Construction Of Knowledge Gragh Based On Law

Since Google proposed "Knowledge Gragh"(KG) in 2012,people from all kinds of work have continued to conduct relevant research.Knowledge Gragh has also attracted the attention of researchers in the field of law.However,there are few relevant literature reports on this field.The purpose of this paper is to study the construction of knowledge gragh in the field of law,to obtain legal judgments from the China Judgment Online,and to conduct relevant research with judgments as data.Since there is no open corpus in the legal field,the judgment must first be obtained from the the China Judgment Online.And then various characteristics of the judgment entity,including word features,part-of-speech features,dictionary features,spelling features,suffix features,etc.Finally,we make unified annotation of the judgment.On this basis,we do the following work:1)Named entity recognition is an important part of knowledge extraction.It is one of the key technologies for constructing knowledge gragh.This paper focuses on the study of named entity recognition.Among them,CRF(Conditional random field) is a statistical-based machine learning method and has a wide range of applications in named entity recognition.Taking CRF as the baseline,a feature function set is defined according to the entity characteristics of the judgment.The weight of CRF is trained by CRF++,and the CRF named entity recognition model is obtained for data testing.The defined feature function set can improve the recognition effect of CRF named entity.2)In recent years,due to the development of big data era and processors,deep learning has gradually shown its advantages and achieved good results in named entity recognition.Bi-LSTM-CRF,which based on character-level features for named entity recognition,is consisted by Bi-LSTM(Bi-directional Long Short-Term Memory neural network) and CRF.CRF can use the transfer matrix to further obtain the relationship between each location tag,making Bi-LSTM-CRF retains both contextual information and the influence between the current location and the previous location.By comparing the experiments,the feature information proposed in this paper can effectively improve the recognition effect of CRF named entities,but the comprehensive effect of Bi-LSTM-CRF is better and simpler.The legal knowledge gragh has great value for upper-level application development,and it is helpful for decision analysis and case reasoning.This paper studies the business problems,data acquisition,text preprocessing,knowledge extraction and knowledge storage of legal knowledge gragh.Finally,this paper summarizes the work done and the shortcomings,and explains what to do next.