User Interactions in Online Travel Communities: A Social Network Perspective

Abstract

Online travel communities (OTCs) enable users to interact and share travel information voluntarily. Extant research has primarily focused on the content generated through user interactions but neglected how user interactions are structured. This study employed exponential random graph models to examine the formation of user interactions and the outcomes of homophily in terms of network structure across levels (actor, dyad, triad, and network). A dataset of 2,926 posts and 25,854 replies involving 9,712 users in an OTC was used. Results reveal that users’ question initiating and replying ties in OTCs exhibit significant positive structural dependencies in terms of reciprocity, activity spread, generalized transitive closure, and multiple connectivity. Homophily serves as the basis of dyadic interactions and homophilous ties evolve after formation. The study advances hospitality and tourism network research and methodology by going beyond traditional dyadic user interactions, and provides insights into user interactions in OTCs from the social network perspective.

Keywords

exponential random graph models (ERGMs)online travel community (OTC)social network user interaction

Highlights

Simultaneous dyadic and triadic user interactions exist in online travel communities.

Homophily in users’ attributes acts as an antecedent of the formation of replying ties.

Users’ replying ties demonstrate significant structural dependencies.

Homophilous ties among users evolve across levels in online travel communities.

Introduction

Online travel communities (OTCs)—interactive platforms for large, voluntary, and collective groups of individual users over the internet—have become influential information sources for tourists making travel decisions (Sharifi, 2019). Although specific types of OTCs may vary in serving a wide range of users, these platforms commonly provide travel reviews, discussion forums, or both, for user interactions (Chang et al., 2020; Nguyen et al., 2022; Xue et al., 2021). Users seek and share information in OTCs, ask questions, and answer other users’ questions. Users also share their own travel experiences, including their detailed itineraries, hotel choices, restaurants, and activities. OTCs serve as an “information hub” for tourists to find the most helpful resources for their travel purchases and activity/experience choices in pre-trip planning and en route (Ai et al., 2019; Pantano et al., 2017).

Along with the rapid development of OTCs, extensive studies have focused on content generated through user interactions in the functional appraisal of hospitality and tourism products and services, as well as their influence on tourists’ decision-making (Hernández, et al., 2021; Pantano et al., 2019). However, very limited research has examined why user interactions occur and how user interactions are structured in OTCs, which largely determines the vibrancy of OTCs. Thus, user interactions entail much more than evaluating products and services in OTCs (Zhou et al., 2021).

The user interaction in OTCs is inherently dyadic, involving two parties: the sender who initiates his/her post (e.g., asking questions) and the receiver who replies to the post (e.g., answering questions). User interaction occurs when a user’s post is replied to by another user, as a replying tie is generated from the question receiver to the question sender. Given all individual users and their question initiating and replying ties (hereafter, “replying ties”) formed in an OTC, dyads are not independent or separate observations, as users are, rather, interdependent through interactions over time (Choi et al., 2019). For example, single dyads may share a common user, making two isolated users indirectly tied via the common third party; or triads may overlap dyads, with the higher-order configurations of triple users and the ties among them (Lomi et al., 2014). Therefore, understanding user interactions at the dyadic level might be biased without considering the context in which they are embedded, which calls for academic investigation on dyads and triads in user interactions.

Given the importance of dyadic interactions among users in OTCs, extant literature considers the homophily or heterogeneity as a basis of dyads, referring to users with similar or dissimilar attributes as the antecedents of tie formation (Kunz & Seshadri, 2015). However, very little attention has been paid to the “consequences of homophily or heterogeneity,” which refer to the outcomes of collective users forming such ties across levels (Ertug, et al., 2022). Thus, how homophilous or heterogeneous relationships evolve across multiple levels (i.e., actor, dyad, triad, and network) and their different types of outcomes in terms of network structure is still largely unknown.

In examining the mechanism of user interactions in the network analysis, prior studies commonly used regression as the statistical method, which assumes data independence and is thus deemed problematic for handling data interdependency in networks (Kim et al., 2016). Unlike conventional statistical methods, exponential random graph models (ERGMs) can be used in the OTCs context to overcome these limitations (Hernández et al., 2021). ERGMs captures structural dependencies generated by individual propensities to form ties (i.e., sending and receiving ties) and by tendencies of certain tie-forming patterns (i.e., open or enclosed in triads), which concurrently shape networks (Lusher & Robins, 2013).

The peer-to-peer user interactions through question-and-answer (Q&A) and experience-sharing are considered more trustworthy and customized to individual travelers’ needs and wants (Moon et al., 2019). These interactive platforms connect users’ physical lives to virtual communities, enabling users to initiate and participate in online activities across time and space (Xiang et al., 2017). As a result, users are embedded in a web of relationships via interactions in online communities (Xiong et al., 2020). Drawing on social network theories and methodologies, the current study aims to examine how user interactions occur in the network context and model their structural dependencies in OTCs using ERGMs. Specifically, the study focuses on question-and-answer forums in OTCs, as they particularly reflect the significant characteristic of the voluntary nature of user participation and interactions in online communities (Jin et al., 2015). Therefore, this study has the following research purposes: first, to empirically analyze how replying ties in the OTC among users with similar or dissimilar attributes are formed; and second, to reveal the consequences of homophily or heterogeneity through modeling user interactions in the OTC by uncovering the underlying structural dependencies formed by users’ replying ties, which may lead to consequential variations in network structure.

