Contributions in Merlin

Research in the area of human information interaction (HII) typically represents viewpoints on debated topics in a binary fashion, as either against or in favor of a given topic (e.g., the feminist movement). This simple taxonomy, however, greatly reduces the latent richness of viewpoints and thereby limits the potential of research and practical applications in this field. Work in the communication sciences has already demonstrated that viewpoints can be represented in much more comprehensive ways, which could enable a deeper understanding of users’ interactions with debated topics online. For instance, a viewpoint's stance usually has a degree of strength (e.g., mild or strong), and, even if two viewpoints support or oppose something to the same degree, they may use different logics of evaluation (i.e., underlying reasons). In this paper, we draw from communication science practice to propose a novel, two-dimensional way of representing viewpoints that incorporates a viewpoint's stance degree as well as its logic of evaluation. We show in a case study of tweets on debated topics how our proposed viewpoint label can be obtained via crowdsourcing with acceptable reliability. By analyzing the resulting data set and conducting a user study, we further show that the two-dimensional viewpoint representation we propose allows for more meaningful analyses and diversification interventions compared to current approaches. Finally, we discuss what this novel viewpoint label implies for HII research and how obtaining it may be made cheaper in the future.

null

On 3 September 2021, the Russian court forbade Google and Yandex to display search results for “Smart Voting,” the query referring to a tactical voting project by the jailed Russian opposition leader Alexei Navalny. To examine whether the two search engines complied with the court order, we collected top search outputs for the query from Google and Yandex. Our analysis demonstrates the lack of compliance from both engines; however, while Google continued prioritizing outputs related to the opposition’s web resources, Yandex removed links to them and, in some cases, promoted conspiratorial claims aligning with the Russian authorities’ anti-Western narrative.

The television broadcasting industry is of crucial economic and social importance. Traditionally, this industry has been dominated by free-to-air TV (FTV) but due to technological progress, subscription-based pay TV (PTV) has emerged as a competing business model. A key question for the PTV broadcasters is whether to air commercials in addition to charging subscription fees. Based on a theoretical model of asymmetric competition between a PTV and an FTV broadcaster, we examine the effects of placing PTV advertising on broadcaster market strategies, viewer demands, broadcaster profits and consumer surplus. We find that introducing advertising on PTV can induce a higher viewer demand on this channel but a lower viewer demand on the FTV channel. Surprisingly, consumers can benefit through the introduction of advertising in PTV and broadcaster profits can increase if the viewer disutility of advertising is sufficiently large. Our study provides an analytical framework for choosing and implementing an optimal PTV strategy when an FTV competitor preexists in the market. Furthermore, our study derives implications for policymakers and regulatory authorities by showing that additional PTV advertising is not necessarily socially undesirable due to the strategic market reactions.

The growing significance of programming languages is manifested by the change in educational curricula, which include programming lectures as early as in primary school. The growing significance also demonstrates that the means to acquire programming skills need to improve. We propose an interactive assistant to help novice programmers acquire knowledge through natural language during their programming tasks. Thus, learning by doing with an assistant that supports the user by providing source code in case of difficulties or missing know-how. Our approach is targeting single-line code generation while considering the natural language intent and an extensive context to provide accurate and relevant recommendations in the form of a single-line of source code. This approach is based on the idea that learning programming languages require solving programming exercises in addition to a fast feedback loop. Our approach aims to provide the learner with more opportunities to learn and understand the programming language by assisting them with minimal source code to help them continue with their task. Thus, only generating single-lines of source code instead of providing full functions. We implemented two models intending to contextualize single-line code generation using a custom context workflow. Our evaluations show that our approach is able to learn strong context representations from our custom context workflow as well as an option to improve context compression. In summary, we contribute a trained context-sensible model that takes natural language input, context, and predicts the next line. Using a custom workflow to deal with contexts of variable size, a new Javabased dataset, of over 200’000 samples tailored to our task of generating single-line Java code. Each sample contains context information, natural language intent, and the target line. Allowing us to contextualize our model, and we compare our work with other state-of-the-art approaches.

Recent studies in the field of localization and identification of persons indoors have shown that there are various possibilities to track individuals indoors. Such localization techniques can be divided into either active or passive indoor localization systems, needing a direct or no direct connection with the system respectively. The corona pandemic unveiled the importance and benefit of applying such techniques to trail possible infections of persons to contain the spread. In other fields, it has also been valuable for marketing analysis. In a recent master’s thesis at UZH, a passive localization and identification tool for tracking devices using only passive Bluetooth signals was explored in collaboration with Livealytics. The solution called BluePIL provided moderate accuracy, which can be considered to work despite the underlying naturally noisy data. The time consuming deployment, calibration and analysis yield a second version of it, called BluePIL 2.0, on which this thesis focuses on. Using new hardware, additional power-banks and by adding a new analysis tool with adjustable parameters, a new version is developed. The experiments have shown that BluePIL 2.0 improves the system by reducing the manual steps to deploy the system and improving mobility. The time to launch the system is 7.32 seconds on average. In addition, the analysis tool contributes to a better understanding of the indoor scenarios by displaying an updating plot of the tracked devices and providing the option to calibrate the path loss coefficient of the localization algorithm during data collection. The near-real-time plot is set to update within 0.5 seconds. In general, BluePIL 2.0 met the goals in a requirements-based evaluation.

