Design And Implementation Of Opinion Risk Assessment System For Judicial Cases Based On Semantic Similarity

With the popularity of Internet applications,as a new channel for the public to express judicial opinion and a new way to participate in the judicial process,the Internet has an important impact on judicial activities and judicial disclosure.While online public opinion plays a role in strengthening judicial supervision,judicial openness will also face the risk of online public opinion.It is a practical problem to be solved that how to effectively assess the risk of public opinion in judicial cases.The traditional risk assessment of public opinion in judicial cases is mostly carried out manually.However,due to the large amount of data to be processed and the lack of uniform criteria for judgement,this method always has the problems of high human costs and low efficiency.In order to addresses this issue,this thesis proposes a method of public opinion risk assessment based on semantic similarity and designs and implements the corresponding system by applying text mining techniques to the assessment of public opinion risk in judicial cases.The work in this paper was supported by the state key research and development program project "Research on the Collaborative Support Technology of Internal and External Consistent Trial Execution and Litigation Services"(2018YFC0831300).The main work of the paper is as follows:(1)A computational model of sentence semantic similarity based on Siamese LSTM is proposed.To solve the problem that traditional sentence semantic similarity computation requires manual extraction of sentence features by constructing feature engineering,the paper proposes a Siamese LSTM-based sentence semantic similarity computation model based on the introduction of deep learning model.The model is constructed by a Siamese network,and the semantic features of the sentences are obtained by using a double-stacked Bi-LSTM with a combined attention mechanism to enhance the validity of the model.(2)The performance of the proposed model was experimentally verified.The accuracy of the model can reach 85.9% with the F1 value reaching 87.1 in the large-scale test set of Chinese sentence pair dataset,which is better than CBOW,CNN,Bi-MPM and other models.In addition,the paper sets up three sets of comparative experiments on the infrastructure and key techniques used in the model.The results of experiments show that using Siamese network,setting a reasonable number of Bi-LSTM stacking layers and introducing attention mechanism can effectivelyimprove the accuracy of the model.(3)A system of risk assessment of public opinion in judicial cases is designed and implemented.Using Flask and the bootstrap framework,a semantic similarity-based public opinion risk assessment system for judicial cases is designed and implemented to provide visual operations throughout the entire process,from data collection to risk assessment.The system makes it possible to conduct an opinion risk assessment of current judicial case and improves the efficiency of the assessment.