Research On Crime Prediction And Law Recommendation Based On Deep Learning

In recent years,with the rapid development of artificial intelligence,machine learning and deep learning have been widely used.The intelligence of legal field has become a hot research issue in artificial intelligence,and judgment prediction is an important way of intelligence in legal field.Judgment prediction Is based on the fact description text in legal documents,through machine learning,deep learning and other technologies to predict the judgment.The realization of judgment prediction can effectively assist judges,lawyers and other legal staff to deal with legal cases more efficiently,and also better help non-legal staff to understand the application of laws and regulations in cases.Crime prediction and law recommendation are two important sub-tasks in judgment prediction.Charge prediction is to predict the charge of the case according to the fact description of the case,while law recommendation is to recommend the law related to the case according to the fact description part.For crime prediction and law recommendation,most of the traditional methods are based on statistical methods,but the effect of this method is not good,the prediction performance is not good.Therefore,this paper realizes the task of crime prediction and law recommendation through deep learning technology,and then realizes the prediction of crime and law recommendation.An experiment was conducted to verify the feasibility of the method by using the data of public criminal judgment documents on the public data set "China Judgment Documents Network".The main research work of this paper includes the following two aspects:（1）Crime prediction task.Aiming at the crime prediction task,a BERT and LSTM-CNN fusion model was proposed to solve the crime prediction task.The BERT model can use the self-attention mechanism to simultaneously obtain the contextual information of legal texts and perform unsupervised pre-training on a large number of unlabeled texts.Firstly,BERT model is used to extract the embedding vectors of words and sentences from the text data in fact description.Then,the feature vectors extracted by BERT model are input to the fusion model of LSTM model and CNN model for feature extraction,and the features extracted from the fusion model are classified by Softmax to obtain the prediction results.The experimental results show that the proposed method achieves the best results compared with the existing methods for related problems.The accuracy rate is97.64%,macro accuracy rate is 96.41%,macro recall rate is 96.57%,and F1 value is96.97%.（2）Tasks recommended by law.Aiming at the law recommendation task,a law recommendation model based on self-attention mechanism and feature fusion was proposed to solve the law recommendation task.Word2 vec can better calculate the similarity between words and effectively obtain the semantic information of text.Therefore,Word2 vec is first used to vectorize the text in the fact description.Considering that the features mentioned by a single model are not comprehensive,this paper uses multiple models to extract features and fuse them to obtain text features.Firstly,BIGRU model is used to extract contextual features of text,and self-attention mechanism is used to extract weighted information after BIGRU feature extraction.Meanwhile,CNN model is used to extract local key information features of text.Finally,the features of self-attention mechanism and CNN are fused,and the result of law recommendation is obtained from the fused features.The experimental results show that the accuracy,accuracy and macro F1 value of the proposed model are significantly improved compared with other models,reaching 96.20%,94.83%,94.85% and94.82%.