Advancements in Knowledge Graph Reasoning: Innovative Approaches to Complex Query Answering and Logical Hypothesis Generation

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Advancements in Knowledge Graph Reasoning: Innovative Approaches to 
Complex Query Answering and Logical Hypothesis Generation"

By

Mr. Jiaxin BAI


Abstract:

Complex logical query answering on knowledge graphs (KGs) has emerged as a 
crucial component of knowledge graph reasoning. This thesis presents a 
comprehensive framework for advancing KG reasoning through multiple novel 
approaches across diverse knowledge graph types. We introduce Query2Particles 
(Q2P), a particle-based embedding method that effectively handles distributed 
answer sets in incomplete KGs by encoding queries into multiple particle 
embeddings. To address numerical reasoning, we develop the Number Reasoning 
Network (NRN), which integrates numerical attribute processing with 
traditional entity-relation reasoning. We further propose Sequential Query 
Encoding (SQE), transforming computational graphs into sequences for more 
efficient and accurate query processing. Our work extends to eventuality 
knowledge graphs through Complex Eventuality Query Answering (CEQA), 
incorporating implicit temporal and logical constraints via Memory-Enhanced 
Query Encoding (MEQE). For practical applications, we develop the Logical 
Session Graph Transformer (LSGT) to understand cross-session user intentions 
in recommendation systems. Additionally, we advance abductive reasoning 
through our Reinforcement Learning from Knowledge Graph (RLF-KG) approach, 
enabling robust logical hypothesis generation. Extensive experiments on 
benchmark datasets demonstrate that our methods consistently achieve 
state-of-the-art performance across various tasks, from basic query answering 
to complex reasoning scenarios. These contributions collectively advance the 
field of knowledge graph reasoning, providing powerful tools for complex 
query answering in real-world applications.


Date:                   Friday, 7 March 2025

Time:                   10:00am - 12:00noon

Venue:                  Room 4472
                        Lifts 25/26

Chairman:               Dr. Jia LIU (MARK)

Committee Members:      Dr. Yangqiu SONG (Supervisor)
                        Prof. Nevin ZHANG
                        Prof. Xiaofang ZHOU
                        Dr. Can YANG (MATH)
                        Prof. Hong CHENG (CUHK)