Refine
Document Type
- Doctoral Thesis (11) (remove)
Keywords
- Data Mining (1)
- Enhanced Representation (1)
- Eye Tracking (1)
- Formale Ontologie (1)
- GazeTheWeb (1)
- Information Retrieval (1)
- Latent Negative (1)
- Link Prediction (1)
- Maschinelles Lernen (1)
- Modellgetriebene Entwicklung (1)
Institute
As a multilingual system,Wikipedia provides many challenges for academics and engineers alike. One such challenge is cultural contextualisation of Wikipedia content, and the lack of approaches to effectively quantify it. Additionally, what seems to lack is the intent of establishing sound computational practices and frameworks for measuring cultural variations in the data. Current approaches seem to mostly be dictated by the data availability, which makes it difficult to apply them in other contexts. Another common drawback is that they rarely scale due to a significant qualitative or translation effort. To address these limitations, this thesis develops and tests two modular quantitative approaches. They are aimed at quantifying culture-related phenomena in systems which rely on multilingual user-generated content. In particular, they allow to: (1) operationalise a custom concept of culture in a system; (2) quantify and compare culture-specific content- or coverage biases in such a system; and (3) map a large scale landscape of shared cultural interests and focal points. Empirical validation of these approaches is split into two parts. First, an approach to mapping Wikipedia communities of shared co-editing interests is validated on two large Wikipedia datasets comprising multilateral geopolitical and linguistic editor communities. Both datasets reveal measurable clusters of consistent co-editing interest, and computationally confirm that these clusters correspond to existing colonial, religious, socio economic, and geographical ties. Second, an approach to quantifying content differences is validated on a multilingual Wikipedia dataset, and a multi-platform (Wikipedia and Encyclopedia Britannica) dataset. Both are limited to a selected knowledge domain of national history. This analysis allows, for the first time on the large scale, to quantify and visualise the distribution of historical focal points in the articles on national histories. All results are cross-validated either by domain experts, or external datasets.
Main thesis contributions. This thesis: (1) presents an effort to formalise the process of measuring cultural variations in user-generated data; (2) introduces and tests two novel approaches to quantifying cultural contextualisation in multilingual data; (3) synthesises a valuable overview of literature on defining and quantifying culture; (4) provides important empirical insights on the effect of culture on Wikipedia content and coverage; demonstrates that Wikipedia is not contextfree, and these differences should not be treated as noise, but rather, as an important feature of the data. (5) makes practical service contributions through sharing data and visualisations.
Folksonomies are Web 2.0 platforms where users share resources with each other. Furthermore, they can assign keywords (called tags) to the resources for categorizing and organizing the resources. Numerous types of resources like websites (Delicious), images (Flickr), and videos (YouTube) are supported by different folksonomies. The folksonomies are easy to use and thus attract the attention of millions of users. Together with the ease they offer, there are also some problems. This thesis addresses different problems of folksonomies and proposes solutions for these problems. The first problem occurs when users search for relevant resources in folksonomies. Often, the users are not able to find all relevant resources because they don't know which tags are relevant. The second problem is assigning tags to resources. Although many folksonomies (like Delicious) recommend tags for the resources, other folksonomies (like Flickr) do not recommend any tags. Tag recommendation helps the users to easily tag their resources. The third problem is that tags and resources are lacking semantics. This leads for example to ambiguous tags. The tags are lacking semantics because they are freely chosen keywords. The automatic identification of the semantics of tags and resources helps in reducing problems that arise from this freedom of the users in choosing the tags. This thesis proposes methods which exploit semantics to address the problems of search, tag recommendation, and the identification of tag semantics. The semantics are discovered from a variety of sources. In this thesis, we exploit web search engines, online social communities and the co-occurrences of tags as sources of semantics. Using different sources for discovering semantics reduces the efforts to build systems which solve the problems mentioned earlier. This thesis evaluates the proposed methods on a large scale data set. The evaluation results suggest that it is possible to exploit the semantics for improving search, recommendation of tags, and automatic identification of the semantics of tags and resources.
