Multi-facet graph mining with contextualized projections

Yang, Carl

Permalink

https://hdl.handle.net/2142/109388 Copy

Description

Title

Multi-facet graph mining with contextualized projections

Author(s)

Yang, Carl

Issue Date

2020-12-01

Director of Research (if dissertation) or Advisor (if thesis)

Han, Jiawei

Doctoral Committee Chair(s)

Han, Jiawei

Committee Member(s)

• Peng, Jian
• Zhai, ChengXiang
• Leskovec, Jure

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Graph mining
• Contextualized projection
• Graph embedding
• Graph generation
• Network applications

Abstract

The goal of my doctoral research is to develop a new generation of graph mining techniques, centered around my proposed idea of multi-facet contextualized projections, for more systematic, flexible, and scalable knowledge discovery around massive, complex, and noisy real-world context-rich networks across various domains. Traditional graph theories largely overlook network contexts, whereas state-of-the-art graph mining algorithms simply regard them as associative attributes and brutally employ machine learning models developed in individual domains (e.g., convolutional neural networks in computer vision, recurrent neural networks in natural language processing) to handle them jointly. As such, essentially different contexts (e.g., temporal, spatial, textual, visual) are mixed up in a messy, unstable, and uninterpretable way, while the correlations between graph topologies and contexts remain a mystery, which further renders the development of real-world mining systems less principled and ineffective. To overcome such barriers, my research harnesses the power of multi-facet context modeling and focuses on the principle of contextualized projections, which provides generic but subtle solutions to knowledge discovery over graphs with the mixtures of various semantic contexts.

Graduation Semester

2020-12

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/109388

Copyright and License Information

Copyright 2020 Ji Yang (Carl)

Owning Collections