Research on Users’ Exercise Behaviors of Online Exercise Community Based on Social Capital Theory

Abstract

Online exercise communities play an important role in their users’ self-health management. The willingness of users to interact and create user-generated content in online communities reflects the vitality of the online exercise community and the positive impact it has on offline users’ health performance. Therefore, based on social capital theory, we study the relationship between three types of social capital and users’ offline exercise behaviors and add off-topics in the community in the model. We select the KEEP health community user group as the research setting and conduct the regression analysis. The results show that owned centrality and reciprocity have a significant positive relationship with users’ exercise behaviors; accessed centrality and trust have a significant negative relationship with users’ exercise behaviors; and common topics and off-topics show a partly significant correlation. As a moderating variable, off-topics have a negative moderating effect on owned centrality and betweenness centrality, but a positive moderating effect on reciprocity and trust among group members. The results enrich and expand social capital theory, deepen the research on users’ exercise behaviors in the online exercise community, and provide a good reference for online exercise community management.

1. Introduction

With the Chinese government’s emphasis on national fitness and people’s growing awareness of health, fitness and sports have become an increasingly popular way of life. The State Council released “the National Fitness Program (2021–2025)” [1], which clearly plans to create a social atmosphere for national fitness, popularize the culture of national fitness and strengthen incentives for national fitness. This marks a new development opportunity for the cause of national fitness. At the same time, the Law of PRC on Physical Culture and Sports was renewed on 24 June 2022 and implemented on 1 January 2023 [2]. Its second Chapter focuses on “National Fitness”, with Article 16 stating that China implements a National Fitness Strategy, establishes a public service system for national fitness, encourages and supports citizens to participate in fitness activities and promotes the deep integration of national fitness and public health. Online exercise communities provide a unique space for fitness enthusiasts to communicate and share. The strength of online exercise communities lies in the fact that they not only provide fitness resources, but also social functions, i.e., resources such as encouragement with each other, sharing experiences, finding exercise partners and establishing social relationships through the exercise social community. These social functions form a certain special resource which is called social capital in the academic field. This social capital drives users’ exercise behaviors and promotes national fitness activities, representing the long-term value of online communities. The support and recognition that users receive in online exercise communities motivate them to be more active in fitness activities and achieve the goal of self-health management.

According to the 51st Statistical Report on the Development Status of China Internet Network released by China Internet Network Information Center (CNNIC) in March 2023, as of December 2022, the size of online medical users was 380 million, accounting for 35.6% of the overall internet users. Among users, 18.9% prefer mobile apps for fitness, and 14.6% participate in online training [3]. Online exercise communities offer exercise information and services to users and facilitate the establishment of social relationships through interactive activities. The various resources and capital that accumulate through these interactions are known as social capital [4]. The promotion of users’ exercise behaviors driven by social capital represents the long-term value of online communities, aiding users in self-health management.

Online exercise communities serve as platforms for communication and interaction, integrating social networks, social support, and health management. Previous scholarly research has predominantly examined the relationship between users’ participation in these communities and social capital [5,6,7]. Various factors, including trust levels, shared topics, and social identification among users, have been found to have differing effects on their contributions to the community [8,9,10]. In particular, the continuous intention of members to contribute to the creation of knowledge within online exercise communities is influenced by three dimensions of social capital: trust, shared topics and network density [10].

The term “offline participation behaviors of online exercise community users” in this research refers to the actions taken by users outside of the online community to foster community development. Users of online exercise communities engage in ongoing and interactive self-health management activities, seeking both informational and emotional support from the community [11]. The final goal of online community should be promoting the offline exercises, which can really improve users’ health conditions. Self-health management is a proactive approach that individuals adopt to take personal responsibility for managing their healthcare. It is effective in improving various health conditions, including weight control, blood pressure management, and enhancing the overall well-being of individuals with chronic illnesses [12,13,14,15,16,17].

In this paper, our primary focus lies in examining the intricate relationship between three types of social capital and users’ offline exercise behaviors. Specifically, we aim to address two fundamental research questions: Firstly, we investigate the mechanisms through which social capital is formed within online communities. Secondly, we explore the subsequent impact of online social capital on users’ self-exercise and self-health management behaviors. To explore these questions comprehensively, we select the KEEP health community user group as our research setting and employ regression analysis as the analytical method. In Figure 1, we present the entire research process, offering a clear overview of the steps taken to address the research objectives. The outcomes of our research hold significance in enriching and expanding the existing social capital theory. By delving into the complexities of social capital formation and its effects on offline exercise behaviors, we make valuable contributions to the theoretical understanding of social capital’s role in shaping individuals’ health-related practices and habits. Our findings not only provide insights into the dynamics of online communities but also offer practical implications for promoting self-exercise and health management behaviors among users in these virtual social spaces.

2. Theoretical Basis and Research Hypothesis

The term “social capital” was initially introduced in 1910s [4]. The accumulation of social capital is essential for the formation of enduring organizations [18]. Social Capital Theory is composed of three dimensions: structural capital, cognitive capital, and relationship capital [19]. This theory has been extensively used to analyze individuals’ utilization of information technology, participation, and knowledge dissemination in groups. Some scholars [20,21] verified the positive impact of social capital on individuals’ motivation and satisfaction in group knowledge sharing. Based on social capital theory, the availability of perceived channels positively influenced users’ engagement in online exercise communities, while their health conditions influenced their willingness to share and seek information [22].

