ANALISIS SENTIMEN PUBLIK INDONESIA TERHADAP KONFLIK ISRAEL-IRAN DI MEDIA SOSIAL MENGGUNAKAN INDOBERT DAN EXPLAINABLE AI (LIME)

Abstract

This study analyzes Indonesian public sentiment toward the Israel–Iran conflict on social media using a fine-tuned IndoBERT model and Explainable AI (XAI) techniques based on LIME. Data were collected from YouTube and X (Twitter) between April 2024 and June 2025, yielding 7,922 unique entries that were preprocessed and automatically annotated. The IndoBERT model achieved an accuracy of 74.90% and an F1-score of 0.7521 on the test set (n = 1,513). LIME-based interpretations reveal trigger words such as “zionist” (negative) and “allahu akbar” (positive), indicating opinion polarization driven by anti-Zionist narratives and religious solidarity. This approach enhances the transparency of black-box models and provides insights for public opinion monitoring. The findings contribute to Indonesian NLP and geopolitical analysis, with limitations related to the quality of automatic annotation. The results are consistent with studies on the Israel–Palestine conflict showing dominant negative sentiment on social media. Similar analyses using BERT for Reddit discussions highlight the importance of hybrid approaches for stance. The integration of XAI techniques such as LIME has been shown to be effective in explaining sentiment predictions. Therefore, this methodology enriches the understanding of opinion dynamics in sensitive geopolitical contexts