Αναγνώριση επιθέσεων κοινωνικής μηχανικής βασισμένων σε συνομιλίες με χρήση τεχνικών βαθιάς μάθησης και επεξεργασίας φυσικής γλώσσας προς επίγνωση της κατάστασης κυβερνοασφάλειας

Περίληψη

Σε μια εποχή που κυριαρχείται από την ψηφιακή επικοινωνία, η αύξηση των Επιθέσεων Κοινωνικής Μηχανικής βασισμένων σε Συνομιλίες (ΕΚΜΣ) είναι προδιαγεγραμμένη. Αυτές οι επιθέσεις, που χαρακτηρίζονται από την ψυχολογική εκμετάλλευση, και την εξαπάτηση, αποτελούν μια σοβαρή απειλή τόσο για τα άτομα όσο και για τις επιχειρήσεις. Για την αντιμετώπιση αυτής της αυξανόμενης απειλής , η παρούσα διδακτορική διατριβή παρουσιάζει ένα σύστημα αναγνώρισης επιθέσεων ΕΚΜΣ, υπό την αιγίδα του Συστήματος Αναγνώρισης Επιθέσεων Κοινωνικής Μηχανικής βασισμένων σε Συνομιλίες (CSE-ARS). Τα θεμέλια αυτής της έρευνας τίθενται ξεκινώντας μια εκτενή εξερεύνηση του σχετικού θεωρητικού υπόβαθρου. Η έρευνα εξετάζει τις βασικές έννοιες και αρχές που είναι απαραίτητες για την κατανόηση του πλαισίου της αναγνώρισης των επιθέσεων ΕΚΜΣ. Τα θέματα που εξετάζονται εκτείνονται από τον ευρύτερο τομέα της κυβερνοασφάλειας αλλά με έμφαση στο πλαίσιο της κοινωνικής μηχανικής, μέχρι τις λεπτομέρειες του κύκλου μιας επίθεσης ΕΚ ...
περισσότερα class="col-auto">:

Περίληψη σε άλλη γλώσσα

In an era dominated by digital communication, the escalating threat of chat-based social engineering (CSE) attacks looms large. These attacks, characterized by manipulation, cheating, and psychological exploitation, pose a grave danger to individuals and organizations alike. To confront this burgeoning menace, this doctoral thesis presents an all-encompassing system for recognizing CSE attacks, under the banner of the Chat-based Social Engineering Attack Recognition System (CSE-ARS). The foundation for this research is laid in an exhaustive exploration of the theoretical landscape. This comprehensive survey delves into the core concepts and principles essential for grasping the context of CSE attack recognition. Topics encompassed here range from the broader realm of cybersecurity, particularly in the context of social engineering, to the intricacies of the attack cycle and the profound impact of social engineering attacks. We further examine the pivotal role of advanced technologies s ...

Όλα τα τεκμήρια στο ΕΑΔΔ προστατεύονται από πνευματικά δικαιώματα.

