A Method for Extracting Documents for Theft Crime Based on Deep Learning

In order to assist judicial staff in analyzing the sentencing of the case, this article uses the named entity identification approach to extract the sentencing elements from the theft records. This study adopts the BERT named entity recognition approach, primarily employing the pre-trained model of BERT-WWM-EXT and CHINESE, and extracts the named entity recognition dataset based on stolen papers in Fuzhou City in recent years. The F1 value of the experimental model findings was 70.05%, which essentially made it possible to extract the punishment components from the theft records. However, the F1 value of experimental data is also lower than the reference evaluation index of Chinese Machine Reading Comprehension Data (CJRC) for the judicial area published online due to manual labeling errors, data processing errors, a small number of datasets, and machine restrictions.