Distributed Decision Models for Structured Data

Abstract

In recent years, many approaches for producing machine-readable and semantically enriched information on web data, expressed as knowledge graphs, have emerged. Nowadays, scalable analysis of large-scale knowledge graphs for assisting applications in multiple domains is a major challenge for researchers due to the rapid expansion of semantic data on the Web. The primary objective of this thesis is to lay the groundwork for developing efficient algorithms that can perform complex tasks on large-scale OWL datasets, encompassing parsing, exploration, inference of hidden knowledge, and mining of semantic knowledge graphs.<br /> Initially, we proposed an innovative approach for parsing large-scale OWL datasets that can scale horizontally.<br /> In addition, we introduced an innovative approach for conducting statistical computations on large-scale OWL datasets, which calculates 50 different statistical measures about OWL datasets in a distributed in-memory environment. Further, we proposed a scalable, distributed approach for RDFS and OWL reasoning over large-scale OWL datasets. Finally, we developed an innovative decentralized approach to enhance the efficiency of concept learning that is achieved by allowing the generation of terminological decision trees using the statistics of the input dataset.<br /> We conducted several empirical assessments to evaluate the scalability, performance, and efficiency of our proposed approaches. The results of this evaluation showed that the proposed approaches can enable efficient processing and analysis of large-scale OWL datasets. We successfully incorporated all the proposed approaches into the SANSA framework - a big data engine for scalable processing of large-scale RDF data.