Development Of Password Attribute Analysis Tool Based On Neural Network

With the development of Internet, information security is gaining more and more attention. Password, as a most common identification security mechanism, is widely used in various security fields. Users’ habits of password setting are reflected in the passwords and have an effect on password security. By studying the password features, we can analyze and get the potential rules of the users’ password habits, which can then be applied to research of password cracking and password recovery.This paper developed a password attribute analysis tool based on neural network(NN), which was designed for the research of massive password data. When using it, we should first perform attribute assignment for a large number of passwords, in order to obtain attribute data that could reflect password features. Then we could use MATLAB to build up a NN model for different password attributes. By using a proper sample, we could perform training on this model to gain the optimal approximation of the real sample relationship model. After that, a simulation test was needed with a proper testing sample. According to the simulation accuracy of the test result, we could find out the potential attribute relationships or rules that are of worth. Finally, brute force dictionaries reflecting the users’ password habits could be generated using these rules, to provide guidance and help for research of password cracking and password recovery.The experiment showed that, this password attribute analysis tool could effectively analyze the users’ password habits of different application types, that is, the potential relationships between password attributes. Therefore we could use the valuable ones to generate corresponding brute force dictionaries, which could accelerate password cracking and recovery.The innovative point of this analysis method was that it was not the same as the common data mining method, nor did it learn things from statistical analysis of the password data. Instead, it used NN to directly build up a network model simulating the real data relationship. By training with a large number of data samples, it could approach closely to the real data relationship model. Therefore we could get the potential rules and relationships of the data. Besides, another point was the password attribute definition and assignment, presented in this paper for the concretization and digitization of password features.