Two-layer visual analytics of truckers’ risk-coping social network

Abstract

Within organizations, managers’ specific responsibilities and domain expertise shape their interests in the output of social network analysis. Our proposed visualization approach is tailored to meet the operation-directed needs and preferences for visual analysis of specific tasks. This method prioritizes an overall geographical map with focal-contextual dynamics within the network. To enable a comprehensive and in-depth understanding of pinpointed focal areas, we customize an analytical framework for analyzing inter-community networks. We extract focal sub-networks from specific nodes to create graph visualization for detailed analysis, represent rich types of domain-specific graphic properties, and provide direct zoom+filtering interactions to allow easy pattern recognition and knowledge discovery. We applied our approach to visualizing the data from interactions among 300 city-based truck communities on the largest occupational platform for truckers in China. We also conduct a case study to demonstrate that our approach is effective in supporting managers’ network analysis and knowledge discovery.

Keywords

Visual analysis data visualization social networks knowledge discovery visual information processing

Introduction

In organizations, managers’ decision-making is constrained by their hierarchical position and available resources. Senior-level managers typically prioritize holistic patterns and long-term trends, whereas junior-level managers often focus on immediate tasks, direct connections, and local concerns. Junior-level managers’ position and their access to limited resources shape their approach to the use of network visual analysis. They analyze data from their department’s viewpoint, using the broader network as a backdrop for decisions. Their analysis aims to identify practical problems, drawing on their experience and domain-specific knowledge. Previous design in visual analysis falls short of fully considering junior-level managers’ needs. To facilitate the targeted uses of visual analysis, we present a framework of innovative interactive visual data analytics to satisfy the needs of junior-level managers and cultivate a deeper understanding of internal organizational relationships.

The following three sub-sections discuss the essential components of this framework. We highlight how junior-level managers’ use of network visual analytics can be improved, then introduce our design proposal and elucidate its contributions to the existing literature.

Visual information processing pipeline

A well-defined procedure of visual analytics can be supported by direct complex statistical analyses and intricate visualizations.^1,2 The path from raw data storage to knowledge discovery through visualization is established on the underpinning analytical frameworks, within which the Data Transformation, Visualization, and Visualization Interaction stages are addressed,^3–6 as shown in Figure 1. In particular, the data transformation stage prepares data to identify key variables and statistical relationships. In this stage, the application of statistical methods, data mining techniques, and algorithms to extract patterns and relationships from the data can be considered as data transformation. Visualization stages map relationships and patterns to human-friendly visual forms. The mapping relationship can be considered as a visual transformation process. The visual interactions stage enhances users’ engagement by including functions such as zooming, filtering, and drilling down. Visual interaction involves flexible view transformation that integrates both data and visual transformations. The framework is applied to accommodate macro, meso, or micro-level analyses,⁴ and support to complete a substantial workload of network data.^7,8 Based on the framework, the adaption to specific tasks, such as optimizing data quality,³ and visualizing egocentric network evolution⁴ is proposed to enhance its applicability and foster fostering users’ knowledge discovery.⁹

Figure 1.

Traditional visual information processing framework.

The process assumes that the three stages as a flow serve the visual analytics, starting from data analysis, moving to interaction, and concluding with visualization. In an ideal condition, sufficient knowledge would be revealed by defining the data from the data analysis stage. However, when this framework is applied to junior-level managers’ analysis, the stages proceed in an undefined sequence. The managers need a comprehensive overview and the exploration of local network visualization in its context. Multiple circles composed of data subset selection, overview, and focus-context relationship analysis would be conducted to achieve their commercial insights. The data, visual, and view transformation supported by the functions in each stage need to be tailored to their needs. Unlike the above-mentioned flow, we propose a two-layer leverage-seeking visual analysis framework. The framework centers on rough visualization, followed by interaction and data analysis, leading to refined visualization. The three types of transformations are carefully tailored to empower the global-scale network analysis.

Network analysis for organizational knowledge discovery

Along the step-by-step progressive framework, the managers refine data to discriminate their path toward a managerial objective. They need to first update their overview of the network, that is, their knowledge discovery context, and then analyze a focal network. The network overview can be presented by either geographic visualizations of physical entities within the social-physical environment,¹⁰ or abstract network visualizations in abstract space.^4,7 Based on the overview, the managers shape their focus, and proceed to analyze focal networks and seek insights guiding actions. In the focal network analysis, the abstract network visualization is widely adopted to support the understanding of multifaceted network attributes. The entire process entails the data transformation from the overall network to the focal network, and view transformation between multi-layered network representations.

In the current visualization literature, Overview+Detail visualization and Focus+Context visualization are the two primary methods addressing the aforementioned needs of managers in network analysis. Overview+Detail visualization provides a global overview and a detailed view, enabling users to switch between the two to obtain both high-level and granular information.^11–13 This approach offers a clear global perspective while allowing interactive access to detailed data.^14,15 However, frequent switching between views can increase cognitive load, especially when dealing with very large networks.¹² On the other hand, the Focus+Context method simultaneously displays the focal area and contextual information within a single view, reducing the need to switch between views and maintaining visual consistency.^16–18 This is typically achieved through distortion techniques such as fisheye lenses, which magnify the focal area while preserving the overall network structure.^19–21 While this method minimizes cognitive load by keeping everything in one view, it is more complex to implement and may not be as intuitive for visual transforming. Both methods have their strengths and limitations and do not fully meet all managerial needs.

Transformations offer a potential solution to the limitations of existing network analysis methods by enabling connections and transitions between different layers of visualization. Visual transformations allow managers to switch between overview and detailed views, ensuring that both global context and local details are adequately represented. This is particularly important for distinguishing focal local networks within a broader context. Previous literature has often overlooked this differentiation, focusing instead on network structures from global, local, or nodal perspectives respectively.^4,7,8,22 While prior studies have emphasized overall network dynamics⁸ and the complex interdependencies among various entities in socio-physical information systems – such as supply chain networks,^10,23 water supply networks,²⁴ road networks,²⁵ economic networks,²⁶ and sector-to-sector relationships²⁷– junior managers are often more concerned with the specific details and context of local networks within their broader operational environment.

