國立虎尾科技大學 |

Applying Graph Neural Networks to Applications : = Brain Disease Detection, Recommendation and Fraud Detection.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Applying Graph Neural Networks to Applications :/
Reminder of title:	Brain Disease Detection, Recommendation and Fraud Detection.
Author:	Meng, Lin.
Description:	1 online resource (129 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 85-07, Section: B.
Contained By:	Dissertations Abstracts International85-07B.
Subject:	Computer science. -
Online resource:	click for full text (PQDT)
ISBN:	9798381411010

Applying Graph Neural Networks to Applications : = Brain Disease Detection, Recommendation and Fraud Detection.
Meng, Lin.

Applying Graph Neural Networks to Applications :Brain Disease Detection, Recommendation and Fraud Detection. - 1 online resource (129 pages)

Source: Dissertations Abstracts International, Volume: 85-07, Section: B.

Thesis (Ph.D.)--The Florida State University, 2023.

Includes bibliographical references

Graph-structured data widely exists in the real world, including biomedical, e-commerce, and social areas. The graph-structured data contains objects and relationships among them, which refer to nodes and edges in the graph, respectively. Typically, to learn the information from graphs, we need to leverage both nodes and edges, which is hard for those traditional methods designed for i.i.d. since they do not capture the dependencies among data instances. Recently, graph neural networks (GNNs) have attracted increasing attention since they solve the drawbacks of traditional neural networks when dealing with graph-structured data. Empowered by the GNNs, research has extended to many areas that use graph data to model the interactions and get better results. Despite the success of GNNs in general tasks like node classification, more work is needed to handle task-specific problems. Typically, when facing various downstream tasks, GNNs often adopt sufficient changes according to the characteristics of the datasets for tasks and incorporate domain knowledge as well. The dissertation includes my research works on applying graph neural networks to brain disease detection, recommendation, and fraud detection. Since GNNs often lack interpretability and require node features, for brain disease detection, I focus on improving the interpretability of GNNs and learning the structural features without node features in Chapter 3. For recommendation, I focus on problems like modeling users' changing preferences based on pure historical interactions between users and items through time and the intrinsical oscillation problem brought by the bipartite nature of user-item graphs. In Chapter 4, I make dynamic recommendations by analyzing the historical data based on the life cycles of users and items. In Chapter 5, I discuss the oscillation problem, varying locality, and fixed propagation pattern for user-item bipartite graphs and provide an analysis of the propagation of GNNs. For fraud detection, I focus on solving camouflage and class imbalance problems. In Chapter 6, I analyze how the camouflage violates the ``homophily'' assumption of GNNs and how it affects the learning process of GNNs first, and then propose a model that decouples the learning process of feature learning and structure learning. In Chapter 7, I alleviate the class imbalance problem by augmenting synthesized nodes that mimic the behaviors of real fraudsters. The nodes and their neighborhoods are automatically generated by the feature and edge generators, respectively. Our experimental results show that adding a few nodes to the original graph can improve the performance of the original GNN-based fraud detector.Lastly, I provide a conclusion about this dissertation in the final chapter.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798381411010Subjects--Topical Terms:

573171
Computer science.
Subjects--Index Terms:

Brain disease detectionIndex Terms--Genre/Form:

554714
Electronic books.

Applying Graph Neural Networks to Applications : = Brain Disease Detection, Recommendation and Fraud Detection.
LDR:04189ntm a22003857 4500 001 1147304
005 20240909100751.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798381411010
035 $a (MiAaPQ)AAI30691802
035 $a AAI30691802
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Meng, Lin. $3 1473001
245 1 0 $a Applying Graph Neural Networks to Applications : $b Brain Disease Detection, Recommendation and Fraud Detection.
264 0 $c 2023
300 $a 1 online resource (129 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: 85-07, Section: B.
500 $a Advisor: Zhang, Jiawei;Zhang, Xiaonan.
502 $a Thesis (Ph.D.)--The Florida State University, 2023.
504 $a Includes bibliographical references
520 $a Graph-structured data widely exists in the real world, including biomedical, e-commerce, and social areas. The graph-structured data contains objects and relationships among them, which refer to nodes and edges in the graph, respectively. Typically, to learn the information from graphs, we need to leverage both nodes and edges, which is hard for those traditional methods designed for i.i.d. since they do not capture the dependencies among data instances. Recently, graph neural networks (GNNs) have attracted increasing attention since they solve the drawbacks of traditional neural networks when dealing with graph-structured data. Empowered by the GNNs, research has extended to many areas that use graph data to model the interactions and get better results. Despite the success of GNNs in general tasks like node classification, more work is needed to handle task-specific problems. Typically, when facing various downstream tasks, GNNs often adopt sufficient changes according to the characteristics of the datasets for tasks and incorporate domain knowledge as well. The dissertation includes my research works on applying graph neural networks to brain disease detection, recommendation, and fraud detection. Since GNNs often lack interpretability and require node features, for brain disease detection, I focus on improving the interpretability of GNNs and learning the structural features without node features in Chapter 3. For recommendation, I focus on problems like modeling users' changing preferences based on pure historical interactions between users and items through time and the intrinsical oscillation problem brought by the bipartite nature of user-item graphs. In Chapter 4, I make dynamic recommendations by analyzing the historical data based on the life cycles of users and items. In Chapter 5, I discuss the oscillation problem, varying locality, and fixed propagation pattern for user-item bipartite graphs and provide an analysis of the propagation of GNNs. For fraud detection, I focus on solving camouflage and class imbalance problems. In Chapter 6, I analyze how the camouflage violates the ``homophily'' assumption of GNNs and how it affects the learning process of GNNs first, and then propose a model that decouples the learning process of feature learning and structure learning. In Chapter 7, I alleviate the class imbalance problem by augmenting synthesized nodes that mimic the behaviors of real fraudsters. The nodes and their neighborhoods are automatically generated by the feature and edge generators, respectively. Our experimental results show that adding a few nodes to the original graph can improve the performance of the original GNN-based fraud detector.Lastly, I provide a conclusion about this dissertation in the final chapter.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer science. $3 573171
650 4 $a Information science. $3 561178
653 $a Brain disease detection
653 $a Fraud detection
653 $a Graph neural networks
653 $a Recommendation
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
690 $a 0723
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a The Florida State University. $b Computer Science. $3 1473002
773 0 $t Dissertations Abstracts International $g 85-07B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30691802 $z click for full text (PQDT)