Group Password Strength Meter Based on Attention Mechanism

Abstract

User authentication is an important means to ensure the security of users' cyber accounts. Although there are various authentication means such as irises and fingerprints, passwords are still the main authentication method for the foreseeable future due to their low cost and easy implementation. Password strength evaluation is to measure the security strength of passwords, which has been widely studied. However, we found that the current password strength evaluation methods ignore the characteristics from password creators and do not consider the impact of regional groups on password generation. In this paper, we propose the concept of group passwords to analyze the password characteristics of different groups. Based on this notion, a group-based password strength evaluation method is proposed. In addition, we analyze the vulnerabilities of largescale real-world password groups leaked in previous security incidents. The analysis results show that different password groups have different characteristics. Then, we use the attention mechanism (AM) in the neural network model to learn the dependence between group characteristics and password context features. A long short-term memory (LSTM) model with natural advantages in processing timing features is used to process the password to achieve a more accurate password strength evaluation. We demonstrate the effectiveness of groupbased password evaluation using the real-world password data sets.