Harnessing heterogeneous association in real-world networks

Shi, Yu

Permalink

https://hdl.handle.net/2142/104857 Copy

Description

Title

Harnessing heterogeneous association in real-world networks

Author(s)

Shi, Yu

Issue Date

2019-04-17

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

Han, Jiawei

Doctoral Committee Chair(s)

Han, Jiawei

Committee Member(s)

• Sundaram, Hari
• Peng, Jian
• Kim, Myunghwan

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)

• Network mining
• Heterogeneity
• Heterogeneous association
• Real-world networks

Abstract

Real-world networks often contain heterogeneity due to the heterogeneous nature of the world. A few examples of such networks include multi-view social networks, heterogeneous bibliographic networks, biomedical networks, etc. Ostensibly the heterogeneity of real-world network appears as the typed essence of nodes and edges. By considering type information, researchers have shown that using the typed networks can achieve performance better than using the homogeneous networks in a wide variety of downstream applications such as classification, clustering, recommendation, and outlier detection. Beyond the low-level heterogeneity in nodes and edges on the surface, their types also naturally induce higher-level typed network components. In my practice mining real-world networks, I identify that the heterogeneity also prevalently lies in the association across different network components, and such heterogeneous association is often important and intrinsic to the information embodied in the networks. In this dissertation, I investigate the necessity of modeling heterogeneous association in real-world networks and develop methodologies to simultaneously leverage the rich information and accommodate the incompatibility in the presence of heterogeneous association. A series of new models along this line are proposed for specific problems including learning network embedding, defining relevance measures, and discovering hypernymy relation, together with the discussion on how the principles reflected by these models can be used in other network mining tasks. These proposed models cannot only achieve better quantitative results but also uncover the semantics hidden in the heterogeneous association of real-world data.

Graduation Semester

2019-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/104857

Copyright and License Information

Copyright 2019 Yu Shi

Owning Collections