In computer and information sciences, an ontology is a formal representation of the concepts, individuals, and their properties and relationships within a specific domain of discourse in the real-world. Forgetting is a non-standard reasoning operation that seeks to form a new ontology from an existing one by eliminating a set of concept and role names from the existing ontology while preserving all logical consequences up to the remaining names using a set of inference rules.
Existing forgetting methods include LETHE and UI-FAME. LETHE is a resolution-based method for eliminating concept and role names from ALCH-TBoxes, while UI-FAME is a hybrid method using both resolution and Ackermann’s Lemma to eliminate concept and role names from ALC-TBoxes. Ackermann’s Lemma allows concept names to be eliminated in a faster and easier way, and thus UI-FAME has better overall performance than LETHE. Because of the incompleteness of forgetting, “success rates” has become a key indicator of assessing the performance of forgetting methods.
This work is concerned with a refinement and improvement of the forgetting method used by UI-FAME. The new method allows more types of forgetting problems in ALC-TBoxes. On the one hand, we have solved a class of cyclic dependency problem by extending the inference rules used by the old forgetting method, and on the other hand, we have exploited the ontology modularization technique, which is a polynomial-time operation, to accelerate the forgetting process. We have implemented a prototype of the improved method in Java, and compared it with LETHE and the old UI-FAME, with the results showing that: the extension of inference rules promote the success rate by 11% and the effect of modularization is moderate when the data in small scale while the average speed is accelerated by 106% when the scale of ontology is bigger than 500.