- : Ms Word, Ms Word Format
- : 70 Pages
- : ₦3,000 | $25 | ₵60 | Ksh 2720
- : 1-5 Chapters
- Click to DOWNLOAD Materials
A MULTI-LEVEL ANALYSIS OF INFORMATION AND SUPPLY FLOWS IN SOCIAL AND BUSINESS NETWORKS
Abstract
Networks are ubiquitous in both the physical world and the cyberspace. They enable the flow of information, materials, services, etc. The common thread running through my dissertation is the macro-level and micro-level analysis of information and supply flows in social and business networks. The overarching research question is “How do network flows relate to network structures and individual entities’ behaviors in networks?”
My dissertation approaches this question in the context of supply-chain networks, online social networks, and inter-organizational networks. Specifically, I explore how changes in supply-chain network topologies affect supply flows, and propose strategies to improve supply-chain networks’ robustness against disruptions. I analyze how the flow of information through individuals’ interactions influences their sentiment in online health communities, and utilize the sentimental influence to identify influential users. I model how the flow of information via organizations’ interactions impacts the structure of their collaboration network, and provide suggestions on how to promote collaboration. To support the research, the dissertation uses various computational and quantitative methodologies, such as network analysis and modeling, data and text mining, agent-based simulation, and statistical analysis.
The goal of my dissertation research is to achieve a deeper understanding of network flows, better prediction of network performance and individual behaviors, as well as new insights into ways to improve the design, management, and utilization of networks. Specifically, the outcome of this dissertation has implications for network disruption management, health care, online community building, humanitarian relief, and so on.
Table of Contents
List of Figures vii
List of Tables x
Acknowledgments xi
Chapter 1
Introduction 1
Chapter 2
Supply Flows in Supply-Chain Networks–A Topological Analysis 6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 A strategy to design supply-chain networks . . . . . . . . . . . . . . 11
2.3.1 New robustness metrics . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 The new hybrid design strategy . . . . . . . . . . . . . . . . 17
2.3.3 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.4 Simulation results for random disruptions . . . . . . . . . . 22
2.3.5 Simulation results for targeted disruptions . . . . . . . . . . 25
2.3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 A re-wiring strategy for supply-chain networks . . . . . . . . . . . . 32
2.4.1 Updated robustness metrics . . . . . . . . . . . . . . . . . . 32
2.4.2 The rewiring model . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.3 Rewiring scale-free networks with RLR . . . . . . . . . . . . 37
2.4.3.1 Simulation setup . . . . . . . . . . . . . . . . . . . 38
2.4.3.2 Simulation results for random supply disruptions . 40
2.4.3.3 Simulation results for targeted supply disruptions . 41
2.4.4 Rewiring Small-world Networks with RLR . . . . . . . . . . 43
2.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4.6 An Experiment for a Retailer’s Distribution Network . . . . 48
2.4.6.1 Simulation setting . . . . . . . . . . . . . . . . . . 48
2.4.6.2 Experimental results . . . . . . . . . . . . . . . . . 50
2.5 Summary and future work . . . . . . . . . . . . . . . . . . . . . . . 52
Chapter 3
Information Flows in Online Health Communities–An Analysis
at the Individual Level 56
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3 The dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4 A classification approach . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4.1 Features and initial results . . . . . . . . . . . . . . . . . . . 62
3.4.2 Improved results with new features and ensemble methods . 65
3.5 A sentiment approach . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.5.1 Sentiment analysis of posts . . . . . . . . . . . . . . . . . . . 70
3.5.2 The new metric based on sentiment dynamics . . . . . . . . 71
3.6 Comparison and integration of the two approaches . . . . . . . . . . 77
3.7 Summary and future work . . . . . . . . . . . . . . . . . . . . . . . 79
Chapter 4
Information Flows in Inter-organizational Networks–Connect
Individual Interactions and Network Topologies 82
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2.1 The multi-relational perspective . . . . . . . . . . . . . . . . 86
4.2.2 Formation of links in networks . . . . . . . . . . . . . . . . . 88
4.2.3 Dissemination in networks . . . . . . . . . . . . . . . . . . . 91
4.3 Proposed approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.4 A case study of the humanitarian sector . . . . . . . . . . . . . . . 97
4.4.1 Configuration of the simulation . . . . . . . . . . . . . . . . 97
4.4.2 An experiment on how to facilitate collaboration among or-
ganizations . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.4.3 Simulation setup and results . . . . . . . . . . . . . . . . . . 105
4.4.4 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.5 Summary and future work . . . . . . . . . . . . . . . . . . . . . . . 111
Chapter 5
Summary 114
Appendix A
Properties of a Good Metric for Average Delivery Efficiency 117
Appendix B Theoretical Analysis of the Degree Distribution for Simplified
RLR Scale-free Networks 119
Appendix C
The list of words and expressions related to death 122
Bibliography 124
List of Figures
| 2.1 A hierarchical supply-chain network. . . . . . . . . . . . . . . . . .