This study is expected to contribute to OTCs as well as the existing body of network research in hospitality and tourism from an innovative perspective. First, the study investigates the under-studied but fast-growing user interactions in OTCs, by highlighting how users’ replying ties are structured over time. Second, the study makes a methodological contribution by modeling the structural dependencies underlying the formation of user interactions through ERGMs, which overcomes the limitation of conventional statistical methods (Hernández et al., 2021), and is of interest to hospitality and tourism scholars in the field of network research.

Theoretical Foundations and Hypotheses Development

From the social network perspective, the present study connects two strands of literature. First, the study employs social network theory regarding network formation. Second, built on relevant literature on OTCs, the study focuses on question initiating and replying interactions among users in question-and-answer forums on OTCs. Hypotheses are then proposed for how user replying ties are formed and structured at the dyad and structure levels.

The Principle of Homophily in Social Network Theory

Social network theory has been increasingly applied in management research and newly emerging and fast-growing online communities and social media websites (Colladon et al., 2019). As a core concept in social network theory, the principle of homophily has been widely applied in various contexts as the antecedent in understanding social interaction activities together with different dimensions of similarity (Currarini et al., 2016). Homophily refers to a continuum that indicates the degree of similarity between a dyad, from perfect correlation to no correlation, as well as to completely negative correlation (i.e., complete heterogeneity), which captures the tendency to form a tie between two actors with similar or dissimilar attributes, such as gender, age, and status (Podolny, 2005). In the specific context of OTCs, homophily/heterogeneity captures the similarity/dissimilarity between users’ attributes such as status and location, which are publicly disclosed by OTCs (Chan et al., 2017). The embeddedness of individuals in social networks is accounted for as the pivotal determinant of the patterns of their social behaviors (Munn, 2020). The presence of homophily has important implications for how individual interactions occur along the social networks and how individual attributes bond with their interactive behaviors. On the one hand, homophily can conduce to more trust and thus better communication between actors (Opper et al., 2015). On the other hand, homophily may cause one-sided opinions, reduce the diversity of the resources in the whole network, and narrow the scope of interactions (Ertug et al., 2022).

Although the antecedents of homophily have been widely documented (Melamed et al., 2020), the consequences of homophily (i.e., how homophilous ties evolve after their formation) have not been adequately scrutinized (Lawrence & Shah, 2020). Further, studies about the consequences of homophily have mainly focused on the influences on actors’ behaviors and performances (Ertug et al., 2022). Relatively few studies have explored the outcomes of homophily in terms of network structure, except some which have investigated the relationship between homophily and network indices (Fang et al., 2018). Yet, what happens after dyads of actors form homophilous ties and their underlying structural process across levels (i.e., actor, dyad, triad, and network) in the network context is still underexamined.

Moreover, since dyads and triads may interact simultaneously in shaping networks, the interdependency of network data may cause violations of the assumption of data independence in traditional statistical methodologies such as regressions (Hernández et al., 2021). Meanwhile, existing ties in networks may further influence the establishment of future ties (Kim et al., 2016). Such endogeneity also calls for more appropriate methodologies for network research. Therefore, simulation-based social network methodologies—ERGMs—are employed to address the limitations of traditional regression methods (Ghahramani et al., 2018).

Since the antecedents and outcomes of homophily may largely depend on actors’ attributes, it is necessary to take specific context into consideration when examining their effects (Ertug et al., 2022). As online interaction has been a fast-growing component in people’s daily lives, particularly in hospitality and tourism (Kim & Kim, 2021), this study takes OTC as the research context to examine the consequences of homophily in terms of network structure across levels. Following Lusher and Robins (2013), the study employs given set of actors A, B, and C to illustrate three types of network structures and the associated structural processes appropriate for the study context (Table 1).

Table 1

Types of Network Structures and Associated Structural Processes.

Type	Diagram	Structural process
I		Activity is often called an “out-two-star” term because of its star-like structures. The “out-two-star” configuration exists in the network when there are two outgoing ties (from actor A to actor B; from actor A to actor C) from a central actor A in a directed network. It captures how often an actor (i.e., actor A) initiates ties by connecting other actors (Kim et al., 2016). The more ties an actor maintains, the higher the level of power the actor has in the network. Activity leads to the tendency of activity spread in network contexts (Lusher & Robins, 2013).
II		Triad closure indicates that two actors (A and C) are likely to form a direct tie (represented by the dotted line) if actor B, as the common actor, maintains direct ties with actors A and C, which creates a triangle connecting all three actors (A, B, and C). The formation of three actor triangle structures may enhance the efficiency and coordination of information exchange within the triad, contributing to the equilibrium of power among three actors (Lee et al., 2014). Triad closure leads to the tendency of generalized transitive closure in network contexts (Lusher & Robins, 2013).
III		A triad is open when an actor (actor B), as the broker, connects others (actors A and C), who had no previous connections. In an open triad, actor B enjoys particular advantages in terms of power in the brokerage position. Actor C is less likely to establish a direct tie with actor A when they both have a tie with actor B, which avoids structural redundancy in the triad (Kim et al., 2016). Open triad leads to the tendency of multiple connectivity in network contexts (Lusher & Robins, 2013).

