Nowadays, the situation of cumulative public opinion risks caused by repeated occurrences of similar events is becoming increasingly common. Therefore, it is necessary to study the key factors in the formation of public opinion on similar events. Currently, there are fewer studies on public opinion on similar events, and the model of the public opinion evolution mechanism is getting more complicated, which raises the cost of making suggestions for public opinion response. In response, integrating the machine learning explanation model, text mining algorithms, and complex network, this study constructed an algorithm for mining key factors of public opinion evolution that considers the characteristics of similar events. The effectiveness of the algorithm was demonstrated through specific cases.