Deep learning-based larceny auxiliary sentencing method

Abstract

The invention discloses a deep learning-based larceny auxiliary sentencing method, and belongs to the field of computers. The method mainly aims to solve the problem that the case semantic representation and the criminal period accurate prediction of a specific criminal name cannot be realized under the condition of less manual annotation, and comprises the following steps of defining the 11-dimensional features for formally describing a theft case from the perspectives of stolen article value, crime subject information, crime fact description and judgment result according to the criminal lawprovisions and sentences provisions about the theft and in combination with a disclosed theft first-review decision-making document; preprocessing the text of the judgment document; integrating into acorpus set, and training the word vectors; extracting the features except the values of the stolen articles and the criminal periods, and constructing a feature generator for each dimension of features by using a recurrent neural network so as to extract the feature values; using the linear regression and a multi-layer neural network model as the predictors, inputting the case feature vectors, and outputting the criminal period prediction results, so that the deep semantic understanding of the case can be realized and a clear criminal period prediction value can be given under the condition of less dependence on the manual annotation.

Description

technical field [0001] The invention belongs to the field of computers, and in particular relates to an auxiliary sentencing method for the crime of theft based on deep learning. Background technique [0002] The amount of data in the judicial field has grown rapidly over the past few years. These data involve judicial documents, laws and regulations, and judicial interpretations of various legal cases. Legal professionals such as judges, lawyers, and prosecutors not only have to deal with a large number of cases, but also need to consult a large number of case-related documents for reference and analysis. This places an increasing burden on legal professionals and can lead to reduced productivity and an increased risk of judicial errors. In order to better defend judicial justice and protect public safety, the auxiliary sentencing method based on artificial intelligence and data mining technology needs to be applied to judicial practice. [0003] The task of auxiliary se...

AI Technical Summary

Problems solved by technology

[0005] The present invention overcomes the deficiencies of the above-mentioned prior art, and provides an auxiliary sentencing method for theft based on deep learning. According to the relevant regulations in the "Criminal Law", 11-dimensional features are defined for the theft case, and the text preprocessing of the first-instance verdict for theft is provided. , to filter the text parts and irrelevant words that interfere with the feature extraction, use the cyclic neural network to build generators for case features of different dimensions, use linear regression and multi-layer neural network models for training, and calculate according to the input case features in months The prediction results of the unit’s sentence effectively solve the technical problem that the semantic representation of the case and the accurate prediction of the sentence of a specific crime cannot be realized in the case of less manual labeling

Images