We propose a two-layer framework that combines elements of Overview+Detail and Focus+Context visualization with transformations to better meet the needs of junior managers in network analysis. Our framework’s first layer focuses on the overview, providing context information to managers, while the second layer delves into the details of local networks for focused analysis. A key contribution of our framework is the emphasis on visual transformation between these two layers, facilitating a seamless transition from global to local perspectives in network data analysis. In our work, visual transformation establishes a connection between the focal entity and its context. We assume that the managers have accumulated rich knowledge in their department or domain, but relatively lack understanding of the overall picture of the business. As per their available knowledge, we designed a focal view in an abstract space to facilitate logical inferences, while we designed a context view of the social-physical environment to provide managers with rich information of a holistic picture. The transformation between physical and abstract visualization empowers the identification of how the features of the focal entity change along with its context.

Visual constancy in complex operational environment

It could be challenging to analyze a focal local network among a complex network.²⁸ To position a focal local network within its context and track its changes along with the context leads to a great cognitive overload for managers. Therefore, it is crucial to reduce their cognitive effort. One approach to reduce cognitive effort in focal+context analysis is to improve the manager’s experience of visual constancy. Visual consistency refers to the ability to maintain a stable perception of objects even when the viewing conditions vary.²⁹ It helps to keep a stable perception of things, despite changes in appearance, position, or angle. When users adopt interactive visualization to analyze the focal local network, visual constancy significantly influences the ease in recognition of the focal entities from their changing context and the experiences of coherent observation.^30–32

The transformation in views makes it challenging to maintain visual constancy. Both highlights³³ and fixed position, facilitating focal object tracking in static view have their constraints when applied to view transformation. The highlight of focal objects could lead to confusion when diversified colors are assigned key information. The fixed position might hinder algorithms from generating fair knowledge across all nodes. The view transformation may exacerbate the disadvantage of highlights and fixed position, especially when view transformation between geographical space and abstract space is applied and between a dataset and subset visualization. To address this challenge, we introduce a system design that allows for controlled highlighting of the focal entity during different view transformations.

Research context: Trukers’ risk-coping networks

Long-distance truck driving, while not highly lucrative, is fraught with unexpected challenges during prolonged trips. Truckers, often faced with the high costs of professional towing services, usually rely on their peers for assistance. This has fostered a self-organized community, where truckers exchange information, and seek and offer support. Traditional communication methods, such as phone calls and group chats, confine assistance resources within a small, closed circle. The advancements in internet technology facilitate an online social network and provide hubs for truckers to share experiences with, seek help from, and offer assistance to a greater number of peers. On these online platforms, truckers actively participate in discussions on road conditions, safety and maintenance, improving the resilience of the trucking industry. Now, the platform connects >2.7 million truckers across 300 cities. Their social interactivities have formed a complex social network. Truckers, upon joining, affiliate with a city based on their residence or frequent delivery destinations.

Community members utilize the platform to request and offer help, creating a trukers’ risk-coping network. This city-based system enables the platform to integrate offline resources, forming a robust, nationwide support network to cope with disasters and accidents. The risk-coping network, with cities as nodes and assistance as relationships, presents complex data characteristics, resulting in millions of edge relationships. A visualization system is essential to help managers understand city connectivity in the network and map various attributes for informed management decisions.

We interviewed the city-level managers of the platform to understand their daily work and their needs when using visual analytics tools. The managers were deeply involved in the platform’s daily operations and interacted with truckers to gather valuable feedback. A junior manager is responsible for managing truckers in an urban community, knowing who our community has been helped by and helps. Their focus of interest is to gain insights into the trucker community network structure attributes and find out where they can leverage the performance of future operations (such as risk-coping efficiency) at their current job position. Managers need to comprehend the dynamics around focal cities, recognizing important connections and unique aspects in inter-city relationships. Their analysis focuses on three main tasks:

T1 – National Network Insights: Observing the national inter-city network to grasp basic patterns.

T2 – Focal City Analysis: Conducting thorough analyses of specific cities, exploring their network links with others.

T3 – Resource Identification: Searching for potential resource opportunities and identifying main resource stakeholders.

Network technology have facilitated the information sharing of trucker communities, and raise the challenges for junior managers at the city-level to effectively manage the vast and complex networks, such as to identify the network resilience when natural hazard approaches, and to reduce network vulnerability by allocating resources to or requesting resources from certain nodes. An increasing number of tasks involves dynamic interaction at both global and local levels. Those tasks cannot be adequately supported by traditional single-layer data views, and call for two-layer visual analytics framework. Our analytics framework would enable deep understanding of the interactions and resource flows between cities, allow managers to comprehend the network structure on a macro level and identify detailed information of their focal cities on a micro level.

Proposed novel visualization

Overall, this research utilizes data visualization techniques to serve junior-level managers’ visual analysis. To match their needs and from their perspective of view, we introduce a tailored analytical framework to support the explorative search for paths to objectives. We prioritize the focal-contextual transformation within the network, facilitating a nuanced understanding of influential relationships. Also, we improved the visual consistency experiences by optimizing the representation of focal entities and local networks in their context. To test whether our setup can help junior-level managers’ knowledge discovery, we conducted a case study to test the effectiveness of our design.

Proposed two-layer visualization system: Framework and transformation

Framework

We introduce a two-layer leverage-seeking visual analysis framework (Figure 2) building upon the generic visualization model (shown as in Figure 1). In line with the prior model, our framework initiates with data and concludes with knowledge discovery. What is tailored to fit the managers’ needs is the two visualization stages, a progression emphasizes ongoing navigation of data and visualization. In the global stage, the user chooses data for transformation, leading to context visualization, performs context-focal interaction, and then selects the visualization object of interest (data in a specific range) to enter the local stage. In the local stage, users conduct detailed, specific visualization of their focused entities. Users are allowed to return to the global stage and reselect the object. The convenient (re)selection of data and view transformation between two stages facilitates users’ knowledge discovery.

Figure 2.

Two-layer interactive visual data analytics framework.

Given junior managers’ managerial knowledge, we have tailored our framework to use geographical visualization for the global stage and abstract visualization for the local stage. Junior managers typically focus on a city and its surrounding areas, lacking senior managers’ broader perspective and comprehensive knowledge. Therefore, geographical visualization in the global stage provide a broad overview of the entire network, aiding junior managers in quickly grasp the national operational environment. In the local stage, junior managers have accumulated implicit knowledge of the local network and familiar with their immediate geographic context, abstract visualization provides deeper insights. The two types of layouts match the junior managers’ needs and facilitate to reveal complex and latent inter-city relationships.