2.1. Interactions between Online Community and Offline Exercise Behavior

The frequency of using Social Networking Sites (SNS) is related to both running behavior and social life satisfaction [23]. A scholar explores how Social Media Information Seeking (SMIS) influences psychological and behavioral progress during COVID-19. SMIS positively affects attitudes and perceived behavioral control, leading to increased behavioral intentions, especially regarding mask-wearing [24]. Social media have a profound impact on human behavior, encompassing emotional experiences, behavioral habits, and self-perception [25]. Users’ offline activities, which influence their exposure to social situations, have a causal impact on their behavior in online social networks [26]. Changes in offline behavior can affect online interests and sentiments, and this relationship can be used to create predictive models for users’ online behaviors. Psychological distance is important to users’ willingness to participate offline in online exercise communities, using the explanatory level theory framework [27]. The findings suggested a positive influence on users’ treatment processes. These studies collectively demonstrate that online exercise communities contribute to effective self-management and improved health outcomes by providing online support. Additionally, the value co-creation theory emphasizes the role of users as active participants and creators of health value within online communities. This highlights the increasing importance of individual engagement in the self-management of health in the healthcare sector.

The above literature summarizes that social media have a profound impact on people’s behavior and psychology, including the important role of positive social relationships on happiness, and the frequency of social network use associated with exercise behavior and social satisfaction. Future directions require further in-depth research into the complex mechanisms of social media’s influence, especially as technology continues to evolve.

2.2. Structural Capital and Users’ Exercise Behaviors

Structural capital pertains to the connections and relationships among members of an organization or community. These connections are formed through social interactions and are closely linked to network density, which refers to the strength of social relationships. Weak Ties Theory proposes that weak ties are beneficial for individuals to access valuable information and integrate into communities [28]. Maintaining strong ties requires more time and effort [29]. However, some studies [30,31,32] argue that weak ties may not be as advantageous as strong ties when it comes to emotional support and conflict resolution. Weak ties may also face challenges in disseminating complex knowledge, while strong ties can lead to knowledge redundancy. Communities with structural holes in their networks are more likely to generate innovative ideas. Highly dense networks tend to propagate repetitive information, whereas groups with lower network density are more inclined towards innovation [31].

Centrality encompasses the extent of individual involvement and assistance in user communication within a social network [33]. It serves as a reflection of members’ positions, importance, and influence within the network. Individuals occupying central positions in a social network maintain direct connections with other members and are more inclined to comprehend and adhere to community rules, consequently contributing to the community. A study exploring the impact of network density and centrality on people’s acceptance of system usage. They categorized centrality into three levels: centrality owned, centrality accessed, and centrality betweenness, drawing on relevant concepts from UCINET social network analysis [34].

Centrality owned pertains to individuals in the community who possess authoritative or leadership characteristics. Users with high centrality owned are active members who maintain direct connections with numerous individuals and provide a substantial amount of information output. On the other hand, centrality accessed represents the popularity and attractiveness of a focal user. Within a community, posts from a popular focal user generate a significant number of replies, indicating a higher centrality accessed. This suggests that the focal user receives a considerable amount of information input. In comparison to marginalized nodes, nodes with higher accessed centrality are more likely to establish direct connections with central figures, enabling them to access high-quality information. Furthermore, centrality betweenness [35] measures the frequency with which a node acts as a bridge in a node-to-node network. Therefore, we propose the following hypotheses:

H1.

Structural capital formed within online exercise communities is positively correlated with users’ exercise behaviors.

H1a.

The centrality owned is positively correlated with users’ exercise behaviors.

H1b.

The centrality accessed is positively correlated with users’ exercise behaviors.

H1c.

The centrality betweenness is positively correlated with users’ exercise behaviors.

2.3. Relationship Capital and Users’ Exercise Behaviors

Relationship capital refers to the nature of mutual connections among organizational members based on their identification with a common identity [36], mutual trust [37], team identity [38], etc., which promotes information exchange and mutual support. This study categorizes relationship capital into two dimensions: connection reciprocity and trust.

Reciprocity refers to the anticipated benefits of future knowledge needs that arise when community members contribute knowledge. It can foster members’ beliefs that investing resources will yield rewards, thereby increasing willingness to contribute. Community members who frequently assist others are more likely to receive timely and abundant help when they require relevant knowledge [39]. Social capital, trust, reciprocity norms, and identification on individuals’ knowledge is important to share in virtual communities [38]. Altruism, identification, reciprocity, and shared topics significantly promote knowledge sharing. While reputation, social interaction, and trust positively impact the quality of shared knowledge, they do not affect its quantity [21]. Social identification influences members’ long-term commitment to the community, manifested as their intention to maintain their identity [22], and a strong sense of social identification fosters loyalty to the community. A study on the Sweet Home website, an online exercise community for diabetes patients, and observed reciprocity and transitivity in the reply network. They found that users were more likely to respond to others who were similar to them, and those with higher social capital received more replies, thus facilitating information exchange [16].

Trust is a fundamental element in fostering participation within online exercise communities, encompassing the relationships between community members and between members and the community itself. It serves as a prerequisite for self-disclosure, as it entails the members’ willingness to accept the actions of other community members, with the expectation that these actions will significantly impact them, rendering the need for monitoring and control unnecessary. There are two different forms of trust: cognitive trust and affective trust. These forms of trust motivate members to assist their peers, actively share personal experiences, and increase interaction [40]. Members with higher levels of trust demonstrate a greater propensity for engaging in community interactions. Trust, perceived similarity and social interaction exert a positive influence on information-seeking behavior [41]. Based on this, hypotheses are made:

H2.

Relationship capital within online exercise communities is positively related to users’ exercise behaviors.

H2a.

Reciprocal behavior among users in online exercise communities is positively related to users’ exercise behaviors.

H2b.

Mutual trust among users in online exercise communities is positively related to users’ exercise behaviors.

2.4. Cognitive Capital and Users’ Exercise Behaviors

Cognitive capital refers to the collective interpretation and expression of meaning within a group, encompassing shared topics, vocabulary, codes, terms, and narrative styles employed during communication [40]). It enhances communication effectiveness and understanding among members. Shared topics and shared vision are key factors influencing cognitive capital. Shared topics involve the use of verbal symbols, terms, and descriptions in communication, enabling mutual understanding and effective knowledge dissemination [40]. Shared vision represents the common goals and strong aspirations of members [42], facilitating resource exchange and the formation of social identification. For post-treatment cancer patients, relationship between perceived social support and psychological adaptation is val. The study revealed a positive correlation, with social support, directly and indirectly, influence emotional adaptation and overall well-being [29].

