The study is to explore the risks in the Internet finance and the factors affecting users' behavior under the background of big data. First, the risks of the Internet finance under the background of big data and the existing risk control modes are analyzed. Then, based on BP neural network (BPNN), an Internet financial fraud identification model is constructed, and corresponding touch rules are made. Its prediction performance is quantitatively compared with that of support vector machine and random forest algorithm. Finally, based on the structural equation model, the influence path of perceived security control on the Internet financial behavior is explored. The results show that, the applicants whose unit addresses are on blacklist have the highest touch fraud rate (14.16%). The precision rate (88.14%), accuracy rate (96.37%), recall rate (70.96%), and F-Score value (16.36) of the financial fraud identification model based on BPNN are the highest versus the other two algorithms, and the error detection rate (7.19%) is the lowest. The perceived security, identity authentication, non-repudiation of transactions, privacy protection, and control strength of data integrity positively affect users’ trust, which further positively affects the attitude and intention of using the Internet finance, and the intention eventually affects users’ behavior. Finally, some suggestions are put forward to improve the supervision of the Internet finance in China. To sum up, the Internet financial fraud identification model based on BPNN demonstrates satisfying performance and is worth of promotion. Additionally, the authentication technology, non-repudiation of transactions, privacy protection, data integrity, and users' sense of trust of the Internet finance have a significant impact on users' behavior.

本研究旨在探讨大数据背景下互联网金融的风险及影响用户行为的因素。首先，分析了大数据背景下互联网金融的风险和现有的风控模式。然后，基于BP神经网络（BPNN）构建互联网金融欺诈识别模型，并制定相应的接触规则。其预测性能与支持向量机和随机森林算法进行定量比较。最后，基于结构方程模型，探索感知安全控制对互联网金融行为的影响路径。结果表明，单位地址在黑名单上的申请人的触摸欺诈率最高（14.16%）。准确率（88.14%），准确率（96.37%），与其他两种算法相比，基于BPNN的金融欺诈识别模型的召回率（70.96%）和F-Score值（16.36）最高，错误检测率（7.19%）最低。感知安全性、身份认证、交易不可否认性、隐私保护、数据完整性控制强度正向影响用户信任度，进而正向影响用户使用互联网金融的态度和意愿，最终影响用户行为. 最后，提出了完善我国互联网金融监管的一些建议。综上所述，基于BPNN的互联网金融欺诈识别模型表现令人满意，值得推广。此外，身份验证技术、交易不可否认性、