Visual transformation at the system level

The user-friendly framework design (shown in Figure 2) is built on the transformations of data and views (Figure 3). How the transformation works is illustrated as follows. In the global stage, the process begins with raw data. The data are transformed into a whole network. The network is presented through geographical visualization, an overview of the social-physical context. In the local stage, the selected entities from the geographical visualization, a set of refined data, are visualized as a local network. The network is then represented through abstract visualization as a focal view. The abstract visualization discards geographical details, and instead addresses statistical attributes of the network, which aids managers in incorporating their knowledge into their analysis.

Figure 3.

Flowchart for visualization transformation. This flowchart illustrates a two-stage data visualization transformation process, transitioning from a comprehensive global overview to a detailed local analysis. Initially, the Global Stage presents an extensive display of all data elements, including geographical and network-wide visualizations for contextual understanding. The central Data Transformation module serves as a critical juncture, refining broad data sets into focused segments. Subsequently, the Local Stage emphasizes specialized, targeted local network that highlight refined data and abstract visualization.

Visualization & interaction design

Our two-layered visualization design incorporates two types of visualization: context visualization for conveying broad geographical information, and focal visualization for emphasizing specific areas. We employ the Attributed Graph Model to map this data into these two perspectives and to align domain-specific attributes. We’ve designed a user interface using Python, offering an integrated visualization canvas and interactive experience. In addition, we describe the transformation processes of data and views within this system, along with further detailed design elements.

Attributed graph model

We visualize entire truckers’ risk-coping network as a graph $G = (V, E)$ which consists of a finite set of V vertices and a finite set of E edges. A vertex represents a city and an edge represents a relation between two cities. The entire G or a subset G’ is visualized through the context view and the focal view, as showed in Table 1 and Figure 4.

Let the graph $G = (V, E)$ represent the entire network, where $V$ is the set of nodes (cities) and $E$ is the set of edges (relationships). The layout of the graph $G$ is typically represented by a mapping function $L : V \to R^{2}$ , where $L (v) = (x_{v}, y_{v})$ denotes the coordinates of node $v$ in the two-dimensional plane.

In the context geographical visualization, the layout of the graph $G$ is geographically mapped as $L_{geo}$ , where $L_{geo} (v_{i}) = (x_{i}, y_{i})$ in which $(x_{i}, y_{i})$ represent the latitude and the longitude coordinates of node $v_{i}$ . Two background images under this visualization are Standard Map ( $M_{strandard}$ ) and Satellite Map ( $M_{satellite})$ .

In the focal graph visualization, a subgraph (subnetwork) $G' = (V', E')$ is extracted from the entire network $G$ , while we have $G' \subseteq G$ , and a force-directed layout algorithm is applied $G'$ to draw this subgraph in a 2D space.

Static and dynamic modes are used to draw the subgraph $G'$ in an abstract space. In the static drawing of $G'$ , the layout function $L'$ comprises algorithms such as the fruchterman-reingold layout ( $L_{fr})$ ,³⁴ spiral layout ( $L_{spiral}$ ), and circular layout ( $L_{circular}$ ). The abstract coordinates $L' (v)$ for node $v$ are determined based on the chosen algorithm: $L' = {L_{fr}, L_{spiral}, L_{circular}, \dots}$ . In the dynamic drawing of $G'$ , a force-directed algorithm is employed.

Table 1.

Mapping model in two stages.

	Global stage	Local stage
Network	Whole network: $G = (V, E)$	Local network: $G' = (V', E'), G' \subseteq G$ ,
Mapping function	$L_{geo} : V \to R^{2}$	$L' : V' \to R^{2}$
Visualization	Geographical visualization	Abstract visualization
View	Standard map ( $M_{strandard}$ )	Static view: $L' = {L_{fr}, L_{spiral}, L_{circular}, \dots}$
	Satellite map ( $M_{satellite})$ .	Dynamic view: force-directed algorithm

Figure 4.

Mapping model and transformation process in two different visualizations. This diagram outlines the data transformation and mapping process across two distinct visualization stages. It begins with data for the whole network which then undergoes a mapping, preparing it for visualization. The geographical visualization stage presents the data in two formats: a standard view and a satellite view, offering diverse perspectives of the same dataset. A subsequent transformation channels the data from the geographical stage to the local stage, refining it into selected data for the local network. This selected data is then visualized through abstract visualization, including static and dynamic views with multiple layout options.

We adopt the attributed graph visualizations in our research. Attributed graph visualizations, commonly used in Collaborative Networks, to map graphic properties of nodes and edges to represent characteristics like categories and connections in Collaborative Networks.³³ Attribute property is a set of domain-specific attributes or properties that are associated with cities and relationships. We use two types of graphical properties: node properties and edge properties that are used to represent the domain-specific attributes of network objects and relationships among objects in the truckers’ risk-coping network. The mapping process is described in the following sections.

Node Size: The node size is configurable to reflect the number of connections, with options including in-degree (see Figure 5(a)), out-degree, or total degree (the sum of in-degree and out-degree) (see Figure 5(d)). In the absence of attribute representation, the node size remains uniform (see Figure 5(c)).

Node Color: Nodes are set to blue by default, with capital cities marked in red. The color can be adapted to highlight specific cities (see Figure 5(a), (b), and (d)) or to represent connection attributes (see Figure 5(c)), where the color intensity indicates the number of connections for each node.

Node Shape: The default shape for nodes is circular. If representing different attributes, the shape can be altered to indicate the province to which a node belongs (see Figure 5(c)).

Edge Color: Edge colors are uniformly set, typically blue (Figure 5(a)) or gray (Figure 5(c)), unless representing specific attributes. In such cases, colors can vary, for instance, edges originating from a focal node in red, those targeting the focal node in blue, and others in gray (Figure 5(c)).

Edge Width: The width of the edges is consistent (Figure 5(c)) unless it is used to represent the weight of connections, in which case it varies to indicate the frequency of help between nodes (see Figure 5(a), (b), and (d)).

Edge Arrow: In this directed network, edge arrows can be employed to indicate the direction of assistance, showing the flow of help between nodes (see Figure 5(d)).

Figure 5.