2.2 Two sample supply-chain networks with the same number of supply (S) and demand (D) nodes and edges. Network in (a) can maintain |
10 |
| flow of supplies better than the network in (b). . . . . . . . . . . . .
2.3 An example of the DLA growth model. ID for the new node is |
12 |
| underscored. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Cumulative degree distributions of supply-chain networks (each has |
19 |
| 1000 nodes). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 20 |
| 2.5 Simulated 70-node supply-chain networks with various topologies. .
2.6 The four networks’ responses to random disruptions. Each data |
21 |
| point is the average of 20 runs. . . . . . . . . . . . . . . . . . . . .
2.7 The four networks’ responses to targeted disruptions. Each data |
23 |
| point is the average of 20 runs. . . . . . . . . . . . . . . . . . . . .
2.8 The distributions of shortest supply-path length in the four net- |
25 |
| works’ largest functional sub-network (15% targeted disruption). . . 2.9 Two simple supply-chain networks with supply nodes (denoted with
S) and demand nodes (denoted with D). Each edge represents a |
28 |
| distance of 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 34 |
| 2.10 Pseudo code for the algorithm to generate the RLR network topology. 38 2.11 The distribution of the three simulated supply-chain networks with
1000 nodes each. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.12 Snapshots of a simulated 70-node 120-edge supply-chain networks with the RLR scale-free topology (Supply nodes are denoted in black and demand nodes are in white.) . . . . . . . . . . . . . . . . 40 2.13 Various military logistic networks’ responses to random supply dis- ruptions. Average of 20 runs. . . . . . . . . . . . . . . . . . . . . . 41 2.14 Various military logistic networks’ responses to targeted supply dis- ruptions. Average of 20 runs. . . . . . . . . . . . . . . . . . . . . . 42 2.15 Various military logistic networks’ responses to random supply dis- ruptions. Average of 20 runs. . . . . . . . . . . . . . . . . . . . . . 44 |
|
| 2.16 Various military logistic networks’ responses to targeted supply dis- | |
| ruptions. Average of 20 runs. . . . . . . . . . . . . . . . . . . . . .
2.17 Degree distributions of simulated supply-chain networks (pr = 1 for the RLR scale-free supply-chain network. All networks have 1000 |
45 |
| nodes and 1815 edges). . . . . . . . . . . . . . . . . . . . . . . . . .
2.18 The ln-ln degree distribution of the retailer’s distribution network |
47 |
| in California. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.19 Various retail distribution networks’ responses to random supply |
49 |
| disruptions. Average of 30 runs. . . . . . . . . . . . . . . . . . . . .
2.20 Various retail distribution networks’ responses to targeted supply |
51 |
| disruptions. Average of 30 runs. . . . . . . . . . . . . . . . . . . . . | 52 |
| 3.1 Distributions of number of posts each user published. . . . . . . . . | 60 |
| 3.2 An example of threaded discussions. . . . . . . . . . . . . . . . . . . | 61 |
| 3.3 The distribution for the number of responding replies in threads . . | 61 |
| 3.4 The distribution for the life span of threads . . . . . . . . . . . . .
