Fig 1 - available from: Applied Network Science
This content is subject to copyright. Terms and conditions apply.
Comparison of random, small-world and scale free networks. Topological structure of benchmark network models. Random and Small-world network topologies do not include hub nodes. In contrast, scale-free topologies are characterised by the presence of small number of highly connected hub nodes and a high number of feebly connected nodes. Presence of distinct hubs in scale-free networks make them more vulnerable to targeted attacks, compared to random and small-world networks
Source publication
Due to the increasingly complex and interconnected nature of global supply chain networks (SCNs), a recent strand of research has applied network science methods to model SCN growth and subsequently analyse various topological features, such as robustness. This paper provides: (1) a comprehensive review of the methodologies adopted in literature fo...
Similar publications
Network motifs are topological subgraph patterns that recur with statistical significance in a network. Network motifs have been widely utilized to represent important topological features for analyzing the functional properties of complex networks. While recent studies have shown the importance of network motifs, existing network models are not ca...
Citations
... Network science enables us to perceive and analyze the intricate web of relationships and dependencies within systems, providing a holistic framework that enhances predictive accuracy, decision-making, and system optimization. This work not only strengthens the role of network science in contemporary analytical domains but also helps broaden its application in other complex systems such as supply chain management [57] and urban planning [58]. As network science continues to develop, it is promising that it would spearhead further innovations, benefiting an even broader spectrum of industries by transforming how we understand and interact with connected data. ...
This dissertation explores the applicability of network science in solving real-life business problems that involve large-scale complex systems. Specifically, the study investigates how network science can be used in financial market forecasting, decision- making in stock market trading, and conversational AI systems.
In the financial market forecasting, the study proposes a new method that utilizes network measurements extracted from S&P 500 networks to improve the predictability of conventional ARIMA models. The study shows that changes in the network strength distributions provide important information on the network’s future movements, and the inclusion of network measurements significantly improves the model accuracy. Moreover, the study proposes a network-based Exponential Moving Aver- age (EMA) method for making trading decisions, which outperforms the traditional EMA with an ability to capture more trading opportunities and yield a bigger return at each trade.
In the conversational AI area, the study proposes a network community detection- based approach to automatically label text data for solving a classification problem, which outperforms human-labeled data by 2.68-3.75% in classification accuracy. This approach helps detect mislabeled and ambiguous data points that negatively affect model performance, reducing development time and cost for industries using conver- sational AI technology.
Overall, the study demonstrates that network science can offer significant insights to some of the most critical modeling problems in business applications by providing an additional and powerful perspective to the traditional methodologies. As this dissertation presents, the ability to integrate network science into various business analytics can lead to more informed decision-making, enhanced development efficiency, and a competitive edge in rapidly changing business environments.
... Kim, Chen, and Linderman (2015) developed four connectivity link models, including block-diagonal, scale-free, centralised, and diagonal link structures. Perera, Bell, and Bliemer (2017) used attachment probability to determine connectivity links based on the fitness of nodes. However, previous studies assume that nodes with an equal degree are added to the SCN, which does not represent real-world SCNs, where firms may have different numbers of links. ...
... The propagation of economic shocks has been traditionally studied using industry-level input-output tables [23], although the importance of firm-level supply data to provide an unbiased assessment of the consequences of individual failures [24,25] has been increasingly recognized [26][27][28]. Network theory is, thus, becoming an increasingly popular tool to analyze production systems [29][30][31][32][33][34][35][36][37][38][39][40], analogously to what has happened within the realm of financial systems in the aftermath of the 2008 crisis [41]. ...
... [10]) or sector-specific (e.g. global automotive [11], aerospace [12], or computer and electronics [13]) supply chains. While firm-centric and industry-specific studies have been important to gather insights into how network features shape the operation of supply chains, it remains hard to generalize these findings, due to the sector-specific and incomplete nature of these datasets. ...
... On a sample of ∼90 k connections between South Korean firms, where 20% of the examples are used as a test set, the authors achieve an AUROC of 0.92. 13 This trajectory of studies reflects a consistent evolution in methodology, with each iteration contributing incremental enhancements in feature integration and model sophistication, partially akin to what we will see now for papers which address supply network reconstruction specifically. ...
