Based off this paper resulted in increased cracking accuracy by 22.5% from John the ripper’s Markov and incremental model created a tool that wass flexible enough to perform n-gram and markov chains based password generation using a cracked password list. It works because a large number of users use some part of their email or username or any other detail in their password and if we can first check passwords that start with n grams containing usernames portions, user emails portions etc, then we can intuitively speed up the accuracy and the experiments proved this intuition right. Another reason for increased accuracy was using a training word list from the same category as that of website i.e adult/relationship websites. These two factors were the main cause of the increase in accuracy.
source: http://fsecurify.com/using-ordered-markov-chains-and-user-information-to-speed-up-password-cracking/