Abstract: In recent years, broad and profound research has been launched on plain passwords cracking and recovery. In general, mainstream approaches have greatly improved the success rate of password cracking and recovery, but there is a lack of comparison and internal correlation between different methods. This paper focuses on the generation of plaintext cipher dictionaries, summarizing three main kinds of competitive models for the past decade：Markov model, Probability Context-Free Grammar（PCFG） and Neural Network（NN） model. In order to evaluate these three mainstream models effectively and objectively, the paper designs and conducts a series of performance verification experiments, explains their occasions, advantages, disadvantages and reasons from the principle level, proposes some ideas to optimize and improve. In the end, it thinks that Neural Network will become the most promising model in the future, and points out that further research and development in this field depend on the establishment of normative data set and the combination of various methods.

Key words: plain password, password generation, Markov model, Probability Context-Free Grammar（PCFG）, Neural Network（NN）

摘要： 近些年来，针对明文口令的破解和恢复等技术展开了广泛而深刻的研究，总的来说，该领域的主流方法已经大大提升了口令破解和恢复的成功率，但多种方法之间缺少比较和内在关联的分析。着眼于明文口令字典的生成问题，总结了近十年来主流的明文口令生成模型：马尔可夫模型、概率上下文无关模型和神经网络模型。为了有效并客观地评估这三种主流模型，设计并进行了一系列性能验证实验，给出了三种模型各自适用的场合及优缺点，并从原理层面解释了三种模型存在缺陷的原因，给出了一些优化改进的思路。最后，认为神经网络将会成为未来最具潜力的模型，并指出该领域的进一步研究和发展依赖于规范数据集的建立和多种方法的结合运用。

关键词: 明文口令, 口令生成, 马尔可夫模型, 概率上下文无关语法, 神经网络