The publication of freely available and machine-readable information has increased significantly in the last years. Especially the Linked Data initiative has been receiving a lot of attention. Linked Data is based on the Resource Description Framework (RDF) and anybody can simply publish their data in RDF and link it to other datasets. The structure is similar to the World Wide Web where individual HTML documents are connected with links. Linked Data entities are identified by URIs which are dereferenceable to retrieve information describing the entity. Additionally, so called SPARQL endpoints can be used to access the data with an algebraic query language (SPARQL) similar to SQL. By integrating multiple SPARQL endpoints it is possible to create a federation of distributed RDF data sources which acts like one big data store.
In contrast to the federation of classical relational database systems there are some differences for federated RDF data. RDF stores are accessed either via SPARQL endpoints or by resolving URIs. There is no coordination between RDF data sources and machine-readable meta data about a source- data is commonly limited or not available at all. Moreover, there is no common directory which can be used to discover RDF data sources or ask for sources which offer specific data. The federation of distributed and linked RDF data sources has to deal with various challenges. In order to distribute queries automatically, suitable data sources have to be selected based on query details and information that is available about the data sources. Furthermore, the minimization of query execution time requires optimization techniques that take into account the execution cost for query operators and the network communication overhead for contacting individual data sources. In this thesis, solutions for these problems are discussed. Moreover, SPLENDID is presented, a new federation infrastructure for distributed RDF data sources which uses optimization techniques based on statistical information.
The Web is an essential component of moving our society to the digital age. We use it for communication, shopping, and doing our work. Most user interaction in the Web happens with Web page interfaces. Thus, the usability and accessibility of Web page interfaces are relevant areas of research to make the Web more useful. Eye tracking is a tool that can be helpful in both areas, performing usability testing and improving accessibility. It can be used to understand users' attention on Web pages and to support usability experts in their decision-making process. Moreover, eye tracking can be used as an input method to control an interface. This is especially useful for people with motor impairment, who cannot use traditional input devices like mouse and keyboard. However, interfaces on Web pages become more and more complex due to dynamics, i.e., changing contents like animated menus and photo carousels. We need general approaches to comprehend dynamics on Web pages, allowing for efficient usability analysis and enjoyable interaction with eye tracking. In the first part of this thesis, we report our work on improving gaze-based analysis of dynamic Web pages. Eye tracking can be used to collect the gaze signals of users, who browse a Web site and its pages. The gaze signals show a usability expert what parts in the Web page interface have been read, glanced at, or skipped. The aggregation of gaze signals allows a usability expert insight into the users' attention on a high-level, before looking into individual behavior. For this, all gaze signals must be aligned to the interface as experienced by the users. However, the user experience is heavily influenced by changing contents, as these may cover a substantial portion of the screen. We delineate unique states in Web page interfaces including changing contents, such that gaze signals from multiple users can be aggregated correctly. In the second part of this thesis, we report our work on improving the gaze-based interaction with dynamic Web pages. Eye tracking can be used to retrieve gaze signals while a user operates a computer. The gaze signals may be interpreted as input controlling an interface. Nowadays, eye tracking as an input method is mostly used to emulate mouse and keyboard functionality, hindering an enjoyable user experience. There exist a few Web browser prototypes that directly interpret gaze signals for control, but they do not work on dynamic Web pages. We have developed a method to extract interaction elements like hyperlinks and text inputs efficiently on Web pages, including changing contents. We adapt the interaction with those elements for eye tracking as the input method, such that a user can conveniently browse the Web hands-free. Both parts of this thesis conclude with user-centered evaluations of our methods, assessing the improvements in the user experience for usability experts and people with motor impairment, respectively.