... The data is provided by Teikoku Databank Ltd a business intelligence company. 13 The authors do not specify the undersampling ratio of their exercise. 14 The embedding usually consists of computing an average distance d between node k and the nodes i and j, and then embedding k in a vector f k ij = δ dd ′ . ...
Network reconstruction is a well-developed sub-field of network science, but it has only recently been applied to production networks, where nodes are firms and edges represent customer-supplier relationships. We review the literature that has flourished to infer the topology of these networks by partial, aggregate, or indirect observation of the data. We discuss why this is an important endeavour, what needs to be reconstructed, what makes it different from other network reconstruction problems, and how different researchers have approached the problem. We conclude with a research agenda.
... Random networks are rare to observe in real-world social and organizational networks as they assume the randomness of link creation between the graph nodes. As such, random networks are usually used as the null hypothesis in categorizing a network and testing the attribution of its connectivity to randomness (Perera et al., 2017). Small-world networks, mathematically formulated by Watts and Strogatz (1998), mimic the natural social and cultural tendency of individuals and organizations to connect based on similar interests or attributes. ...
... There are CI two main implications of the scale-free network structure of the CLT industry. First, the industry supply chain may be subject to extreme disruptions due to its vulnerable centralization around few stakeholders (Kereri and Harper, 2019;Perera et al., 2017), mainly the timber contractors, engineers and manufacturers. Second, the scale-free structure of the collaborative network enables efficient and fast knowledge diffusion due to the existence of influencing network hubs to accumulate and disseminate information rapidly (Lin and Li, 2010;Manshadi et al., 2020;Tang et al., 2010). ...
Purpose
Cross-laminated timber (CLT) is an innovative construction material that provides a balanced mix of structural stiffness, fabrication flexibility and sustainability. CLT development and innovation diffusion require close collaborations between its supply chain architectural, engineering, construction and manufacturing (AECM) stakeholders. As such, the purpose of this study is to provide a preliminary understanding of the knowledge diffusion and innovation process of CLT construction.
Design/methodology/approach
The study implemented a longitudinal social network analysis of the AECM companies involved in 100 CLT projects in the UK. The project data were acquired from an industry publication and decoded in the form of a multimode project-company network, which was projected into a single-mode company collaborative network. This complete network was filtered into a four-phase network to allow the longitudinal analysis of the CLT collaborations over time. A set of network and node social network analysis metrics was used to characterize the topology patters of the network and the centrality of the companies.
Findings
The study highlighted the scale-free structure of the CLT collaborative network that depends on the influential hubs of timber manufacturers, engineers and contractors to accelerate the innovation diffusion. However, such CLT supply collaborative network structure is more vulnerable to disruptions due to its dependence on these few prominent hubs. Also, the industry collaborative network’s decreased modularity confirms the maturity of the CLT technology and the formation of cohesive clusters of innovation partners. The macro analysis approach of the study highlighted the critical role of supply chain upstream stakeholders due to their higher centralities in the collaborative network. Stronger collaborations were found between the supply chain upstream stakeholders (timber manufacturers) and downstream stakeholders (architects and main contractors).
Originality/value
The study contributes to the field of industrialized and CLT construction by characterizing the collaborative networks between CLT supply chain stakeholders that are critical to propose governmental policies and industry initiatives to advance this sustainable construction material.
... Network models are broadly categorized as Generative models (e.g., ER model for random graphs (Karoński and Ruciński 1997), WS model for small-world graphs (Humphries and Gurney 2008)) and Evolving models (e.g., BA Model (Barabási et al. 1999;Barabasi and Albert 1999)). However, static models like ER and small-world are less applicable to real networks, while evolving models, like the BA Model, can simulate various network types through a growth process (Perera et al. 2017). Scale-free models are apt for homogeneous networks, often seen in IoT, as they exhibit a power-law distribution of node degrees, making them robust against random failures but susceptible to targeted attacks ( Melthis et al. 2016;Sohn 2018;Mozafari and Khansari 2019). ...
