IMPLEMENTATION OF TEXT PROCESSING FOR SENTIMENT ANALYSIS OF TAX PAYMENT INTEREST AFTER THE "RUBICON" PHENOMENON

Abstract

In February 2023, an incident occurred involving the child of an official from the Indonesian Directorate General of Taxes who committed violence against a member of the GP Ansor organization. The news spread widely and brought a new issue, namely suspicious reporting of the official's wealth with an amount of up to 56 billion Indonesian Rupiahs. In order to determine public sentiment towards the "RUBICON" case, which was receiving attention, sentiment analysis of tax payment interest was conducted using text mining techniques. Data processing was done using the R language and RStudio application, taking a dataset of 23,785 tweets from the public about paying taxes on Twitter. Next, text cleaning was done to remove numbers, symbols, and URLs, as well as text processing using stemming, tokenizing, stopword removal, and TF-IDF methods. The TF-IDF method shows that the words "rafael" and "case" are the top keywords. This study used a supervised model by comparing SVM, KNN, and Naive Bayes algorithms, and evaluation was done using a confusion matrix with accuracy results in descending order of 0.8922, 0.8049, and 0.7369. The conclusion of this study is that the SVM algorithm successfully classified sentiment with the highest level of accuracy and obtained the highest negative sentiment of 5,616 sentences.