Digital trust and reputation: Applications and security issues
Digital Trust borrows the trust concept in sociology to indicate that, in distributed computing and communication systems, one party evaluates whether other parties are trusted to perform a certain action or have a certain property. This dissertation investigates the security issues in digital trust and reputation systems in the context of online social networks and explores a new application of digital trust in biomedical sensor systems. We propose TAUCA, short for joint Temporal And User Correlation Analysis. It identifies malicious users and recovers reputation scores from a novel angle: combination of temporal analysis and user correlation analysis. Benefiting from the rich information in the time-domain, TAUCA identifies the products under attack, the time when attacks occur, and malicious users who insert dishonest ratings. TAUCA and two other representative schemes are tested against real user attack data collected through a cyber competition. TAUCA demonstrates significant advantages. It largely improves the detection rate and reduces the false alarm rate in the detection of malicious users. It also effectively reduces bias in the recovered reputation scores. Defending multiple-user-multiple-target attacks in online reputation systems. Driven by the huge profits of online markets, reputation manipulations have evolved rapidly. Nowadays, some powerful attacks are conducted by companies that make profit through manipulating the reputations of online items for their customers. We define this type of attacks as multiple-user-multiple-target attacks. Compared to the single-target attacks addressed in the previous part, the multiple-target attack usually has larger attack power while it is relatively new to the research community. Very limited work has been done to address it. To address these attacks, we propose a defense scheme that (1) sets up heterogeneous thresholds for detecting multiple suspicious items and (2) identifies target items based on correlation analysis among suspicious items. The proposed scheme and two other comparison schemes are evaluated by a combination of real user data and simulation data. The proposed scheme demonstrates significant advantages in detecting malicious users, recovering reputation scores of target items, and reducing interference to normal items. Note that the concept of "correlation" in this thesis is not as same as the strict definition of correlation in statistics. Given two users, we consider these two users' rating behaviors as two random variables, compute the Euclidean distance between these two random variables and take its inverse as the correlation value between these two given users. Furthermore, the item correlation is calculated by combining user correlation values among users who rate these two items. The concepts of user correlation and item correlation in this thesis are actually introduced to measure the similarity among users and among items. Trust sensor interface for improving reliability of EMG-based user intent recognition. In the above two parts, we studied how to build up secure and reliable digital trust and reputation systems. The next question is that: can we apply this secure and reliable trust and reputation systems in other fields, such as biomedical sensor systems? In such systems, signals obtained from sensors are the basis for monitoring patients' status and making medical decisions. Therefore, ensuring the reliability of sensors is critically important. The digital trust mechanism which dynamically evaluates the reliability of a given object inherently fits in this field. Specifically, we apply the digital trust approach on a neural-machine interface for artificial legs. To achieve natural and smooth control of prostheses, Electromyographic (EMG) signals have been investigated for decoding user intent. However, EMG signals can be easily contaminated by diverse disturbances, leading to errors in user intent recognition and threatening the safety of prostheses users. To address this problem, we propose a trust sensor interface (TSI) that contains 2 modules: (1) abnormality detector that detects diverse disturbances with high accuracy and low latency and (2) trust evaluation that dynamically evaluates the reliability of EMG sensors. Based on the output of the TSI, the user intention recognition (UIR) algorithm is able to dynamically adjust their operations or decisions. Our experiments on five able-bodied subjects and two amputee subjects have demonstrated that the proposed TSI can effectively detect two types of disturbances, motion artifacts and baseline shifts, and improve the reliability of the UIR. (Abstract shortened by UMI.)
"Digital trust and reputation: Applications and security issues"
Dissertations and Master's Theses (Campus Access).