Abstract: 【Purpose/significance】With the continuous expansion of the network scale, strengthening the feature mining and trend pre⁃diction of public opinion on public health security, and exploring the early warning mechanism of public opinion have become the key issues to be solved urgently in the field of public health security governance.【Method/process】Based on the understanding of the early warning characteristics of public opinion on public health safety network, this paper adopts the weak correlation mining method to construct an early warning model of public opinion on public health safety network including three dimensions of public opinion co⁃hesion, thermality and trend. And taking the public opinion data of "Changchun Changsheng Vaccine Incident" as an example, the in⁃fluencing factors of public opinion on public health safety network were analyzed in detail. At the same time, the K-Means clustering algorithm and the grey relational analysis method are comprehensively used to carry out early warning and classification of the public opinion development time of the event, which provides support for subsequent countermeasures and suggestions.【Result/conclusion】The results show that the weak association mining method based on K-Means clustering algorithm has strong applicability in the pub⁃lic opinion of public health and safety.【Innovation/limitation】Based on the public opinion characteristics of public health safety net⁃work, this paper uses K-Means clustering algorithm and grey relational analysis method to divide the early warning grade, and realizes
the innovation of research method. However, there are still some deficiencies, such as the research on automatic contact points of early warning classification, etc., which need to be further explored in the future.