The Web contains some extremely valuable information; however, often poor quality, inaccurate, irrelevant or fraudulent information can also be found. With the increasing amount of data available, it is becoming more and more difficult to distinguish truth from speculation on the Web. One of the most, if not the most, important criterion used to evaluate data credibility is the information source, i.e., the data origin. Trust in the information source is a valuable currency users have to evaluate such data. Data popularity, recency (or the time of validity), reliability, or vagueness ascribed to the data may also help users to judge the validity and appropriateness of information sources. We call this knowledge derived from the data the provenance of the data. Provenance is an important aspect of the Web. It is essential in identifying the suitability, veracity, and reliability of information, and in deciding whether information is to be trusted, reused, or even integrated with other information sources. Therefore, models and frameworks for representing, managing, and using provenance in the realm of Semantic Web technologies and applications are critically required. This thesis highlights the benefits of the use of provenance in different Web applications and scenarios. In particular, it presents management frameworks for querying and reasoning in the Semantic Web with provenance, and presents a collection of Semantic Web tools that explore provenance information when ranking and updating caches of Web data. To begin, this thesis discusses a highly exible and generic approach to the treatment of provenance when querying RDF datasets. The approach re-uses existing RDF modeling possibilities in order to represent provenance. It extends SPARQL query processing in such a way that given a SPARQL query for data, one may request provenance without modifying it. The use of provenance within SPARQL queries helps users to understand how RDF facts arederived, i.e., it describes the data and the operations used to produce the derived facts. Turning to more expressive Semantic Web data models, an optimized algorithm for reasoning and debugging OWL ontologies with provenance is presented. Typical reasoning tasks over an expressive Description Logic (e.g., using tableau methods to perform consistency checking, instance checking, satisfiability checking, and so on) are in the worst case doubly exponential, and in practice are often likewise very expensive. With the algorithm described in this thesis, however, one can efficiently reason in OWL ontologies with provenance, i.e., provenance is efficiently combined and propagated within the reasoning process. Users can use the derived provenance information to judge the reliability of inferences and to find errors in the ontology. Next, this thesis tackles the problem of providing to Web users the right content at the right time. The challenge is to efficiently rank a stream of messages based on user preferences. Provenance is used to represent preferences, i.e., the user defines his preferences over the messages' popularity, recency, etc. This information is then aggregated to obtain a joint ranking. The aggregation problem is related to the problem of preference aggregation in Social Choice Theory. The traditional problem formulation of preference aggregation assumes a I fixed set of preference orders and a fixed set of domain elements (e.g. messages). This work, however, investigates how an aggregated preference order has to be updated when the domain is dynamic, i.e., the aggregation approach ranks messages 'on the y' as the message passes through the system. Consequently, this thesis presents computational approaches for online preference aggregation that handle the dynamic setting more efficiently than standard ones. Lastly, this thesis addresses the scenario of caching data from the Linked Open Data (LOD) cloud. Data on the LOD cloud changes frequently and applications relying on that data - by pre-fetching data from the Web and storing local copies of it in a cache - need to continually update their caches. In order to make best use of the resources (e.g., network bandwidth for fetching data, and computation time) available, it is vital to choose a good strategy to know when to fetch data from which data source. A strategy to cope with data changes is to check for provenance. Provenance information delivered by LOD sources can denote when the resource on the Web has been changed last. Linked Data applications can benefit from this piece of information since simply checking on it may help users decide which sources need to be updated. For this purpose, this work describes an investigation of the availability and reliability of provenance information in the Linked Data sources. Another strategy for capturing data changes is to exploit provenance in a time-dependent function. Such a function should measure the frequency of the changes of LOD sources. This work describes, therefore, an approach to the analysis of data dynamics, i.e., the analysis of the change behavior of Linked Data sources over time, followed by the investigation of different scheduling update strategies to keep local LOD caches up-to-date. This thesis aims to prove the importance and benefits of the use of provenance in different Web applications and scenarios. The exibility of the approaches presented, combined with their high scalability, make this thesis a possible building block for the Semantic Web proof layer cake - the layer of provenance knowledge.
