An Unsupervised Model for Identifying and Characterizing Dark Web Forums

Abstract

Dark Web forums are significantly exploited to trade confidential information and illicit products by criminals. This paper addresses the problem of how to identify the cluster of discussion forums and their characteristics on the Dark Web. Exiting methods are mostly dependent on the continuous labeled contents, which are expensive and not feasible due to the nature of Dark Web data. Therefore, an approach that does not need a continuous availability of labeled forum and related knowledge is required. To this end, we propose an unsupervised model to identify and characterize Dark Web forums by combining clustering algorithm and decision tree algorithm. The proposed method presents the characteristics in an explainable form that can be used by the cyber threat intelligence system and law enforcement as scientific evidence to analyze any data breach or illicit activities in the Dark Web forums. To evaluate the performance of our model comprehensive experiments were conducted using real Dark Web forum data. The proposed approach achieves 98% accuracy and F1 score of 98% validating the efficacy of our proposed model to successfully characterize Dark Web forums. The experimental results suggest that the proposed model could be useful to the cyber threat intelligence and law enforcement community for building an intelligent source of knowledge that can be used for detecting data breach and illicit activities happening in the Dark Web forums.

Dark Web Financial Fraud Identification Using Mathematical Models in Healthcare Domain

“…At the end of cycle, the inventory completed a most elevated quantity of deficiency S and once that brand new request is placed to zero-in the excess [16] [17]. The adjustment in inventory degree I (t) as for point eventually is as given: d/dt I = − a + bt , 0 ≤ t ≤ μ h (22) d/dt I + θ t I t = − a + bt , μ h ≤ t ≤ t (23) In this section, the objective is to create probably the most excellent estimations of T as well as t1 that limit Z (T, t1). For a foreordained estimation of T, taking second and ist request subsidiaries of Z (T, t1) regarding t1,…”

“…A linear regression coefficient over this time period [21] shows how the company's approach has changed. Next, hospitals will be grouped hierarchically [22]. First, second, and third clusters contain medical facilities in order (5,6,7,9,10,11).…”

Dark Side of the Web: Dark Web Classification Based on TextCNN and Topic Modeling Weight

“…Additional approaches include employing term extraction methods such as TF-IDF and bag of words (BOW), where classification was conducted using naive Bayes (NB), support vector machines (SVM), and linear regression (LR) [1,17]. Another research focused on classifying marketplace attributes via group clustering and decision trees (DT) [18]. There have been endeavors to detect the Dark Web using edge computing and feature weighting [19], and to generate software quality metrics for classifying Dark Web characteristics [20].…”

“…Additional research classified Dark Web forums by integrating TF-IDF, random projection, and PCA [42], extracting features according to class, and applying machine learning [43]. This study also employed TF-IDF for feature extraction and executed classification using Kmeans and DT [18]. Numerous analyses of the Dark Web have been conducted using word2vec [44], and there is active research on transforming extracted text features into matrices via embedding and implementing deep learning [45,46].…”

“…Numerous analyses of the Dark Web have been conducted using word2vec [44], and there is active research on transforming extracted text features into matrices via embedding and implementing deep learning [45,46]. In the Dark Web context, text classification is predominantly utilized for preprocessing and feature extraction [16][17][18][40][41][42][43][44].…”