Research On Element Extraction And Crime Prediction Based On Legal Documents

In recent years,with the development of the legal society and the gradual improvement of the law,the demand for the justice,efficiency,intelligence and other aspects of case judgment began to seek artificial intelligence technology to solve.The widespread application of artificial intelligence in all walks of life also makes it possible to use artificial intelligence technology to solve the above problems in the judicial field.In the judicial field,how to extract the elements of the case from the unstructured legal documents and prejudge the crimes committed by the parties according to the description of the legal documents are two important tasks in the legal intelligence.Which elements of the extract to the case violates the law has the close relation,and in the legal document data according to one over ten of the total charges,all charges data distribution is not balanced,has a great influence on charges of anticipation performance,at the same time,considering the case elements is a main and important information in the legal document,is actively promoting significance to the anticipation of the case.Information,therefore,this paper proposes a fusion rules and the elements of context prediction information extraction method and fusion cases elements in capsule network model to predict how crime method,purpose is through the crawler frame extracting criminal cases the legal document data from the Internet,and then on the basis of experimental data for the data set to training factors identification and prediction model on charges,to automatically extract from the legal document to define the elements of information,and further make use of information elements guide charges forecasting task.This paper mainly completed the following research work:(1)This paper collects and labels criminal case data,obtains the element labels data of criminal cases and the prediction data of charges with multiple charges labels.Due to the lack of labeling data of the categories of criminal cases and the absence of labeling data of multiple charges in the existing crime label data.Therefore,firstly,lawyers and other rich cases on the Internet are used to collect the data of these pages through web crawler technology,and the crawling content is mainly the factual description,legal articles,verdict,etc.(sentence term,charge,fine)of the case.Then the data were preprocessed,data cleaning,crime normalization,word segmentation and word vector training were carried out in order,and 58,000 300 dimensional word vector sets were obtained through word vector pretraining.Finally,the data were marked according to the defined labeling rules,and 1000 element labeling data and 2.2 million crime labeling data were obtained.(2)Proposed the method of extracting elements of criminal cases integrated with regulatory information,and realized the automatic extraction of elements in legal documents.The research on the description of crime in the legal documents of criminal cases has a relatively standard structure,so the names of people,places and organizations are easier to identify than the general text.The difficulty is to identify the specific elements in the legal documents.These element statements are often related to the violated laws and regulations,so we consider incorporating the regulatory information into the text to assist the identification of the elements.According to this idea,we first construct a statutory dictionary,namely a keyword dictionary.And then through the way of attention mechanism will regulatory information into text representation,and the statement regulations labels,then each sentence of laws and regulations after the label can be seen as a statement of the implicit state,will predict regulations label as context into the inside,finally USES the text characterization of joining together of two softmax classification.(3)Put forward the method of multi-charge prediction of criminal cases combined with element information,which realized the prediction of charges in legal documents.The data distribution of multiple charges in criminal cases is unbalanced and there is a close relation between charges.The elements in the text of criminal cases are important information in the text and have important guiding significance for the prediction of charges.In this paper,a multi-charge prediction method based on capsule network is proposed.Firstly,the description text of the case is extracted at sentence level through LSTM network,and the information of the elements is integrated into the hidden layer by means of attention mechanism to improve the weight of the elements.Secondly,the sentence level representation of the obtained text is used as the input of the capsule layer,and then the characteristic capsules are sent to different charges respectively through the dynamic routing protocol.Finally,the prediction results are obtained from the vector of different charges in the prediction layer.Experiments show that the self-built data set and the public data set in this paper have been improved to some extent,especially in the performance of multi-charge prediction.(4)Set up the prototype system of element extraction and charge judgment of criminal cases,and completed the automatic extraction of elements and automatic prediction of charges.The system first takes the text of criminal cases as input and then processes the text.(4)Set up the prototype system of element extraction and charge judgment of criminal cases,and completed the automatic extraction of elements and automatic prediction of charges.The system first takes the text of criminal cases as input and then processes the text.The key elements of the case are extracted automatically through the element extraction module and displayed in the form of list.Finally,through the element information and case description text,the case is automatically prejudged and all possible crimes are displayed.