Cognitive capital plays a crucial role in promoting community cohesion and development, particularly through shared topics. Firstly, shared topics facilitate mutual understanding among members, enabling the effective sharing of resources. Secondly, the utilization of specialized terms and vocabulary within the community indicates a higher level of expertise, increasing the likelihood of generating high-quality information and providing valuable assistance to members. Lastly, shared topics nurture shared cognition and shared vision, fostering a sense of mutual identification with the community. In this study, cognitive capital is operationalized as the frequency of users’ participation in or initiation of shared topics within the group. Therefore, we propose the following hypothesis:

H3a.

A positive correlation exists between shared topics among user members in online exercise communities and exercise behaviors.

2.5. Irrelevant Topics and Users’ Exercise Behaviors

Irrelevant topics encompass discussions that deviate from the main theme of an online community, often characterized by chitchat or off-topic conversations. While irrelevant topics may reduce the overall information quality within a community, they can foster interpersonal relationships and enhance the cohesion of smaller subgroups. For user participation behavior in online communities, there is a positive correlation between off-topic discussions and emotional support behaviors, but there is no significant relationship with knowledge-sharing networks [43]. This study proposes that irrelevant topics have a negative impact on users’ exercise behaviors. Engaging in off-topic discussions tends to emphasize the connections within smaller subgroups, potentially leading to a disconnection from the broader community and causing marginal or new users to gradually fade away due to difficulties integrating into these subgroups. Additionally, the intense connections formed through irrelevant topics require considerable time and energy. This study raises questions about the sustainability of such connections and the possibility of transforming irrelevant topics into actionable knowledge outcomes. However, irrelevant topics may indirectly influence exercise behavior through their potential positive effects on relationship capital. Thus, in this study, irrelevant topics are considered a moderating variable, indicated by the frequency of users initiating or participating in off-topic discussions. The hypotheses are as follows:

H4a.

In online exercise communities, irrelevant topics among users are negatively correlated with users’ exercise behaviors.

H4b.

Irrelevant topics among users in an online exercise community have a moderating effect on the relationship between various elements of social capital and users’ exercise behaviors.

Based on the above, the theoretical framework model of this study is shown in Figure 2.

3. Study Design

3.1. Sample Selection and Data Sources

The research data for this study were collected from the professional exercise community KEEP (https://www.gotokeep.com/, accessed on 4 August 2023) and the chat histories of users in a WeChat group associated with the app. This online community primarily aims to pursue a healthy lifestyle through exercise, encourage, learn from, and motivate each other, and promote scientific exercise and regular exercises to prevent or alleviate health issues such as obesity, hypertension, and fatty liver. The WeChat group of community members has two main characteristics: maintaining activity and cultivating habits. Activity refers to the hundreds of daily chat messages exchanged within the group, including casual conversations, knowledge sharing, Q&A, and seeking assistance, while consistently maintaining a relatively high level of activity. Cultivating habits stems from the shared goal of mutual encouragement for exercise within the group. Group members utilize methods such as screenshots, smartwatches, exercise apps, and mini programs to share their progress in the group.

This study selected a total of 8800 chat records (excluding emojis and symbol messages) generated by 150 members of the community from 17 March 2018 to 15 April 2018, spanning a period of 30 days. The chat records were analyzed and categorized, and based on user interaction behaviors, a social network was constructed (as shown in Figure 3). The study aimed to explore the network structure, node attributes, and the impact of chat content on users’ exercise behaviors. To achieve this, a social capital model was constructed and empirical analysis was conducted to examine the effects of social capital and irrelevant topics on users’ exercise behaviors, in line with the formulated hypotheses.

3.2. Encoding Process

This study employed content analysis to examine the chat records, involving statistical word frequency analysis and content categorization. The content was divided into semantically similar and quantifiable subcategories. Following Zhou Junjie’s social support behavior coding (SSBC) framework [43], the chat content of the group users was classified into six types: informational support, emotional support, off-topic discussions, instrumental support, esteem support, and network support. Given the specific characteristics of the WeChat group under investigation, this research categorized the text content into four subcategories based on user-initiated topics: check-ins, irrelevant topics, common topics, and self-disclosure.

Among these, common topics encompassed three types of support behaviors from the SSBC framework: network support, informational support, and instrumental support, emphasizing their relevance to the group theme. Network support referred to shared interests and concerns related to physical exercise. Informational support involved knowledge sharing, specifically providing information related to scientific exercise, equipment selection, healthy diets, and other knowledge content relevant to the group theme. Instrumental support referred to guiding problem solving, such as how to use mini-programs for check-ins. This study considered expressions of care, encouragement, and supervision as indicators of emotional support reflected in the variable of reciprocity, while expressions of respect and trust toward others were seen as indicators of esteem support, reflecting the variable of trust.

In this study, the data underwent a preprocessing step where the conversations were segmented into different fragments based on their occurrence time and topic. Conversations that had no response for more than thirty minutes were considered ended. An example of the coding scheme can be found in

Table 1. Code Examples.

User Behaviors	Extracted topics	Chat Record
Checked-in	Exercise Checked-in Record	(1) Check-in in the KEEP applet
		(2) Check-in in the form of a smartwatch photo that records exercise duration and calorie consumption
		(3) Swim for one hour, burpee 30, check in
Self-disclosure	Describe and share physical conditions	(4) After the birth of my baby, I have fluid in my knee, I don’t know what causes it and the medication doesn’t help
		(5) I measured around 19% body fat on the body fat scale and 15.5% at the gym
		(6) Upload personal photos or pictures of health data
Irrelevant topics	Post off-topic links, advertisements, etc.	(7) Please vote for me, No. 14, thank you
	Initiate daily event discussions	(8) I will go to Russia to watch the World Cup in a few days, and the air tickets and hotels are all booked.
Common topics	Initiate an event (network support)	(9) Is there anyone who signed up with me for Spartan on 5.19 in Beijing?
	Inquiry about knowledge (informational support)	(10) What kind of running shoes do you wear?
	Initiate help (physical support)	(11) How do I check in for a run? I don’t see the interface
	Provide guidance (physical support)	(12) You can ask me if you have questions about the choice of protein powder

.

