Research on Scenario Classification of Burglary Cases via BERT

Case descriptive files record the details of the crimes. For burglary cases, criminal scenarios are crucial for case analysis. This paper focuses on scenario classification task. We regard the task as a text classification problem. The BERT model is applied to tackle the problem. We additionally propose a stop-words processing algorithm which can effectively reduce the interference caused by irrelevant information in the text and speed up the model to convergence. Experiments show that the BERT model with the help of stop-words processing algorithm achieves higher performance than other four benchmarks. The F1 score finally reaches 98.83%.