الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 147 |  |  |
| | | from | 147 | | |
Abstract
The profusion of existing ontologies in different domains has made reusing ontologies a best practice when developing new ontologies. The ontology reuse process reduces the expensive ontology development cost, in terms of time and effort, and supports semantic interoperability. Existing ontology development tools do not assist in the recommendation of ontologies or their concepts to be reused. Also, existing ontology recommendation tools could suggest whole ontologies covering the set of input keywords without referring to which parts of them (e.g., concepts) can be reused.
In this thesis, we first propose a novel context-based semantic matching algorithm that identifies pairs of equivalent concepts in terms of the context of their ontologies. The proposed algorithm takes the correspondences obtained by existing matchmakers and checks the correctness of those mappings based on the semantics encoded in the given ontologies. Then, an effective ontology recommendation system is proposed which helps the user in the iterative ontology development and reuse process. We develop a utility-based recommender system that uses the Multi-Attribute Utility Theory (MAUT). MAUT considers that every alternative (item) has a bundle of attributes, and it helps the decision- maker to choose the alternative that yields the greatest utility. The system allows the user to provide his explicit preferences about the preferred new ontology and iteratively guides the user to parts from existing ontologies matching his preferences for reuse. To personalize each user’s recommendations, the system builds a user preference model for each user using his provided explicit preferences. We utilize the reinforcement learning technique to learn the user’s implicit preferences. It records the
user’s feedback (i.e., previous ontology selections) to detect the user’s implicit preferences which could indicate if the user has a change of interests. If so, the system should optimize and balance between quickly updating the user’s preference model or waiting for a few other interactions with the user to be sure of this change in the user’s interests.
To demonstrate the effectiveness of our ontology recommendation system a prototype was implemented and used to conduct a user-based evaluation. Finally, we evaluate the performance of our proposed recommender system using different evaluation metrics (e.g., mean absolute error, mean reciprocal rank, mean average precision and normalized discounted cumulative gain) which prove its effectiveness in generating and ranking ontology recommendations to the user.