Abstract: As complex systems jointly constituted by physical space and human activity, cities are characterized by continuously evolving, feedback-driven operational processes. The rapid development of the Internet and the Internet of Things in recent years has made it possible to capture real-time urban activities; however, extracting interpretable knowledge structures from fragmented perceptions remains a central challenge for achieving dynamic urban cognition. This study employs knowledge graphs as a representational framework, integrates multi-source spatiotemporal data, and constructs a cognition methodology for urban human activities centered on the semantic structure of "human–object–place–time–event–effect," enabling multidimensional networked cognition and structured expression of urban space and human activity. Taking a university campus as a representative micro-scenario, a knowledge graph comprising approximately 54,000 entity nodes and 118,000 relational edges was constructed to perform relation extraction and causal tracing analysis. Results demonstrate that this approach effectively supports activity attribution and spatial governance response in complex environments through activity backtracking and semantic linkage. It also exhibits strong cross-scale scalability and transferability, providing a feasible pathway and methodological foundation for behavior-driven planning and governance in the era of emerging urban technologies.