This monograph has presented a comprehensive approach to multimodal event detection on social media by designing three complementary models that systematically address information fragmentation, cross-platform heterogeneity, and open-world discovery. These challenges are not isolated engineering issues; they are structural properties of real-world social media streams. Posts are short and context-poor, platforms encode events with different conventions, and new event types continually emerge. Consequently, effective event detection requires more than a stronger classifier: it requires mechanisms for recovering missing context, transferring capability across platforms, and discovering new categories without destabilizing old knowledge.