Multi-attribute graphics in different views. By adjusting the shape, size, and color of nodes, as well as the width, color, and arrow styles of edges, these visualizations (a-d) effectively represent various domain-specific attributes.

TruckerNet: An interactive visualization user interface of trucker networks

To visualize and interact with the aforementioned model at the terminal level, we’ve developed a terminal system. Figure 6 shows the main visual user interface (UI) of our system, consisting of three key sections: the Network Generation panel, the Network Canvas, and the Parameters panel (more details in Table 2).

Figure 6.

The user interface designed for trucker networks. The UI comprises three main components: the Network Generation panel (A), the Network Canvas (B), and the Parameters panel (C). The Network Generation panel offers controls for network creation, selection options, and a data overview. The Network Canvas serves as the central interactive visualization interface, allowing user engagement. The Parameters panel provides adjustable settings to fine-tune the network visualization.

Table 2.

Components of user interface designed for trucker networks.

Components	Description
A. Network Generation	Provides tools for generating networks and exploring network data.
A1. Time Range	Allows users to define the start and end dates for data analysis.
A2. Select and Filter	Users can select specific nodes or network items and apply filters based on properties.
A3. Data Overview	Displays a summary of the data points within the selected range, including details like names and counts.
B. Network Canvas	The central interactive map showing network activity among cities, with tools to zoom and adjust the view.
C. Parameters	Provides options to manage display settings such as city marks, geographical settings, layout, edge, node, layout, and etc.

The Network Generation panel (see A in Figure 6) consists of the Time Range section (A1) for generating the whole network, the Select and Filter section (A2) for generating the sub-network, and the Data Overview section (A3).

• Time Range (see A1 in Figure 6): This section generates the whole network based on a specified time range, representing the initial step of the interactive interface for network generation.

• Select and Filter (see A2 in Figure 6): The panel generates the sub-network based on the whole network generated in the initial panel.

■ Selection: A specific range of nodes can be chosen through a drop-down window. Upon clicking the “Select” button, these selected nodes and their connected networks are generated.

■ Filtering: This section establishes filtering conditions through three distinct windows. The first window allows for the selection of network items, including nodes and edges. Following this, a “select a property” column is generated, corresponding to the properties of the chosen node or edge, thus providing a range of further choices. For example, selecting a node might offer properties such as the node’s origin, destination, province, and other relevant attributes. The third window then presents value options based on the selected property. Activating the filter button generates a network that reflects these specified conditions.

• Data Overview (see A3 in Figure 6): The third panel serves as the data preview panel of the network, encompassing details such as the date and time, the sending node (from), the receiving node (to), and the relationship count.

The Network Canvas panel (see B in Figure 6) comprises two views. Each view allows for panning across the canvas by dragging and supports zooming in or out with the mouse wheel. The views can be switched by clicking on the button above.

Context View: The context view provides the contextual geographical overview of the entire dataset. City nodes are labeled based on real geographical coordinates, and the network relationships between nodes are depicted. The geo view incorporates two visualization modes. (1) Standard map (refer to Figure 5(a)), displaying administrative boundaries, railways, and rivers, offering rich social-physical environmental information. (2) Satellite map (refer to Figure 5(b)), revealing terrain and topographic features, providing authentic physical environmental characteristics.

Focal View: The focal view allows detailed analysis of selected nodes, employing refined data analysis and visualization methods on fewer nodes. The focal view encompasses two visualization modes: (1) Static view, enabling users to display relationships between nodes based on different layout algorithms (refer to Figure 5(c)). (2) Dynamic view, where dynamic presentation is based on force-directed algorithms, allowing users to observe and interact by dragging nodes (refer to Figure 5(d)).

The Parameters Panel (see C in Figure 6) comprises graphical and domain parameter settings. Clicking on different parameter categories expands detailed parameter settings. This encompasses settings for the map mode in the geo view, various layouts in the focal view, nodes and edges, as well as markers and selections in the interaction category.

Context to focal data transformation

From a computational power perspective, the calculation of large-scale networks involves a greater number of parameters, necessitating higher computer performance and more sophisticated algorithms. In designing such networks, the system’s visualization rendering incorporates relatively fewer parameters to ensure user-friendliness in terms of performance and rendering time. This is crucial as decision-makers require a basic understanding of the environment for rapid knowledge discovery. Furthermore, the system is designed with a more extensive array of visualization parameters and interactive modes for local networks, enabling users to explore areas of interest in greater depth. These requirements necessitate a transformation from global visualization to local visualization in the network.

The focal point of the junior manager’s attention is the city under their supervision and its surroundings. Once the entire network is comprehended, the transition from the contextual geographical view to a more focused perspective is necessary (from (a) to (d) in Figure 7). As depicted in Figure 4, this shift involves a clear transition between two views: moving from the geographical view to the abstract view, driven by the fundamental logic of data transformation. Table 3 shows two methods for generating local networks.

Figure 7.

The process of transforming a geographic view to a focal view. View (a) represents the whole network within the geographical view. By utilizing select and filter operations (shown in (b)) or by making selections on the canvas (illustrated in (c)), a local network can be generated, which is depicted as an abstract view in (d).

Table 3.

Method for generating local networks.

User Interface	Method	Reference
Network Canvas	Interactive graphical interaction	B in Figure 6
	- Node selection by clicking specific nodes
	- Box/lasso selection to specify nodes	Figure 7(c)
Network Generation	Data selection through interface	A2 in Figure 6
	- Selection: choose specific target cities	Figure 7(b)
	- Filtering: apply specific attributes or conditions	Figure 7(b)

First, the interactive analysis can be conducted based on the canvas. Once the main graph is generated, the graphics serve as the most direct and intuitive results. The canvas allows interactive exploration, enabling the selection of data for a focus on cities or local networks of interest.

Node selection via clicking to generate sub-networks: Basic information is obtained by hovering and clicking on a node. Upon identifying a node of interest, the system allows the activation of Select Mode, enabling the user to designate it as a candidate for the next phase of analysis. This process can be repeated for multiple nodes, generating a local network in the focal view to observe relationships among the selected nodes.

Node selection via box/lasso selection (see Figure 7(c)): An alternative method involves drawing a circle around the cities of interest. For general selections, a rectangular box suffices, while a more purposeful selection can be achieved using the lasso tool to draw a circle within a specific range.

