Fine-Grained Forgetting for Expressive Description Logics: Deductive, Semantic, and Query Forgetting in One Framework

Abstract

The aim of this thesis is to study different notions of forgetting in the context of the description logic ALC. Forgetting is an engineering task that eliminates subsets of the vocabulary from an ontology, and preserves the information relative to the remainder of the vocabulary. Forgetting offers solutions to many important applications such as ontology creation, reuse, debugging, information hiding, abduction, and agent communication. The main contribution of this thesis is a fine-grained forgetting framework, that is capable of forgetting concept names from ALC ontologies, and obtaining representations of the deductive, query and semantic forgetting views. The framework also computes sets of axioms that indicate the differences between the contents of these forgetting views, which allows for a new understanding of the different forms of forgetting and the relationships between them. Moreover, we develop an algorithm whereby the content of the forgetting views can be enhanced with some of these axioms according to user requirements in a way that reveals a spectrum of fine-grained forgetting views in-between the deductive, the query, and the semantic forgetting views. These views can be used to address more applications and use cases than currently anticipated. Another contribution of this thesis is a semantic forgetting method for ALC ontologies that preserves the syntactic structure of the input ontology, and improves the readability of the forgetting view. The deductive customization of the forgetting framework has been evaluated against the forgetting system Lethe in six different evaluations. In most experiments our deductive customization successfully terminated in less than half of the time consumed by Lethe, despite computing more information. The semantic forgetting method was evaluated on its own, and in comparison with the semantic forgetting system Fame. In two evaluations out of three, our semantic forgetting method terminated successfully in less than 42% of the time consumed by Fame.

Date of Award	1 Aug 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Lucas Cordeiro (Supervisor) & Renate Schmidt (Supervisor)