Learning Structured Knowledge from Real-World Data without Excessive Annotations Público

Abstract

In a world where vast quantities of data are continually generated by humans every day, the majority of the data remains unstructured, posing a significant challenge to knowledge discovery and insight generation. Unleashing the full potential of these valuable information sources requires organizing the data with interconnections and contexts. This dissertation delves into the fundamental task of transforming unstruc- tured real-world data into structured knowledge, all without an excessive reliance on manual annotations. Particularly, I investigate three areas of research, including: (1) Constructing concept maps from unstructured text data. We first develop an inno- vative unsupervised concept map construction method by utilizing syntactic parsing techniques [48]. Then we further study how to translate the initial parsing-based concept maps into more concise task-oriented concept maps under the guidance of weak supervision signal from downstream tasks [50]. (2) Aligning and completing taxonomic knowledge graphs (KGs). Given the widely available KGs scattered in different sites, it is urgent to integrate them into a comprehensive knowledge base to harness knowledge-centric applications. We propose a novel perspective to lever- age expert-curated taxonomies as the backbone to aligning various KGs [52] under a few-shot manner. We further study how to complete taxonomic KGs after initial alignment between them [49]. (3) Empowering downstream applications with struc- tured knowledge. Finally, we explore how to harness the performance of downstream applications with learned structured knowledge. For instance, we utilize similarity- based communities for multiclass classification [51]. Together, these works cover the whole life cycle of construction, integration, completion, and utilization of structured knowledge.

Table of Contents

Table of Contents

1 Introduction .................................................................................1

2 Learning to Construct Concept Maps ..............................................5

3 Learn to Aligning and Completing Taxonomic Knowledge Graphs ...37

4 Applications of Structured Knowledge ..........................................63

5 Conclusion and Future Work ........................................................75

Bibliography ..................................................................................78

About this Dissertation

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Computer Science and Informatics
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Artificial Intelligence
Palavra-chave	Large Language Models Structured Knowledge Unsupervised Learning Graph Neural Networks Weakly-supervised Learning
Committee Chair / Thesis Advisor	Carl Yang, Emory University
Committee Members	Fei Liu, Emory University Yuanzhe Xi, Emory University Jiaming Shen, Google Research

Última modificação

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Learning Structured Knowledge from Real-World Data without Excessive Annotations ()	2024-04-17 22:49:30 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions