Exploring Isis’ Takfir Discourse: A BERT-Based Entity Level Sentiment Analysis Approach

Abstract

The Islamic State of Iraq and Syria (ISIS) significantly influenced the lives of many people during and after the Syrian civil conflict, especially civilians. Analyzing social media discussions about ISIS can provide valuable insights into the group’s beliefs and attitudes. In this paper, we examine ISIS’s takfir discourse—their practice of labeling other Muslims as unbelievers to justify exclusion or violence—in Telegram groups. We collected 14,500 Telegram messages (2015–2017) using snowball sampling, API-based crawling, language filtering, and time-window selection. We then integrated a BERT-based Named Entity Recognition (NER) model with two layers of the Span ASTE (Aspect-Based Sentiment Analysis) model. We also used the Span ASTE as an end-to-end baseline for comparison. Based on Precision, Recall, and F1-scores, our hybrid model outperformed the baselines, demonstrating its effectiveness in sentiment analysis of extracted named entities. © 2025 Elsevier B.V., All rights reserved.