The Prediction of Public Risk Perception by Internal Characteristics and External Environment: Machine Learning on Big Data

Presently, the public’s perception of risk in terms of topical social issues is mainly measured quantitively using a psychological scale, but this approach is not accurate enough for everyday data. In this paper, we explored the ways in which public risk perception can be more accurately predicted in the era of big data. We obtained internal characteristics and external environment predictor variables through a literature review, and then built our prediction model using the machine learning of a BP neural network via three steps: the calculation of the node number of the implication level, a performance test of the BP neural network, and the computation of the weight of every input node. Taking the public risk perception of the Sino–US trade friction and the COVID-19 pandemic in China as research cases, we found that, according to our tests, the node number of the implication level was 15 in terms of the Sino–US trade friction and 14 in terms of the COVID-19 pandemic. Following this, machine learning was conducted, through which we found that the R(2) of the BP neural network prediction model was 0.88651 and 0.87125, respectively, for the two cases, which accurately predicted the public’s risk perception of the data on a certain day, and simultaneously obtained the weight of every predictor variable in each case. In this paper, we provide comments and suggestions for building a model to predict the public’s perception of topical issues.