Password Dictionary Generating With Long Short-term Memory

The importance of password is questioned.We use password to protect our information security.At the same time,users' passwords are cracked and leaked.In the study of password security,attack on victim's password is an important research issue of computer security.Studying computer password is of great value for investigating network crime,strengthening financial management and ensuring information security.Dictionary attack is the main way of password attack.Although some researches have been done in password dictionary,improving dictionary guessing efficiency is always an important task in information security research.At present,the neural network model based on the text analysis mechanism has achieved outstanding results in Natural Language Processing.Some researchers began to study the cryptographic characteristics from this perspective.However,the selection and training of neural network models often spend a lot of time,and there is not enough research on the generation of password.As the password feature learning tools using Long Short-Term Memory recurrent neural networks model of deep learning tools and research on model parameters selection,through Long Short-Term Memory recurrent neural networks model with memory and prediction ability,the initial completion of password dictionary generating and password guessing task.In previous studies,the guessing efficiency of a password dictionary depends on the size of the dictionary and the content contained in the dictionary.The training efficiency of the Long Short-Term Memory recurrent neural networks model is also influenced by the amount of input data and the tools.In the training model,firstly,the removal of the low probability,focus on training the password,and then using the Keras programming based on Python that features of deep learning framework is easy to use,the use of relatively high efficiency CNTK deep learning tool to train model,by Dropout regularization and Nadam optimization algorithm to obtain the best training effect,the neural network model after training what takes only 30 minutes to generate 3235 passwords.Experimental results show that the proposed recognition method is efficient.In cryptographic dictionary guess matching,the passwords that Long Short-Term Memory recurrent neural networks generated,only 53 exist in training password set.4%-9%'s password can be matched to the corresponding real password in the password that the website leaks.In the test contrast experiment with other dictionary dictionaries,the password dictionary generated by the long and short term memory neural network is more than 4 times more than the other dictionaries.