Second, the interactive analysis can be conducted based on the network generation panel (A2 in Figure 6). Purposeful filtering and selection are conducted to generate a sub-network through this method, considering that the manager possesses accumulated experience and prior knowledge (see Figure 7(b)).

Selection: The select window allows the user to choose specific target cities. For instance, selecting cities like Taiyuan, Shijiazhuang, and others will generate a network depicting interactions among these cities.

Filtering: The filtering window enables observation based on specific attributes or conditions. For example, the network can be filtered to identify nodes with more than 20 connections using the method illustrated in Figure 7(b).

Zoom+filtering view transfomation

In both stage, the system enables a detailed analysis of each network by different views through view change and zoom+filter interaction. As illustrated in the process diagram in Figure 8, at this stage, the data remains unchanged, and transformations in views of the same data are achieved through zoom+filtering view interaction.

Figure 8.

Zoom+Filter interaction. When certain nodes are selected in view (a), other nodes will be filtered in gray, and the selected window will zoom into the view (b).

The visualization system in geographical visualization employs two views: a standard map (Figure 5(a)) and a satellite map (Figure 5(b)). This allows users to toggle between the two views to display the national network through different backgrounds. Additionally, the system also enables users to examine specific local networks using zoom and filter interactions (Figure 7(c)).

The system facilitates view transformation through the provision of two distinct visualization modes within the focal view: static view and dynamic view. In the static view, relationships between nodes can be displayed using various layout algorithms, as exemplified in Figures 5(c), (d), 7(d), and 8. In the dynamic view, the presentation is dynamic and based on a force-directed algorithm, allowing users to observe and interact by dragging nodes (see Figure 5(d)).

Views within a layout can be transformed through zooming and filtering interactions. In Figure 8(a), focus on nodes in the top-left corner can be achieved through box selection, causing others to fade into gray. Alternatively, if the connections in the top-left corner appear too dense to observe, the system allows users to zoom in on this selected area to enlarge the view (refer to Figure 8(b)).

Force-directed drawing of sub-graph G′

In the dynamic focal mode, the system employs the force-directed algorithm,³⁵ enabling users to observe and interact by dragging nodes. This view supports zooming in and out, as well as interactions like clicking on nodes and edges. The force-directed algorithm’s main goal is to smoothly reposition nodes after users adjust them, strategically minimizing edge crossings for visually appealing graph layouts. Spring forces maintain optimal distances between elements, simulating the graph as a physical system.

The algorithm involves calculating forces on each element and strategically placing them to avoid edge crossings in three iterative steps. Here, d denotes the distance between two nodes and k the optimal distance between nodes.

Calculate the effect of attractive forces $f_{a} (d) = d^{2} / k$ between adjacent nodes;

Calculate the effect of repulsive forces $f_{r} (d) = k^{2} / d$ between all pairs of nodes;

Finally, stop the iteration if f_a and f_r tend not to be changed.

Maintain visual constancy in complex visual analyses

Maintaining visual consistency becomes challenging during view transformations. Both highlights and fixed positions, which facilitate tracking focal objects in a static view, have limitations when applied to view transformations. The transition between geographical space and abstract space or between a dataset and subset visualization can exacerbate these limitations. In response to this challenge, we propose a design to help the user track their focal entity during view transformations.

This design concept, depicted in Figure 9, demonstrates the system’s capacity to manage the visual transition between geographical and abstract spaces or between dataset and subset visualizations. For instance, in a network of 20 nodes (A–T), where node A represents a city managed by a user, the system can employ positioning (central location as shown in Figure 9(b) and (e)) or color marking (node A highlighted in red in Figure 9(c) and (f)) to facilitate quick identification of the focal city. This feature allows for efficient localization of specific entities within the network. Conversely, if deemed necessary, the system allows for the removal of these markers, enabling the observation of all entities under uniform parameters, thus preventing any bias in cognitive judgment.

Figure 9.

Managed city’s mark. Assuming that the managed city is A, the network (a) is shown without any settings, in (b) A is set centrally located, and in (c) A is marked as red, the second row of three diagrams continues to show all three scenarios through another layout.

The proposed system includes a configuration panel specifically for managed cities, as shown in Figure 6 (in the Parameter panel (C), labeled as “Managed City’s Mark”). This panel offers two annotation options: marking the managed city in red and placing it in a central position. The system’s default settings are tailored to suit the characteristics of the view. In geographical views, where node positions correspond to geographic coordinates, color marking is suggested. In abstract views, where the layout algorithm highlights relationships between nodes, color marking is also recommended. However, if node colors represent domain-specific attributes, position marking is advised. This flexibility allows managers to choose markings that enhance their understanding without influencing their judgment. The system also provides an option to remove these markings, ensuring unbiased observation.

Case study

The case study illustrates how context-focal network analysis can reveal valuable insights, evolving goals, and refined knowledge in the field of information visualization. In examining the risk-coping networks of truckers in Taiyuan, the manager begins by establishing specific tasks and objectives. The progression of these objectives becomes more defined through iterative navigation, as detailed in the subsequent sections focusing on the shift from context visualization to focal analysis. By employing visual transformations and interactions, the manager strategically positions Taiyuan within the wider organizational context.

Managerial objectives

As the capital of Shanxi province in northern China, Taiyuan occupies a distinctive position in the network dynamics. While not at the forefront, the city offers growth opportunities. The Taiyuan manager analyzes network data to identify potential resource opportunities and uncover cities with critical resources. This exploration aims to strategically allocate limited resources for future decisions, enhancing overall performance. The manager has set provisional network analysis objectives:

Examine connections among cities in the final 10 days of 2020 to understand the overall network’s status.

Understand the contextual distribution of the network, focusing on various cities, especially Taiyuan, within the broader socio-organizational landscape.

Conduct a detailed analysis of cities closely linked to Taiyuan, revealing distinctive features and patterns within a more refined subset of the network.

Global geographical visualization

To examine city connections in the final 10 days of 2020, the manager starts data analysis and visualization. He uses the network generation panel, choosing the date range from December 22 to December 31, 2020. This creates a detailed network that includes all cities in the platform, as shown in Figure 6. The data preview graph gives a snapshot of the data quantity during this period. Both the visual representation and accompanying data overview confirm the extensive and complex nature of the overall network.