Based on the above classification method, a total of 730 conversations initiated by 137 participants were identified and recorded, as presented in

Table 2. Basic Information Statistics of Conversations’ Classification.

Checked-In	Self-Disclosure	Irrelevant Topics	Common Topics
1070 times	37 times	230 times	512 times

. In this study, the act of “check-in“ is considered the dependent variable, reflecting users’ exercise behaviors. It would be beneficial for the authors to further explore and discuss the relationship between offline exercise behavior and the dynamics of online group chat.

Understanding how users’ offline exercise behaviors correspond to their engagement in an online group chat can provide valuable insights into the influence of social interactions on exercise habits. By analyzing the content and patterns of conversations within the online health community, the authors could potentially uncover correlations between participants’ exercise routines, their participation in group discussions, and the supportive nature of the community. Furthermore, investigating the impact of the online group chat on users’ exercise behaviors can shed light on the motivational aspect of social support within the community. The discussions and interactions within the group may play a role in encouraging individuals to adhere to their exercise goals and maintain a consistent fitness routine. Thus, delving into the interplay between offline and online exercise behaviors can offer a comprehensive understanding of how social capital formed in online communities may influence users’ self-exercise and self-health management.

By addressing the relationship between offline exercise habits and the dynamics of the online group chat, the authors can enhance the relevance and applicability of their findings, potentially uncovering valuable implications for promoting exercise engagement and fostering a supportive virtual environment within the health community.

3.3. Data Processing Based on UCINET Social Network Analysis Method

Social networks can be classified into directed and undirected networks, and a comprehensive social network includes various structures, such as dyadic relationships, triadic relationships, subgroups, and positions. The interactions among members within a WeChat group are considered a type of directed network. Each member is represented by a node, and the connections established through interactions are represented by edges. Therefore, the interactions among members can be visualized as shown in Figure 4.

We converted the dataset into a matrix format for social network analysis, as shown in

Table 3. Examples of Social Network Matrix Construction.

Member	A	B	C	D
A	--	1	0	2
B	1	--	1	1
C	0	2	--	1
D	3	1	1	--

. The valid data of our WeChat group comprise 137 people. The meaning of A–D in Table 3 is not to label these 137 users as A–D, but the name of the user, i.e., user A, user B, user C, user D, as an example for demonstration: the rows represent the sender of the conversation, and the columns represent the recipients of the conversation. Consider number “2” in the fifth column of the second row, for example; it means that user A has sent two messages to user D. The number “3” in the first column and the fifth row means that user D has sent three messages to user A. From this, the number of replies between users can be counted, and thus the independent variable can be calculated. We generated the social network graph of the community, depicted in Figure 5, where larger nodes indicate higher centrality.

The directed social relationship network graph depicted in Figure 4 illustrates the connections between nodes with specific directions. Outdegree refers to the number of outward connections a node has with other nodes, while indegree represents the number of inward connections a node receives from other nodes. The “outdegree” metric reflects an individual’s sociability, indicating the extent to which they pay attention to others. Individuals with higher outdegree tend to exhibit greater information output within the group, such as sharing knowledge, expressing opinions, or engaging in social interactions. Consequently, individuals with higher outdegree demonstrate a higher level of authority. Conversely, “indegree” signifies the degree to which an individual receives replies or receives recognition in the form of being “liked”, indicating their reception of information from others.

Furthermore, the members within the social relationship network graph can be categorized into cohesive subgroups based on their level of closeness and the direction of their interaction. Cliques, which are fundamental concepts in UCINET social network analysis, represent cohesive subgroups where the distance between any two nodes within a clique is equal to or less than $n$ and we set $n$ = 2. Based on the records of each member in the WeChat group, we adopted UCINET software and built the social network diagram as shown in Figure 5 (different numbers represent different member). The number in each circle represents a WeChat member and the size of the circle represents the closeness of the member. The larger the circle, the stronger the closeness and the greater the influence.

Based on the existing studies [44,45,46], we put up our way of calculation mode. In this study, the dependent variable, user’s exercise continuity, is represented by the “number of check-ins” made by the users. It is calculated as the total number of check-ins by user $i$ in the group within a certain period, denoted as $Y_i$.

3.3.1. Structural Capital: Centrality Owned, Centrality Accessed, Centrality Betweenness

We chose to use “outdegree” to measure centrality owned (Li et al., 2016). In this paper, the centrality of point $i$ is set as $CE{N}_i$, and the outdegree of $i$ is $Out{ \mathrm{deg}}_i$. If point $i$ replies to point $j$ for $N$ times, then ${𝜕}_{ji}=N$. If point $j$ replies to point $i$ for $N$ times, then ${𝜕}_{ji}=N$. Therefore, we obtain:

$$CE{N}_i={{\displaystyle \sum }}_{j=1}^N{𝜕}_{ji}=Out{ \mathrm{deg}}_i$$

(1)

Centrality accessed is represented by the “indegree” of the point, where the centrality accessed of point $i$ is defined as $CEN\_AC{E}_i$, the indegree of $i$ is defined as $In{\mathrm{deg}}_i={\displaystyle \sum }_{j=1}^N{𝜕}_{ji}$. All other factors being equal, points with higher centrality owned in the network are more likely to have higher centrality accessed due to their greater number of direct connections. In order to eliminate the interference caused by centrality owned, this study considers the frequency and proportion of interactions for different nodes and introduces a weight factor, denoted as $w_i$.

