فهرست نشریات علمی دسترسی آزاد ایران

The rapid growth of Twitter has transformed it into a critical real-time sensor for world events, often surfacing information about disasters, political upheaval, and public health crises ahead of traditional sources. While detecting major events is valuable, the ability to identify sub-events—fine-grained, evolving components—is crucial for deeper situational awareness. This survey provides a comprehensive review of NLP techniques for sub-event prediction and evolution on Twitter. We introduce a novel taxonomy that categorizes methods from traditional text-based and graph-based approaches to modern deep learning and transformer-based architectures, specifically evaluating their capacity to capture sub-event dynamics. Our analysis covers widely used benchmarks (e.g., CrisisNLP, CrisisBench, COVID-Twitter datasets) and evaluation protocols (e.g., precision, recall, F1, clustering metrics). The findings indicate that while significant advances have been made, the fusion of multimodal data, the application of large language models, and the adoption of privacy-preserving frameworks like federated learning represent the most promising pathways for robust sub-event detection. By synthesizing methodological advances and evaluation practices, this paper underscores the central role of sub-event analysis in advancing research and its critical importance for real-time, high-stakes applications in disaster response, public health, and security.