Johannes Sänger

Trust and Big Data: A Roadmap for Research

We are currently living in the age of Big Data coming along with the challenge to grasp the golden opportunities at hand. This mixed blessing also dominates the relation between Big Data and trust. On the one side, large amounts of trust-related data can be utilized to establish innovative data-driven approaches for reputation-based trust management. On the other side, this is intrinsically tied to the trust we can put in the origins and quality of the underlying data. In this paper, we address both sides of trust and Big Data by structuring the problem domain and presenting current research directions and inter-dependencies. Based on this, we define focal issues which serve as future research directions for the track to our vision of Next Generation Online Trust within the FORSEC project.

SoDA: Dynamic visual analytics of big social data

In this work we apply dynamic visual analytics on big social data by the example of microblogs from Twitter. Thereby, we address current challenges like real-time analytics as well as analyses of unstructured data. To this end, we propose SoDA - a concept enabling the integrated analysis of the dimensions: message, location and time. Furthermore, we introduce a novel design for tag cloud visualizations, the weighted tag network, offering enhanced semantic insights. All concepts are fully implemented and evaluated by a comprehensive software prototype in different application scenarios.

Reusability for Trust and Reputation Systems

Reputation systems have been extensively explored in various disciplines and application areas. A problem in this context is that the computation engines applied by most reputation systems available are designed from scratch and rarely consider well established concepts and achievements made by others. Thus, approved models and promising approaches may get lost in the shuffle. In this work, we aim to foster reuse in respect of trust and reputation systems by providing a hierarchical component taxonomy of computation engines which serves as a natural framework for the design of new reputation systems. In order to assist the design process we, furthermore, provide a component repository that contains design knowledge on both a conceptual and an implementation level.

Visualizing Unfair Ratings in Online Reputation Systems

Reputation systems provide a valuable method to measure the trustworthiness of sellers or the quality of products in an e-commerce environment. Due to their economic importance, reputation systems are subject to many attacks. A common problem are unfair ratings which are used to unfairly increase or decrease the reputation of an entity. Although being of high practical relevance, unfair rating attacks have only rarely been considered in literature. The few approaches that have been proposed are furthermore quite non-transparent to the user. In this work, we employ visual analytics to identify colluding digital identities. The ultimate benefit of our approach is the transparent revelation of the true reputation of an entity by interactively using both endogenous and exogenous discounting methods. We thereto introduce a generic conceptual design of a visual analytics component that is independent of the underlying reputation system. We then describe how this concept was implemented in a software prototype. Subsequently, we demonstrate its proper functioning by means of an empirical study based on two real-world datasets from eBay and Epinions. Overall, we show that our approach notably enhances transparency, bares an enormous potential and might thus lead to substantially more robust reputation systems and enhanced user experience.

Look Before You Leap: Improving the Users' Ability to Detect Fraud in Electronic Marketplaces

Reputation systems in current electronic marketplaces can easily be manipulated by malicious sellers in order to appear more reputable than appropriate. We conducted a controlled experiment with 40 UK and 41 German participants on their ability to detect malicious behavior by means of an eBay-like feedback profile versus a novel interface involving an interactive visualization of reputation data. The results show that participants using the new interface could better detect and understand malicious behavior in three out of four attacks (the overall detection accuracy 77% in the new vs. 56% in the old interface). Moreover, with the new interface, only 7% of the users decided to buy from the malicious seller (the options being to buy from one of the available sellers or to abstain from buying), as opposed to 30% in the old interface condition.

TRIVIA: visualizing reputation profiles to detect malicious sellers in electronic marketplaces

Reputation systems are an essential part of electronic marketplaces that provide a valuable method to identify honest sellers and punish malicious actors. Due to the continuous improvement of the computation models applied, advanced reputation systems have become non-transparent and incomprehensible to the end-user. As a consequence, users become skeptical and lose their trust toward the reputation system. In this work, we are taking a step to increase the transparency of reputation systems by means of providing interactive visual representations of seller reputation profiles. We thereto propose TRIVIA - a visual analytics tool to evaluate seller reputation. Besides enhancing transparency, our results show that through incorporating the visual-cognitive capabilities of a human analyst and the computing power of a machine in TRIVIA, malicious sellers can be reliably identified. In this way we provide a new perspective on how the problem of robustness could be addressed.

Reusable components for online reputation systems

Reputation systems have been extensively explored in various disciplines and application areas. A problem in this context is that the computation engines applied by most reputation systems available are designed from scratch and rarely consider well established concepts and achievements made by others. Thus, approved models and promising approaches may get lost in the shuffle. In this work, we aim to foster reuse in respect of trust and reputation systems by providing a hierarchical component taxonomy of computation engines which serves as a natural framework for the design of new reputation systems. In order to assist the design process we, furthermore, provide a component repository that contains design knowledge on both a conceptual and an implementation level. To evaluate our approach we conduct a descriptive scenario-based analysis which shows that it has an obvious utility from a practical point of view. Matching the identified components and the properties of trust introduced in literature, we finally show which properties of trust are widely covered by common models and which aspects have only rarely been considered so far.

Reusable Defense Components for Online Reputation Systems

Attacks on trust and reputation systems (TRS) as well as defense strategies against certain attacks are the subject of many research papers. Although proposing valuable ideas, they all exhibit at least one of the following major shortcomings. Firstly, many researchers design defense mechanisms from scratch and without reusing approved ideas. Secondly, most proposals are limited to naming and theoretically describing the defense mechanisms. Another issue is the inconsistent denomination of attacks with similar characteristics among different researchers. To address these shortcomings, we propose a novel taxonomy of attacks on TRS focusing on their general characteristics and symptomatology. We use this taxonomy to assign reusable, clearly described and practically implemented components to different classes of attacks. With this work, we aim to provide a basis for TRS designers to experiment with numerous defense mechanisms and to build more robust systems in the end.

Biometric Identity Trust: Toward Secure Biometric Enrollment in Web Environments

The nonrepudiation of a biometric authentication depends on the authenticity of the corresponding biometric profile. If the enrollment process is not controlled by some trusted entity, a user’s biometric data might be misleadingly linked to another person’s digital identity. To secure the biometric enrollment in open Web-based environments, we propose the biometric observer principle: An arbitrary trustworthy person observes an individual’s enrollment at a biometric identity provider and confirms this to the system. The concept rests on a specified trust model, which assesses the trustworthiness of both the observer and the authenticity of an observed biometric profile. Trust relations between observer and observed persons are managed by the authentication system. We implemented a cloud-based biometric identity provider to validate and demonstrate the proposed concept.

Interactive Visualization of Recommender Systems Data

Recommender systems provide a valuable mechanism to address the information overload problem by reducing a data set to the items that may be interesting for a particular user. While the quality of recommendations has notably improved in the recent years, the complex algorithms in use lead to high non-transparency for the end user. We propose the usage of interactive visualizations for presenting recommendations. By involving the user in the information reduction process, the quality of recommendations could be enhanced whilst keeping the systems transparency. This work gives first insights by analyzing recommender systems data and matching them to suitable visualization and interaction techniques. The findings are illustrated by means of an example scenario based on a typical real-world setting.