Abstract: 【Purpose/significance】In the scene of epidemic situation,portrait and grouping users in online health communities will help to raise the quality of community services and contribute to broadening the channels of Internet epidemic risk perception【. Method/pro⁃cess】Take "COVID-19" as an example,combine the characteristics of community data to construct portrait labels from five perspec⁃tives:basic user characteristics,user interest topics,emotional tendencies,user consultation needs,and user interaction network roles,and use DBSCAN clustering to achieve portraits. The portrait result presents the user profile; the AP algorithm is used to achieve user grouping based on the portrait,and the user group with the most epidemic risk discovery value is found through social network analysis.【Result/conclusion】The case study shows that the model can effectively generate user portraits of OHC,which can summarize and map users' original appearance.The results showed that there were five groups of community users:patients,suspected patients,doctors, devotees and community administrators.Social network analysis showed that suspected patients and devotees were the most valuable user groups for epidemic risk detection【. Innovation/limitation】The amount of instance analysis data is not up to the standard of severe "big data",and the granularity of image construction still has room for further improvement.