A Novel Collapsed GIBBS Sampling Methodology for Various Categories of Corpus in Collaborative Cybercriminal Network Discovery

摘要

In the recent past, there has been a continuous increase in the amount of cybercrimes that source for considerably financial loss to several organizations. Current investigations expose that cybercriminals are likely to work together or even transact cyber-attack tools through the "dark markets" established in online social media and don't mine syntactic and semantic terms for online social media messages in collaborative cybercrime activities. The foremost role of the work is to solve this problem by means of proposing a weakly supervised cybercriminal network mining called Collapsed Gibbs Sampling method for the extensively utilized Latent Dirichlet Allocation (CGSLDA) model in order to facilitate cybercrime forensics. Here, at the beginning the syntactic and semantically text corpus is extracted from the online social media text documents by means of the shallow-parsed corpus and lexico-syntactic associations. This method results in considerable speedups on real world text corpora. Subsequently, Kernel based Support Vector Machine (KSVM) classification is carried out on the results of CGSLDA in order to categorize the cybercriminal activities of the association labels of messages into cybercriminal and regular activities. CGSLDA is evaluated in accordance with two social media corpora, it reveals that proposed CGSLDA scheme drastically outperforms the Context sensitive Latent Dirichlet Allocation (CSLDA) based scheme and other schemes in terms of Precision, Recall, F-measure, Accuracy and Region Of Convergence (ROC) Curve respectively.