3.5 Distribution of the number of discussion boards to which a user |
62 |
| contributed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6 An example of a user network with two sub-communities (IUs are |
66 |
| in gray. Users’ values on feature f are below their IDs). . . . . . . . | 68 |
3.7 A flow chart of the automatic sentiment analysis using classification. 71
3.8 Sentiment change of thread originators by number of posts. A point represents the average sentiment of thread originators’ n-th posts in threads they initiated. As the 2nd post from the originator is the 1st self-reply, the 2nd data point from the left-hand side denotes
| the average sentiment of originators’ first self-replies. . . . . . . . .
3.9 Change in originators’ sentiment as a function of the average sen- |
72 |
| timent of responding replies. . . . . . . . . . . . . . . . . . . . . . .
3.10 An example of how a responding reply influences the thread originator’s sentiment (happy and sad faces illustrate the sentiment of |
74 |
| a post). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.11 Cumulative distribution of the intervals between initial posts and |
74 |
| their first/last self-replies. . . . . . . . . . . . . . . . . . . . . . . .
3.12 Examples of how to identify IRRs from a thread. Note that Pr > 0.5 means positive sentiment (denoted with happy faces); Pr ≤ 0.5 |
75 |
| indicates negative sentiment (denoted with sad faces). . . . . . . . . | 76 |
| 4.1 An example bipartite graph for collaboration. . . . . . . . . . . . . | 90 |
| 4.2 An example inter-organizational collaboration network. . . . . . . . | 90 |
| 4.3 Pseudo code for the agent-based model. . . . . . . . . . . . . . . . . | 94 |
4.4 The process of simulation configuration and calibration. . . . . . . . 98 4.5 The communication network among 30 member organizations of
GlobalSympNet as of May 2008. . . . . . . . . . . . . . . . . . . . . 99
4.6 The collaboration networks among 30 organizations in GlobalSymp-
Net as of October 2009. . . . . . . . . . . . . . . . . . . . . . . . . 100
4.7 The predicted collaboration networks among 30 organizations in
GlobalSympNet. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.8 The communication network among the 95 humanitarian organizations as of October 2009. Node colors denote organization types. . . 104
4.9 Degree distribution of the communication network among the 95 humanitarian organizations. . . . . . . . . . . . . . . . . . . . . . . 104
4.10 Effectiveness of different strategies to promote collaboration. . . . . 107
4.11 The total number of unique candidate projects that agents evaluates. 108
B.1 Inferred degree distributions of scale-free and three simplified RLR rewired scale-free networks. . . . . . . . . . . . . . . . . . . . . . . 121
List of Tables
| 2.1 Some standard metrics for network robustness. . . . . . . . . . . . . | 12 |
| 2.2 Taxonomy of the new robustness metrics for supply-chain networks
2.3 95% confidence intervals for supply availaibility rates(%) in random |
16 |
| disruptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 95% confidence intervals for supply availaibility rate(%) in targted |
23 |
| disruptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Numerical analysis for supply availability rate (10% targeted node |
26 |
| removal) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6 Taxonomy of the updated robustness metrics for supply-chain net- |
31 |
| works. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 36 |
3.1 Summary of statistics for the CSN forum dataset used in this research. 62
| 3.2 Summary of basic user features . . . . . . . . . . . . . . . . . . . .
3.3 Compare the top-150 recall of 5 individual classifiers on 3 datasets |
63 |
| (using IU List-1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . | 65 |
| 3.4 Compare top-150 recalls of various approaches. . . . . . . . . . . . . | 69 |
| 3.5 Examples of posts in CSN and their sentiment classes. . . . . . . .
