Contributions published at Computation and Economics (Sven Seuken)

Competition can increase the incentive a subject has towards performing an action. This thesis shows, that competition can be used to enhance motivation for learning course content during a semester. In order to transmit this motivational influence to learning software, we concentrated on the definition of badges and rankings. We conducted an experiment to test, whether the determined achievements have an influence on the usage characteristics of consumers of the learning software. To test the badges, we implemented a mobile quiz application, including the opportunity to collect achievements or improve the position in a ranking. We show, that users are willing to spend more time using the application when competing with other participants through badges or rankings.

Combinatorial auctions have received increasing attention during the last years. Their allocational power, which allows for highly complex bidder valuations, has been utilized in many multi billion dollar public-sector auctions. Despite their frequent use, many questions about combinatorial auction still remain unanswered. The nature of these questions is often not only economical but also computational. In this thesis we focus on payments in combinatorial auction. Recent literature has suggested that one should select payments that are in the core of the auction. In the first part of the thesis we introduce new theoretical results on core constraints. We then use this insight to improve the state of the art algorithm for computing core payments. Our experiments show that these new algorithmic ideas can compute core payments in hard instances at least 75 faster than the currently fastest known algorithm. In the second part of the thesis we propose an algorithm that can approximate BNEs in core-selecting combinatorial auctions. We then use this algorithm to analyze several known core-payment rules. Additionally we introduce new core payment rules that, in our analysis, have better incentives.

We examine how the incentive compatibility property of payment rules and differing risk preferences affect effort elicitation and truthful reporting on monetary crowdsourcing markets. Therefore we conduct two experiments on Amazon's Mechanical Turk. We show that the two approaches for assessing risk preferences proposed by Eckel and Grossmann (2008) and Holt and Laury (2002) provide results which are not corresponding to each other and even exhibit different ground distributions. Furthermore we provide empirical evidence of participants being able to exploit incentive incompatible payment rules by reporting untruthfully. Also, we show that whether incentive incompatible nor incentive compatible payment rules necessarily result in an increase of performance and can even have negative effects on effort elicitation. Our research resulted in no visible effects of risk preferences on truthful reporting or effort elicitation.

Online recommender systems have become an important help for users of various ap- plications. They help users deal with the information overload often encountered in today's online world and filter out relevant alternatives. Social recommender systems moreover try to use information contained in social networks to enhance the performance of conventional recommendation methods. This thesis focuses on the collaborative filtering recommendation approach and ana- lyzes its performance in different environments. A close look is taken at the incorporation of social network information into the collaborative filtering algorithm. Further, concepts from network theory are used to analyze if there exist users who influence the taste of others and therefore could be better predictors.

We study one-sided matching mechanisms where agents have vNM utility functions and report ordinal preferences. We first show that in this domain strategyproof mechanisms are characterized by three intuitive axioms: swap monotonicity, upper invariance, and lower invariance. Our second result is that dropping lower invariance leads to an interesting new relaxation of strategyproofness, which we call partial strategyproofness. In particular, we show that mechanisms are swap monotonic and upper invariant if and only if they are strategyproof on a restricted domain where agents have sufficiently different valuations for different objects. Furthermore, we show that this domain restriction is maximal and use it to define a single-parameter measure for the degree of strategyproofness of a manipulable mechanism. We also provide an algorithm that computes this measure. Our new partial strategyproofness concept finds applications in the incentive analysis of non-strategyproof mechanisms, such as the Probabilistic Serial mechanism, different variants of the Boston mechanism, and the construction of new hybrid mechanisms.

For the design of distributed work systems like P2P file-sharing networks it is essential to provide incentives for agents to work for each other rather than free ride. Several mechanisms have been proposed to achieve this goal, including currency systems, credit networks, and accounting mechanisms. It has proven particularly challenging to provide robustness to sybil attacks, i.e., attacks where an agent creates and controls multiple false identities. In this paper, we consider accounting mechanisms for domains in which (1) transactions cannot be bound to reports, (2) transactions are bilateral and private, and (3) agents can only form trust links upon successful work interactions. Our results reveal the trade-offs one must make in designing such mechanisms. We show that accounting mechanisms with a strong form of transitive trust cannot be robust against strongly beneficial sybil attacks. However, we also present a mechanism that strikes a balance, providing a weaker form of transitive trust while also being robust against the strongest form of sybil attacks. On the one hand, our results highlight the role of strong social ties in providing robustness against sybil attacks (such as those leveraged in credit networks using bilateral IOUs), and on the other hand our results show what kind of robustness properties are possible and impossible in domains where such pre-existing trust relations do not exist.