The semantic web and model-driven engineering are changing the enterprise computing paradigm. By introducing technologies like ontologies, metadata and logic, the semantic web improves drastically how companies manage knowledge. In counterpart, model-driven engineering relies on the principle of using models to provide abstraction, enabling developers to concentrate on the system functionality rather than on technical platforms. The next enterprise computing era will rely on the synergy between both technologies. On the one side, ontology technologies organize system knowledge in conceptual domains according to its meaning. It addresses enterprise computing needs by identifying, abstracting and rationalizing commonalities, and checking for inconsistencies across system specifications. On the other side, model-driven engineering is closing the gap among business requirements, designs and executables by using domain-specific languages with custom-built syntax and semantics. In this scenario, the research question that arises is: What are the scientific and technical results around ontology technologies that can be used in model-driven engineering and vice versa? The objective is to analyze approaches available in the literature that involve both ontologies and model-driven engineering. Therefore, we conduct a literature review that resulted in a feature model for classifying state-of-the-art approaches. The results show that the usage of ontologies and model-driven engineering together have multiple purposes: validation, visual notation, expressiveness and interoperability. While approaches involving both paradigms exist, an integrated approach for UML class-based modeling and ontology modeling is lacking so far. Therefore, we investigate the techniques and languages for designing integrated models. The objective is to provide an approach to support the design of integrated solutions. Thus, we develop a conceptual framework involving the structure and the notations of a solution to represent and query software artifacts using a combination of ontologies and class-based modeling. As proof of concept, we have implemented our approach as a set of open source plug-ins -- the TwoUse Toolkit. The hypothesis is that a combination of both paradigms yields improvements in both fields, ontology engineering and model-driven engineering. For MDE, we investigate the impact of using features of the Web Ontology Language in software modeling. The results are patterns and guidelines for designing ontology-based information systems and for supporting software engineers in modeling software. The results include alternative ways of describing classes and objects and querying software models and metamodels. Applications show improvements on changeability and extensibility. In the ontology engineering domain, we investigate the application of techniques used in model-driven engineering to fill the abstraction gap between ontology specification languages and programming languages. The objective is to provide a model-driven platform for supporting activities in the ontology engineering life cycle. Therefore, we study the development of core ontologies in our department, namely the core ontology for multimedia (COMM) and the multimedia metadata ontology. The results are domain-specific languages that allow ontology engineers to abstract from implementation issues and concentrate on the ontology engineering task. It results in increasing productivity by filling the gap between domain models and source code.
Semantic Web technologies have been recognized to be key for the integration of distributed and heterogeneous data sources on the Web, as they provide means to define typed links between resources in a dynamic manner and following the principles of dataspaces. The widespread adoption of these technologies in the last years led to a large volume and variety of data sets published as machine-readable RDF data, that once linked constitute the so-called Web of Data. Given the large scale of the data, these links are typically generated by computational methods that given a set of RDF data sets, analyze their content and identify the entities and schema elements that should be connected via the links. Analogously to any other kind of data, in order to be truly useful and ready to be consumed, links need to comply with the criteria of high quality data (e.g., syntactically and semantically accurate, consistent, up-to-date). Despite the progress in the field of machine learning, human intelligence is still essential in the quest for high quality links: humans can train algorithms by labeling reference examples, validate the output of algorithms to verify their performance on a data set basis, as well as augment the resulting set of links. Humans —especially expert humans, however, have limited availability. Hence, extending data quality management processes from data owners/publishers to a broader audience can significantly improve the data quality management life cycle.
Recent advances in human computation and peer-production technologies opened new avenues for human-machine data management techniques, allowing to involve non-experts in certain tasks and providing methods for cooperative approaches. The research work presented in this thesis takes advantage of such technologies and investigates human-machine methods that aim at facilitating link quality management in the Semantic Web. Firstly, and focusing on the dimension of link accuracy, a method for crowdsourcing ontology alignment is presented. This method, also applicable to entities, is implemented as a complement to automatic ontology alignment algorithms. Secondly, novel measures for the dimension of information gain facilitated by the links are introduced. These entropy-centric measures provide data managers with information about the extent the entities in the linked data set gain information in terms of entity description, connectivity and schema heterogeneity. Thirdly, taking Wikidata —the most successful case of a linked data set curated, linked and maintained by a community of humans and bots— as a case study, we apply descriptive and predictive data mining techniques to study participation inequality and user attrition. Our findings and method can help community managers make decisions on when/how to intervene with user retention plans. Lastly, an ontology to model the history of crowd contributions across marketplaces is presented. While the field of human-machine data management poses complex social and technical challenges, the work in this thesis aims to contribute to the development of this still emerging field.
Through the increasing availability of access to the web, more and more interactions between people take place in online social networks, such as Twitter or Facebook, or sites where opinions can be exchanged. At the same time, knowledge is made openly available for many people, such as by the biggest collaborative encyclopedia Wikipedia and diverse information in Internet forums and on websites. These two kinds of networks - social networks and knowledge networks - are highly dynamic in the sense that the links that contain the important information about the relationships between people or the relations between knowledge items are frequently updated or changed. These changes follow particular structural patterns and characteristics that are far less random than expected.
