用“M-Pajak”的公众舆情与参与度分析。数据通过网络爬虫采集自谷歌应用商店（Google Play Store）的用户评论，对应应用包名为id.go.pajak.djp，包含用户名、评论内容、评分（1-5星）及发布日期等匿名化属性字段。最终数据集共包含1500条记录，用以反映公众对DJP电子税务服务的感知评价。

所有数据均经过文本清洗、归一化及匿名化处理，以移除个人身份标识。本数据集后续被用于开展情感分类（分类标签为正面、中性、负面）任务，所用机器学习模型包括多项式朴素贝叶斯（Multinomial Naïve Bayes）、支持向量机（SVM）及印尼BERT（IndoBERT）；同时还被用于基于带外生变量的自回归移动平均模型（ARIMAX）的预测建模，以分析舆情、参与度与漏斗转化率之间的关联关系。

本数据集可支撑印尼税务征管语境下的电子政务服务评估、数字参与行为及政策有效性相关的量化研究。