Constructing a Personal Knowledge Graph from Disparate Data Sources

This thesis revolves around the idea of a Personal Knowledge Graph as a uniform coherent structure of personal data collected from multiple disparate sources: A knowledge base consisting of entities such as persons, events, locations and companies interlinked with semantically meaningful relationships in a graph structure where the user is at its center. The personal knowledge graph is intended to be a valuable resource for a digital personal assistant, expanding its capabilities to answer questions and perform tasks that require personal knowledge about the user. We explored techniques within Knowledge Representation, Knowledge Extraction/ Information Extraction and Information Management for the purpose of constructing such a graph. We show the practical advantages of using Knowledge Graphs for personal information management, utilizing the structure for extracting and inferring answers and for handling resources like documents, emails and calendar entries. We have proposed a framework for aggregating user data and shown how existing ontologies can be used to model personal knowledge. We have shown that a personal knowledge graph based on the user's personal resources is a viable concept, however we were not able to enrich our personal knowledge graph with knowledge extracted from unstructured private sources. This was mainly due to sparsity of relevant information, the informal nature and the lack of context in personal correspondence