The social network data contain a great deal of unstructured knowledge which is useful to understand public safety incidents. The structure of social network data, such as news, statistical statements, social network messages, government reports and so on, is complicated. So, it is difficult to obtain knowledge from social network data because social network data are multi-source heterogeneous. Knowledge graph is one of the effective tools for processing such multi-source heterogeneous data. However, the previous works didn’t notice the particularity of entities in public safety incidents and the linguistic characteristics of Chinese. Moreover, previous works on event knowledge graph mainly focus on single data source which make it is difficult to process multi-source social network data. Thus, the construction of a public safety incident knowledge graph using complex social network data in the Chinese environment is still a difficult problem. In this paper, a framework for public safety event knowledge graph construction towards Chinese social network data, namely EventSKG, is proposed. EventSKG contains a direct mapping module for knowledge extraction from structured data, an Event-Specific FLAT model for entity extraction from unstructured data, a manual template for relation extraction from unstructured data, and an integration surgery to connect the triples obtained from different data source. An event knowledge graph related to the flood disaster in 2020 in China is constructed via EventSKG framework, which contains 14397 entities and 43168 relations. The event knowledge graph will form associations around network data from various sources for the same event, not only enabling humans to better understand the event, but also providing basic knowledge base for future intelligent understanding of events.