3.6 Compare the Top-K recall from various single-metric user rankings |
70 |
| (using IU List-1). . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.7 Compare the recalls and precisions of the IRR ranking and an en- |
77 |
| semble classifier (using IU List-2). . . . . . . . . . . . . . . . . . . . | 78 |
4.1 List of missions and focus regions for organizations in GlobalSympNet 99
4.2 Statistics of the simulated and actual collaboration network . . . . 102
4.3 Comparison of four communication networks. . . . . . . . . . . . . 108
Chapter 1
Introduction
Networks are pervasive in our life. Social networks, transportation networks, power grids, and supply-chain networks are just some of the important networks around us. Therefore, the emerging multi-disciplinary network science, which “develops theoretical and practical approaches and techniques to increase our understanding of natural and man-made networks” [1] , has drawn the attention of many researchers from various disciplines, including mathematics, sociology, physics, biology, computer science, information science, operations research, and so on.
Research on networks started in the 18th century and revived in the late 1990s. Its inception can be traced back to 1736, when Euler proposed the famous Seven Bridges of Konigsberg problem [2] . Since then, networks have been studied in areas related to graph theory, discrete mathematics, and optimization. Network metrics (e.g., node centrality [3] ) and algorithms (e.g., finding shortest path [4] ) have also been developed to aid the analysis of networks. However, lacking data of large-scale networks, research was often limited to relatively small networks. It was also often believed that random network [5] is the dominant topology for many real-world networks.
The rising of Internet, advances in computing technologies, and the recent prevalence of online social networks and social media all contributed to the revitalization of research on networks. Internet has made it possible and easier to collect “big data” of large-scale networks, such as the World Wide Web, citation networks, email networks, and so on. Moreover, such “big data” also includes details on how individuals interact with each other through these networks, and even the content of the interactions. As a result, researchers are now able to study networks at a scale and granularity that were not possible before.
While today’s network research covers many different topics, studies in this area can be grouped into micro-level and macro-level analysis, although it is difficult to find a clear-cut boundary between research at the two levels. At the micro level, the focus is on individuals in networks. Research topics include centrality measures for nodes [6] , reciprocity [7] , link prediction [8] , and so on. Up to the macro level, research emphasizes more on network structures and phenomena at the system level. For example, scientists have built models for the topologies and evolution of networks, such as the beta re-wiring model for small-world networks [9] , and the preferential-attachment model for scale-free networks [10] . Also, algorithms have been developed to discover community structures in networks [11] , optimize or reveal flows in networks [12] [13] , etc.
One of the important topics related to research at both levels is the flow through a network. Examples include the flow of goods through supply-chain networks, the flow of information through Internet, the flow of influence through social networks, and so on. At the macro level,to understand how the structure of a network is related to the network’s performance, it is often necessary to examine how flows through the network are enabled, facilitated, or constrained by the network structure. At the micro level, to study the dynamics or evolution of individuals’ behaviors or opinions, one may need to focus on the flow of information or influence among individuals in a network context.
This dissertation conducts a multi-level analysis of information and supplies flows in social and business networks. It not only studies flows at both macroscopic and microscopic levels, but also tries to connect network structures at the macro level and the individual behaviors at the micro level using network flows.
First, at the macro level, the dissertation studies how topological changes caused by disruptions affect the flows of supplies in supply-chain networks, which are responsible for distributing or disseminating supplies [14] . With a special focus on the heterogeneous roles that different types of nodes play (as providers, distributors, and consumers) and the characteristic of supply flows in supply-chain networks, this study first proposes new performance metrics at the topological level, such as supply availability, network connectivity, delivery efficiency, etc.
Then two customizable heuristic strategies are proposed to improve or balance the robustness of distribution networks against disruptions. The first strategy, Degree-Locality-based Attachment, is a hybrid network growth model to build or design supply-chain networks. The second strategy, Randomized Local Re-wiring, adjusts topologies of existing supply-chain networks that may be difficult or expensive to re-build. In simulations based on synthesized and real-world supply-chain networks, both strategies lead to supply-chain network topologies that can preserve more supply flows than traditional topologies after random (e.g., natural disasters and power outages) and targeted (e.g., from terrorists and cyber-attacks) disruptions, especially when it is not possible to predict which type of disruption will occur.