$$CEN\_AC{E}_i=\{\begin{array}{c}\frac{{{\displaystyle \sum }}_{i=1}^N\left(a_{ji}\times w_{ij}\right)}{{{\displaystyle \sum }}_{i=1}^N{𝜕}_{ji}},w_i>0\\ 0,w_i\le 0\end{array}$$

(2)

$$w_i=\{\begin{array}{c}\frac{{𝜕}_{ji}\times {{\displaystyle \sum }}_{j=1}^N{𝜕}_{ij}-{𝜕}_{ij}\times {{\displaystyle \sum }}_{i=1}^N{𝜕}_{ij}}{{{\displaystyle \sum }}_{i=1}^N{𝜕}_{ji}\times {{\displaystyle \sum }}_{i=1}^N{𝜕}_{ij}},{\displaystyle \sum }_{i=1}^N{𝜕}_{ji}\ast {\displaystyle \sum }_{j=1}^N{𝜕}_{ij}\ne 0\\ 0,{\displaystyle \sum }_{i=1}^N{𝜕}_{ji}\times {\displaystyle \sum }_{j=1}^N{𝜕}_{ij}\le 0\end{array}$$

(3)

The betweenness centrality measures the extent to which a node serves as a bridge or intermediary between other nodes in terms of their connections. Nodes with high betweenness centrality may not have a high number of direct connections themselves but play a crucial role in controlling the flow of information and resources between other nodes in the network. These nodes have more pathways to access useful information and social support and tend to utilize community resources more frequently. In this study, the betweenness centrality is calculated using UCINET social network analysis software, and it is represented as $BET{W}_i$.

$$b_{jk}\left(i\right)=\frac{g_{jk}\left(i\right)}{g_{jk}}$$

(4)

Here, $b_{jk}\left(i\right)$ is the probability that $i$ is on the path between points $j$ and $k$, $g_{jk}\left(i\right)$ is the number of paths that exist between $j$ and $k$, and $g_{jk}\left(i\right)$ denotes the number of paths that exist between points $j$ and k that pass through point $i$.

3.3.2. Relationship Capital: Reciprocity and Trust

In this study, trust is measured based on whether users engage in self-disclosure behaviors within the group. If a user has exhibited self-disclosure behaviors in the community by posting selfies, disclosing personal health data, or describing their physical condition, it is considered an indication of trust and reliance on other members of the community. This study takes trust into account. However, if a user has engaged in self-disclosure behaviors on other publicly accessible social media platforms, the user’s self-disclosure behaviors within the group may not necessarily stem from trust. In reference to users’ activities on their individual accounts in the KEEP platform during the same period, trust is represented as $TR{U}_i$. User $i$‘s self-disclosure behaviors within the target WeChat group are represented as $d_i$, and the user’s self-disclosure behaviors in KEEP are represented as $d_k$, $d_i$ and $d_k$, which are binary variables. We referred to users’ personal account activities on the KEEP app during the same period and set trust as TRU_i. The self-disclosure behavior of user i in the target WeChat group is represented by $d_i$, while the self-disclosure behavior in the KEEP app is represented by $d_k$. Both $d_i$ and $d_k$ are binary variables, with 0 indicating no self-disclosure and 1 indicating self-disclosure. If user i make self-disclosure both in WeChat and KEEP, then $TR{U}_i$ equals to 0 and indicates self-disclosure, and vice versa. Therefore, the relationship can be expressed as follows:

$$TR{U}_i=\{\begin{array}{c}{d}_i-d_k,d_k\le d_i\\ 0,d_k\ge d_i\end{array}$$

(5)

In this study, the concept of cliques is used to measure reciprocity. Cliques are fundamental concepts of cohesive subgroups in UCINET social network analysis. In a binary-directed relationship network, the relationships among members within a clique are reciprocal and exclusive, meaning that adding any other member would alter its nature. The number of cliques a member belongs to reflects the level of reciprocity within the overall network. Let us assume that user $i$ belongs to n cliques, represented as $n\times c_i$. Reciprocity ($REC{P}_i$) can be expressed as follows:

$$REC{P}_i=n\times c_i$$

(6)

3.3.3. Cognitive Capital: Common Topics

Cognitive capital refers to the shared topics and vocabulary that are generated and used within a collective. In this study, the concept of shared group topics is used to represent cognitive capital, and it is denoted as $SHA\_TPC$. $SHA\_TPC$ represents the number of times a user participates in or initiates shared topics within the group.

3.3.4. Irrelevant Topic

Many users perceive online communities as spaces for making friends and for relaxation, resulting in a significant number of irrelevant topics. In this study, the variable of irrelevant topics is included in the model as both an independent variable and a moderating variable. It is denoted as $OFF\_TP{C}_i$, representing the number of times a user initiates or participates in discussions on irrelevant topics.

3.4. Users’ Exercise Behaviors Model

To visually study the influence of social capital formed among group members in online communities on individuals’ offline exercise behaviors, this study employs the method of ordinary least squares regression to test the hypotheses and construct the following econometric models (where the meanings of each variable are described in

Table 4. Description of Variables.

