Towards Practical Ontology-based Data Access for Existential Rules

Abstract

Nowadays, knowledge reasoning is gaining more attentions in Artificial Intelligence, which stimulates the development of modern information systems. As the key ingredient of new generation of information systems, ontology-based data access (OBDA), which employs domain-specific knowledge provided by an ontology to reason over data, has received considerable attention in recent years. However, current research on ontological reasoning is not sufficient to establish practical OBDA with regards to scalability, feasibility, and usability. The primary aim of this thesis is to promote the applications of OBDA, by addressing the typical limitations of several research problems that are important to the practicality of OBDA systems. In this thesis, we focus on studying these problems in existential rules, a prominent family of ontological languages that proves to be both expressive and tractable. Observing that current query rewriting techniques are not scalable over expressive ontologies, we propose a novel datalog rewriting approach for existential rules based on the notion of unfolding. While datalog rewritability cannot be guaranteed in general existential rules, we propose a novel abstract class called weakly-separable rules for datalog rewriting and show that it can generalize several combinations of existing well-accepted classes. We develop a prototype of query answering system called Drewer based on our proposed datalog rewriting method and evaluations show that our system has superior performance to state-of-the-art systems. While query answering is the essential reasoning task for OBDA, it is necessary to provide appropriate explanations to query answers. We study the problem of query abduction, which is the underlying problem of explaining negative query answers. To make the abduction process more user-oriented, we present a novel abduction framework that discriminates between predicates expressing high-level and low-level concepts. We also develop an efficient algorithm for its computation based on first-order rewriting in existential rules, which shows scalability over large databases in experiments. Forgetting is a well-known mechanism that can have a variety of potential applications to the manipulation of ontologies. However, current studies about forgetting cannot handle inconsistent ontologies, which hinders its applications to OBDA scenarios where errors of knowledge might occur. We present the first study of inconsistency-tolerant forgetting, that is, forgetting with the presence of inconsistencies. Three different definitions based on inconsistency-tolerant query answering are proposed and their rationality is illustrated by comparing to other possible solutions. We explored their properties and computation methods in a light-weight Description Logic language, DL-lite.core.