Skip to Main Content
Frequently Asked Questions
Submit an ETD
Global Search Box
Need Help?
Keyword Search
Participating Institutions
Advanced Search
School Logo
Files
File List
Thesis.pdf (29.05 MB)
ETD Abstract Container
Abstract Header
Community Structure in Co-Location Networks
Author Info
Xi, Wenna
ORCID® Identifier
http://orcid.org/0000-0002-5427-5689
Permalink:
http://rave.ohiolink.edu/etdc/view?acc_num=osu1566156023255678
Abstract Details
Year and Degree
2019, Doctor of Philosophy, Ohio State University, Biostatistics.
Abstract
Co-location networks are a type of two-mode networks that capture individuals' activity patterns. The two modes are individuals and geographic locations, with ties indicating the relationship between them (e.g., the fact that the individual visits the location, the number of times the individual visits the location). The community structure in co-location networks -- patterns of individuals' with overlapping activity spaces -- provides important information about the functioning of a city and its neighborhoods. Because of its potential implications in social sciences and public health, in this dissertation, we study the community structure and detect communities of individuals in co-location networks. Specifically, we identify latent activity pattern profiles, which, for each community, summarize its members' probability distribution of visiting each location, and community assignment vectors, which, for each individual, summarize his/her probability distribution of belonging to each community. We introduce two statistical methods to do so: 1) latent Dirichlet allocation (LDA), a well-developed method used primarily in text data mining, and 2) hierarchical Bayesian non-negative matrix factorization (NMF), which is similar to LDA but can be adapted to make use of sparse finite mixture techniques to automatically determine the number of communities in a co-location network. We show that the Bayesian NMF model is a more general version of a Bayesian mixture model and, therefore, also suffers from the label switching problem. We propose two methods to address this problem: the posterior rotation method, which is based on the Procrustes transformation of posterior samples, and the point-process representation clustering method, which is based on the k-means clustering of posterior samples of component-specific parameters. The performance of the proposed Bayesian NMF model is evaluated via simulation studies. In detecting the number of communities, close attention is paid to the robustness of the result to the initial values of the Dirichlet concentration parameters. Assuming the true number of communities is detected, the posterior results from the posterior rotation method and the point-process representation clustering method are compared in terms of average bias, mean squared errors (MSE), and the 95% credible interval (CI) coverage rate. Using the caregivers' activity data from the Adolescent Health and Development in Context (AHDC) Study, we employ our methods to identify activity pattern profiles and communities. For the LDA method, we further explore differences across neighborhoods in the strength and within-neighborhood consistency of community assignment, and find that, among white neighborhoods, the stronger the residents are attached to one community, the higher the likelihood that they all belong to the same community; however, this pattern disappears among neighborhoods with more heterogeneity of residents in terms of race.
Committee
Catherine Calder, Ph.D. (Advisor)
Christopher Browning, Ph.D. (Committee Member)
Eloise Kaizar, Ph.D. (Committee Member)
Subhadeep Paul, Ph.D. (Committee Member)
Bo Lu, Ph.D. (Committee Member)
Pages
217 p.
Subject Headings
Biostatistics
Recommended Citations
Refworks
EndNote
RIS
Mendeley
Citations
Xi, W. (2019).
Community Structure in Co-Location Networks
[Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1566156023255678
APA Style (7th edition)
Xi, Wenna.
Community Structure in Co-Location Networks.
2019. Ohio State University, Doctoral dissertation.
OhioLINK Electronic Theses and Dissertations Center
, http://rave.ohiolink.edu/etdc/view?acc_num=osu1566156023255678.
MLA Style (8th edition)
Xi, Wenna. "Community Structure in Co-Location Networks." Doctoral dissertation, Ohio State University, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=osu1566156023255678
Chicago Manual of Style (17th edition)
Abstract Footer
Document number:
osu1566156023255678
Download Count:
165
Copyright Info
© 2019, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.