Variable		Variable Description	Variable Symbol	Type of Data
Dependent Variable	Exercise Behaviors	Times of Check-in (user participation)	Y	integer
Independent Variable	Centrality owned	users’ social network: Centrality owned	CENi	continuous variable
	Centrality accessed	users’ social network: Centrality accessed	CEN_ACEi	continuous variable
	Centrality betweenness	users’ social network: Centrality betweenness	BETWi	continuous variable
	Reciprocity	Network Nodes: Number of belonged Cliques	RECPi	continuous variable
	Trust	self-disclosure	TRUi	binary variable
	Common topics	users initiate or participate in discussions on common topics	SHA_TPCi	continuous variable
Moderator variable & independent variable	irrelevant topic	Users initiate or participate in irrelevant discussions	OFF_TPCi	continuous variable

). The model without interaction effects is presented in Equation (7) and the model with interaction effects is presented in Equation (8). ${\alpha }_i$ and ${\beta }_i$ are coefficients of the following two regression equations.

$$\begin{array}{l}{Y}_i={\alpha }_0+{\alpha }_{1}CE{N}_i+{\alpha }_{2}CEN\_AC{E}_i+{\alpha }_{3}BET{W}_i+{\alpha }_{4}REC{P}_i+ {\alpha }_{5}TR{U}_i+\\ \hspace{1em}\hspace{1em}\hspace{1em} {\alpha }_{6}SHA\_TP{C}_i+{\alpha }_{7}OFF\_TP{C}_i\end{array}$$

(7)

$$\begin{array}{l}{Y}_i={\beta }_0+{\beta }_{1}CE{N}_i+{\beta }_{2}CEN\_AC{E}_i+{\beta }_{3}BET{W}_i+{\beta }_{4}REC{P}_i+{\beta }_{5}TR{U}_i+\\ \hspace{1em}\hspace{1em}\hspace{1em} {\beta }_{6}SHA\_TP{C}_i+{\beta }_{7}OFF\_TP{C}_i+{\beta }_{8}CE{N}_i*OFF\_TP{C}_i+\\ \hspace{1em}\hspace{1em}\hspace{1em} {\beta }_{9}CEN\_AC{E}_i*OFF\_TP{C}_i+{\beta }_{10}BET{W}_i*OFF\_TP{C}_i\\ \hspace{1em}\hspace{1em}\hspace{1em} {\beta }_{11}REC{P}_i*OFF\_TP{C}_i+{\beta }_{12}TR{U}_i*OFF\_TP{C}_i+\\ \hspace{1em}\hspace{1em}\hspace{1em} {\beta }_{13}SHA\_TP{C}_i*OFF\_TP{C}_i\end{array}$$

(8)

4. Empirical Results and Analysis

4.1. Data Sources and Description

Table 4 provides variable descriptions, and

Table 5. Results for Descriptive Statistics of Variables.

Variable Symbol	Mean	Standard Deviation	Y	CENi	CEN_ACEi	BETWi	RECPi	TRUi	SHA_TPCi	OFF_TPCi
Y	7.81	15.04	1.000
CENi	17.92	54.92	0.471 **	1.000
CEN_ACEi	5.11	42.06	0.540 **	0.638 **	1.000
BETWi	98.76	550.74	0.471 **	0.765 **	0.674 **	1.000
RECPi	24.07	23.24	0.359 **	0.883 **	0.558 **	0.719 **	1.000
TRUi	0.07	0.26	0.027	0.257 **	0.244 **	0.226 **	0.222 **	1.000
SHA_TPCi	3.85	12.78	0.473 **	0.710 **	0.626 **	0.729 **	0.628 **	0.125	1.000
OFF_TPCi	1.74	4.59	0.252 **	0.561 **	0.611 **	0.565 **	0.506 **	0.199 *	0.540 **	1.000

Note: the correlation is significant; ** means p < 0.05; * means p < 0.1.

displays descriptive statistics for the research variables. The variable $TR{U}_i$ measures the level of trust and is a binary categorical variable, where 0 indicates self-disclosure behaviors based on trust within the group, and 1 indicates the absence of self-disclosure behaviors based on trust within the group. Centrality owned, centrality accessed, centrality betweenness, reciprocity, trust, shared topics, irrelevant topics, and check-in behaviors are continuous variables. Additionally, based on the statistical data for irrelevant and shared topics, it can be observed that, on average, out of every 10 topics initiated, approximately 7 topics are relevant to the main theme, while 3 topics are irrelevant chit-chat.

4.2. Analysis of Regression Result

Considering that the variables have different units, which may lead to bias in various statistical measures, in order to reduce multicollinearity among the variables in the regression equation, the variables are first centered. Centering the variables helps eliminate the influence of scale on the data structure. To address potential bias in statistical measures caused by variables with different units, it is necessary to reduce multicollinearity in the regression equation. This can be achieved by centering the variables. The process of centering mitigates the impact of scale variation on the data structure.

Table 6. Regression Result of User’s Exercise Behavior (without Moderating Effect).

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9
Constant	8.098 ***	8.122 ***	8.231 ***	6.723 ***	8.256 ***	4.981 ***	8.265 ***	8.378 ***	8.380 ***
CENi	0.297 ***				0.300 ***	0.342 ***			0.328 ***
CEN_ACEi	−0.094 ***				−0.092 ***	−0.086 ***			−0.084 ***
BETWi	−0.014 ***				−0.014 ***	−0.017 ***			−0.015 ***
RECPi		0.293 ***			0.138 **		0.281 ***		0.159 ***
TRUi		−5.384			−9.280 **		−7.514		−7.433 *
SHA_TPCi			0.465 ***		−0.170			0.434 ***	−0.163
OFF_TPCi				0.768 ***		−0.576 *	0.345	0.175	−0.494
R2	0.371	0.189	0.154	0.054	0.422	0.391	0.209	0.156	0.441
Adjusted R2	0.356	0.176	0.147	0.046	0.393	0.371	0.189	0.142	0.408

Note: *** means p < 0.01, the correlation is significant; ** means p < 0.05; * means p < 0.1.

presents regression models 1–4, where structural capital, relational capital, cognitive capital, and irrelevant topics are used as independent variables separately. Model 5 shows the regression model with structural capital, relational capital, and cognitive capital as independent variables together. Model 9 represents the regression model with all variables as independent variables.

Table 7. Regression Result of User’s Exercise Behavior (with Moderating Effect).

