Computer Assisted Sentencing Based On Convolutional Neural Network

In recent years,the research on machine learning and deep learning in the field of natural science is in full swing.With the continuous development of computer technology,people have gradually transferred their research to the field of social science,especially the field of justice.Sentencing is an important part of the judgment system in the field of justice.It determines whether to punish the suspect and the corresponding degree of punishment.In the traditional judicial judgment,the determination of sentencing often takes a lot of time and manpower,which causes a certain pressure to the trial judgment,and the introduction of computer-aided sentencing can effectively alleviate this pressure,but at present the application of computer technology in the field of sentencing is not much.Therefore,both the industry and the general public are looking to use computer technology to reform the sentencing system.This paper mainly considers the study of the criminal cases in judicial cases and the sentence and fine in the sentencing system.Computer assisted sentencing measurement refers to the prediction of the specific value of sentence and fine in a case by modeling given the case facts.This study mainly uses the convolutional neural network model in deep learning,and compares three integrated models in traditional machine learning,namely random forest,GBDT and XGBoost,in order to find a more effective computer assisted sentencing model.First,the case fact text in the sentence prediction and fine prediction data set is preprocessed,and the Jieba word segmentation tool is improved on the basis of the case fact text data cleaning,and the improved word segmentation tool is used for text segmentation of the case facts.Secondly,the text features of the text data of the case facts were analyzed to obtain the corresponding word frequency statistics and word cloud map,from which the high-frequency words and low-frequency words were obtained and the text data after word segmentation was cleaned and filtered.Thirdly,the text word vector is constructed based on Word2 Vec.Finally,the corresponding sentence prediction model and penalty prediction model are established according to the word vector,and the evaluation system of the model is constructed according to the distribution characteristics of sentence and penalty to evaluate the model,and some conclusions are obtained from the evaluation results of the model.he results show that the prediction of sentence and fine is not good in three traditional machine learning algorithms.Among them,XGBoost model is slightly better than randomforest model and GBDT model in predicting sentence and penalty,with scores of 60.83 and59.89 respectively.The random forest model has the lowest scores in the prediction of prison term and fine,which are 59.76 and 56.93 respectively.With the introduction of convolutional neural network,the effect of sentence prediction model and penalty prediction model has been significantly improved.The convolutional neural network scores 78.99 in the sentence prediction model and 76.73 in the penalty prediction model.It can be seen that compared with machine learning,convolutional neural network has a very good effect in improving computer-aided sentencing.