The market for cyber insurance seems to be at a turning point in the recent years. Premiums are surging at record rates with new claims continuing to be driven by ransomware attacks, as well as by insider threats. The cyber insurance capacity is becoming limited and it is clear that many challenges still need to be tackled in order to avoid market failure. For one, it is clear that traditional risk assessment methods currently applied do not sufficiently address the issues of information asymmetries and the adverse selection and moral hazard that go hand-in hand with them. Therefore, new approaches to audit and assess the level of self-protection of the insureds need to be developed. Taking this into account, this thesis focuses on the assessment of operational cyber risks related to failed internal processes and proposes a novel approach to apply the methods of process mining in cyber insurance. For this purpose, MeritMiner4CI approach was designed and developed. Also, the fundamental challenges and requirements of cyber insurance at the coverage level were clearly mapped to specific process mining methods. The MeritMiner4CI approach was evaluated by conducting a survey with experts and also by considering case study scenarios. The results of this survey provide strong indication that the analyses of the proposed method can be applied by practitioners and have an impact on the ratings of confidence factors that are applied in the industry. Furthermore, quantitative evaluations were conducted to evaluate the performance of the applied methods. Finally, this thesis also highlights how MeritMiner4CI can be integrated with other cybersecurity risk assessment approaches, such as SecRiskAI and MENTOR.

Ride-hailing services such as Uber and Lyft have become globally pervasive in the last decade, revolutionizing the taxi sector for both customers and drivers. This business model breaks many barriers of entry for new drivers and makes commuting by taxi cheaper and more convenient for customers. Nevertheless, it is also affected by drastic income inequalities and weak job security. The goal of this thesis is to investigate whether we can reduce income inequalities by using ML-based demand prediction to strategically dispatch drivers. As such, I first develop a machine learning model to predict customer demand using a real-world ride-hailing dataset collected by the city of Chicago. Secondly, I integrate the real-world data within an agent-based model to make an initial exploration of the potential of using the predictions for decreasing the income inequality of drivers. The results show that (a) the prediction model is able to fairly accurately predict demand, and (b) one of the three implemented rule-based naive dispatchers successfully used the predictions in order to increase the level of fairness.

Open Set Recognition (OSR) is a promising field that aims to adapt deep neural models to a real-world scenario, i.e., the presence of unknown data distributions at inference time. Recent approaches address OSR by including into the training set negative classes that a classifier learns to reject, expecting that these data increase the robustness of the classifier on unknown classes. Furthermore, a critical question on OSR is how to select representative negative classes. In this work, we study the performance an OSR approach, the Entropic open-set and Objectosphere, two recently proposed loss functions to deal with unknowns. We show that these losses can be used in complex custom scenarios that use natural images and several levels of similarity between the negative classes. Additionally, we investigate a method to adapt adversarial training into OSR scenarios. Our experimental results show that it is possible to use synthetic adversarial examples to increase the robustness of a classifier in OSR.

Die Risiken von Immobilien scheinen in der Bevölkerung wegen der hohen Renditen der letzten 20 Jahre immer mehr in Vergessenheit zu geraten. Jedoch bahnen sich im Jahr 2022, als geldpolitische Reaktion zur ansteigenden Inflation, mehrere Zinserhöhungen an, welche die meisten Wertanlagen empfindlich treffen könnten. Für Wohneigentümer existieren keine Kursverläufe wie für Aktien. Aufgrund dessen wird angenommen, dass sich Eigentümer an Immobilienindizes orientieren, um den Wertver- lauf ihres Objekts zu verfolgen. Dieses Verhalten ist eines der Motive für diese Arbeit, da die Repräsentativität der Immobilienindizes trügerisch sein kann. Das Ziel der vorliegenden Arbeit ist, zu beantworten, ob der Tracking Error von Wohneigentum ein unterschätzter Faktor ist und ob somit die Sicherheit von Wohneigentum illusorisch sein könnte. Um die Forschungsfrage zu beantworten, wurde bestehende, adäquate Theorie analysiert. Ergänzt mit eigenen Beispielen und Berechnungen konnte Evidenz zur Proposition gesammelt werden. Es hat sich herausgestellt, dass gewisse Personengruppen das Risiko des Tracking Errors erkennen, jedoch schmälert dies nicht die Gefahr, die von ihm ausgeht. Die Sicherheit von Wohneigentum wird von den Gefahren der Unterdiversifikation, der Opportunitätskosten des Eigenkapitals und des Tracking Errors beeinträchtigt. Des Weiteren hat sich gezeigt, dass die kleinen Stichprobengrössen von Immobilienindizes den Tracking Error noch verstärken können. Dies führt zur Mahnung, die Repräsentativität von Immobilienindizes für Einzelobjekte nicht als gegeben zu betrachten.