Filtern
Erscheinungsjahr
- 2016 (4) (entfernen)
Dokumenttyp
- Dissertation (2)
- Habilitation (1)
- Masterarbeit (1)
Schlagworte
- Articles for Deletion (1)
- Function Words (1)
- I-messages (1)
- Wikipedia (1)
- You-messages (1)
- description logic (1)
- reasoning (1)
Institut
One of the main goals of the artificial intelligence community is to create machines able to reason with dynamically changing knowledge. To achieve this goal, a multitude of different problems have to be solved, of which many have been addressed in the various sub-disciplines of artificial intelligence, like automated reasoning and machine learning. The thesis at hand focuses on the automated reasoning aspects of these problems and address two of the problems which have to be overcome to reach the afore-mentioned goal, namely 1. the fact that reasoning in logical knowledge bases is intractable and 2. the fact that applying changes to formalized knowledge can easily introduce inconsistencies, which leads to unwanted results in most scenarios.
To ease the intractability of logical reasoning, I suggest to adapt a technique called knowledge compilation, known from propositional logic, to description logic knowledge bases. The basic idea of this technique is to compile the given knowledge base into a normal form which allows to answer queries efficiently. This compilation step is very expensive but has to be performed only once and as soon as the result of this step is used to answer many queries, the expensive compilation step gets worthwhile. In the thesis at hand, I develop a normal form, called linkless normal form, suitable for knowledge compilation for description logic knowledge bases. From a computational point of view, the linkless normal form has very nice properties which are introduced in this thesis.
For the second problem, I focus on changes occurring on the instance level of description logic knowledge bases. I introduce three change operators interesting for these knowledge bases, namely deletion and insertion of assertions as well as repair of inconsistent instance bases. These change operators are defined such that in all three cases, the resulting knowledge base is ensured to be consistent and changes performed to the knowledge base are minimal. This allows us to preserve as much of the original knowledge base as possible. Furthermore, I show how these changes can be applied by using a transformation of the knowledge base.
For both issues I suggest to adapt techniques successfully used in other logics to get promising methods for description logic knowledge bases.
This habilitation thesis collects works addressing several challenges on handling uncertainty and inconsistency in knowledge representation. In particular, this thesis contains works which introduce quantitative uncertainty based on probability theory into abstract argumentation frameworks. The formal semantics of this extension is investigated and its application for strategic argumentation in agent dialogues is discussed. Moreover, both the computational as well as the meaningfulness of approaches to analyze inconsistencies, both in classical logics as well as logics for uncertain reasoning is investigated. Finally, this thesis addresses the implementation challenges for various kinds of knowledge representation formalisms employing any notion of inconsistency tolerance or uncertainty.
“Did I say something wrong?” A word-level analysis of Wikipedia articles for deletion discussions
(2016)
Diese Arbeit beschäftigt sich damit, linguistische Erkenntnisse auf Wortebene über schriftlichen Diskussionen zu gewinnen. Die Unterscheidung zwischen Botschaften, welche sich förderlich auf Diskussionen auswirken und jene, welche diese unterbrechen, spielte dabei eine besondere Rolle. Hierbei lag ein Schwerpunkt darauf, zu ermitteln, ob Ich- und Du-Botschaften charakteristisch für die beiden Kommunikationsarten sind. Diese Botschaften sind über Jahre hinweg zu Empfehlungen für erfolgreiche Kommunikation avanciert. Ihre zugeschriebene Wirkung wurde zwar mehrfach bestätigt, jedoch geschah dies stets in kleineren Studien. Deshalb wurde in dieser Arbeit mithilfe der Löschdiskussionen der englischen Wikipedia und der Liste gesperrter Nutzer eine vollautomatische Erstellung eines annotierten Datensatzes entwickelt. Dabei wurden Diskussionsbotschaften entweder als förderlich oder schädlich für einen konstruktiven Diskussionsverlauf markiert. Dieser Datensatz wurde anschließend im Rahmen einer binären Klassifikation verwendet, um charakteristische Worte für die beiden Kommunikationsarten zu bestimmen. Es wurde zudem untersucht, ob anhand von Synsemantika (auch bekannt als Funktionswörter) wie Pronomen oder Konjunktionen eine Entscheidung über die Kommunikationsart einer Botschaft getroffen werden kann. Du-Botschaften wurden, übereinstimmend mit ihrer zugeschriebenen negativen Auswirkung auf Kommunikation, als schädlich in den durchgeführten Untersuchungen identifiziert. Entgegen der zugeschriebenen positiven Auswirkung von Ich-Botschaften, wurde bei diesen ebenfalls eine schädlich Wirkung festgestellt. Eine klare Aussage über die Relevanz von Synsemantika konnte anhand der Ergebnisse nicht getroffen werden. Weitere charakteristische Worte konnten nicht festgestellt werden. Die Ergebnisse deuten darauf hin, dass ein anderes Modell textliche Diskussionen potentiell besser abbilden könnte.
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.