| After years of development,the informatization of China’s court has made great achievements and electronic judgment document is one of the important achievements.The special analysis of electronic judgment documents has become the new direction of judicial big data.Through the current literature read and research,there are no research to be found about the construction of knowledge graph for the judgment documents in the vertical industry.Therefore,it is of great research significance and practical value to the work of acquiring knowledge and integrating knowledge through the judgment documents.In this paper,we carry out the following work on the text classification for judgment documents:First of all,through the research and analysis of the structure of the judgment documents,we use the word2 vec technology to transform the text information into vector form,which lays a good foundation for the next clustering of the judgment documents and the classification of the judgment documents based on ULMFi T model.Secondly,we summarized the advantages and disadvantages of the three classic K-means,Birch,and DBSCAN clustering algorithms.Then,these three algorithms are used to perform clustering comparison experiments on the information of the subject matter and industry feature words in the referee documents to find a more applicable one.Algorithm for clustering referee documents.Experimental results show that the clustering effect of the Birch algorithm is better than the other two.Last,combined with deep learning and transfer learning,this paper proposes a classification method of judgment documents based on ULMFi T model.Experiments are carried out from three aspects: pre-training universal language model,fine tuning of target task language model and fine tuning of target task classifier.The results show that the fine-tuning method based on language model can effectively solve the classification task of judgment documents.The accuracy of this method is 94.53% and 95.17%. |
PDF Full Text Request