Using the geographical view context, the manager can understand the basic social-physical environment of the platform for the chosen period. This includes an overview of interconnections among all organizations. It also provides insights into the positions of different cities and allows an examination of Taiyuan’s placement and its key associations within the network. Managers can engage in knowledge discovery through interactive methods in the complex network distribution. The manager can perform view transitions through various map modes (Figure 10(a) and (c)). Zooming is possible using the mouse wheel and zoom buttons at the top left of the interface, as shown in Figure 10. Figure 10(a)–(c) demonstrate the process of zooming into the geographic area where Taiyuan is located. Hovering over a node’s marker displays the city name, and clicking on the marker reveals basic information about the node, including province, number of users, and number of helpers in the specified period, as shown in Figure 10(c).

Figure 10.

Geo visualization: (a) the standard map, (b) the satellite map, and (c) interaction: zoom, hover, and click.

Through interaction and observation, the manager has gained valuable insights. (1) The overall distribution trend shows dense networks in the southeast and sparse networks in the northwest, with a clear regional division between the east and west. (2) The manager hypothesizes that the network distribution trend, distinguished by the Hu-line (also known as the Hu Huanyong Line, revealing the distribution pattern of population density across the country, serving as a vital reference point for understanding enduring demographic and geographic dynamics in China³⁶), is linked to geographical features. Switching to terrain map mode, the manager confirms that the entire network aligns with the Hu-line, where density corresponds to the terrain. The western yellow region, representing plateaus or deserts, has minimal nodes and connections, while the eastern green region, particularly the North China Plain, is denser. The southern hilly region has fewer interconnected lines. (3) Taiyuan’s position on the map is intriguing – it lies almost on the Hu-line, connecting the densest truck driver network in China on one side and relatively sparse cities in the northwest on the other.

Focused analysis

After transforming, the manager turns the subgraph from geographical to abstract visualization. In this view, the manager can use different interactions and switch layouts to examine node and connection details. By zooming, dragging, and selecting, the manager can choose specific node ranges for further analysis in focal analysis.

In focal analysis, the manager can understand various node relationship characteristics through different layouts. He can first observe the subgraph shown with abstract geographical layouts, where node positions indicate distances based on latitude and longitude (Figure 11(a)). However, due to dense connections in the bottom-left, causing unclear relationships, the manager can switch to different layouts for more information, as shown in Figure 11. Opening different layouts annotates the basis or meaning of the layout calculation at the bottom of the interface. Additionally, the manager can adjust the graphical representation by setting different domain attributes through the parameter panel. In Figure 10, the size of geographic view nodes reflects the number of connections, with all nodes, except Taiyuan, in blue. In Figure 12, the manager uniformly sets node sizes, using color intensity to reflect the number of connections. To observe provincial information, the manager represents different provinces with node shapes.

Figure 11.

Geo-coordinate-based display of focal views and their interactions. (a) Displays a local network in the focal view, based on a geo-coordinate layout, where the red node (marked city) represents the managed city. This view is interactive, allowing users to select and magnify specific nodes using a mouse box selection, as shown in (b), which is then expanded for detailed display in (c). Additionally, users can click on nodes to reveal detailed information about the cities, as depicted within the red box in (c).

Figure 12.

Different static layout modes. The three diagrams illustrate three static layouts of a local network. (a) Features the fruchterman-reingold layout, (b) shows a spiral layout, and (c) depicts a circular layout. In each diagram, the shape of the nodes represents the provinces of the cities, while the colors indicate the node connections. Annotations below each figure explain the significance of each layout.

In the focal view, the manager obtains detailed network information through interactions. In static views with various layouts, clicking on nodes reveals more detailed information. As shown in Figure 11(b) and (c), clicking on Taiyuan discloses that the node is in Shanxi Province, has provided help 195 times, sought help 340 times, and is connected to 10 nodes. Alternatively, the manager can switch to dynamic interaction mode for a nuanced understanding of connections between nodes. Using line color settings and node dragging, the Taiyuan manager can gain a specific understanding of connections between Taiyuan and other cities. As shown in Figure 13, the manager can see cities pointed to by Taiyuan (in red) and cities pointing to Taiyuan (in blue). Furthermore, clicking on edges allows exploration of the relationship between the two cities.

Figure 13.

Dynamic Interaction Mode. (a) Displays the dynamic mode of a local network, where users can adjust the layout by dragging a node and thicken the display of connected edges for emphasis. (b) is a zoomed-in version of (a), where clicking on an edge thickens its display and adds a label that details the connection, such as “Edge from [Taiyuan] to [Jinzhong].” In both figures, red nodes signify the Marked Managed City (Taiyuan), red edges indicate connections emanating from the City, blue edges show connections directed toward the City, and gray edges represent connections between other cities.

Using various methods, the Taiyuan manager conducted a focal analysis of the local network of selected cities and made interesting discoveries: (1) Cities in Hebei, Shaanxi, and Shanxi provinces are closely connected, forming abundant networks. In contrast, cities in the relatively remote areas of Ningxia and Inner Mongolia have fewer connections, likely limited to neighboring cities within their provinces, but they actively engage with cities in the first three provinces. (2) Among these 21 cities, the Baotou has more connections and is relatively important but is not connected to Taiyuan. Although Hohhot does not appear very important, being a provincial capital city, it is also not connected to Taiyuan. Both cities will be important development targets for the manager in the future. (3) In this local network, Taiyuan is not the central node, nor does it have the most connections. The manager discovered that Taiyuan has two development directions: learning from central nodes like Shijiazhuang or playing a bridge role between remote western cities and central cities.

Evaluation

To evaluate the effectiveness of the proposed visualization system for the truckers’ network, we conducted a survey based on real management tasks. We invited 12 graduate and doctoral students from Chinese universities to act as junior managers and use the visualization system. They were required to analyze network structures at both global and local levels, and identify resource distribution and key assistance relationships. By guiding the managers through role-specific tasks, we gathered feedback on the system’s ability to support the three main tasks outlined in Section 1.4: national network insights, focal city analysis, and resource identification. The evaluation included questionnaires and interviews to assess the user experience with the system’s main interface, geographical mode, and abstract mode, and their effectiveness in supporting management decisions.

