Research On Crime Prediction Based On Deep Neural Network

With the progress of the times and the development of society,it can be seen from the data alone that China can produce tens of thousands of legal cases every year,however,China’s legal resources are limited,plus legal aid must be the relevant legal practitioners or professionals to provide services,which may lead to the existence of some legal resources do not cover the place,some legal aid may not be implemented to the actual Therefore,it is possible to forecast legal judgments with the help of computers,which can effectively alleviate the current legal situation,implement the strategic policy of the rule of law,promote justice,appropriately relieve the pressure of relevant judicial practitioners,and provide a convenient and fast form of legal aid for those in need.In this paper,we explore the problem that the models in the current charge prediction task cannot fully obtain the information contained in the text and how to further improve the prediction accuracy,aiming to appropriately solve the problems related to the lack of judicial resources in our society with the help of current computer technology.In order to solve the problem of insufficient prediction accuracy in the current crime prediction task,this paper designs and proposes some research on crime prediction based on natural language processing technology of deep learning and judicial related knowledge as follows.Firstly,this paper presents a study of charge prediction combining BiGRU and Attention mechanism.In order to obtain more comprehensive textual semantic information,the model uses a two-way gated recurrent neural network to extract information features from textual data,and appropriately introduces the Attention mechanism to filter the extracted information.In order to obtain more comprehensive text semantic information,the model uses a two-way gated recurrent neural network to extract information features from the text data,and appropriately introduces the Attention mechanism to filter the extracted information,eliminating meaningless or secondary information to improve the purpose of obtaining the text information intended to be expressed in the text.The final experimental results obtained good results compared with the baseline model.The experimental results obtained the highest experimental scores compared to the baseline model,proving the effectiveness of the proposed method in solving the charge prediction task.Secondly,this paper proposes a deep neural network model incorporating ALBERT and TextCNN for charge prediction.The model is a pre-trained language model,which can show good performance on legal long text information.The model automatically learns to obtain the semantic information in the input text data and transforms it to improve the performance of the model,using ALBERT can reduce the number of parameters generated during the training process without losing the model accuracy compared with the previously used deep neural network model.This allows the training speed of the model to be improved,which saves the corresponding resources and achieves good performance,and then combined with convolutional neural network for charge prediction.The experimental results show that the model performs up to 3.7% and 14.6% higher on the micro-average macro-average compared with other mainstream neural network models on the charge prediction task,which effectively improves the effect of the charge prediction task.