Research On Entity Recognition And Entity Relation Extraction Of Internet Fraud Cases Based On Semi-supervised Learning Analysis

At present,network fraud not only has a wide variety of fraudulent means to update and update iteratively,but also has a wide range of fraudulent areas throughout the country,resulting in huge property losses.In order to solve cases quickly and protect people’s property,it is necessary to analyze fraud methods from a large number of fraud cases,find out their characteristics and take precautions against them.Faced with the huge amount of data,it is not enough to rely on manual analysis of human resources,which needs to be realized by means of existing technology and methods.This paper uses the method of semi-supervised learning and neural network to extract and analyze the relationship between entities in network fraud.This paper consists of the following aspects:Firstly,there is no standard,authoritative and wide-ranging network fraud corpus available on the Internet.This project crawls data from Hungqian and other news reporting websites,builds data1 and data2 datasets,and manually annotates some cases by BIO annotation to build the corpus of this paper.Secondly,on the basis of Bi-LSTM+CRF entity recognition model,in order to overcome the problem of small corpus,semi-supervised recognition model(SRM)is constructed by introducing the idea of semi-supervised learning.Thirdly,the Semi-supervised Recognition Model(SRM)is modified by using the joint annotation strategy.By replacing the original CRF layer with the software Max layer and adding location tags,a new relationship extraction model,namely the semi-supervised relationship model-T with location tags,is constructed to extract the relationship.And through comparative experimental analysis,the method proposed in this paper has achieved good results in entity recognition and entity relationship extraction of network fraud cases.