Crime Prediction Based on Element Recognition and Entity Boundary Correction

In recent years, crime prediction has become a focal point in intelligent legal systems. Intellectual property crime prediction is especially complex due to its intricate criminal elements and the significant impact of amounts involved on sentencing. This paper introduces the first Intellectual property entity recognition dataset and proposes an innovative model for crime prediction based on element recognition and entity boundary correction. To tackle the issue of ambiguous boundaries in span identification during prediction, we incorporate a boundary detection task into joint training to refine entity boundaries. After recognizing entities and their categories, they are matched with organized rule tables to predict corresponding crime. Our proposed method demonstrates superior performance compared to other entity recognition models on this dataset.