We conducted a survey using a structured 7-point Likert scale to assess various aspects of the system’s performance. Participants were asked if the global view provided a clear overall structure of the network (Global – Structure) and effectively highlighted key node information (Global – Key Nodes). They were also queried on whether the local view helped identify important connections and relationships (Local – Connections) and highlighted key node information (Local – Key Nodes). Additionally, we evaluated if the transition from the global geographical network to the local abstract network helped them understand different network relationships at various levels (Visualization Transformation). The survey further assessed the system’s effectiveness in identifying important flow paths (Resource – Flow Paths) and key resource nodes (Resource – Key Nodes). Participants were also asked if the system facilitated knowledge discovery (Knowledge Discovery) and, overall, how well the system supported their network analysis and knowledge discovery (Overall Experience).

Based on the survey assessing various aspects of the system, participants generally expressed high satisfaction with the system’s aid in their network analysis and knowledge discovery processes (results are shown in Figure 14). For global network insights, the clarity of the overall structure and key node information scored 6.08 and 5.92, respectively. In local network analysis, the system’s effectiveness in identifying important connections and key nodes scored 6.00 and 6.33. The transition between global geographical views and local abstract views, helping to understand different network relationships, scored 6.25. For resource identification, the system’s tools for identifying important flow paths and key resource nodes scored 5.75 and 6.50. The system’s effectiveness in facilitating knowledge discovery scored 6.08. Overall, the system’s support for network analysis and knowledge discovery received a score of 6.00.

Figure 14.

Survey results of evaluation. The labels in the chart correspond to the following aspects of the survey: “Global” represents Global Network, “Local” stands for Local Network, and “Resource” signifies Resource Identification.

We also surveyed participants with open-ended questions to gather feedback on the system’s features and suggested improvements. Most participants found the geographical zoom and dynamic drag functions particularly helpful for knowledge discovery. For instance, one participant noted, “The zoom function in the geographical view is very useful,” and another mentioned, “The dynamic drag function helps a lot.” The dynamic layout of the local network, shown in Figure 13, was highlighted for its intuitive representation using node size, color, and edge thickness, which aids in identifying key nodes and improving resource identification. One participant stated, “The identification of key nodes in the local network is very clear.” Another observed, “The differentiation of node shapes and colors based on provincial features in the static and dynamic layouts enriches the information and clearly presents inter-city interactions.” However, participants also suggested improvements, mainly focusing on the dynamic layout and system stability and operability. One participant suggested, “Can the dynamic layout reflect changes in the network over time?” Others recommended enhancing the differentiation of edge relationships, particularly to highlight strong inter-city connections. For example, one participant said, “Consider further differentiating edge relationships in the local network by using thicker or differently colored edges for closely connected cities.” Some also mentioned the need for better linkage between the global and local maps for easier navigation to specific node information. Overall, participants provided positive feedback on the system, acknowledging its effectiveness in identifying key nodes and supporting resource allocation, while also noting areas for potential improvement to enhance user experience and functionality.

Discussion

Theoretical contribution

This research offers a tailored analytical framework, enriching the visual analysis experience for junior-level managers. The contributions encompass the development of an interactive framework, a refined approach to network analysis, and a focus on visual constancy in complex operational environment. These contributions collectively aim to empower managers in their knowledge-discovery endeavors.

The first contribution is the introduction of a two-layer interactive leverage-seeking visualization framework, catering specifically to the needs of junior-level managers. This framework offers a structured path from raw data storage to knowledge discovery through visualization. Building upon traditional flows, it emphasizes a two-layer approach, focusing on rough visualization, followed by interaction and data analysis, leading to refined visualization. The tailored transformations within each stage address the managers’ requirements, allowing them to achieve comprehensive overviews and explore local network visualizations in context. This innovative framework enhances the visual analysis experience for junior-level managers, providing a nuanced and effective approach to global-scale network analysis.

Our work also contributes to network analysis by integrating Overview+Detail and Focus+Context visualization methods with visual transformations to better meet the needs of junior managers. By emphasizing visual transformation, we connect the focal entity with its broader context. Recognizing that managers possess extensive departmental knowledge but may lack a comprehensive understanding of the overall business, we designed transformations specifically for networks, linking macro-level geographic attributes with local-level abstract visualizations. This includes creating a context view that offers a holistic perspective on the socio-physical environment and a focal view in an abstract space to support logical inferences. This transformation between views enables managers to understand how the features of the focal entity interact with its context, helping them better discern dynamic relationships and fostering a more comprehensive understanding of business dynamics.

The third contribution addresses the challenge of analyzing focal local networks within a complex network while minimizing cognitive overload for managers. We emphasize the importance of maintaining a stable perception of objects during view transformations. The proposed system design allows for controlled highlighting of focal entities, addressing the limitations of traditional methods like highlights and fixed positions. This contribution enhances the visual consistency experience, facilitating coherent observation and recognition of focal entities in dynamic visual contexts.

Research limitations and future directions

While this study provides valuable insights into the information processing framework of junior-level managers in visual analysis, certain limitations should be acknowledged. Firstly, the study’s focus on a specific organizational context may limit the generalizability of its findings. Future research could broaden its scope to encompass various industries and organizational structures to validate and extend these insights. Additionally, this study heavily relies on quantitative data and visualization techniques; incorporating qualitative insights through in-depth interviews or surveys with junior-level managers could offer a more comprehensive understanding of their information processing needs. Lastly, the study primarily employs established network visualization and analysis techniques without introducing groundbreaking algorithmic improvements. Exploring the integration of artificial intelligence, machine learning, and augmented reality in visual analysis tools could be a promising avenue for future research.

Conclusion

In conclusion, our research is focused on meeting the specific needs of junior-level managers in organizational settings through social network analysis. We have tailored our approach to match their responsibilities and domain expertise, presenting a comprehensive design to address their operational requirements and preferences in visual analysis. Our key contributions revolve around using data visualization techniques to elevate the visual analysis experience for junior-level managers. With the introduction of a customized analytical framework, we have emphasized the focal-contextual transformation within the network, enabling a nuanced understanding of influential relationships. Additionally, our efforts have concentrated on enhancing visual consistency by refining the representation of focal entities and local networks within their contextual environment.

To validate our design’s effectiveness, we applied our approach to visualize interactions among truckers from 300 city-based communities on China’s largest occupational platform for truckers. Additionally, we conducted a case study, demonstrating how our approach supports junior-level managers in network analysis and knowledge discovery processes. Our design collectively aims to empower junior-level managers, equipping them with effective tools for knowledge discovery within their organizational roles.