The proliferation of the Internet of Things (IoT) in healthcare necessitates networks that are robust against potential node or link failures to ensure uninterrupted patient care. This research concentrates on enhancing the topology robustness of IoT-based smart healthcare networks. Utilizing Schneider R as a robustness metric, a system model is developed, where IoT nodes’ geographic information is centrally stored. Introducing the Convergence And Robustness Efficiency (CARE) solution, we exploit a geometric approach within genetic algorithms (GA) to enhance the network’s topology robustness. CARE’s nested strategy tackles the slow convergence and high computational costs seen in contemporary techniques, leading to optimized results. Simulations show that CARE surpasses conventional GA by an impressive 21% in Schneider R and only degrades by 7.5% (compared to 16% of existing solutions) when node density surges to 200 in healthcare settings. By tapping into efficient computing methods, this research paves the way for robust healthcare IoT networks, ensuring patient safety and data integrity.
... The propagation of economic shocks has been traditionally studied using industry-level input-output tables [23], although the importance of firm-level supply data to provide an unbiased assessment of the consequences of individual failures [24,25] has been increasingly recognized [26][27][28]. Network theory is, thus, becoming an increasingly popular tool to analyze production systems [29][30][31][32][33][34][35][36][37][38][39][40], analogously to what has happened within the realm of financial systems in the aftermath of the 2008 crisis [41]. ...
Production networks arise from supply and customer relations among firms. These systems are nowadays gaining attention as a consequence of the supply chain disruptions due to natural or man-made disasters that happened in the last few years, such as the Covid-19 pandemic or the Russia-Ukraine war. Recent empirical evidence shows that production networks are shaped by functional modules arising from 'complementary relationships' between firms. However, data constraints force the few, available studies to consider only country-specific production networks. In order to fully capture the crosscountry structure of modern supply chains, here we focus on the global automotive industry as represented by the 'MarkLines Automotive' dataset. After representing this data as a network of manufacturers, suppliers, and products, we perform a pattern-detection exercise using a statistically grounded validation technique based on the maximum entropy principle. We reveal the presence of a significantly large number of V-shaped and square-shaped motifs, indicating that manufacturing firms compete and are seldom engaged in a buyer-supplier relationship, while they typically have many suppliers in common. Interestingly, 'generalist' and 'specialist' suppliers coexist in the network. Additionally, we unveil the presence of geographical patterns, with manufacturers clustering around groups of suppliers; for instance, Chinese firms constitute a disconnected community , likely an effect of the protectionist policies promoted by the Chinese government. We also show the tendency of suppliers to organize their production by targeting specific 'functional modules' of a vehicle. Besides shedding light on the self-organising principles shaping production networks, our findings open up the possibility of designing realistic generative models of supply chains, to be used for testing the resilience of the interconnected global economy.
... The parameters of the supply chain networks' degree distribution vary somewhat across regions, industries, and datasets studied. The authors of Perera et al. (2017) consulted 10 sources and reported that the parameters of the degree distribution of the corporate SCN ranged from 1.0 to 3.3, depending on the literature. Meanwhile, Fujiwara and Aoyama (2008) estimated the parameters of the cumulative degree distribution. ...
This study constructs an agent-based model suitable for analyzing the propagation of economic shocks based on a macroeconomic agent-based model structure that covers major economic entities. Instead of setting an upstream and downstream structure of firms in the inter-firm networks, our model includes a mechanism that connects each firm through supplier–customer relationships and incorporates interactions between firms mutually buying and selling intermediate input materials. It is confirmed through the proposed model’s simulation analysis that, although a firm’s sales volume temporarily falls due to an economic shock of the type that causes a sharp decline in households’ final demand, the increase in assets held by households as they refrain from spending rather expands their capacity for consumption. As a result, after the economic shock ceases to exist, the firm’s sales volume tends to be even greater than that of the preceding periods of the shock. Furthermore, we found that when the sales volume of products in a final consumer goods sector falls during the shock, the falls in sales in the non-final consumer goods sectors are suppressed due to replacement demand, and the increase in sales volume for the non-final consumer goods sectors is moderated after the shock ceases to exist.
