Research Of Online P2P Lending Behavior Based On Text Mining

Online P2P lending is the important constitution of Internet finance in China,which has fully absorbed the long tail flow of bank deposits and loans.Compared to traditional finance,the Internet financial develops rapidly and has strong penetrability,but because of its late start,there are still a lot of big or small problems in the industry.For example,P2P platforms often occurr bankruptcy,borrowers overdue default,lenders lack investment,borrowers failed to raise money and so on.These problems not only cause losses to the interests of the borrowers and investors,but also are not conducive to the development of the entire P2P industryand the continued prosperity of Internet Finance.This article will focus on the difficultly of funding and investment on P2P lending,trying to explore the influence of the factors affecting the loan results and repayment results.Eventually,this article will build machine learning model to predict borrowing and lending results,which has a certain theoretical and practical significance for the development of online P2P lending industry.On the lending market,after a borrower issue a loan request,the most concerned is the success of financing;after lending funds,investors care most about the timely recovery of principal and obtain income.Existing research has proved that the borrower's hard information will significantly affect the loan results and repayment results,but whether the text soft information has impact on it remains to be explored.On the basis of previous studies,this paper will explore the factors that affect loan and repayment results from the perspective of loan texts.By using the topic model,the borrower's borrowing purpose is summed up from the loan title,and the borrowing urgency is calculated quantitatively from the loan description by using the emotional analysis method.,this paper will apply learning and semi-supervised learning algorithm and add text information as input features,which will establish a richer and more comprehensive model to predict loan and repayment results.As a result,the borrower can predict the probability of successfully raising funds before issuing a loan requirement.Investors can also be informed of the possibility of default by borrowers before lending money.For the P2P lending platform,on the one hand this research will improve the volume of business,on the other hand,it will reduce the default rate of repayment.It plays a certain role to ensure healthy and permanent development of online P2P lending platform.The innovations of this paper are as follows:1.This paper applies the text mining method to the empirical research of online P2P lending behavior,and explores the influencing factors of online P2P lending behavior from a new point of view.Previous studies on the influencing factors of online P2P lending behavior mainly explore the impact of borrowers' personal basic information,financial information and historical loan information on lending results.While fewer literatures focus on borrowing texts.2.This paper uses the emotional analysis method to analyze the borrower's borrowing urgency degree from the loan description.Research on online P2P lending behavior from the perspective of text literature is relatively less,and most of them analyzed the linguistic expressions of text,such as grammatical errors,typos or tone etc.This paper uses the emotional analysis method to define the loan urgency degree variable for the first time,and quantitatively analyze the impact of the loan urgency on the online P2P lending behavior.3.This paper adds text information prediction variables,and constructs the ensemble learning supervised classifier to predict borrowing results Most of the traditional credit prediction models use financial information,demographic information,history of borrowing and other hard information as predictive variables..While there are many cases of unsuccessful borrowing in the field of credit,in order to make better use of these text soft information information which refuse users,this paper utilizes semi-supervised algorithm to construct a semi-supervised classifier to predict the results of repayment,whichis better than traditional supervised models.