Conventional closed-corpus adaptive information systems control limited sets of documents in predefined domains and cannot provide access to the external content. Such restrictions contradict the requirements of today, when most of the information systems are implemented in the open document space of the World Wide Web and are expected to operate on the open-corpus content. In order to provide personalized access to open-corpus documents, an adaptive system should be able to maintain modeling of new documents in terms of domain knowledge automatically and dynamically. This dissertation explores the problem of open-corpus personalization and semantic modeling of open-corpus content in the context of e-Learning.

Information on the World Wide Web is not without structure. Many collections of online instructional material (tutorials, electronic books, digital libraries, etc.) have been provided with implicit knowledge models encoded in form of tables of content, indexes, headers of chapters, links between pages, and different styles of text fragments. The main dissertation approach tries to leverage this layer of hidden semantics by extracting and representing it as coarse-grained models of content collections. A central domain ontology is used to maintain overlay modeling of students’ knowledge and serves as a reference point for multiple collections of external instructional material. In order to establish the link between the ontology and the open-corpus content models a special ontology mapping algorithm has been developed.

The proposed approach has been applied in the Ontology-based Open-corpus Personalization Service that recommends and adaptively annotates online reading material. The domain of Java programming has been chosen for the proof-of-concept implementation. A controlled experiment has been organized to evaluate the developed adaptive system and the proposed approach overall. The results of the evaluation have demonstrated several significant learning effects of the implemented open-corpus personalization. The analysis of log-based data has also shown that the open-corpus version of the system is capable of providing personalization of similar quality to the close-corpus one. Such results indicate that the proposed approach successfully supports open-corpus personalization for e-Learning. Further research is required to verify if the approach remains effective in other subject domains and with other types of instructional content.