L3S Research Center

Abstract Artificial Intelligence (AI)‐based systems are widely employed nowadays to make decisions that have far‐reaching impact on individuals and society. Their decisions might affect everyone, everywhere, and anytime, entailing concerns about potential human rights issues. Therefore, it is necessary to move beyond traditional AI algorithms optimized for predictive performance and embed ethical and legal principles in their design, training, and deployment to ensure social good while still benefiting from the huge potential of the AI technology. The goal of this survey is to provide a broad multidisciplinary overview of the area of bias in AI systems, focusing on technical challenges and solutions as well as to suggest new research directions towards approaches well‐grounded in a legal frame. In this survey, we focus on data‐driven AI, as a large part of AI is powered nowadays by (big) data and powerful machine learning algorithms. If otherwise not specified, we use the general term bias to describe problems related to the gathering or processing of data that might result in prejudiced decisions on the bases of demographic features such as race, sex, and so forth. This article is categorized under: Commercial, Legal, and Ethical Issues > Fairness in Data Mining Commercial, Legal, and Ethical Issues > Ethical Considerations Commercial, Legal, and Ethical Issues > Legal Issues

Distributional semantic models derive computational representations of word meaning from the patterns of co-occurrence of words in text. Such models have been a success story of computational linguistics, being able to provide reliable estimates of semantic relatedness for the many semantic tasks requiring them. However, distributional models extract meaning information exclusively from text, which is an extremely impoverished basis compared to the rich perceptual sources that ground human semantic knowledge. We address the lack of perceptual grounding of distributional models by exploiting computer vision techniques that automatically identify discrete “visual words” in images, so that the distributional representation of a word can be extended to also encompass its co-occurrence with the visual words of images it is associated with. We propose a flexible architecture to integrate text- and image-based distributional information, and we show in a set of empirical tests that our integrated model is superior to the purely text-based approach, and it provides somewhat complementary semantic information with respect to the latter.

The run-time binding of web services has been recently put forward in order to support rapid and dynamic web service compositions. With the growing number of alternative web services that provide the same functionality but differ in quality parameters, the service composition becomes a decision problem on which component services should be selected such that user's end-to-end QoS requirements (e.g. availability, response time) and preferences (e.g. price) are satisfied. Although very efficient, local selection strategy fails short in handling global QoS requirements. Solutions based on global optimization, on the other hand, can handle global constraints, but their poor performance renders them inappropriate for applications with dynamic and real-time requirements. In this paper we address this problem and propose a solution that combines global optimization with local selection techniques to benefit from the advantages of both worlds. The proposed solution consists of two steps: first, we use mixed integer programming (MIP) to find the optimal decomposition of global QoS constraints into local constraints. Second, we use distributed local selection to find the best web services that satisfy these local constraints. The results of experimental evaluation indicate that our approach significantly outperforms existing solutions in terms of computation time while achieving close-to-optimal results.

Tagging systems have become major infrastructures on the Web. They allow users to create tags that annotate and categorize content and share them with other users, very helpful in particular for searching multimedia content. However, as tagging is not constrained by a controlled vocabulary and annotation guidelines, tags tend to be noisy and sparse. Especially new resources annotated by only a few users have often rather idiosyncratic tags that do not reflect a common perspective useful for search. In this paper we introduce an approach based on Latent Dirichlet Allocation (LDA) for recommending tags of resources in order to improve search. Resources annotated by many users and thus equipped with a fairly stable and complete tag set are used to elicit latent topics to which new resources with only a few tags are mapped. Based on this, other tags belonging to a topic can be recommended for the new resource. Our evaluation shows that the approach achieves significantly better precision and recall than the use of association rules, suggested in previous work, and also recommends more specific tags. Moreover, extending resources with these recommended tags significantly improves search for new resources.

In addition to the actual content Web pages consist of navi-gational elements, templates, and advertisements. This boil-erplate text typically is not related to the main content, may deteriorate search precision and thus needs to be detected properly. In this paper, we analyze a small set of shallow text features for classifying the individual text elements in a Web page. We compare the approach to complex, state-of-the-art techniques and show that competitive accuracy can be achieved, at almost no cost. Moreover, we derive a simple and plausible stochastic model for describing the boilerplate creation process. With the help of our model, we also quantify the impact of boilerplate removal to re-trieval performance and show significant improvements over the baseline. Finally, we extend the principled approach by straight-forward heuristics, achieving a remarkable accuracy.

