基于中文自由文本击键特征的自动欺骗检测模型

Automatic deception detection model based on keystroke features of Chinese free text

摘要:

研究表明欺骗行为在一定程度上会影响用户击键模式的变化。在互联网社交应用领域，通过击键特征对欺骗行为的检测对网络信息安全建设具有重要意义。然而，现有的欺骗行为检测模型侵入性强，实时性差等问题，限制了其在互联网社交应用领域的应用。针对以上问题，本研究设计了一个实验从短文本中收集了广泛的用户击键特征（单键特征、内容特征、双键特征），分别采用遗传算法（GA）和支撑向量机（SVM）完成特征选择和模型建立，开发出一个用以预测用户欺骗行为的模型（GA-SVM）。研究结果表明：该模型能够有效地检测出用户的欺骗行为，获得82.86%的分类准确率；三类击键特征对欺骗行为的检测都有贡献。此外，欺骗者认知负荷和心理压力对击键模式影响也被探讨。

Abstract:

Research has found that human’s deceptive behaviors would affect their keystroke patterns. Detecting deceptive behaviors through keystroke patterns is a critical step toward building a cyber information security system in the field of social networking. However, the existing models detecting deceptive behaviors still suffered from the problems of high invasion and low real-time performance. To solve the problems, we first designed an experiment to collect a wide range of stroke features (i.e., single-key features, content features and double-key features) from users’ typing process of short text and then developed a predictive model to detect the deceptive behaviors by using Genetic Algorithms (GAs) and Support Vector Machines (SVMs) as feature selection and model building methods, respectively. The results showed that the developed model could effectively detect the deceptive behaviors with accuracy of 82.86%; all the three categories of keystroke features had contributions to detecting deceptive behaviors. In addition, the effects of cognitive workload and pressure on keystroke pattern of deceivers had also been explored.