	Model 10	Model 11	Model 12	Model 13
Constant	4.236 ***	1.825	6.212 ***	2.927 *
CENi	0.359 ***			0.518 ***
CEN_ACEi	0.157			0.511 ***
BETWi	−0.010 *			−0.008 ***
RECPi		0.257 ***		0.171 ***
TRUi		−2.334		−7.383 *
SHA_TPCi			0.775 ***	−1.534 ***
OFF_TPCi				−0.365
CENi*OFF_TPCi	−0.003			−0.055 ***
CEN_ACEi*OFF_TPCi	−0.008			−0.028 *
BETWi*OFF_TPCi	−0.002			0.001 ***
RECi*OFF_TPCi		0.029 ***		0.075 **
TRUi*OFF_TPCi		−2.165 ***		−1.520 *
SHA_TPCi*OFF_TPCi			−0.034 **	0.142 **
R2	0.441	0.293	0.192	0.570
Adjusted R2	0.413	0.270	0.179	0.520

Note: *** means p < 0.01, the correlation is significant; ** means p < 0.05; * means p < 0.1.

displays models 10–13, which include irrelevant topics as both independent and moderating variables in the analysis of users’ exercise behaviors. Comparing the results, it is observed that the inclusion of irrelevant topics as moderating variables has improved the R² values. Treating $OFF\_TP{C}_i$ as a moderating variable enhances the explanatory power of the models when considering individual structural capital, relational capital, and cognitive capital variables, indicating a significant moderating effect of irrelevant topics.

4.3. Moderating Effects of Irrelevant Topics

In Figure 6 and Figure 7, the relationship between irrelevant topics and various research variables is examined, with the horizontal axis representing levels of irrelevant topics (low and high) and the vertical axis representing the dispersion of the variables. It is observed that irrelevant topics have a negative moderating effect on centrality owned and centrality betweenness, while they have a positive moderating effect on reciprocity and trust among group members.

The higher the degree of irrelevant topics, meaning the less relevance between irrelevant topics and community themes, the more negative the moderating effect on centrality owned and centrality betweenness of user participation, as shown in Figure 6. It illustrates that as the degree of irrelevant topics increases, the moderating effect on centrality owned and centrality betweenness becomes more negative. This interference-type moderating effect suggests that excessively irrelevant content reduces the positive influence of centrality owned on users’ exercise behaviors. This effect is more pronounced as the degree of irrelevance decreases. Additionally, for communities with high centrality betweenness, a higher degree of irrelevant topics leads to lower levels of user engagement in exercise behaviors. Two possible reasons for this effect are identified. Firstly, excessive irrelevant topics diminish the quality of information, resulting in a decreased proportion of “useful” information and a decline in perceived expertise exhibited by users, consequently impacting their behaviors. Secondly, irrelevant topics consume users’ time and energy, which counteracts their engagement in exercise behaviors.

The higher the degree of irrelevant topics, meaning the less relevance between irrelevant topics and community themes, the stronger the positive influence of reciprocity and trust among group members on user engagement, as shown in Figure 7. This indicates the presence of a reinforcement-type moderating effect. It indirectly enhances users’ exercise behaviors. In online exercise communities, members with “close relationships” show higher tolerance and participation in initiating irrelevant topics, which creates emotional bonds that promote exercise behaviors.

Furthermore, irrelevant topics have no moderating effect on visited centrality and related topics. Therefore, hypothesis H4b is supported.

5. Discussion

Based on social capital theory, this study has developed a comprehensive model to explore the factors influencing the exercise behaviors of users in online exercise communities. The empirical research results have supported most of the hypotheses in the model. These meaningful findings allow us to gain a deeper understanding of the exercise behaviors of individual users in online exercise communities.

5.1. Structural Capital

Structural capital plays a crucial role in shaping the exercise behaviors of users within online exercise communities, confirming hypothesis H1a. In such communities, certain individuals with high centrality act as influential figures or team leaders, wielding significant expertise and impacting the adoption of successful exercise patterns through their knowledge and experience. This finding aligns with the research [47] which observed that certain members, such as mentors, have a stronger influence on social networks compared to others. From a resource perspective, members with high centrality owned often have access to more relevant information, advice, experiences, and knowledge related to the group’s objectives. For exercise behaviors, timely and accurate encouragement and guidance play a crucial role.

H1b did not receive significant positive validation; instead, the removal of centrality owned and the inclusion of centrality accessed showed a significant negative effect on users’ exercise. Additionally, H1c was not supported either, as centrality betweenness did not have a significant impact on users’ exercise and showed only a moderate negative correlation. The negative effects of visited centrality and centrality betweenness on the continuity of exercise suggest that this may be attributed to the externality of knowledge. In other words, members with high centrality accessed and centrality betweenness may perceive the exercise group more as a social and recreational environment [48], lacking in exercise-related knowledge or self-health management behaviors. Alternatively, these users may have a greater psychological distance from the behaviors of tracking exercise.

5.2. Relationship Capital

H2a is strongly supported by the findings, revealing a significant positive correlation between reciprocity and users’ exercise behaviors within online exercise communities. This indicates that the sense of identity users cultivate within the community positively impacts their self-management behaviors. Reciprocal relationships among group members emphasize the importance of maintaining emotional support and camaraderie [47]. Actively forming connections, making friends, and expressing viewpoints create a culture of mutual support with the expectation of receiving assistance when needed, thereby enhancing users’ motivation to take action.

In online exercise and fitness groups, interactions among members foster a supportive environment that encourages sustained exercise behaviors. Reciprocal behaviors promote the sharing of valuable resources and exercise-related information, aiding members in better understanding and incorporating exercise and health knowledge. Moreover, users are committed to sustaining exercise, providing mutual encouragement, sharing tracking among group members and creating an informal contract and collaborative atmosphere [47], which significantly contributes to the collective continuity of exercise behaviors.

