Recon: Ontology Recommendation For Structureless Queries

Ontology search is becoming increasingly important as the number of available ontologies on the Web steadily increases. Ontology recommendation is done by analyzing various properties of ontologies, such as syntax, structure, and usage, to find and recommend high-quality matches for a user defined query. Only a few ontology libraries and search engines facilitate this task for a user who wants to find an ontology that models all or some of the concepts she is looking for. In this paper, we introduce RecOn, a framework that helps users in finding the best matching ontologies to a multi-keyword query. Our approach recommends a ranked list of relevant ontologies using metrics that include the matching cost of a user query to an ontology, an ontology's informativeness, and its popularity. Based on these metrics two versions of RecOn are implemented: RecOnln, where the metrics are combined in a linear model to find the relevance score of an ontology to a query, and RecOnopt that formalizes ontology recommendation as an optimization problem to recommend ontologies to the user that are as informative and popular as possible while incurring the least matching costs. We compare both versions of RecOn with the state-of-the-art approach in ontology ranking by conducting a user study over the CBRBench ontology collection. Our experimental results show that both versions of the proposed approach are promising: they identify high-quality matches for keyword queries over real-life ontologies, and outperform the state-of-the-art ranking method significantly regarding effectiveness, while RecOnopt is more effective than RecOnln. We further test the scalability of our proposed approach, and results show RecOnopt is more efficient than RecOnln.