abstract |
The invention relates to a multi-cause classification method for aviation safety reports based on a BERT model, comprising the following steps: S1, constructing a historical data set of aviation unsafe accidents; S2, performing data preprocessing on the historical data set; S3, constructing a BERT model, Including model fine-tuning, model training and model evaluation; S4, real-time input data set for the trained BERT model and predict accident cause categories at different levels. Compared with the prior art, the present invention can effectively improve the performance of accident cause classification and predict accident cause categories at different levels. |