However, the study revealed an unexpected and noteworthy finding concerning trust, a key element of social capital. Contrary to expectations, trust exhibited a significant negative relationship with exercise behaviors across all models. This result might be attributed to the particular measurement of trust used in this study, which assessed trust based on whether users publicly showcased their exercise activities, such as written or visual records, on their personal accounts within the KEEP app. One plausible explanation for this finding is that users accustomed to displaying their exercise activities on personal social media accounts may not feel the same need to engage in regular check-in behaviors within the exercise group.

5.3. Cognitive and Relationship Capital

The impact of common interests on users’ exercise behaviors can be observed from two perspectives. On one hand, there is a significant positive correlation between common interests and exercise behaviors in Model 3, where it is treated as a single independent variable, as well as in Model 8, which includes common interests and irrelevant variables. In an online environment, users perceive their psychological distance from other group members based on textual and visual content. When common interests bring users closer to the target event psychologically, they engage in more specific thinking, resonate more easily, and respond to the emotions of the post’s author [27]. In a health and fitness community, common interests related to exercise and health create a relaxed and enjoyable atmosphere among group members, thereby motivating more users to maintain their exercise behaviors.

On the other hand, there is no significant correlation between common interests and exercise behaviors in Model 9, which includes all variables. One possible explanation is that when considering the combined effects of all variables on users’ exercise behaviors, the influence of common interests is not pronounced. The role of cognitive capital is not significant in a social group, but if it is in a professional group within a health and fitness community, it will be significant. This suggests that the health and fitness group is more akin to a social group rather than a professional one.

5.4. Irrelevant Topics as Independent and Moderator Variables

In Model 4, the presence of irrelevant topics as a standalone independent variable demonstrates a significant positive correlation with users’ exercise behaviors. However, when all variables are considered in Model 9, the significant positive correlation diminishes, and instead, a certain degree of negative correlation emerges. Interestingly, irrelevant topics exhibit a moderating role in the model, enhancing its explanatory power. The regression results of Models 9 and 13 further illustrate this moderating effect of irrelevant topics on different variables, warranting a detailed analysis. In Model 13, irrelevant topics by themselves do not exert a significant impact on users’ exercise behaviors. However, they do exert a significant influence on other variables in the model, thereby shaping users’ exercise engagement indirectly.

Specifically, irrelevant topics demonstrate a significant negative moderating effect on centrality owned, centrality by degree, and trust. This suggests that the presence of irrelevant discussions can diminish the perceived “authority” and expertise of central nodes within the online exercise community. Furthermore, these irrelevant topics not only waste the time and energy of central members but also impact other group members, potentially diverting their attention away from meaningful exercise-related discussions.

Conversely, irrelevant topics display a positive moderating effect on centrality betweenness, reciprocity, and common topics, aligning with the study’s initial expectations. The presence of irrelevant topics can lead to increased communication frequency among community members, fostering stronger connections between them. This enhanced communication and interaction, in turn, influence users’ exercise behaviors positively, potentially through increased motivation and support.

6. Conclusions

The investigation of the relationship between online user networks and offline self-health management behaviors holds paramount importance in promoting the effectiveness of online exercise communities. This study adopts social network analysis methods and draws on social capital theory to examine the structural characteristics of online user networks. By integrating content analysis, the study identifies user behaviors and investigates the impact of structural capital, relational capital, and cognitive capital on users’ offline self-health management. Additionally, the moderating effect of irrelevant topics in this process is examined. These research findings contribute to enriching and expanding social capital theory, thereby making valuable theoretical contributions.

Based on the findings of this study, the following suggestions are provided for the design and operation of online exercise websites:

• Encouraging users to express professional viewpoints. The positive influence of users’ centrality owned on their exercise behaviors highlights the importance of showcasing users’ expertise within the community. To capitalize on this, online exercise communities should encourage users to express their professional viewpoints and knowledge. Implementing incentive mechanisms, appointing knowledgeable moderators, and organizing themed discussions can stimulate engaging and informative conversations, ultimately motivating users to actively lead and participate in group discussions.
• Fostering relationship bonds. The significant impact of connection reciprocity on users’ exercise behaviors emphasizes the significance of cultivating strong relationship bonds within the community. Online exercise platforms should promote a supportive environment where users actively assist and respond to one another. Providing features such as diary functionalities that encourage users to consistently document and provide feedback on their health-related activities will reinforce relationship bonds and positively impact exercise persistence.
• Managing irrelevant topics. While irrelevant topics may have a negative influence on users’ exercise behaviors, they do contribute to the development of trust and reciprocity among community members. Therefore, administrators should adopt a balanced approach in managing irrelevant topics. Excessive control may alienate users and erode their sense of identity within the community [49]. Instead, administrators can guide new users to help them integrate better into the community and create specialized subgroups based on shared interests, such as “30-Day Exercise Challenge Group.” This approach allows users with common goals and encourages newcomers to establish a sense of shared identity, facilitate their integration into the community and foster tolerance for irrelevant topics over time.

In conclusion, this study focuses on the interplay between online user networks and offline self-health management behaviors. It offers valuable insights for the optimization of online exercise communities. By incorporating the suggestions provided above, administrators can create a more engaging and supportive virtual space that motivates users to actively participate, share their expertise and cultivate meaningful relationships. These improvements will ultimately contribute to the community’s cohesion and effectiveness in promoting users’ self-health management and exercise behaviors.

Future research can explore the transition from online to offline social relationships and its impact on users’ self-health management in online exercise communities. As members become acquainted, offline activities and further communication may enhance self-health management practices. Including longitudinal data and qualitative interviews can provide a deeper understanding of users’ experiences and the sustainability of online engagement. Studying diverse online communities targeting various chronic diseases can offer valuable insights into tailored interventions. Additionally, investigating how technology, such as mobile apps and wearables, facilitates the online-to-offline transition can optimize digital health platforms. In conclusion, examining the shift from online to offline interactions in online exercise communities can inform effective strategies for promoting self-health management and improving overall health outcomes.