With the increasing complexity and refinement of public security forensics, the legal compliance review mechanism plays an important role in the process of public security forensics.In this paper, we propose a semantic role annotation and legal domain oriented entity-relationship extraction method based on BERT-BiLSTM-CRF. In legal text processing, this paper introduces the BERT model with powerful semantic understanding capability on the basis of BiLSTM-CRF model, which further enhances the semantic role annotation model’s ability to understand the terminology of semantic structure of legal text. In addition, the models of legal information enhancement module, legal potential relationship and global correspondence model and decoder are constructed for entity relationship extraction in legal domain. The study shows that the semantic role labeling algorithm in this paper has different degrees of improvement in F1, P and R indicators, while the entity relationship extraction method has an extraction accuracy of more than 78% in multiple cycles, and the extraction accuracy is close to 100% on individual legal relationships. And the application of legal knowledge graph under the method of this paper in public security forensics provides rich legal entity relations for public security forensics, reduces the time of manual review and improves the reliability of the review results.
