Creation, Enrichment and Application of Knowledge Graphs
Simon Gottschalk · Institutional Repository of Leibniz Universität Hannover (Leibniz Universität Hannover) · 2021
The world is in constant change, and so is the knowledge about it. Knowledge-based systems - for example, online encyclopedias, search engines and virtual assistants - are thus faced with the constant challenge of collecting this knowledge and beyond that, to understand it and make it accessible to their users. Only if a knowledge-based system is capable of this understanding - that is, it is capable of more than just reading a collection of words and numbers without grasping their semantics - it can recognise relevant information and make it understandable to its users. The dynamics of the world play a unique role in this context: Events of various kinds which are relevant to different communities are shaping the world, with examples ranging from the coronavirus pandemic to the matches of a local football team. Vital questions arise when dealing with such events: How to decide which events are relevant, and for whom? How to model these events, to make them understood by knowledge-based systems? How is the acquired knowledge returned to the users of these systems? A well-established concept for making knowledge understandable by knowledge-based systems are knowledge graphs, which contain facts about entities (persons, objects, locations, ...) in the form of graphs, represent relationships between these entities and make the facts understandable by means of ontologies. This thesis considers knowledge graphs from three different perspectives: (i) Creation of knowledge graphs: Even though the Web offers a multitude of sources that provide knowledge about the events in the world, the creation of an event-centric knowledge graph requires recognition of such knowledge, its integration across sources and its representation. (ii) Knowledge graph enrichment: Knowledge of the world seems to be infinite, and it seems impossible to grasp it entirely at any time. Therefore, methods that autonomously infer new knowledge and enrich the knowledge graphs are of particular interest. (iii) Knowledge graph interaction: Even having all knowledge of the world available does not have any value in itself; in fact, there is a need to make it accessible to humans. Based on knowledge graphs, systems can provide their knowledge with their users, even without demanding any conceptual understanding of knowledge graphs from them. For this to succeed, means for interaction with the knowledge are required, hiding the knowledge graph below the surface. In concrete terms, I present EventKG - a knowledge graph that represents the happenings in the world in 15 languages - as well as Tab2KG - a method for understanding tabular data and transforming it into a knowledge graph. For the enrichment of knowledge graphs without any background knowledge, I propose HapPenIng, which infers missing events from the descriptions of related events. I demonstrate means for interaction with knowledge graphs at the example of two web-based systems (EventKG+TL and EventKG+BT) that enable users to explore the happenings in the world as well as the most relevant events in the lives of well-known personalities.