互联网金融作为传统金融的重要补充，为金融欧亿·体育（中国）有限公司增添新的活力。然而，由于经济 交流中的信息不对称和个人信用报告的不完整等问题，互联网金融信用风险一直备受金 融机构和国家监管部门的关注。P2P 网贷作为互联网金融中的小额个人借贷平台，不仅 惠及了传统金融机构难以惠及的小额资金需求者，而且有效调动了个人闲散资金，提高 了金融市场的公民参与度，进一步活化了金融市常然而，由于 P2P 网贷欧亿·体育（中国）有限公司前期准入 门槛低、相关法律法规不完善以及平台方风控技术不足等问题，中国 P2P 欧亿·体育（中国）有限公司经历了从 繁荣到低迷的一系列时期。近几年，随着一系列互联网金融相关政策的施行和互联网技 术的改进，P2P 网贷的欧亿·体育（中国）有限公司风险和平台风险都被有效地降低，以小额借贷为主的中国 P2P 网贷欧亿·体育（中国）有限公司的未来也呈现大好趋势。 根据动态信用风险理论，未来中国 P2P 网贷的风险将主要集中在借款人的信用风险 方面。因此，认识 P2P 网贷与传统机构之间的差异，挖掘可以评估 P2P 网贷借款人信用 风险的变量，建立可以有效识别 P2P 网贷借款人信用风险的模型，将成为 P2P 网贷平台 和相关金融机构的主要关注方向。与传统线下借贷不同的是，P2P 网贷平台上的借款人 地域跨度较大，借款人“硬信息”的真实性难以验证，单纯依靠“硬信息”建立的信用 风险识别系统可能难以反映借款人真实的信用风险情况。鉴于此，本文从评估中国 P2P 网贷平台借款人信用风险出发，以丰富借款人信用风险评价指标和建立优越的借款人信 用风险评价模型为主要目的，以挖掘借款人“软信息”中的信用风险识别因子为主要方 向，以预测准确度更高的机器学习方法为评估路径，在控制重要“硬信息”的基础上， 探索“软信息”对借款人信用风险的识别效果，构建基于借款人“软信息”的 BP 神经 网络信用风险评价模型。 本文利用数据挖掘方法获取了中国运转良好的 P2P 网贷平台——人人贷平台上的 借款人相关数据。实证部分首先使用二元 Logistic 回归对借款人的信息指标进行初步筛 选，然后使用合适的实证样本建立借款人 BP 神经网络信用风险评估模型，用测试样本 检验模型的评估效果，并比较 BP 神经网络信用风险评价模型与 Logistic 回归模型的信 用风险预测效果。结果显示，借款人的声誉资产和个人陈述文本等“软信息”中包含有 大量的信用风险识别因子，如成功借款次数、描述文本长度、文本情感倾向等都可以从 一定程度反映借款人的信用风险状况，在传统的“硬信息”指标中加入合适的软信息可 以有效提升信用风险模型的预测准确率。本研究还证明了 BP 神经网络模型在信用风险 评估方面的优越性。 关键词：P2P 网贷; 信用风险; 软信息; BP 神经网络; Logistic 回归ii Abstract As an important supplement to traditional finance, Internet finance adds new vitality to the financial industry. However, due to information asymmetry in economic communication and incomplete personal credit report, Internet financial credit risk has always been the focus of financial institutions and national regulatory authorities. As a small amount of personal loan platform in Internet finance, P2P lending not only benefits the small amount of money demanders who are hard to benefit from traditional financial institutions, but also further mobilizes the individual idle funds, improves the participation of citizens in the financial market, and further activates the financial market. Because of the problems of low access threshold, imperfect laws and regulations and insufficient risk control technology of platform side, P2P industry in China has experienced a series of periods from prosperity to downturn. With the implementation of a series of Internet finance related policies and the improvement of Internet technology, the industry risk and platform risk of P2P online lending have been effectively reduced, and the future of China's P2P online lending industry, which mainly focuses on small loans, also presents a good trend. According to the theory of dynamic credit risk, the future risk of P2P network loan in China will focus on the borrower's credit risk. Therefore, recognizing the difference between P2P network loan and traditional commercial banks, mining the variables that can identify the credit risk of P2P network loan borrowers, and establishing a model that can effectively identify the credit risk of P2P network loan borrowers will become the main focus of P2P network loan platform and related financial institutions. Different from the traditional offline lending, the borrower's regional span on P2P online lending platform is large, and the authenticity of the borrower's "hard information" is difficult to verify. The credit risk identification system relying on "hard information" may be difficult to get the real credit risk of the borrower. In view of this, this paper starts from the evaluation of the credit risk of the borrowers on the P2P network loan platform in China, with the main purpose of enriching the evaluation indexes of the borrowers' credit risk and establishing a superior evaluation model of the borrowers' credit risk, with the main direction of mining the credit risk identification factors in the "soft information" of the borrowers, and with the machine learning method with higher prediction accuracy as the evaluation path, On the basis of controlling important "hard information", this paper explores the recognition effect of "soft information" on the borrower'siii credit risk, and constructs a BP neural network credit risk evaluation model considering the borrower's "soft information". In this paper, we used data mining method to obtain the data of borrowers on Renren dai, a typical P2P platform in China. In the empirical part, first of all, we used binary logistic regression to screen the borrower's information indicators, then used appropriate empirical sample to establish the borrower's BP neural network credit risk assessment model, tested the assessment effect of the model with test samples, and compare the prediction effect of BP neural network credit risk assessment model and logistic regression model. The results show that there are a large number of credit risk identification factors in the borrower's reputation assets and personal statement texts, such as the number of successful loans, the length of description texts, the emotional tendency of texts, etc., which can reflect the borrower's credit risk to a certain extent. And adding appropriate soft information into the traditional "hard information" index can effectively improve the prediction accuracy of credit risk model. This study also proves the superiority of BP neural network model in credit risk assessment. Keywords: P2P Lending; Credit risk; Soft information; BP neural network; Logistic regression湖南科技大学硕士学位论文 目 录 摘要............................................................................................................................................i Abstract.......................................................................................................................................ii 第 1 章 绪论...............................................................................................................................1 1.1 研究背景与意义...........................................................................................................1 1.1.1 研究背景............................................................................................................. 1 1.1.2 研究意义............................................................................................................. 2 1.2 文献综述.......................................................................................................................3 1.2.1 P2P 网贷平台发展研究.......................................................................................3 1.2.2 P2P 网贷平台风险研究.......................................................................................4 1.2.3 借款人信用风险评估指标研究......................................................................... 4 1.2.4 借款人信用风险评估方法研究......................................................................... 6 1.2.5 文献评述............................................................................................................. 7 1.3 研究基本框架.......................................................................................................