Federated knowledge base debugging in DL-Lite A

Nolle, Andreas

PDF
nolle-2021-dissertation.pdf - Veröffentlichte Version
Download (1MB)

URL:	https://madoc.bib.uni-mannheim.de/60344
URN:	urn:nbn:de:bsz:180-madoc-603441
Dokumenttyp:	Dissertation
Erscheinungsjahr:	2021
Ort der Veröffentlichung:	Mannheim
Hochschule:	Universität Mannheim
Gutachter:	Stuckenschmidt, Heiner
Datum der mündl. Prüfung:	2 August 2021
Sprache der Veröffentlichung:	Englisch
Einrichtung:	Fakultät für Wirtschaftsinformatik und Wirtschaftsmathematik > Practical Computer Science II: Artificial Intelligence (Stuckenschmidt 2009-)
Fachgebiet:	004 Informatik
Freie Schlagwörter (Englisch):	ontology-based information integration , LOD , federated knowledge base , knowledge base debugging , inconsistency detection , repair generation , trust assessment
Abstract:	Due to the continuously growing amount of data the federation of different and distributed data sources gained increasing attention. In order to tackle the challenge of federating heterogeneous sources a variety of approaches has been proposed. Especially in the context of the Semantic Web the application of Description Logics is one of the preferred methods to model federated knowledge based on a well-defined syntax and semantics. However, the more data are available from heterogeneous sources, the higher the risk is of inconsistency – a serious obstacle for performing reasoning tasks and query answering over a federated knowledge base. Given a single knowledge base the process of knowledge base debugging comprising the identification and resolution of conflicting statements have been widely studied while the consideration of federated settings integrating a network of loosely coupled data sources (such as LOD sources) has mostly been neglected. In this thesis we tackle the challenging problem of debugging federated knowledge bases and focus on a lightweight Description Logic language, called DL-LiteA, that is aimed at applications requiring efficient and scalable reasoning. After introducing formal foundations such as Description Logics and Semantic Web technologies we clarify the motivating context of this work and discuss the general problem of information integration based on Description Logics. The main part of this thesis is subdivided into three subjects. First, we discuss the specific characteristics of federated knowledge bases and provide an appropriate approach for detecting and explaining contradictive statements in a federated DL-LiteA knowledge base. Second, we study the representation of the identified conflicts and their relationships as a conflict graph and propose an approach for repair generation based on majority voting and statistical evidences. Third, in order to provide an alternative way for handling inconsistency in federated DL-LiteA knowledge bases we propose an automated approach for assessing adequate trust values (i.e., probabilities) at different levels of granularity by leveraging probabilistic inference over a graphical model. In the last part of this thesis, we evaluate the previously developed algorithms against a set of large distributed LOD sources. In the course of discussing the experimental results, it turns out that the proposed approaches are sufficient, efficient and scalable with respect to real-world scenarios. Moreover, due to the exploitation of the federated structure in our algorithms it further becomes apparent that the number of identified wrong statements, the quality of the generated repair as well as the fineness of the assessed trust values profit from an increasing number of integrated sources.