東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Statistical Tools for Directed and B...

Yoo, Hyesun,

FindBook

Google Book

Amazon

博客來

Statistical Tools for Directed and Bipartite Networks /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Statistical Tools for Directed and Bipartite Networks // Hyesun Yoo.
作者:	Yoo, Hyesun,
面頁冊數:	1 electronic resource (123 pages)
附註:	Source: Dissertations Abstracts International, Volume: 82-07, Section: B.
Contained By:	Dissertations Abstracts International82-07B.
標題:	Statistics. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28240385
ISBN:	9798684619250

Statistical Tools for Directed and Bipartite Networks /
Yoo, Hyesun,

Statistical Tools for Directed and Bipartite Networks /Hyesun Yoo. - 1 electronic resource (123 pages)

Source: Dissertations Abstracts International, Volume: 82-07, Section: B.

Directed networks and bipartite networks, which exhibit unique asymmetric connectivity structures, are commonly observed in a variety of scientific and engineering fields. Despite their abundance and utility, most network analysis methods only consider symmetric networks. In this thesis, we develop statistical methods and theory for directed and bipartite networks.The first chapter focuses on matched community detection in a bipartite network. The detection of matched communities, i.e. communities that consist of nodes of two types that are closely connected with one another, is a fundamental and challenging problem. Most widely used approaches for matched community detection are either computationally inefficient or prone to non-ideal performance. We propose a new two-stage algorithm that uses fast spectral methods to recover matched communities. We show that, for bipartite networks, it is critical to adjust for the community size in matched community detection, which had not been considered before. We also provide theoretical error bounds for the proposed algorithm on the number of mis-clustered nodes under a variant of the stochastic block model. Numerical studies indicate that the proposed method outperforms existing spectral algorithms, especially when the sizes of the matched communities are proportionally different between the two types.The second chapter of the thesis introduces a new preference-based block model for community detection in a directed network. Unlike existing models, the proposed model allows different sender nodes to have different preferences to communities in the network. We argue that the right singular vectors of a graph Laplacian matrix contain community structures under the model. Further, we propose a spectral clustering algorithm to detect communities and estimate parameters of the model. Theoretical results show insights on how the heterogeneity of preferences and out-degrees contribute to an upper bound of the number of mis-clustered nodes. Numerical studies support the theoretical results and illustrate the outstanding performance of the proposed method. The model can also be naturally extended to bipartite networks. In the third chapter, we propose a dyadic latent space model which accommodates the reciprocity between a pair of nodes in directed networks. Nodes in a pair in directed networks often exhibit strong dependencies with each other, though most widely used approaches usually account for this phenomenon with limited flexibility. We propose a new latent space model for directed networks that incorporates the reciprocity in a flexible way, allowing for important characteristics such as homophily and heterogeneity of the nodes. A fast and scalable algorithm based on projected gradient descent has been developed to fit the model by maximizing the likelihood. Both simulation studies and real-world data examples illustrate that the proposed model is effective in various network analysis tasks including link prediction and community detection.

English

ISBN: 9798684619250Subjects--Topical Terms:

517247
Statistics.
Subjects--Index Terms:

Network analysis

Statistical Tools for Directed and Bipartite Networks /
LDR:04680nmm a22005293i 4500 001 2432019
005 20250626110544.5
006 m o d
007 cr|nu||||||||
008 261011s2020 miu||||||m |||||||eng d
020 $a 9798684619250
035 $a (MiAaPQD)AAI28240385
035 $a (MiAaPQD)umichrackham003393
035 $a AAI28240385
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Yoo, Hyesun, $e author. $3 3821096
245 1 0 $a Statistical Tools for Directed and Bipartite Networks / $c Hyesun Yoo.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 1 electronic resource (123 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 82-07, Section: B.
500 $a Advisors: Zhu, Ji Committee members: Kang, Jian; Levina, Elizaveta; Xu, Gongjun.
502 $b Ph.D. $c University of Michigan $d 2020.
520 $a Directed networks and bipartite networks, which exhibit unique asymmetric connectivity structures, are commonly observed in a variety of scientific and engineering fields. Despite their abundance and utility, most network analysis methods only consider symmetric networks. In this thesis, we develop statistical methods and theory for directed and bipartite networks.The first chapter focuses on matched community detection in a bipartite network. The detection of matched communities, i.e. communities that consist of nodes of two types that are closely connected with one another, is a fundamental and challenging problem. Most widely used approaches for matched community detection are either computationally inefficient or prone to non-ideal performance. We propose a new two-stage algorithm that uses fast spectral methods to recover matched communities. We show that, for bipartite networks, it is critical to adjust for the community size in matched community detection, which had not been considered before. We also provide theoretical error bounds for the proposed algorithm on the number of mis-clustered nodes under a variant of the stochastic block model. Numerical studies indicate that the proposed method outperforms existing spectral algorithms, especially when the sizes of the matched communities are proportionally different between the two types.The second chapter of the thesis introduces a new preference-based block model for community detection in a directed network. Unlike existing models, the proposed model allows different sender nodes to have different preferences to communities in the network. We argue that the right singular vectors of a graph Laplacian matrix contain community structures under the model. Further, we propose a spectral clustering algorithm to detect communities and estimate parameters of the model. Theoretical results show insights on how the heterogeneity of preferences and out-degrees contribute to an upper bound of the number of mis-clustered nodes. Numerical studies support the theoretical results and illustrate the outstanding performance of the proposed method. The model can also be naturally extended to bipartite networks. In the third chapter, we propose a dyadic latent space model which accommodates the reciprocity between a pair of nodes in directed networks. Nodes in a pair in directed networks often exhibit strong dependencies with each other, though most widely used approaches usually account for this phenomenon with limited flexibility. We propose a new latent space model for directed networks that incorporates the reciprocity in a flexible way, allowing for important characteristics such as homophily and heterogeneity of the nodes. A fast and scalable algorithm based on projected gradient descent has been developed to fit the model by maximizing the likelihood. Both simulation studies and real-world data examples illustrate that the proposed model is effective in various network analysis tasks including link prediction and community detection.
546 $a English
590 $a School code: 0127
650 4 $a Statistics. $3 517247
650 4 $a Applied mathematics. $3 2122814
650 4 $a Computer science. $3 523869
650 4 $a Technical communication. $3 3172863
650 4 $a Engineering. $3 586835
650 4 $a Systems science. $3 3168411
650 4 $a Information technology. $3 532993
653 $a Network analysis
653 $a Directed network
653 $a Bipartite network
653 $a Community detection
653 $a Asymmetric connectivity structures
653 $a Engineering fields
653 $a Homophily
653 $a Link prediction
690 $a 0463
690 $a 0364
690 $a 0790
690 $a 0489
690 $a 0643
690 $a 0984
690 $a 0537
710 2 $a University of Michigan. $b Statistics. $e degree granting institution. $3 3816509
720 1 $a Zhu, Ji $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 82-07B.
790 $a 0127
791 $a Ph.D.
792 $a 2020
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28240385