Jürgen Bock

Automatic reasoner selection using machine learning

The ability to draw logical conclusions in ontologies from explicitly given axioms and facts is one of the key advantages of using semantic technologies. Based on the W3C recommendation of the Web Ontology Language (OWL) a variety of reasoners have been developed for this task. Different language profiles, reasoning algorithms, and special-purpose optimisation techniques have brought up reasoners with various strengths and weaknesses. Selecting the most suitable reasoner for a given reasoning scenario thus is a challenge. This paper presents an automatic reasoner selection approach based on machine learning techniques. The most important ontology and query features are identified and used for learning a model that can be used to predict the best performing reasoner for a given request. The approach is implemented as a strategy in a reasoning broker framework called HERAKLES. Using a training set consisting of 187 real-world ontologies found on the Internet, we evaluated four different machine learning techniques. The results show that a machine learning based reasoner selection strategy can predict the best performing reasoner for a given reasoning request with more than 77% accuracy.

Advancement of MOOCs with Learning Pathways

In this paper, educational and technical challenges in the field of Massive Open Online Courses (MOOCs), such as cultural adaptation, consideration of learning habits or the efficient construction of educational content, are outlined. We argue that learning pathways in combination with learner-centered metadata are optimal methods to meet those challenges and hence optimize learning efficiency and learning experience substantially by creating personalized learning recommendations and feedback. This is illustrated from a conceptual and technological perspective, as is currently in development in the EU-project INTUITEL. We believe that this approach not only opens up new possibilities for MOOCs, but also provides a variety of new dimensions for eLearning in general.

A Case Study on the Use of Semantic Web Technologies for Learner Guidance

Personalized learning pathways have been advocated by didactic experts to overcome the problem of disorientation and information overload in technology enhanced learning (TEL). They are not only relevant for providing user-adaptive navigational support, but can also be used for composing learning objects into new personalized courses (sequencing and assembly). In this paper we investigate, how Semantic Web technologies can effectively support these tasks, based on a proper representation of learning objects and courses according to didactic requirements. We claim that both eLearning tasks, adaptive navigation and course assembly, call for a representational model that can capture the syntax and semantics of learning pathways adequately. In particular: (1) a new type of navigation that takes into account ordering information and the hierarchical structure of an eLearning course complemented with adaptive constraints; (2) closely tied to it, a semantic layer to guarantee interoperability and validation of the correctness of the learning pathway descriptions. We investigate to what extend Semantic Web Languages like RDF/S and OWL are expressive enough to handle different aspects of learning pathways. While both share a structural similarity with DAGs, only OWL ontologies - formally underpinned by description logics (DLs) - are expressive enough to validate the correctness of the data and infer semantically related learning resources on the pathway. For tasks that are more related to the syntax of learning pathways, in particular navigation similar to a guided tour, we test the time efficiency on various synthetic OWL ontologies using the HermiT reasoner. Experimental results show that the course structure and the density of the knowledge graph impact on the performance. We claim that in a dynamically changing environment, where the computation of reachability of a vertex is computed on demand at run-time, OWL-based reasoning does not scale up well. Using a real-world case study from the eLearning domain, we compare an OWL 2 DL implementation with an equivalent graph algorithm implementation with respect to time efficiency.

Reasoning Brokerage: New Reasoning Strategies

Once a formal representation of data is available an important issue is to infer additional aspects out of this knowledge base. But according to the type of representation scheme chosen, different techniques can be applied or gain better or more accurate results. A reasoning broker system offers the possibility to apply strategies for selecting the best reasoning system, or for letting run different reasoners in parallel. In this article a reasoning broker system enabling the usage and integration of remote reasoners is presented. Additionally, the new reasoning capability of anytime reasoning has been developed and integrated into the reasoning broker.