Penerapan Local Interpretable Model Agnostic Explanation Lime pada Analisis Sentimen Pengguna TWITTER terhadap Kebijakan Pajak

Abstract

This study aims to explore additional information from the Twitter user sentiment analysis model on tax issues that develop by applying one of the machine learning model interpretation methods, namely LIME (Local Interpretable Model-agnostic Explanation). The data used to build the model was obtained from Twitter in the period of October 2021, with two keywords used is "pajak" and "tax amnesty" in Indonesian. The data that has been taken using the data crawling process using the Twitter API (application Programming Interface) by employing Python programming language is cleaned of unnecessary words, fonts and symbols. The data were then manually labeled positive and negative. There are 1000 texts that will be used with 132 positive sentiments and 868 negative data. The magnitude of the difference in labels or classes makes the data need to be balanced. In this study, the data balancing process uses the SMOTE (Synthetic Minority Over-sampling Technique) method which is applied to the data to be trained to build a model and produces the same amount of data in each class. The process of developing the sentiment analysis model uses the Random Forest classification method with the best accuracy value of 94.5%. The interpretation of the model using LIME provides information that “taxamnesty”, “wajib”, “pajak”, “uuhpp” and “bayar” are the top 5 words that affect the topic of taxes to be positive and “tajakperasrakyat”, “rakyat”, “kemplang”, “kaya” and “pajak” became the top 5 words that influenced the topic of tax to become negative sentiment.