Year: 2024
Type: dissertation
Abstract: Artificial intelligence (AI) has been transforming the way we live, work, and interact with the world.Neural symbolic AI has emerged in recent years.By combining the power of deep learning with symbolic reasoning, it promises the next-generation AI systems that are more explainable, trustworthy, and versatile.It is expected to revolutionize a myriad of high-impact applications, ranging from code generation, question answering to drug discovery.To unleash the full potential of neural-symbolic reasoning, a critical cornerstone is to represent the symbolic knowledge and to further integrate that knowledge with neural models.Among others, knowledge graphs, as a structured representation of knowledge that captures relationships between entities and concepts in a graph-like format, provide a powerful and versatile tool for organizing and connecting real world information.A knowledge graph (KG) is a graph-based data structure that represents a collection of real-world facts in the form of triples, where each triple contains a subject, predicate, and object.Knowledge graphs find use in various applications such as search engines, recommender systems, and question-answering (KGQA for short).The primary aim of knowledge graph reasoning is to extract meaningful insights from the available knowledge graph data which involves discovering and explaining existing knowledge or inferring new knowledge from the existing information in the knowledge graph.Reasoning on knowledge graph with neural network models can naturally leverage the advantages of neural symbolic AI.Despite remarkable progress in knowledge graph reasoning, there are still several challenges that need to be addressed.First ( Background Knowledge Graph), the background knowledge graph is often vast and incomplete, posing difficulties to the KG reasoning problem.Existing knowledge is explicitly stored as triplets in the knowledge graph, while implicit knowledge is hidden within the graph structure, paths, or subgraphs.Besides, knowledge graph is large, and how to reason efficiently remains a key issue.Second ( Reasoning Input), inputs to different tasks vary, and the input may be ambiguous or iterative, making the reasoning process more challenging.Third ( Reasoning Model), relations in KG have rich properties, and how to support these properties, including transitivity, symmetry, and asymmetry, remains an active area of research.Furthermore, it is crucial to ensure the generalization ability of the model, e.g., whether it can solve multiple tasks, and whether it can transfer to other knowledge graphs.Addressing these challenges is crucial for further advancements in neural symbolic knowledge graph reasoning.In this thesis proposal, we study the problem knowledge graph reasoning that aims to First and foremost, I want to express my deepest gratitude to my advisor Dr. Hanghang Tong for his adequate guidance, invaluable advice, continuous support, encouragement and patience.This dissertation is impossible without his help.Dr. Tong directed me to the exciting data mining research and mentored me through every step along the way towards conducting excellent research: from discovering research problems to coming up with novel solutions, from writing well-polished papers to giving well-prepared presentations, with dedication to every detail.Every research discussion is a great pleasure for me and I have marvelously benefited from his sharp insights, enlightened suggestions and broad vision.More than that, he is considerate and cares for others.His encouragement during the moments when my research was not going as planned, when my paper was rejected, and when I was suffering from the arduous job search, always brightens my Ph.D. journey.His gentle personality and stringent research principle will motivate me in my future career.I would also like to say thank you to all the other thesis committee members, Yizhou Sun, Hari Sundaram and ChengXiang Zhai for giving me valuable feedback and asking insightful questions during my thesis proposal and defense.Specifically, I am grateful to Dr. Zhai for sharing his deep and broad knowledge in the field of information retrieval in his CS 510 Advanced Information Retrieval.I would like to thank Dr. Sun for his great works on knowledge graph reasoning and mining, which inspired me a lot in this research area.I would like to express my sincere gratitude to Dr.
Author: Liu, Lihui
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7220775363
