Research On Quality Inspection Technology For Court Judgement Documents

With the gradual implementation of the online disclosure policy for court judgement documents.The quality of judgement documents has became an important issue for judicial workers.Traditionaly,in order to guarantee the quality of the overt judicial documents,judicial workers often need to manually proofread the judicial documents.However,in the background of such few workers and such amount of cases,the method of manual review is very inefficient.With the progress of the national judicial informatization construction,some scholars have carried out some researches on the quality inspection of the judgement documents.But these previous researches often concentrate on the universal words proofreading technology and so on.And without deep understanding of the judicial practice,they haven't studied on the criterion inspection and reasoning quality inspection which are most concerned in the judicial practice.In this paper,in view of the shortage of field research,19 testing subjects are studied in three aspects: the division of text structure,the criterion of text structure,the criterion of sentence structure and the criterion of the reference of the law.At the same time,for more in-depth judgment documents reasoning quality detection,firstly we quantificationally define the inspection for reasoning comprehensiveness and the inspection for accuracy of reasoning legitimate basis.Then we propose the technology of applying XGBoost algorithm to solve the problem of inspection for reasoning comperhensiveness.And we use the deep-learning model Resnet to model and solve the problem of inspection for accuracy of reasoning legitimate basis.And this paper tests and validates the effectiveness of the proposed judicial documents criterion inspection and reasoning quality inspection technology through a number of comparison experiments.At last,more than 85% of the F1 score has been obtained on multiple standard criterion inspection subjects.Also a 88.26% F1 score is reached on the problem of judgement respond to the claim in the reasoning comprehensiveness inspection.And 81.41% F1 score has been reached in the more complex justification and referee respond to the claim problem whose dataset is more lopsided.The results of training and testing on the 280 thousand civil referee data sets have reached a macro average F1 value of 83.3% and a micro mean F1 of 82.5% in the problem of inspection for accuracy of reasoning legitimate basis.Finally,this paper uses the PyQt5 framework to design and implement the quality inspection system for referee documents.And we use Django to design the Web system for the quality inspection of the court judgement documents based on the B\S architecture,and integrates the research technology organically.