One of the main concerns in centralized and decentralized OSNs is related to the fact that OSNs users establish new relationships with unknown people with the result of exposing a huge amount of personal data. This can attract the variety of attackers that try to propagate malwares and malicious items in the network to misuse the personal information of users. Therefore, there have been several research studies to detect specific kinds of attacks by focusing on the topology of the graph [159, 158, 32, 148, 157]. On the other hand, there are several solutions to detect specific kinds of attackers based on the behavior of users. But, most of these approaches either focus on just the topology of the graph [159, 158] or the detection of anomalous users by exploiting supervised learning techniques [157, 47, 86, 125]. However, we have to note that the main issue of supervised learning is that they are not able to detect new attacker's behaviors, since the classifier is trained based on the known behavioral patterns. Literature also offers approaches to detect anomalous users in OSNs that use unsupervised learning approaches [150, 153, 36, 146] or a combination of supervised and unsupervised techniques [153]. But, existing attack defenses are designed to cope with just one specific type of attack. Although several solutions to detect specific kinds of attacks have been recently proposed, there is no general solution to cope with the main privacy/security attacks in OSNs. In such a scenario, it would be very beneficial to have a solution that can cope with the main privacy/security attacks that can be perpetrated using the social network graph. Our main contribution is proposing a unique unsupervised approach that helps OSNs providers and users to have a global understanding of risky users and detect them. We believe that the core of such a solution is a mechanism able to assign a risk score to each OSNs account. Over the last three years, we have done significant research efforts in analyzing user's behavior to detect risky users included some kinds of well known attacks in centralized and decentralized online social networks. Our research started by proposing a risk assessment approach based on the idea that the more a user behavior diverges from normal behavior, the more it should be considered risky. In our proposed approach, we monitor and analyze the combination of interaction or activity patterns and friendship patterns of users and build the risk estimation model in order to detect and identify those risky users who follow the behavioral patterns of attackers. Since, users in OSNs follow different behavioral patterns, it is not possible to define a unique standard behavioral model that fits all OSNs users' behaviors. Towards this goal, we propose a two-phase risk assessment approach by grouping users in the first phase to find similar users that share the same behavioral patterns and, then in the second phase, for each identified group, building some normal behavior models and compute for each user the level of divergency from these normal behaviors. Then, we extend this approach for Decentralized Online Social Networks (i.e., DOSNs). In the following of this approach, we propose a solution in defining a risk measure to help users in OSNs to judge their direct contacts so as to avoid friendship with malicious users. Finally, we monitor dynamically the friendship patterns of users in a large social graph over time for any anomalous changes reflecting attacker's behaviors. In this thesis, we will describe all the solutions that we proposed.

Risk assessment in centralized and decentralized online social network / Laleh, Naeimeh. - (2016).

Risk assessment in centralized and decentralized online social network.

Laleh, Naeimeh
2016-01-01

Abstract

One of the main concerns in centralized and decentralized OSNs is related to the fact that OSNs users establish new relationships with unknown people with the result of exposing a huge amount of personal data. This can attract the variety of attackers that try to propagate malwares and malicious items in the network to misuse the personal information of users. Therefore, there have been several research studies to detect specific kinds of attacks by focusing on the topology of the graph [159, 158, 32, 148, 157]. On the other hand, there are several solutions to detect specific kinds of attackers based on the behavior of users. But, most of these approaches either focus on just the topology of the graph [159, 158] or the detection of anomalous users by exploiting supervised learning techniques [157, 47, 86, 125]. However, we have to note that the main issue of supervised learning is that they are not able to detect new attacker's behaviors, since the classifier is trained based on the known behavioral patterns. Literature also offers approaches to detect anomalous users in OSNs that use unsupervised learning approaches [150, 153, 36, 146] or a combination of supervised and unsupervised techniques [153]. But, existing attack defenses are designed to cope with just one specific type of attack. Although several solutions to detect specific kinds of attacks have been recently proposed, there is no general solution to cope with the main privacy/security attacks in OSNs. In such a scenario, it would be very beneficial to have a solution that can cope with the main privacy/security attacks that can be perpetrated using the social network graph. Our main contribution is proposing a unique unsupervised approach that helps OSNs providers and users to have a global understanding of risky users and detect them. We believe that the core of such a solution is a mechanism able to assign a risk score to each OSNs account. Over the last three years, we have done significant research efforts in analyzing user's behavior to detect risky users included some kinds of well known attacks in centralized and decentralized online social networks. Our research started by proposing a risk assessment approach based on the idea that the more a user behavior diverges from normal behavior, the more it should be considered risky. In our proposed approach, we monitor and analyze the combination of interaction or activity patterns and friendship patterns of users and build the risk estimation model in order to detect and identify those risky users who follow the behavioral patterns of attackers. Since, users in OSNs follow different behavioral patterns, it is not possible to define a unique standard behavioral model that fits all OSNs users' behaviors. Towards this goal, we propose a two-phase risk assessment approach by grouping users in the first phase to find similar users that share the same behavioral patterns and, then in the second phase, for each identified group, building some normal behavior models and compute for each user the level of divergency from these normal behaviors. Then, we extend this approach for Decentralized Online Social Networks (i.e., DOSNs). In the following of this approach, we propose a solution in defining a risk measure to help users in OSNs to judge their direct contacts so as to avoid friendship with malicious users. Finally, we monitor dynamically the friendship patterns of users in a large social graph over time for any anomalous changes reflecting attacker's behaviors. In this thesis, we will describe all the solutions that we proposed.
2016
Online social network, risk assessment, anomaly detection, graph processing, unsupervised Learning.
Risk assessment in centralized and decentralized online social network / Laleh, Naeimeh. - (2016).
File in questo prodotto:
File Dimensione Formato  
Phd_Thesis_Lalehnaeimeh_completa.pdf

accesso aperto

Descrizione: testo completo tesi
Tipologia: Tesi di dottorato
Licenza: Non specificato
Dimensione 5.27 MB
Formato Adobe PDF
5.27 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11383/2090533
 Attenzione

L'Ateneo sottopone a validazione solo i file PDF allegati

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact