信息网络安全 ›› 2017, Vol. 17 ›› Issue (9): 143-146.doi: 10.3969/j.issn.1671-1122.2017.09.033

• • 上一篇    下一篇

基于梯度提升模型的行为式验证码人机识别

欧阳志友1,2(), 孙孝魁1,2   

  1. 1.南京邮电大学先进技术研究院,江苏南京 210023
    2.南京邮电大学自动化学院,江苏南京210023
  • 收稿日期:2017-08-01 出版日期:2017-09-20 发布日期:2020-05-12
  • 作者简介:

    作者简介: 欧阳志友(1982—),男,湖南,实验师,博士研究生,主要研究方向为机器学习、电力大数据分析;孙孝魁(1991—),男,河南,硕士,主要研究方向为机器学习和电力负荷预测与分析。

  • 基金资助:
    国家自然科学基金重点项目[61533010];南京邮电大学实验室工作研究课题重点项目[2014XSG03]

Human-machine Behavior Recognition for CAPTCHA Based on Gradient Boosting Model

Zhiyou OUYANG1,2(), Xiaokui SUN1,2   

  1. 1. Institute of Advanced Technology, Nanjing University of Posts and Telecommunications, Nanjing Jiangsu 210023, China
    2. School of Automation, Nanjing University of Posts and Telecommunications, Nanjing Jiangsu 210023, China
  • Received:2017-08-01 Online:2017-09-20 Published:2020-05-12

摘要:

通过使用非正常手段模拟人类操作行为,绕过验证码系统,黑客工具就可以向系统后台发起批量请求,实现对系统的攻击,从而给系统的正常运行带来很大的风险,轻则影响系统运行,重则产生巨大的经济损失。而传统的验证码方法,在易用性和人机识别率方面都存在不足,行为式验证码应运而生。文章提出了一种基于行为式验证码的行为轨迹信息来构建特征工程,并运用梯度提升模型来进行人机行为识别的方法,在10万真实的行为轨迹样本上可以获得90%以上的识别准确率。

关键词: 梯度提升, 验证码, 机器学习, 人机识别

Abstract:

By using the abnormal means to simulate human behavior operation and bypass the CAPTCHA system, hacking tools can then sent a large batch of requests to the background system to achieved the hacking goals, which may bring to big risk of delay response of system operation, or even produce huge economic losses. However, the traditional verification code method has shortcomings in both ease of use and man-machine recognition rate. In this paper, a new behavior trajectory of the CAPTCHA system based feature engineering, with utilizes the gradient boosting models, for human-machine behavior recognition is proposed. Performance in 100000 samples of real CAPTCHA system can obtain a more than 90% recognition accuracy.

Key words: gradient boosting, CAPTCHA, machine learning, human-machine recognition

中图分类号: