Research On Criminal Aided Judgment Based On TextCNN+RF(0.5) Model

In recent years,with the rapid development of modern high-tech technologies such as artificial intelligence at home and abroad,a large number of companies have emerged at home and abroad that provide big data support for legal systems and legal artificial intelligence systems.These companies not only serve law firms,but even domestic and foreign judicial authorities have begun to introduce relevant technologies.The continuous growth and development of the legal artificial intelligence system will play a positive role in promoting judicial cases and legal social services.In the judicial big data service network,there is still room for improvement and improvement as a case-aided judgment function for cases that is important to judges,the public,and legal practitioners.As one of the typical tasks of legal intelligence,automatic judgment prediction has decades of research history.From machine learning to deep learning,scholars have proposed a variety of models to improve this supervised learning task.From the perspective of the nature of text classification,how to extract text features has always been a key task,and the selection of models is of course crucial.This article starts from the essential problem of text classification and combines the particularity of the application of text classification principles in legal issues.For the prediction of auxiliary judgments in criminal cases,the following work is mainly done: First,based on fully mining and extracting the textual features of criminal case facts,this paper introduces first-level criminal case reasons and confounding judgment attributes as the mapping between criminal case facts and criminal charges.For criminal law cases,the primary criminal case determines the criminal charges in the final criminal sentence.Therefore,combined with the logic of accepting legal cases,this paper considers the method of multi-level label classification.Based on the accurate classification of the first-level criminal cases of criminal cases,it trains the criminal charge prediction models of each class of criminal cases for classification.At the same time,in order to better classify confusion judgment pairs under the same criminal case,this article introduces 10 general confusion judgment attributes.Confusion decisions not only have a good effect on distinguishing confusion decisions,but also share information from high-frequency decisions to low-frequency decisions.At the same time,through sample collection,this paper realizes the focus on small samples,and to a certain extent,improves the prediction results of small samples.Through experiments on real data sets,the results show that the prediction effect based on attribute extraction and first-level criminal case performance performs well on criminal judgment classification tasks.Secondly,in terms of model selection,this paper selects textCNN as the first-class criminal case classification and confusion judgment attribute classification prediction model,and proposes an improved weighted voting random forest algorithm based on F(0.5)score as the final prediction of criminal charges.In view of the excellent performance of textCNN on short text classification tasks,this paper proposes to use textCNN to predict the first-level criminal case,and at the same time,extract confusion judgment attributes based on the facts of the case.Output the first-level criminal case reason and the attribute strength of each confounding judgment attribute.At the same time,it simulates the judge’s judgment logic and uses the if-then rule set idea when the random forest decision tree is generated to classify the final criminal charges.Through experiments on real data sets,the results show that the textCNN + RF(0.5)model proposed in this paper has achieved better results than the single model and other baseline models.