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Natural Science Foundation of China under Grant [72072169; 72293572].

ORCID iDs

Qi Huang

Mao Lin Huang

Yi-Na Li

References

Zhang

Robert

, et al. Visual analysis of the opinion flow among multiple social groups. J Vis 2020; 23: 507–521.

Brunker

Nguyen

Maeder

, et al. A time-based visualization for web user classification in social networks. In: Proceedings of the 7th international symposium on visual information communication and interaction, 2014, pp.9–18.

Liu

Andrienko

, et al. Steering data quality with visual analytics: the complexity challenge. Vis Inform 2018; 2: 191–197.

Pitipornvivat

Zhao

, et al. EgoSlider: visual analysis of egocentric network evolution. IEEE Trans Vis Comput Graph 2016; 22: 260–269.

de Oliveira

MCF

Levkowitz

. From visual data exploration to visual data mining: a survey. IEEE Trans Vis Comput Graph 2003; 9: 378–394.

Chi

. A taxonomy of visualization techniques using the data state reference model. In: IEEE visualization 2000, 2000, pp.69–75.

Linhares

CDG

Ponciano

Pedro

, et al. LargeNetVis: visual exploration of large temporal networks based on community taxonomies. IEEE Trans Vis Comput Graph 2023; 29: 203–213.

Reda

Tantipathananandh

Johnson

, et al. Visualizing the evolution of community structures in dynamic social networks. Comput Graph Forum 2011; 30: 1061–1070.

Liu

Cui

, et al. A survey on information visualization: recent advances and challenges. Vis Comput 2014; 30: 1373–1393.

10.

Basole

Bellamy

Park

. Visualization of innovation in Global Supply Chain Networks: Visualization of innovation in global supply. Decis Sci 2017; 48: 288–306.

11.

van den Elzen

van Wijk

. Multivariate network exploration and presentation: from detail to overview via selections and aggregations. IEEE Trans Vis Comput Graph 2014; 20: 2310–2319.

12.

Hornbæk

Hertzum

. The notion of overview in information visualization. Int J Hum Comput Stud 2011; 69: 509–525.

13.

Coffey

Malbraaten

, et al. Interactive slice WIM: navigating and interrogating volume data sets using a multisurface, multitouch VR interface. IEEE Trans Vis Comput Graph 2012; 18: 1614–1626.

14.

Burigat

Chittaro

. On the effectiveness of overview+detail visualization on mobile devices. Pers Ubiquitous Comput 2013; 17: 371–385.

15.

Burigat

Chittaro

. Visualizing references to off-screen content on mobile devices: a comparison of arrows, wedge, and overview+detail. Interact Comput 2011; 23: 156–166.

16.

Pietriga

Appert

. Sigma lenses: focus-context transitions combining space, time and translucence. In: Proceedings of the SIGCHI conference on human factors in computing systems, 2008, pp.1343–1352.

17.

Hauser

. Generalizing focus+context visualization. In: Bonneau

Ertl

Nielson

(eds) Scientific visualization: the visual extraction of knowledge from data, 2006, pp.305–327. Berlin/Heidelberg: Springer-Verlag pp.305–327.

18.

Baudisch

Good

Bellotti

, et al. Keeping things in context. In: Proceedings of the SIGCHI conference on human factors in computing systems, 2002, pp.259–266. New York, NY: ACM.

19.

Tao

Wang

Shene

, et al. A deformation framework for focus+context flow visualization. IEEE Trans Vis Comput Graph 2014; 20: 42–55.

20.

Gasteiger

Neugebauer

Beuing

, et al. The flowlens: A focus-and-context visualization approach for exploration of blood flow in cerebral aneurysms. IEEE Trans Vis Comput Graph 2011; 17: 2183–2192.

21.

Popescu

. Multiperspective focus+context visualization. IEEE Trans Vis Comput Graph 2016; 22: 1555–1567.

22.

McGee

Ghoniem

Melançon

, et al. The state of the art in multilayer network visualization. Comput Graph Forum 2019; 38: 125–149.

23.

Park

Bellamy

Basole

. Visual analytics for supply network management: system design and evaluation. Decis Support Syst 2016; 91: 89–102.

24.

Rao

, et al. WaterExcVA: a system for exploring and visualizing data exception in urban water supply. J Vis 2023; 26: 957–976.

25.

Feng

Zeng

, et al. Topology density map for urban data visualization and analysis. IEEE Trans Vis Comput Graph 2021; 27: 828–838.

26.

Zhou

Lin

Liu

, et al. A survey of visualization for smart manufacturing. J Vis 2019; 22: 419–435.

27.

Leite

Arleo

Sorger

, et al. Hermes: guidance-enriched visual analytics for economic network exploration. Vis Inform 2020; 4: 11–22.

28.

Nguyen

Nelmes

Huang

, et al. Interactive visualization for patient-to-patient comparison. Genom Inform 2014; 12: 21–34.

29.

Vuilleumier

Henson

Driver

, et al. Multiple levels of visual object constancy revealed by event-related fMRI of repetition priming. Nat Neurosci 2002; 5: 491–499.

30.

Sedig

Parsons

. Interaction design for complex cognitive activities with visual representations: a pattern-based approach. AIS Trans Hum Interact 2013; 5: 84–133.

31.

Ceneda

Gschwandtner

Miksch

. A review of guidance approaches in visual data analysis: a multifocal perspective. Comput Graph Forum 2019; 38: 861–879.

32.

MacEvoy

Paradiso

. Lightness constancy in primary visual cortex. Proc Natl Acad Sci 2001; 98: 8827–8831.

33.

Huang

Nguyen

Hintz

. Attributed graph visualization of collaborative workspaces. In: International conference on computer graphics, imaging and visualization (CGIV’05), 2005, pp.155–161. New York, NY: IEEE.

34.

Fruchterman

TMJ

Reingold

. Graph drawing by force-directed placement. Softw Pract Exp 1991; 21: 1129–1164.

35.

Hua

Huang

Nguyen

. Drawing large weighted graphs using clustered force-directed algorithm. In: 2014 18th international conference on information visualisation (IV), 2014, pp.13–17. New York, NY: IEEE.

36.

Chen

Gong

, et al. Population distribution and urbanization on both sides of the Hu Huanyong line: answering the premier’s question. J Geogr Sci 2016; 26: 1593–1610.