Second, at the micro level, my research identifies influential individuals in an online health community (OHC) by leveraging the flows of information among OHC members. While flows in a supply-chain network may be planned or managed, the flows of information among autonomous individuals in social networks are often driven by distributed inter-personal interactions. Such flows may influence people’s behaviors, attitudes, or emotions. Thus to identify influential users in OHCs, this research leverages the large-scale data of individual users’ interactions in a popular OHC among more than 27,000 cancer survivors.
Two complementary approaches are used, compared, and eventually integrated. The first approach builds classifiers to incorporate multiple metrics, which could reflect one’s influence in different ways. The structure of the online social network among OHC users is also utilized to generate neighborhood-based and clusterbased features that help to improve the performance of classifiers. The second approach tries to measure one’s influence directly by focusing on the flow of information and support among OHC users. A new and intuitive metric, the number of influential responding replies, is proposed to identify influential users. The influential user ranking by the new metric outperforms rankings by many traditional metrics based on users’ contributions (e.g., numbers of posts and active days) and social network centralities (e.g., degree centrality and PageRank). The performance of the new metric also dominates the classification-based approach, which utilizes 60 metrics of users’ contributions and centralities. Integrating the two approaches further improves the accuracy of the identification of influential users.
Last, the dissertation connects micro-level interactions among individuals and macro-level network topologies using the flow of information. Using a case study of inter-organizational networks [15] among humanitarian agencies, the research starts with a network influence model for how organizations influence others’ decisions about collaborations. Combining such exogenous network influence with organizations’ endogenous evaluation of incoming information, a novel model is proposed for how the information about candidate collaborative projects disseminates through organizations’ interactions. The dissemination of project information allows organizations to learn about, evaluate, and keep in their to-do lists several projects supported by others. When several organizations identify a mutually beneficial project and agree to collaborate on it, an event-based approach adds multiple edges simultaneously among them in the collaboration network. Agentbased simulations are used to predict how changes to the communication network affect the flow of information, which in turn alters the topology of the resulting collaboration network. Simulations also suggest a strategy to promote collaboration by encouraging interactions among organizations at the peripheral of the communication network, which will better facilitate the flow of information and collaboration.
Major contributions of the dissertation are two-fold.
First, it contributes to the utilization and management of social and business networks in various domains. For example, the study on supply-chain networks offers strategies to improve or balance network robustness against disruptions. The work on online health communities helps to build an active, supportive, and sustainable community for patients and their caregivers. The research on interorganizational networks provides suggestions on how to promote collaboration among humanitarian agencies, which will eventually benefit disaster victims.
Second and more importantly, research presented in this dissertation has general implications for the study of online and offline networks in social and business contexts. The research on supply-chain network robustness illustrates that incorporating the heterogeneous roles of nodes can provide a new perspective to the analysis of network flows. Considering such heterogeneity can also have a significant impact on the evaluation of a network’s performance. The exploration on online health communities shows analyzing the content of individuals’ interactions and conducting sentiment analysis on large-scale data can help to understand individuals’ behavioral dynamics and the impact of social influence in online settings. The investigation on inter-organizational networks demonstrates the potential to use network flows to connect micro-level and macro-level network phenomena. It also advocates the multi-relational perspective of network analysis by showing how dynamics in one network could affect another.
The remainder of the dissertation is organized as follows. In Chapter 2, I will introduce the topological study on supply flows and the robustness of supply-chain networks against disruptions. Chapter 3 will cover the individual-level analysis of information flows in OHCs and the identification of influential users. In Chapter 4, the dissertation will describe the research on how information flows connects individual interactions and network topologies in inter-organizational networks. Finally, Chapter 5 provides a summary of the dissertation, as well as implications and contributions of this dissertation.
A MULTI-LEVEL ANALYSIS OF INFORMATION AND SUPPLY FLOWS IN SOCIAL AND BUSINESS NETWORKS
