Criminal case criminal name prediction method based on sequence enhanced capsule network

Abstract

The invention relates to the field of intelligent law, in particular to a criminal case criminal name prediction method based on a sequence enhanced capsule network. The method comprises the followingsteps: S1, constructing a training data set, and obtaining fact description of a case and a criminal name penalty result as training data; S2, constructing a sequence enhanced capsule network model and carrying out training through training data; and S3, through the sequence enhanced capsule network model trained in the step S2, inputting the fact description text of the new case into the sequence enhanced capsule network model, and enabling the model to automatically predict a corresponding criminal name as a criminal name prediction result. The model provided by the invention not only can better capture the significant characteristics and semantic information of legal texts, but also has better competitiveness on the low-frequency criminal name prediction problem; and a final loss function is introduced to serve as a loss function of the sequence enhanced capsule network model, so that the problem of high imbalance of the criminal names of the low-frequency criminal name predictiontask is further relieved.

Description

technical field [0001] The invention relates to the field of intelligent law, in particular to a method for predicting crimes in criminal cases based on sequence-enhanced capsule networks. Background technique [0002] In recent years, artificial intelligence technology represented by deep learning and natural language processing has made great breakthroughs, and has begun to emerge in the field of intelligent law, attracting widespread attention from academia and industry. Smart laws endow machines with the ability to understand legal texts, analyze cases, and handle cases intelligently based on cases. [0003] Automatic crime prediction, as one of the most representative subtasks in smart law, plays an important role in legal assistant systems and has wide applications in real life. For example, it can provide legal experts (such as lawyers and judges) with references to the defendant's charges in the case, so as to assist judges in adjudicating cases and improve work eff...

AI Technical Summary

Problems solved by technology

However, there are still deficiencies in the existing methods: (1) Most of the existing work [9,10] ignores the low-frequency crime scene of the automatic crime prediction task, and only considers the high-frequency crime scene, so it cannot solve the low-frequency crime scene very well. crime prediction problem

(2) Hu et al. [11] used artificially generated auxiliary information to achieve good results in low-frequency crime scenes. However, manual labeling information wasted a lot of time and could not achieve an end-to-end deep learning model.

[0007] The national invention patent application "A method for predicting crimes in criminal cases based on memory neural network" (public date: 2019.02.22) uses standard case descriptions and crimes as training data to build a training data set. The memory neural network model is trained, and the "case description feature vector"-"crime code" pair is converted into a key-value pair stored in the memory neural network model, and a multi-layer perceptron classifier is used to judge the crime of a criminal case. This method proposes Although the model can also predict low-frequency crimes, the memory module needs to compare the relationship between the real crime and the predicted crime. However, the amount of data for low-frequency crimes is small, and there are only a few cases in some crimes. Good results are achieved in low-frequency crime prediction scenarios

Images