Web service composition enables seamless and dynamic integration of business applications on the web. The performance of the composed application is determined by the performance of the involved web services. Therefore, non-functional, quality of service aspects are crucial for selecting the web services to take part in the composition. Identifying the best candidate web services from a set of functionally-equivalent services is a multi-criteria decision making problem. The selected services should optimize the overall QoS of the composed application, while satisfying all the constraints specified by the client on individual QoS parameters. In this paper, we propose an approach based on the notion of skyline to effectively and efficiently select services for composition, reducing the number of candidate services to be considered. We also discuss how a provider can improve its service to become more competitive and increase its potential of being included in composite applications. We evaluate our approach experimentally using both real and synthetically generated datasets.

This paper introduces a methodology for determining polarity of text within a multilingual framework. The method leverages on lexical resources for sentiment analysis available in English (SentiWordNet). First, a document in a different language than English is translated into English using standard translation software. Then, the translated document is classified according to its sentiment into one of the classes "positive" and "negative". For sentiment classification, a document is searched for sentiment bearing words like adjectives. By means of SentiWordNet, scores for positivity and negativity are determined for these words. An interpretation of the scores then leads to the document polarity. The method is tested for German movie reviews selected from Amazon and is compared to a statistical polarity classifier based on n-grams. The results show that working with standard technology and existing sentiment analysis approaches is a viable approach to sentiment analysis within a multilingual framework.

We present the final results of the spin asymmetries ${A}_{1}$ and the spin structure functions ${g}_{1}$ of the proton and the deuteron in the kinematic range $0.0008<x<0.7$ and $0.2<{Q}^{2}<100{\mathrm{GeV}}^{2}.$ For the determination of ${A}_{1},$ in addition to the usual method which employs inclusive scattering events and includes a large radiative background at low x, we use a new method which minimizes the radiative background by selecting events with at least one hadron as well as a muon in the final state. We find that this hadron method gives smaller errors for $x<0.02,$ so it is combined with the usual method to provide the optimal set of results.

An analysis of the social video sharing platform YouTube reveals a high amount of community feedback through comments for published videos as well as through meta ratings for these comments. In this paper, we present an in-depth study of commenting and comment rating behavior on a sample of more than 6 million comments on 67,000 YouTube videos for which we analyzed dependencies between comments, views, comment ratings and topic categories. In addition, we studied the influence of sentiment expressed in comments on the ratings for these comments using the SentiWordNet thesaurus, a lexical WordNet-based resource containing sentiment annotations. Finally, to predict community acceptance for comments not yet rated, we built different classifiers for the estimation of ratings for these comments. The results of our large-scale evaluations are promising and indicate that community feedback on already rated comments can help to filter new unrated comments or suggest particularly useful but still unrated comments.

The inherent ambiguity of short keyword queries demands for enhanced methods for Web retrieval. In this paper we propose to improve such Web queries by expanding them with terms collected from each user's Personal Information Repository, thus implicitly personalizing the search output. We introduce five broad techniques for generating the additional query keywords by analyzing user data at increasing granularity levels, ranging from term and compound level analysis up to global co-occurrence statistics, as well as to using external thesauri. Our extensive empirical analysis under four different scenarios shows some of these approaches to perform very well, especially on ambiguous queries, producing a very strong increase in the quality of the output rankings. Subsequently, we move this personalized search framework one step further and propose to make the expansion process adaptive to various features of each query. A separate set of experiments indicates the adaptive algorithms to bring an additional statistically significant improvement over the best static expansion approach.

Abstract Climate records over the last millennium place the twentieth-century warming in a longer historical context. Reconstructions of millennial temperatures show a wide range of variability, raising questions about the reliability of currently available reconstruction techniques and the uniqueness of late-twentieth-century warming. A calibration method is suggested that avoids the loss of low-frequency variance. A new reconstruction using this method shows substantial variability over the last 1500 yr. This record is consistent with independent temperature change estimates from borehole geothermal records, compared over the same spatial and temporal domain. The record is also broadly consistent with other recent reconstructions that attempt to fully recover low-frequency climate variability in their central estimate. High variability in reconstructions does not hamper the detection of greenhouse gas–induced climate change, since a substantial fraction of the variance in these reconstructions from the beginning of the analysis in the late thirteenth century to the end of the records can be attributed to external forcing. Results from a detection and attribution analysis show that greenhouse warming is detectable in all analyzed high-variance reconstructions (with the possible exception of one ending in 1925), and that about a third of the warming in the first half of the twentieth century can be attributed to anthropogenic greenhouse gas emissions. The estimated magnitude of the anthropogenic signal is consistent with most of the warming in the second half of the twentieth century being anthropogenic.

