Social media has evolved from a communication channel into a real-time sensor of the physical world, often surfacing the earliest signals of disasters, outbreaks, and social unrest. However, social streams are noisy, incomplete, and multimodal, which makes reliable event detection difficult in deployment. In this chapter, we first motivate why event detection must move beyond traditional news-centric pipelines and formalize the modern problem setting with streaming, multimodal inputs. We then clarify key concepts, including event instance versus event type, and summarize the practical outputs of detection systems: early detection, coherent grouping, and compact event description. Next, we articulate three structural challenges that repeatedly undermine real-world performance: information fragmentation, cross-platform heterogeneity, and open-world event discovery, and explain how these challenges interact and create trade-offs. Finally, we outline the aim, contributions, and organization of this monograph, introducing a three-tiered, specialized decomposition strategy that progressively addresses grounding, transfer, and novelty discovery.