Penerapan Explainable AI LIME pada Klasifikasi Sentimen Isu IKN, PPN, dan MBG Menggunakan IndoBERTweet
Abstrak
Social media, particularly the X platform (formerly Twitter), has become a primary space for the public to express their opinions on public policy. This study aimed to classify public sentiment towards three strategic issues in Indonesia: the relocation of the national capital (IKN), the 11% VAT increase policy, and the Free Nutritious Meals (MBG) program by applying the IndoBERTweet model combined with a LIME-based Explainable Artificial Intelligence (XAI) approach. A dataset of 2,845 tweets was obtained through scraping using relevant keywords. The sentiment distribution in the data reveals a dominance of negative sentiment, with 1,575 tweets, followed by positive sentiment, with 1,005 tweets, and neutral sentiment, with 265 tweets. The IndoBERTweet model, trained for three epochs, produced an accuracy of 89% and an average macro F1-score of 0.78. The highest performance was achieved in both the positive and negative classes, with an F1-score of 0.92, while neutral sentiment remained challenging, with an F1-score of 0.49. To improve prediction transparency, the LIME method was employed and successfully displayed consistent visualizations that identified the dominant words influencing the classification. This study demonstrates the potential application of explainable Indonesian language transformer models for sentiment analysis of national policies in the digital realm.
Identifier
- DOI: https://doi.org/10.24252/shift.v6i1.236
- OpenAlex: https://openalex.org/W7167375982