Research And Implementation Of Event Entity Extraction In Intelligent Legal System Based On Machine Learning

With the continuous development of artificial intelligence technology,new technologies have brought more possibilities for the upgrading of all walks of life.Among them,"intelligent justice" is the focus of in-depth research by major institutions.In judicial practice,large-scale judicial data,recorded information and judicial documents provide empirical guidance for the study of historical cases,the analysis of criminal situations and the adjudication of case results.How to efficiently and accurately use these data to help legal professionals in the judicial field read and analyze texts has become a key problem that needs to be solved at present.Based on actual needs,this paper carries out relevant research on extraction methods of legal text event entities,and the main work and innovation points are as follows:(1)A data set containing seven legal event entities was proposed and constructed.After analyzing and sorting out the judgment documents of a large number of cases,seven key entities of representative persons,time,place,organization,amount,injury and guilt are summarized in judicial documents.Accurate understanding of these representative entity elements in advance is conducive to improving the work efficiency of each link of cases.Due to the small number of existing data sets in the judicial field and the lack of substantive pertinence,this paper marked the 2018 and 2019 CAIL Legal Research Cup and some online public judgments,and combined with the corpus of the first half of 1998 released by the Institute of Computational Linguistics of Peking University as the data set.(2)An extraction method based on the feature vectorization combined model of word embedding is proposed.Legal matters for entity extraction process of the complexity of the entity and the characteristics of the context,a lot of fusion based on word embedded feature vectorization and two-way network combination way of extracting both short-term and long-term memory,and according to the result of the problem to increase the state transition matrix layer optimization output as a result,the accuracy of the model is verified by experiments.(3)An extraction method based on the combination model of dynamic representation word vector is proposed.By analyzing the limitations of the feature vectorization combination model for complex entities such as people and places,this paper proposes a combination model based on dynamic representation word vector,which dynamically adjusts the actual word vector according to different context information and improves the accuracy of event entity recognition.(4)A model extraction system based on dynamic representation word vector combination is implemented.Based on the analysis of the functional requirements of the legal event entity extraction system,the overall architecture and each functional module of the system are designed,and the system functions are demonstrated by combining the actual entity extraction examples.This paper adopted BERT(Bidirectional Encoder Representations from Transformers)vector model to obtain the dynamic characterization of word,the combination BERT and intelligent direction of legal study is less,in view of the traditional training model,its main characteristic is that can solve the problem of polysemy in different language environment,but also can obtain the characteristics of the long distance between words,so as to better understand the complex text meaning.The combination model based on dynamic representation word vector Bert-Bilstm-CRF adopted in this paper achieved a high accuracy,and the final FB1 also improved to the average value of 94.95%.