Author: Chen, Shengyuan
Title: Knowledge graph reasoning : algorithms and applications
Advisors: Huang, Xiao (COMP)
Li, Qing (COMP)
Degree: Ph.D.
Year: 2024
Subject: Knowledge representation (Information theory)
Artificial intelligence
Reasoning
Hong Kong Polytechnic University -- Dissertations
Department: Department of Computing
Pages: xiv, 95 pages : color illustrations
Language: English
Abstract: Knowledge graphs (KGs) serve as powerful tools in various domains by capturing and organizing complex relationships between entities. Yet, despite their utility, KGs often remain incomplete, limiting their full potential. KG completion and KG alignment emerge as critical solutions to fill these gaps, and at the heart of both lies KG reasoning. Traditional reasoning methods, whether neural or symbolic, often falter when tasked with large-scale KGs due to challenges in scalability and interpretability. Hence, advancing the efficiency and effectiveness of KG reasoning algorithms is paramount for unlocking their broader applications.
This thesis presents several groundbreaking approaches designed to enhance KG reasoning. First, we introduce DiffLogic, a differentiable neuro-symbolic reasoning framework that fuses the strengths of symbolic rule-based inference with the flexibility of neural networks. By modeling the joint probability of truth scores through weighted logic rules embedded within a Markov random field, DiffLogic achieves efficient rule grounding, allowing it to perform scalable inference over expansive KGs with precision. Building upon this, we propose NeuSymEA, a robust entity alignment method that bridges the gap between symbolic and neural reasoning through variational inference. NeuSymEA is uniquely equipped to operate in low-resource environments, aligning entities even when limited labeled data is available. Through logic decomposition and deduction, NeuSymEA scales symbolic reasoning to handle long rule inferences, all while ensuring interpretability and improving precision in alignment tasks. Lastly, we introduce LLM4EA, a label-free entity alignment framework that harnesses the power of large language models (LLMs) to address noisy annotations. By actively selecting key entities for annotation and refining label accuracy through probabilistic reasoning, LLM4EA enhances alignment robustness, minimizing the effect of noise while maximizing performance.
The research presented in this thesis pushes the boundaries of what’s possible in KG reasoning, offering scalable and efficient solutions for constructing and exploiting large KGs. The advancements presented in this work open new doors for robust and interpretable applications across fields such as information retrieval, social computing, health informatics, and bioinformatics, enabling the extraction of deeper insights from the vast and interconnected data that knowledge graphs offer.
Rights: All rights reserved
Access: open access

Files in This Item:
File Description SizeFormat 
7827.pdfFor All Users1.64 MBAdobe PDFView/Open


Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: https://theses.lib.polyu.edu.hk/handle/200/13406