Abstract:

In order to explore the public's risk perception of safety accidents, text mining technology was used to obtain accident micro-blog comment data, emotions like surprise and fear were identified by adopting BERT-RPC. Then, accident probability perception bias was measured by frequency surprise, and accident loss perception bias was measured by fear and loss surprise. Finally, influence of micro-blog form and content on risk perception was investigated by binary logic regression. The results show that the underestimation of safety accident risks is very common, and it is more serious in loss perception. The "capture-analysis" technology based on BERT-RPC model can monitor risk perception bias of the public in the whole network with high efficiency and low delay. Moreover, accident probability and loss in transportation industry are significantly underestimated, and so are the probability of general accidents and loss of major ones. While accident pictures and videos can help correct perception bias of accident loss, their effect on probability perception is limited. Accident reports in the early stage has the best correction effect on risk perception, and then in stages where survey results and people held accountable are announced.

Key words: emotion analysis, accident risk, perception bias, bidirectional encoder representations from transformers (BERT), risk perception-Chinese(RPC)