Análise de posts maliciosos na Dark Web usando aprendizado de máquina não supervisionado

Abstract

This work presents an analysis of malicious posts extracted from Dark Web forums using unsupervised learning techniques, aiming to identify the predominant themes associated with cyber threats. A methodology was employed based on clustering algorithms, such as K-means, DBSCAN, and KNN, in addition to applying Latent Dirichlet Allocation (LDA) to identify latent thematic patterns. The results demonstrated that the K-means algorithm excelled in structuring the data into three main clusters, identifying predominant themes such as data security, search for sensitive information, and hacking communities. This approach enabled the labeling and interpretation of content based on observed patterns, contributing to the understanding of cybercriminals’ tactics and intentions. For future work, it is suggested to expand the dataset to include environments such as the Deep Web, Surface Web, and social networks, as well as to incorporate advanced deep learning algorithms and real-time monitoring tools, aiming for continuous improvement in threat detection and categorization.