The goal of this thesis is to predict three characteristic link patterns for the two network types of interest: the addition of new links, the removal of existing links and the presence of latent negative links. First, we show that the prediction of link removal is indeed a new and challenging problem. Even if the sociological literature suggests that reasons for the formation and resolution of ties are often complementary, we show that the two respective prediction problems are not. In particular, we show that the dynamics of new links and unlinks lead to the four link states of growth, decay, stability and instability. For knowledge networks we show that the prediction of link changes greatly benefits from the usage of temporal information; the timestamp of link creation and deletion events improves the prediction of future link changes. For that, we present and evaluate four temporal models that resemble different exploitation strategies. Focusing on directed social networks, we conceptualize and evaluate sociological constructs that explain the formation and dissolution of relationships between users. Measures based on information about past relationships are extremely valuable for predicting the dissolution of social ties. Hence, consistent for knowledge networks and social networks, temporal information in a network greatly improves the prediction quality. Turning again to social networks, we show that negative relationship information such as distrust or enmity can be predicted from positive known relationships in the network. This is particularly interesting in networks where users cannot label their relationships to other users as negative. For this scenario we show how latent negative relationships can be predicted.
This thesis presents novel approaches for integrating context information into probabilistic models. Data from social media is typically associated with metadata, which includes context information such as timestamps, geographical coordinates or links to user profiles. Previous studies showed the benefits of using such context information in probabilistic models, e.g.\ improved predictive performance. In practice, probabilistic models which account for context information still play a minor role in data analysis. There are multiple reasons for this. Existing probabilistic models often are complex, the implementation is difficult, implementations are not publicly available, or the parameter estimation is computationally too expensive for large datasets. Additionally, existing models are typically created for a specific type of content and context and lack the flexibility to be applied to other data.
This thesis addresses these problems by introducing a general approach for modelling multiple, arbitrary context variables in probabilistic models and by providing efficient inference schemes and implementations.
In the first half of this thesis, the importance of context and the potential of context information for probabilistic modelling is shown theoretically and in practical examples. In the second half, the example of topic models is employed for introducing a novel approach to context modelling based on document clusters and adjacency relations in the context space. They can cope with areas of sparse observations and These models allow for the first time the efficient, explicit modelling of arbitrary context variables including cyclic and spherical context (such as temporal cycles or geographical coordinates). Using the novel three-level hierarchical multi-Dirichlet process presented in this thesis, the adjacency of ontext clusters can be exploited and multiple contexts can be modelled and weighted at the same time. Efficient inference schemes are derived which yield interpretable model parameters that allow analyse the relation between observations and context.
The distributed setting of RDF stores in the cloud poses many challenges. One such challenge is how the data placement on the compute nodes can be optimized to improve the query performance. To address this challenge, several evaluations in the literature have investigated the effects of existing data placement strategies on the query performance. A common drawback in theses evaluations is that it is unclear whether the observed behaviors were caused by the data placement strategies (if different RDF stores were evaluated as a whole) or reflect the behavior in distributed RDF stores (if cloud processing frameworks like Hadoop MapReduce are used for the evaluation). To overcome these limitations, this thesis develops a novel benchmarking methodology for data placement strategies that uses a data-placement-strategy-independent distributed RDF store to analyze the effect of the data placement strategies on query performance.
With this evaluation methodology the frequently used data placement strategies have been evaluated. This evaluation challenged the commonly held belief that data placement strategies that emphasize local computation, such as minimal edge-cut cover, lead to faster query executions. The results indicate that queries with a high workload may be executed faster on hash-based data placement strategies than on, e.g., minimal edge-cut covers. The analysis of the additional measurements indicates that vertical parallelization (i.e., a well-distributed workload) may be more important than horizontal containment (i.e., minimal data transport) for efficient query processing.
Moreover, to find a data placement strategy with a high vertical parallelization, the thesis tests the hypothesis that collocating small connected triple sets on the same compute node while balancing the amount of triples stored on the different compute nodes leads to a high vertical parallelization. Specifically, the thesis proposes two such data placement strategies. The first strategy called overpartitioned minimal edge-cut cover was found in the literature and the second strategy is the newly developed molecule hash cover. The evaluation revealed a balanced query workload and a high horizontal containment, which lead to a high vertical parallelization. As a result these strategies showed a better query performance than the frequently used data placement strategies.