Automated Knowledge Base Quality Assessment and Validation based on Evolution Analysis

In recent years, numerous efforts have been put towards sharing Knowledge Bases (KB) in the Linked Open Data (LOD) cloud. These KBs are being used for various tasks, including performing data analytics or building question answering systems. Such KBs evolve continuously: their data (instances) and schemas can be updated, extended, revised and refactored. However, unlike in more controlled types of knowledge bases, the evolution of KBs exposed in the LOD cloud is usually unrestrained, what may cause data to suffer from a variety of quality issues, both at a semantic level and at a pragmatic level. This situation affects negatively data stakeholders – consumers, curators, etc. –. Data quality is commonly related to the perception of the fitness for use, for a certain application or use case. Therefore, ensuring the quality of the data of a knowledge base that evolves is vital. Since data is derived from autonomous, evolving, and increasingly large data providers, it is impractical to do manual data curation, and at the same time, it is very challenging to do a continuous automatic assessment of data quality. Ensuring the quality of a KB is a non-trivial task since they are based on a combination of structured information supported by models, ontologies, and vocabularies, as well as queryable endpoints, links, and mappings. Thus, in this thesis, we explored two main areas in assessing KB quality: (i) quality assessment using KB evolution analysis, and (ii) validation using machine learning models. The evolution of a KB can be analyzed using fine-grained “change” detection at low-level or using “dynamics” of a dataset at high-level. In this thesis, we present a novel knowledge base quality assessment approach using evolution analysis. The proposed approach uses data profiling on consecutive knowledge base releases to compute quality measures that allow detecting quality issues. However, the first step in building the quality assessment approach was to identify the quality characteristics. Using high-level change detection as measurement functions, in this thesis we present four quality characteristics: Persistency, Historical Persistency, Consistency and Completeness. Persistency and historical persistency measures concern the degree of changes and lifespan of any entity type. Consistency and completeness measures identify properties with incomplete information and contradictory facts. The approach has been assessed both quantitatively and qualitatively on a series of releases from two knowledge bases, eleven releases of DBpedia and eight releases of 3cixty Nice. However, high-level changes, being coarse-grained, cannot capture all possible quality issues. In this context, we present a validation strategy whose rationale is twofold. First, using manual validation from qualitative analysis to identify causes of quality issues. Then, use RDF data profiling information to generate integrity constraints. The validation approach relies on the idea of inducing RDF shape by exploiting SHALL constraint components. In particular, this approach will learn, what are the integrity constraints that can be applied to a large KB by instructing a process of statistical analysis, which is followed by a learning model. We illustrate the performance of our validation approach by using five learning models over three sub-tasks, namely minimum cardinality, maximum cardinality, and range constraint. The techniques of quality assessment and validation developed during this work are automatic and can be applied to different knowledge bases independently of the domain. Furthermore, the measures are based on simple statistical operations that make the solution both flexible and scalable.