Knowledge Discovery with Knowledge Graphs: From Structural Embeddings to Generative Reasoning

Abstract

Knowledge graphs provide a structured modeling of human knowledge, which is essential to many modern knowledge discovery systems. Knowledge graph embedding and knowledge graph representation learning techniques convert entities and relations into feature vectors, thereby facilitating their effective integration into contemporary data mining and machine learning applications. Along with these structural embedding approaches, recent studies on knowledge graphs also explore their applications with large language models, such as improving the Retrieval-Augmented Generation (RAG) based on the graph-structured data, referred to as GraphRAG. This tutorial provides a review of seminal works in knowledge graph representation learning, including knowledge graph embedding with multimodal data, inductive reasoning on knowledge graphs, knowledge graph foundation models, and representation learning on hyper-relational knowledge graphs. Furthermore, the tutorial also investigates various ways to use knowledge graphs with large language models. This Lecture-Style Tutorial consists of six lectures, including some demonstrations of representative knowledge graph representation learning models.