On behalf of the organizing committee, it is our great pleasure to welcome you to the 22nd ACM International Conference on Information and Knowledge Management (CIKM 2013) in San Francisco! CIKM is a premier ACM conference in the areas of information retrieval, knowledge management and databases. Since 1992, it has successfully brought together leading researchers and developers from the three communities. The purpose of the conference is to identify challenging problems facing the development of future knowledge and information systems, and to shape future research directions through the publication of high quality applied and theoretical research findings. In CIKM 2013, we continue the tradition of promoting collaboration among multiple areas and providing a leading forum in which experts from academia, industry, and government gather to exchange ideas, research results, and technical developments in multidisciplinary research areas. As one of the world's most recognized conferences in the field, this year CIKM received 848 valid full paper submissions, 233 poster submissions, and 57 demonstration submissions. Among them, we accepted 143 full papers (16.86% acceptance rate), 107 short papers, 81 posters and 21 demos. In addition to regular research tracks, CIKM 2013 features 4 keynote speakers, a panel on Big Data, dedicated Industry events featuring 10 leading industrial practitioners, 10 tutorials from nprestigious researchers and 14 workshops on cutting-edge areas of research. This is a great demonstration of the lively research areas that contribute to the CIKM area. We are proud of our final program and gratefully thank all authors, invited speakers and organizers who chose to contribute their time and research to CIKM 2013. We are honored to present four distinguished keynote speakers to attendees: Ronald Fagin, Lee Giles, Carlos Guestrin, and Alon Halevy. Their valuable, insightful and interdisciplinary talks will guide us to a better understanding of the field.

We have measured the spin-dependent structure function ${g}_{1}^{p}$ in inclusive deep-inelastic scattering of polarized muons off polarized protons, in the kinematic range $0.003<x<0.7$ and $1{\mathrm{GeV}}^{2}<{Q}^{2}<60{\mathrm{GeV}}^{2}.$ A next-to-leading order QCD analysis is used to evolve the measured ${g}_{1}^{p}{(x,Q}^{2})$ to a fixed ${Q}_{0}^{2}.$ The first moment of ${g}_{1}^{p}$ at ${Q}_{0}^{2}=10{\mathrm{GeV}}^{2}$ is ${\ensuremath{\Gamma}}_{1}^{p}=0.136\ifmmode\pm\else\textpm\fi{}0.013$ (stat) $\ifmmode\pm\else\textpm\fi{}0.009$ (syst) $\ifmmode\pm\else\textpm\fi{}0.005$ (evol). This result is below the prediction of the Ellis-Jaffe sum rule by more than two standard deviations. The singlet axial charge ${a}_{0}$ is found to be $0.28\ifmmode\pm\else\textpm\fi{}0.16.$ In the Adler-Bardeen factorization scheme, $\ensuremath{\Delta}g\ensuremath{\simeq}2$ is required to bring \ensuremath{\Delta}\ensuremath{\Sigma} in agreement with the quark-parton model. A combined analysis of all available proton, deuteron, and ${}^{3}\mathrm{He}$ data confirms the Bjorken sum rule.

The Open Directory Project is clearly one of the largest collaborative efforts to manually annotate web pages. This effort involves over 65,000 editors and resulted in metadata specifying topic and importance for more than 4 million web pages. Still, given that this number is just about 0.05 percent of the Web pages indexed by Google, is this effort enough to make a difference? In this paper we discuss how these metadata can be exploited to achieve high quality personalized web search. First, we address this by introducing an additional criterion for web page ranking, namely the distance between a user profile defined using ODP topics and the sets of ODP topics covered by each URL returned in regular web search. We empirically show that this enhancement yields better results than current web search using Google. Then, in the second part of the paper, we investigate the boundaries of biasing PageRank on subtopics of the ODP in order to automatically extend these metadata to the whole web.

Currently, the retrieval of wind fields from synthetic aperture radar (SAR) images suffers from inadequate knowledge of the wind direction. State-of-the-art spectral analysis works fine on open seas, but is limited in spatial resolution. The method described here is based on the local gradients computed with standard image processing algorithms. It handles image features not caused by wind and can be applied to irregularly shaped regions. The new method has already been applied to many images from the European Remote sensing Satellite SARs and RADARSAT-1 ScanSAR, usually supplying reasonable wind fields. The spatial sampling most frequently used was 20 /spl times/ 20 and 10/spl times/10 km/sup 2/. In some cases, samplings down to 1/spl times/1 km/sup 2/ were tested. This paper describes the local gradients method including the filtering of nonwind generated image features and gives some application examples.

The catalytic mechanisms of transition-metal compounds during the hydrogen sorption reaction of magnesium-based hydrides were investigated through relevant experiments. Catalytic activity was found to be influenced by four distinct physico-thermodynamic properties of the transition-metal compound: a high number of structural defects, a low stability of the compound, which however has to be high enough to avoid complete reduction of the transition metal under operating conditions, a high valence state of the transition-metal ion within the compound, and a high affinity of the transition-metal ion to hydrogen. On the basis of these results, further optimization of the selection of catalysts for improving sorption properties of magnesium-based hydrides is possible. In addition, utilization of transition-metal compounds as catalysts for other hydrogen storage materials is considered.

