A Password Cracking Method Based On Structure Partition and BiLSTM Recurrent Neural Network
Mengli Zhang, Qihui Zhang, Xuexian Hu, Wenfen Liu · 2018
Identity authentication is an important line of defense for network security, and passwords are still the mainstream of identity authentication. Password attacking is an important means of password security research. Probabilistic context-free grammar (PCFG) is the most effective password structure partitioning method at present. The string generation method based on neural network has powerful generalization ability. They effectively characterize the passwords on the substructure level and the character level respectively. In this paper, based on the merits of the above two models, we propose a password attacking method based on structure partition and bidirectional long short-term memory (BiLSTM) recurrent neural network, which is denoted as SPRNN model. Firstly, passwords are divided into abstract substructures. Then substrings of characters, digits and symbols in substructures are generated by using BiLSTM model to take account of the accuracy and generalization ability of the model. Finally, the method is verified by experiment on six real Chinese and English password datasets. The results show that in the context of a fixed number of guessing trials, the SPRNN model breaks the password 25% -30% more than Narayanan's method, about 10% than Weir et al.'s method password between the cross datasets.