We consider the problem of adaptive home heating in the smart grid, assuming that real-time electricity prices are being exposed to end-users with the goal of realizing demand-side management. To lower the burden on the end-users, our goal is the design of a smart thermostat that automatically heats the home, optimally trading o the user’s comfort and cost. This is a challenging problem due to two sources of uncertainty: future weather conditions and future electricity prices. Our main technical contribution is a general technique that uses predictive distributions obtained from Gaussian Process (GP) regressions to compute the state transition probabilities of an MDP, such that the solution to the resulting MDP constitutes a sequentially optimal policy. We apply this general approach to the home-heating problem, where we use the predictive distributions of the GPs for the day-ahead external temperatures and electricity prices. The solution to the home-heating MDP constitutes an optimal heating policy that maximizes the user’s utility given the probability information gathered by the Gaussian process model. Via simulations we show that our MDP-based approach outperforms various benchmarks, especially for cost-sensitive users.

We conducted a pilot experiment to investigate whether cognitively challenging matching experiments on Amazon Mechanical Turk can provide significant results. Our results showed that this is possible. In our experiment, subjects had to learn one of two popular mechanisms (the Boston Mechanism and the Probabilistic Serial Mechanism) and try to find beneficial manipulations. We showed that workers can be selected appropriately and educated to understand complex mechanisms. We also investigated the search paths which human subjects used to beneficial manipulations in fixed scenarios of one sided matchings. We showed that humans exhibited bounded rational strategic behaviour in the sense that they more frequently found local rather than global manipulations in both mechanisms. To conduct our experiment, we developed software which is extendible and can be used for further research.

Simulation performance may be evaluated according to multiple quality measures that are in competition and their simultaneous consideration poses a conflict. In the current study we propose a practical framework for investigating such simulation performance criteria, exploring the inherent conflicts amongst them and identifying the best available tradeoffs, based upon multiobjective Pareto optimization. This approach necessitates the rigorous derivation of performance criteria to serve as objective functions and undergo vector optimization. We demonstrate the effectiveness of our proposed approach by applying it to a specific Artificial Neural Networks (ANN) simulation, with multiple stochastic quality measures. We formulate performance criteria of this use-case, pose an optimization problem, and solve it by means of a simulation- based Pareto approach. Upon attainment of the underlying Pareto Frontier, we analyze it and prescribe preference-dependent configurations for the optimal simulation training.

Haptic technologies provide physical sensations in the interaction with a computing system, by exploiting the human sense of touch and by applying forces, vibrations, or motions to the user hands or body. Considering their features, they can be a useful tool in life-science teaching, especially when molecules are involved. For this purpose, a framework composed of an haptic device and a visual interface for molecular exploration has been developed to simulate molecular and intermolecular interactions . Furthermore, this work evaluates the visual and haptic tool for molecular exploration in a didactic context, performing tests and interviews with students. The final aim is to properly develop the features of the tool, in order to make it suitable for the introduction in chemistry education. Preliminary results show positive and effective responses and learning gains from the tasks. It has also been noticed that the use of such an innovative instrument raises the interest of students in the learning process, which is one of the main benefits of the haptic device.

A key issue for the realization of the smart grid vision is the implementation of effective demand-side management. One possible approach involves exposing dynamic energy prices to end-users. In this paper, we consider a resulting problem on the user’s side: how to adaptively heat a home given dynamic prices. The user faces the challenge of having to react to dynamic prices in real time, trading off his comfort with the costs of heating his home to a certain temperature. We propose an active learning approach to adjust the home temperature in a semiautomatic way. Our algorithm learns the user’s preferences over time and automatically adjusts the temperature in real-time as prices change. In addition, the algorithm asks the user for feedback once a day. To find the best query time, the algorithm solves an optimal stopping problem. Via simulations, we show that our algorithm learns users’ preferences quickly, and that using the expected utility loss as the query criterion outperforms standard approaches from the active learning literature.

The goal of this thesis was to create an application, with which people could check factual statements with the help of Amazon Mechanical Turk. The application was supposed to be setup like a game, that punishes false and rewards right statements. Another goal of the research, was to test the influence of the task parameters on the turkers work. As our main contribution we present Bet Me, a novel web application for betting and fact checking. Users of Bet Me can bet with each other on the answer of any question, they can formulate with the answers yes and no. The bet is then sent to Amazon Mechanical Turk, where workers vote on, what is the right answer. The user, whose answer got the most votes, wins the bet. We conducted a functional test, where we tried out several different parameters to the tasks, we put on Mechanical Turk. We also made an usability study, where we asked potential users to try our application. The results of the functional test and the usability study showed, that it is possible to use Amazon Mechanical Turk to answer arbitrary yes/no questions, though workers have a harder time with trick and subjective questions.

