Towards neural graph databases
Abstract/Contents
- Abstract
- Graph databases are the primary workhorse for storing and organizing structured information over real-world entities. The core task on graph databases is query answering. The objective is to execute a given query of interest and retrieve answers to the query from the underlying graph. However, a major challenge is that data in graph databases are often massive, noisy, and incomplete. Relationships between entities keep changing, which is extremely hard to keep track of. Execution of a query over noisy and missing data leads to answers of poor quality. This significantly limits the application of graph databases in many scenarios. Neural graph query answering algorithms aim to address the challenge via embeddings and representation learning. The idea is to map complex queries and graph data to a latent space. This field has witnessed substantial growth, with extensive research focusing on both theoretical and practical dimensions, addressing various queries and graph types with efficient systems. In this thesis, we introduce the concept of Neural Graph Databases (NGDBs) where neural graph query answering is at its core. NGDB consists of a Neural Graph Storage and a Neural Graph Engine, extending the idea of graph databases. Under NGDB, we provide a collection of neural graph query answering methods, datasets, systems, metrics, and broad applications. In Chapter 2, we start with the definitions of different types of graphs, graph queries, fuzzy logic, and graph representation learning. This serves as the foundation and background of the thesis. In Part I, we discuss two works on query embeddings, Query2box and BetaE. Query2box introduces the first neural graph query answering algorithm that models existential quantification, disjunction, and conjunction operations. BetaE further extends its capabilities to include the negation operation. In Part II, we design the first and most scalable system, SMORE, which allows for the training of neural graph query answering algorithms on extremely large graphs. In Part III, we introduce the overall framework and components of NGDB. We further provide a unified taxonomy of existing neural graph query answering algorithms from the perspectives of graphs, modeling, queries, and datasets. Together the thesis introduces the task of neural graph query answering and the framework of NGDB, presents two novel algorithms, and points out promising directions, unsolved problems, and applications of NGDB.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2023; ©2023 |
Publication date | 2023; 2023 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Ren, Hongyu |
---|---|
Degree supervisor | Leskovec, Jurij |
Thesis advisor | Leskovec, Jurij |
Thesis advisor | Liang, Percy |
Thesis advisor | Ma, Tengyu |
Degree committee member | Liang, Percy |
Degree committee member | Ma, Tengyu |
Associated with | Stanford University, School of Engineering |
Associated with | Stanford University, Computer Science Department |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Hongyu Ren. |
---|---|
Note | Submitted to the Computer Science Department. |
Thesis | Thesis Ph.D. Stanford University 2023. |
Location | https://purl.stanford.edu/sp577dw3968 |
Access conditions
- Copyright
- © 2023 by Hongyu Ren
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...