Research And Application Of Credit Risk Assessment Based On Multi-source Text

Nowadays,natural language processing technology is changing with each passing day,and its application field is gradually expanding to the currently known research field.Scholars in various research fields are also studying how to use natural language processing technology to make the research results in the current field to a new stage.In the credit risk research of SMEs,the importance of the current development of SMEs is gradually increasing in the national economic status.However,the support for commercial banks and investors for SMEs is not in line with their status.This is due to the inability to obtain small and medium-sized enterprises.The business information of the enterprise and its credit level lead to the failure of commercial banks and investors to correctly assess the credit level of SMEs,resulting in credit risk.By combing the research on credit risk of SMEs in recent years,it is found that research focusing only on financial data will construct an imperfect credit risk assessment system,and most of its methods are traditional machine learning algorithms,using deep neural networks for small and medium-sized enterprises.The literature on corporate credit risk assessment is relatively small.Therefore,this thesis focuses on how to use multisource text data and deep learning methods to construct an effective credit risk assessment system,and expounds the definition and characteristics of SMEs and their credit risks.It also clarifies the concepts and principles of the core technologies related to natural language processing.In-depth analysis of the core processes of natural language processing:task types,data processing,build models,and model evaluation.When researching and analyzing data processing,the data are written in Chinese and English respectively.When building the model for English data,based on the analysis of existing models,this thesis uses convolutional neural network with gate mechanism and innovatively proposed the SGCN model.By collecting and collating the public opinion and reporting information related to SMEs and forming a credit risk data set in a specific format,the data is analyzed layer by layer using Chinese data processing method,and the SMEs appearing in the data are taken as the object of investigation.The data is deeply studied and trained to construct an SGCN model adapted to the credit risk data set,and the feasibility of the model is verified by 10 SMEs.The experimental results show that for the SMEs' public opinion and report information,the SGCN model has an accuracy rate of 87.69% for identifying SME credit risk,which is close to two percentage points higher than the 85.68% accuracy rate of the traditional identification model.SMEs in the industry still have certain generalization capabilities.For the credit risk assessment of unknown industries,SGCN also has an accuracy rate of up to 70%,which has achieved an ideal forecasting effect.Therefore,when commercial banks and investors identify SME credit risk,SGCN model provides a method for accurately and conveniently identifying SME credit risk based on multi-source text,which provides a reference for preventing and controlling credit risk.