In an era dominated by digital communication, the escalating threat of chat-based social engineering (CSE) attacks looms large. These attacks, characterized by manipulation, cheating, and psychological exploitation, pose a grave danger to individuals and organizations alike. To confront this burgeoning menace, this doctoral thesis presents an all-encompassing system for recognizing CSE attacks, under the banner of the Chat-based Social Engineering Attack Recognition System (CSE-ARS). The foundation for this research is laid in an exhaustive exploration of the theoretical landscape. This comprehensive survey delves into the core concepts and principles essential for grasping the context of CSE attack recognition. Topics encompassed here range from the broader realm of cybersecurity, particularly in the context of social engineering, to the intricacies of the attack cycle and the profound impact of social engineering attacks. We further examine the pivotal role of advanced technologies such as artificial intelligence, deep learning, and natural language processing. Notably, this investigation scrutinizes the metrics used to evaluate the performance of recognition models, including accuracy, precision, recall, and the F1 score. The aim is to establish a strong theoretical grounding, emphasizing the significance of deep learning models in identifying and addressing the multifaceted challenges of CSE attacks. The identified enablers of successful CSE attacks are then thoroughly examined. One key enabler lies in personality traits, as social engineers strategically exploit their understanding of human behavior to manipulate their targets. Understanding the dynamics of persuasion is also crucial for defense, with machine learning algorithms leveraged to recognize persuasive strategies and enhance resilience against CSE attacks. Persistent behavior, including paraphrasing, is another central strategy used by social engineers to manipulate their targets. Recognizing and characterizing this behavior is crucial for developing effective defenses. Deception is a vital enabler, and investigating deception cues and developing machine learning models for recognition is an essential component of defense. Additionally, recognizing speech acts and the role of chat history in providing insights into the structure and context of conversations is emphasized. Deep learning models are deployed to enhance the accuracy of CSE attack recognition and prevention by studying speech acts and incorporating chat history analysis too. The creation of the CSE Corpus serves as a fundamental resource for studying and understanding CSE attacks. This meticulous process begins with data source selection, dialogues collection, enrichment, linguistic analysis, and finally annotation. The CSE Corpus serves as a valuable asset for researchers and practitioners alike, facilitating the development and evaluation of robust models and methodologies for recognizing and mitigating social engineering attacks. Next, each enabler recognizer is introduced starting with a specialized recognition model, CRINL-R, for the identification of critical information leakage in CSE attacks. By employing deep learning techniques and a carefully curated dataset, CRINL-R demonstrates promising performance in identifying instances of critical information leakage. Personality traits remain at the forefront of the investigation in the development of the PERST-R model. This model leverages a pre-trained BERT model and a rich corpus of labeled text data to specialize in the accurate recognition of individual traits. This recognition plays a pivotal role in understanding social engineering tactics and further fortifying defenses. The recognition of persuasion techniques in CSE attacks takes center stage with the introduction of the PERSU-R model. This model integrates persuasion principles and convolutional neural networks to identify and categorize persuasive elements within textual interactions. Its efficacy in characterizing persuasion techniques contributes significantly to bolstering defenses against social engineering attacks. Recognition of persistence in CSE attacks is addressed through the PERSI-R model, which leverages natural language processing techniques and neural networks. This model accurately identifies and characterizes persistence cues within textual interactions, underlining the significance of recognizing persistence as a critical factor in social engineering attacks. The culmination of this research is presented with the introduction of the Chat-based Social Engineering Attack Recognition System (CSE-ARS). CSE-ARS leverages a late fusion approach to identify and recognize CSE attacks by combining multiple sources of information. By integrating individual recognizers specialized in different facets of CSE attacks, such as critical information leakage, personality traits, dialogue acts, persuasion techniques, and persistence, CSE-ARS achieves a comprehensive understanding of chat-based interactions. The system's performance is rigorously evaluated across various chat-based scenarios, demonstrating its potential real-world applicability. This doctoral thesis endeavors to provide a comprehensive framework for recognizing and mitigating social engineering attacks in the realm of digital communication. The integration of deep learning techniques, multimodal information fusion, and ethical considerations underscores the potential for advanced defense mechanisms against the pervasive challenges of social engineering threats. This interdisciplinary approach empowers individuals and organizations to counteract these attacks effectively, enhancing security and preserving personal and organizational integrity in the digital age. Future research may continue to refine and expand upon these models, contributing to practical deployment and wider adoption in real-world scenarios.
περισσότερα

Όλα τα τεκμήρια στο ΕΑΔΔ προστατεύονται από πνευματικά δικαιώματα.

DOI
10.12681/eadd/54739
Διεύθυνση Handle
http://hdl.handle.net/10442/hedi/54739
ND
54739
Εναλλακτικός τίτλος
Utilizing deep learning and natural language processing to recognise chat-based social engineering attacks for cyber security situational awareness
Συγγραφέας
Τσίγγανος, Νικόλαος (Πατρώνυμο: Βάιος)
Ημερομηνία
2023
Ίδρυμα
Πανεπιστήμιο Μακεδονίας. Σχολή Επιστημών Πληροφορίας. Τμήμα Εφαρμοσμένης Πληροφορικής
Εξεταστική επιτροπή
Μαυρίδης Ιωάννης
Γκρίτζαλης Δημήτριος
Φουληράς Παναγιώτης
Ρεφανίδης Ιωάννης
Ράντος Κωνσταντίνος
Πρωτοπαπαδάκης Ευτύχιος
Στεργιόπουλος Γεώργιος
Επιστημονικό πεδίο
Φυσικές ΕπιστήμεςΕπιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική ➨ Επιστήμη ηλεκτρονικών υπολογιστών
Φυσικές ΕπιστήμεςΕπιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική ➨ Τεχνητή νοημοσύνη
Λέξεις-κλειδιά
Κυβερνοασφάλεια; Μηχανική μάθηση; Βαθιά μάθηση; Σώμα κειμένων; Σχολιασμός αποσπασμάτων; Μεταφορά γνώσης
Χώρα
Ελλάδα
Γλώσσα
Αγγλικά
Άλλα στοιχεία
εικ., πιν., σχημ., γραφ.