Abstract: 【Purpose/significance】From the perspective of intelligent communication, clarifying the generation mechanism and driving mode of network public opinion risk clusters can provide in-depth theoretical support for the guidance and control of network public opinion risks.【Method/process】Based on the stimulus-response theory, this paper analyzes the generation mechanism of network public opinion risk clusters, and explores their driving patterns through configurational path analysis.【Result/conclusion】The genera⁃ tion and development process of network public opinion risk clusters can be explained through the mechanism of triggering risk sources in matters, mechanism of artificial intelligence power manipulation, mechanism of group risk processing, mechanism of risk cycle feedback. The configuration path can be summarized into three driving modes: irrational interaction risk dominant type, extreme viewpoint risk-artificial intelligence risk dual driving type, and irrational interaction risk-extreme viewpoint risk-artificial intelli⁃ gence risk collaborative type. Subsequent research can focus on the risk clusters of online public opinion, encompassing studies on risk identification, risk assessment, and targeted guidance and control.【Innovation/limitation】In the process of mining network public opinion risk clusters, artificial intelligence risks are taken into consideration, providing a new perspective for the governance frame⁃ work of network public opinion risks from the perspective of intelligent communication. This paper focuses on theoretical exploration, and further research is still needed to explore risk identification, risk assessment, and targeted guidance and control of network public opinion risk clusters based on large-scale datasets.