Inductive Representation Learning and Natural Language Question Answering on Temporal Knowledge Graphs
MCML Authors
Abstract
Abstract
Real-world applications such as recommendersystems, socialnetworks, andprotein-protein interactions often involve relational data. In recent years, there has been increasing interest in machine learning on such data, particularly in the context of knowledge graphs (KGs). KGs are structured relational data that store multi-relational information as directed graphs, where each node corresponds to an entity and each labeled edge represents a factual relationship between entities, e.g., (Oxford, located in, the United Kingdom). Traditional KGs assume time-invariant relationships. However, real-world relationships are dynamically evolving over time. For example, the chancellor of Germany in 2020 was Angela Merkel, but in 2022 it became Olaf Scholz. This necessitates the use of temporal knowledge graphs (TKGs), where temporal facts are introduced by coupling stationary facts with additional time identifiers, e.g., (Angela Merkel, is chancellor of, Germany, 2020). TKGs are more expressive than KGs as they model the temporal evolution of knowledge. Consequently, recent research has paid more attention to machine learning on TKGs. In this thesis, we focus on two machine learning problems: inductive knowledge representation learning and natural language question answering (QA) on TKGs. (Shortened)
BibTeXKey: Din25