Research On Similarity Of Judgment Documents Based On Siamese Neural Network

At present,the writing of judgement documents is becoming more and more standardized,and with the popularization of electronic judgement documents,judgement documents have become an important object of justice studies.Based on siamese neural network and combined with text representation,word embedding,pre-trained language model,text feature extraction and other technologies,this paper studies the similarity of judgement documents to provide reference for judicial cases and realize the demand of "same case and same judgment" in the judicial field.The main work of this paper is as follows:Aiming at the domain,preciseness and dynamic change of judgement documents,a text representation model of judicial documents based on TinyBERT-CNN is proposed.The word embedding layer based on TinyBERT solves the problem that the context of judgment documents affects the meaning of words;the feature extraction layer based on CNN extracts the semantic information of the judgment document,simplifies the text representation results,and improves the processing efficiency of subsequent tasksBased on the characteristics of the task in this paper,a model for judging the similarity of judgment documents based on siamese neural network is proposed.The model maps two judgments into the same domain,enhances the connection between the judgments to be matched,improves the accuracy of the matching results,and solves the problem of too many classifications and complex judgment caused by the similarity judgment after the similarity is directly calculated.Based on the text representation model of judgement documents,the siamese neural network is applied to the feature extraction layer,and the judgement documents data provided by CAIL2019 are used for the experiment.By applying ELMO,TinyBERT,GRU,Transformer and CNN to different layers of the model,it is concluded that the TinyBERT-based siamese convolutional neural network model has a better effect in judging the similarity of judging documents;by comparing the siamese neural network model based on TinyBERT with the existing model,the effectiveness of the proposed model is verified.