Files
Abstract
A knowledge graph (KG) provides a framework for data representation, integration, analytics by expressing sets of linked descriptions of entities and places data in a context via semantic metadata, and it helps to enrich the data with computer-processable semantics. In many domains, the KG aids researchers blend related information to a single source for effortless and efficient investigations. External resources and datasets, usually Web documents, are acquired by software programs for the purpose of creating or evolving a KG create or evolve a KG. New findings lead to changes in the original data sources relentlessly; therefore, the generated KG should comply with the changes. The introduced changes can range from individual entities and their relations to more significant changes in the KG schema. In both cases, the domain expert or KG engineer should employ mechanisms to track them and take proper actions. The structure and connectivity among entities in graph-like data make researchers curious about finding new associations by visualizing or querying the data. With the rise of diverse machine learning techniques, this process can be more efficient and achievable. Thus, the link prediction task becomes one of the priorities on KGs, especially in domains such as biology, social networks, andrecommender systems. It generally aims to discover unknown linkage between existing entities in the KG. Machine learning techniques for link prediction have become popular solutions for link prediction, especially deep learning (DL) methods. The scalability issue of these approaches for large graphs calls for an alternative direction. Toward addressing these issues, this dissertation investigates scalable approaches for evaluating and using KG for knowledge discovery. First, we present our work, KGdiff, for tracking the evolution of KGs by discovering meta-data information from KGs and then we introduce RegPattern2Vec for the link prediction problem and its successful application on a large biological dataset.