Real social streams are open and evolving: novel event types emerge continually, and closed-world detectors often misclassify unknown events with misleading confidence. In this chapter, we introduce Dynamic Augmentation with Entropy Optimization (DAEO), a framework for open-world event discovery that couples generalized category discovery with uncertainty-aware learning. We first formalize the shift from closed-set monitoring to open-world discovery and identify the requirements of dual performance, meaningful uncertainty, temporal coherence, and coherent grouping of unknowns. We then present DAEO, including multimodal feature extraction, dynamic multimodal augmentation to strengthen separability under distribution shift, and adaptive entropy optimization to calibrate confidence and stabilize novelty decisions over time. Next, we introduce the Multimodal Social Event Detection (MSED) dataset and an evaluation protocol that measures unknown detection, grouping quality, and stability in evolving event spaces. Finally, comprehensive experiments compare DAEO with strong baselines and include generalization validation on CrisisMMD, demonstrating more reliable separation of known versus unknown events and more coherent discovery of emerging categories.