After the occurrence of social hot events, it is highly prone to trigger a series of public opinion risks. Given the diversity of factors contributing to the generation of public opinion risks in social hot events and their complex coupling relationships, this paper aims to explore the underlying mechanism of such risk generation. Based on 155 cases of public opinion related to social hot events, this study applies the Fuzzy-Set Qualitative Comparative Analysis (fsQCA) to examine the complex causal mechanisms driving the formation of public opinion risks from the configuration perspective. Firstly, a “three-degree” indicator system comprising “concentration degree”, “organization degree”, and “criticality degree” is constructed to quantitatively represent the public opinion risk index. Secondly, antecedent variables are extracted based on the characteristics of public opinion in social risk events, and machine learning methods are employed to filter these variables. Finally, three configuration are obtained through fsQCA. In the general event configuration of high risk index, high topic count and short video dissemination channel are the key reasons; In the configuration of pan-sensitive group events of high risk index, the vulnerable groups and sensitive occupational groups and short video dissemination channel are the key reasons; In the configuration of vulnerable group events of high risk index, the vulnerable groups, traditional media and online opinion leaders are key reasons. This research offers valuable insights for predicting the trends of public opinion risks associated with social hot events and guiding their management and response strategies.