Sequential-Spatial-Aware Attribute Classification of Ultra-Short Password

The composition of passwords varies considerably depending on the configuration strategy employed and the individual user's preferences. To this end, we try to analyse the characteristics of the text that constitutes the password through the process of text classification. The issue of password attribute classification is particularly challenging due to its inherent lexical polysemy, the scarcity of text features, the lack of context and the difficulty in explicitly identifying semantics. To address the aforementioned issues, we propose a sequential-spatial-aware attribute classification method. Our method firstly adopts BiLSTM for sequence modeling of password text, and then further employs convolutional operation for spatial local modeling on the basis of sequence modeling, which is ultimately concatenated into the unified representation and then inputted into the classifier. The validation is conducted on seven authentic password datasets, including 17173, CSDN, Rockyou, and so on. Our method demonstrated a substantial improvement over the baseline method with single-dimension modelling, as well as a notable enhancement in four metrics: accuracy, precision, recall and F1 score.