... Especially the importance of scale-free networks should be in focus. The power-law exponent of γ = 2.40 for the tree-mesh network analysed in this study, confirms the values described in the literature (Perera et al., 2017). Future studies need to show where exactly the industry-specific thresholds lie or how the synchronisation transitions are affected by the scale of the networks. ...
In this paper, we examine the transition process between synchronous and desynchronous states in supply chain networks, which is strongly correlated to network topology. While most research has focused on the merits of steady-state synchronization behavior, very little has been published on the interdependencies between structural and dynamic synchronization properties in supply chain networks. Synchronization processes are described by a variety of coherent state dynamics that exhibit very different transition paths. In this research, we developed a new coupled oscillator model to analyze and evaluate these transient dynamics in more detail. Canonical network models are first described, including hub-and-spoke, tree-mesh, and tree topologies, which are then applied to real-world problems in the automotive industry. We found specific synchronization transitions in automotive supply networks such as explosive and delayed synchronization. In particular, we show that (i) in a hub-and-spoke supply network, the number of suppliers per hub affects synchronization transitions only at larger frequency spreads, (ii) scale-free networks with short path lengths generated by the power law connectivity are unable to synchronize as long as a critical cluster coefficient is exceeded, and (iii) interruptions in tree networks cause nonlinear synchronization losses for upstream suppliers as a function of the interruption time.
... The initial investigative work on random graphs in the '50s by Erdós and Renyi shaped the formation of Network Science. The coining of the term 'small-world' in 1998 by Watts and Strogatz and the introduction of the 'scale-free' property of networks in 1999 by Barabási and Albért has fuelled the research work in this field of science [22]. Based on various properties and prospects for studying network topology robustness, broad categorization of network models is done into two groups, i.e Generative models (examples: ER model for random graphs [23] and WS model for small-world graphs [24]) and Evolving models (example: BA Model [25,26]). ...
... It is due to this reason that utilization of both these models is negligible in the case of real networks. Moreover, evolving network models can be adapted to simulate a wide range of generative models by setting a growth process [22]. Given its ability to simulate the growth and evolution of real-world networks, BA model is widely recognized as a valuable tool for creating and modeling various types of networks. ...
The Internet of Things (IoT) has become a pivotal component of modern technology, enabling the interconnection of numerous intelligent devices. It is envisioned to revolutionize various fields through the realization of numerous applications across industries, including healthcare, precision farming, transportation, and smart cities. However, the practical problem of node and link failures due to various constraints, such as low power backup, adverse environmental conditions, fragile metallurgy, and targeted attacks, can negatively impact the overall functionality of IoT networks. To address these challenges, soft and intelligent computing techniques, such as classical heuristic and genetic algorithms, can be utilized to improve the topology robustness of future IoT networks. In this study, we present a system model for a smart healthcare network and evaluate the effectiveness of optimization techniques, including the newly proposed Optimization of Topology for Efficient Convergence (OTEC), which combines a geometric approach in a genetic algorithm with two mechanisms, Efficient Edge Swap (EES) and Node Removal based on Threshold (NRT), to address key limitations in existing techniques. The metric of Schneider R is used to assess the robustness of the topology, with geographic information about IoT nodes and their neighbors stored on a central big data server. The OTEC approach utilizes a nested approach, demonstrating significant improvement over classical genetic algorithms by achieving a 21% increase in Schneider R. Additionally, OTEC effectively addresses limitations present in traditional heuristic algorithms such as ROSE, Simulated Annealing and Hill Climbing. The concept of trustworthiness is also considered from an overall perspective of system functionality. The optimized topology not only improves the robustness of the network but also enhances its trustworthiness, ensuring that the system can operate reliably and ethically. By leveraging soft and intelligent computing techniques, this study provides a valuable contribution to the development of future IoT networks that can handle the challenges of node and link failures while remaining trustworthy and dependable.
