This paper proposes an unsupervised model for identifying and characterizing clusters of discussion forums on the Dark Web, which are often used for trading confidential information and illicit products. Unlike existing methods that rely on continuously labeled data, the proposed approach combines clustering algorithms with decision tree algorithms to provide an explainable characterization of Dark Web forums. Evaluated with real Dark Web data, the model achieved 98% accuracy and an F1 score of 98%, demonstrating its effectiveness for cyber threat intelligence and law enforcement purposes in detecting data breaches and illicit activities.