We ask if in a one-sided matching setting, global manipulability does imply local manipulability for an agent with a fixed type. If this was the case, it would theoretically be possible for a boundedly rational agent to find a global manipulation through a sequence of successful local manipulations. We show that global manipulability does not necessarily imply local manipulability for the Probabilistic Serial mechanism and the Boston mechanism in the same setting. This indicates that computational-complexity constraints could be more useful than previous results from Carroll (2012) suggest. We also show that clear Contracting Equilibria do occur when using the Boston mechanism and the Probabilistic Serial mechanism in iterative matching scenarios. We compare the welfare of the truthful allocation and the allocation in equilibrium of those Contracting Equilibria and come to the conclusion that when considering rank distribution, the truthful allocation is always better in our samples. Considering social welfare, the case is less clear. All results were obtained using a purpose-built simulation software, which is extensible and is intended to be used for further research into the assignment problem.

We conducted a study with two experiments on the crowdsourcing platform Amazon Mechanical Turk. One experiment had the goal to examine the execution and the quality of the results of a complex decision-making experiment on Amazon Mechanical Turk. The other analysed the influence of an individual behavioural optimised market user interface on the performance of the participants. The first experiment was essential in order to prove the quality of the results of the second experiment. It can be shown that Amazon Mechanical Turk is a useful tool to conduct behavioural studies. The second experiment continued a study originally performed by Seuken et al. [2012]. In this experiment the participants had to play a game repeatedly which represents a potential market situation. They tried to increase the performance of their participants through a behavioural optimised market user interface design. In our study, we extended this approach to an individual behavioural optimisation. Although the optimisation did not lead to a significant increase in the performance of all participants, it indicated that the participants can be split up into three main groups with different behaviours, where one of these groups profited from the optimisation. Thus, this thesis may be seen as a starting point for a more detailed research of designing personalised market user interfaces.

Crowdsourcing markets like Amazon’s Mechanical Turk or Crowdflower are quickly growing in size and popularity. The allocation of workers and compensation approaches in these markets are, however, still very simple. In particular, given a set of tasks that need to be solved within a specific time constraint, no mechanism exists for the requestor to (a) find a suitable set of crowd workers that can solve all of the tasks within the time constraint, and (b) find the “right” price to pay these workers. In this paper, we provide a solution to this problem by introducing CrowdManager – a framework for the combinatorial allocation and pricing of crowdsourcing tasks under budget, completion time, and quality constraints. Our main contribution is a mechanism that allocates tasks to workers such that social welfare is maximized, while obeying the requestor’s time and quality constraints. Workers’ payments are computed using a VCG payment rule. Thus, the resulting mechanism is efficient, truthful, and individually rational. To support our approach we present simulation results that benchmark our mechanism against two baseline approaches employing fixed-priced mechanisms. The simulation results illustrate that our mechanism (i) significantly reduces the requestor’s costs in the majority of settings and (ii) finds solutions in many cases where the baseline approaches either fail or significantly overpay. Furthermore, we show that the allocation as well as VCG payments can be computed in a few seconds, even with hundreds of workers and thousands of tasks.

Despite the pervasiveness of markets in our lives, little is known about the role of user interfaces (UIs) in promoting good decisions in market domains. How does the way we display market information to end users, and the set of choices we offer, influence users' decisions? In this paper, we introduce a new research agenda on "market user interface design." Our goal is to find the optimal market UI, taking into account that users incur cognitive costs and are boundedly rational. Via lab experiments we systematically explore the market UI design space, and we study the automatic optimization of market UIs given a behavioral (quantal response) model of user behavior. Surprisingly, we find that the behaviorally-optimized UI performs worse than the standard UI, suggesting that the quantal response model did not predict user behavior well. Subsequently, we identify important behavioral factors that are missing from the user model, including loss aversion and position effects, which motivates follow-up studies. Furthermore, we find significant differences between individual users in terms of rationality. This suggests future research on personalized UI designs, with interfaces that are tailored towards each individual user's needs, capabilities, and preferences.

null

null

The most prominent way to establish trust between buyers and sellers on online auction sites are reputation mechanisms. Two drawbacks of this approach are the reliance on the seller being long-lived and the susceptibility to whitewashing. In this paper, we introduce so-called escrow mechanisms that avoid these problems by installing a trusted intermediary which forwards the payment to the seller only if the buyer acknowledges that the good arrived in the promised condition. We address the incentive issues that arise and design an escrow mechanism that is incentive compatible, efficient, interim individually rational and ex ante budget-balanced. In contrast to previous work on trust and reputation, our approach does not rely on knowing the sellers' cost functions or the distribution of buyer valuations.

null