CONTEXT: The exchange of health information on the Internet has been heralded as an opportunity to improve public health surveillance. In a field that has traditionally relied on an established system of mandatory and voluntary reporting of known infectious diseases by doctors and laboratories to governmental agencies, innovations in social media and so-called user-generated information could lead to faster recognition of cases of infectious disease. More direct access to such data could enable surveillance epidemiologists to detect potential public health threats such as rare, new diseases or early-level warnings for epidemics. But how useful are data from social media and the Internet, and what is the potential to enhance surveillance? The challenges of using these emerging surveillance systems for infectious disease epidemiology, including the specific resources needed, technical requirements, and acceptability to public health practitioners and policymakers, have wide-reaching implications for public health surveillance in the 21st century. METHODS: This article divides public health surveillance into indicator-based surveillance and event-based surveillance and provides an overview of each. We did an exhaustive review of published articles indexed in the databases PubMed, Scopus, and Scirus between 1990 and 2011 covering contemporary event-based systems for infectious disease surveillance. FINDINGS: Our literature review uncovered no event-based surveillance systems currently used in national surveillance programs. While much has been done to develop event-based surveillance, the existing systems have limitations. Accordingly, there is a need for further development of automated technologies that monitor health-related information on the Internet, especially to handle large amounts of data and to prevent information overload. The dissemination to health authorities of new information about health events is not always efficient and could be improved. No comprehensive evaluations show whether event-based surveillance systems have been integrated into actual epidemiological work during real-time health events. CONCLUSIONS: The acceptability of data from the Internet and social media as a regular part of public health surveillance programs varies and is related to a circular challenge: the willingness to integrate is rooted in a lack of effectiveness studies, yet such effectiveness can be proved only through a structured evaluation of integrated systems. Issues related to changing technical and social paradigms in both individual perceptions of and interactions with personal health data, as well as social media and other data from the Internet, must be further addressed before such information can be integrated into official surveillance systems.

As decision-making increasingly relies on Machine Learning (ML) and (big) data, the issue of fairness in data-driven Artificial Intelligence (AI) systems is receiving increasing attention from both research and industry. A large variety of fairness-aware machine learning solutions have been proposed which involve fairness-related interventions in the data, learning algorithms and/or model outputs. However, a vital part of proposing new approaches is evaluating them empirically on benchmark datasets that represent realistic and diverse settings. Therefore, in this paper, we overview real-world datasets used for fairness-aware machine learning. We focus on tabular data as the most common data representation for fairness-aware machine learning. We start our analysis by identifying relationships between the different attributes, particularly w.r.t. protected attributes and class attribute, using a Bayesian network. For a deeper understanding of bias in the datasets, we investigate the interesting relationships using exploratory analysis.

Dynamic selection of Web services at runtime is important for building flexible and loosely-coupled service-oriented applications. An abstract description of the required services is provided at design-time, and matching service offers are located at runtime. With the growing number of Web services that provide the same functionality but differ in quality parameters (e.g., availability, response time), a decision needs to be made on which services should be selected such that the user's end-to-end QoS requirements are satisfied. Although very efficient, local selection strategy fails short in handling global QoS requirements. Solutions based on global optimization, on the other hand, can handle global constraints, but their poor performance renders them inappropriate for applications with dynamic and realtime requirements. In this article we address this problem and propose a hybrid solution that combines global optimization with local selection techniques to benefit from the advantages of both worlds. The proposed solution consists of two steps: first, we use mixed integer programming (MIP) to find the optimal decomposition of global QoS constraints into local constraints. Second, we use distributed local selection to find the best Web services that satisfy these local constraints. The results of experimental evaluation indicate that our approach significantly outperforms existing solutions in terms of computation time while achieving close-to-optimal results.

Crowdsourcing is increasingly being used as a means to tackle problems requiring human intelligence. With the ever-growing worker base that aims to complete microtasks on crowdsourcing platforms in exchange for financial gains, there is a need for stringent mechanisms to prevent exploitation of deployed tasks. Quality control mechanisms need to accommodate a diverse pool of workers, exhibiting a wide range of behavior. A pivotal step towards fraud-proof task design is understanding the behavioral patterns of microtask workers. In this paper, we analyze the prevalent malicious activity on crowdsourcing platforms and study the behavior exhibited by trustworthy and untrustworthy workers, particularly on crowdsourced surveys. Based on our analysis of the typical malicious activity, we define and identify different types of workers in the crowd, propose a method to measure malicious activity, and finally present guidelines for the efficient design of crowdsourced surveys.