User Interactions in OTCs

OTCs are regarded as the third place for tourists to build connections with others, a phenomenon that has been attracting increasing academic attention in recent years (Xu et al., 2021). The extant literature has primarily focused on content generated through user interactions in online communities (e.g., information searching and acquisition) and their effects on travel decisions (Hou et al., 2019). High-quality content demonstrates persuasive power and influences potential readers’ decision-making (Zhou et al., 2021). Filieri and McLeay (2014) emphasized the importance of information completeness, timeliness, accuracy, relevancy, understandability, and value addition on content adoption. Fang and colleagues (2018) also pinpointed the positive impact of content length and vividness of sharing content on users’ replying behaviors. Some scholars suggest that OTCs function as social platforms where users can interact with each other and mainly focus on the dyadic interactions between users (i.e., readers and reviewers; Chan et al., 2017). Users can aid others by writing posts to answer questions, which, in turn, increases their affinity with and commitment to the OTCs (Chang et al., 2020). Xu and colleagues (2021) investigated the positive role of user interactions in social wellbeing and band attachment. Users acquire pleasure from these interaction activities, resulting in enhanced loyalty and belongingness to the OTCs (Choi et al., 2019). However, the existing literature has neglected to examine how user interactions occur and how their structures evolve, which is essential in user-generated content sharing in OTCs. The underlying mechanism of the occurrence and structure of user interactions in OTCs, especially from the network perspective (i.e., in the dyadic and structural level), is still largely unknown (Williams & Hristov, 2018).

In the current study, we focus on user interactions in terms of replying to others’ questions or information inquiries (i.e., question asking and replying in users’ Q&A forum) in the OTC. Specifically, user interactions occur across two levels according to the principle of homophily in social network theory. At the dyad level, homophily acts as the mechanism for dyads, referring to a higher tendency of tie formation between a user and similar others (Lawrence & Shah, 2020), as homophily facilitates smoother coordination and mutual trust (Ertug et al., 2022). When we consider collective users in the network context, the principle of homophily goes beyond dyadic homophilous interactions, and acts as a second mechanism that captures the underlying social process (e.g., network closure, structural hole) and the consequences derived from homophily (Ertug et al., 2022). Thus, the hypotheses are proposed regarding the formation of user replying ties from the dyad and structure levels based on social network theory and related hospitality and tourism literature.

User Interactions at Dyad Level

The dyadic interaction occurs in a dyad of users, which refers to a replying tie between two users in question-and-answer forums in OTCs. According to the principle of homophily in social network theory, similarities in individual attributes, such as social class, region, or other identities, are likely to breed interpersonal connections (Melamed et al., 2020). OTCs grant users the opportunities to find others with similar interests and characteristics based on their profile information (Kunz & Seshadri, 2015). Users are more likely to reply to similar others, as similar attributes (e.g., background) are perceived to be more trustworthy and attractive in OTCs (Chan et al., 2017). Thus, users tend to pay more attention to questions posted by similar others.

Prior studies suggest that perceived similarity in user-specific characteristics (e.g., a user’s demographic characteristics, ranking level, and number of followers) affect user interactions in OTCs, as users are more likely to share enthusiasm and interests with similar others (Fang et al., 2018). In the context of the question-and-answer forum in this study, dyadic interactions capture the homophily by user-specific characteristics as follows. First, homophily by users’ status, which reflects the similarity in users’ past involvement and experience, helps users generate positive feelings of credibility and helpfulness (Chan et al., 2017). Thus, it is more likely to form replying ties between users with similar status in OTCs. Second, an individual is inclined to reply to posts sent by other users who have similar preferences in OTCs (Chan et al., 2017). Their similar preferences imply a sense of shared enthusiasm for and knowledge of a specific activity, which would stimulate their interactions (Choi et al., 2019). For example, users who often write travelogues have a similar interest in sharing their travel involvement and experience (Akhtar et al., 2019), which increases the likelihood of forming user replying ties (Fang et al., 2018). Travelogue writing also reflects the way users narrate their travel experience, express themselves, and manage their self-image (Xu & Zhang, 2021). Authentic travelogue writing exerts influence on other users, enabling them to be empathic, generating similar travel interests, and converting others to visit the same travel destination (Duffy, 2019). Consequently, users with similar travelogue writing preferences can accumulate mutual empathy and curiosity (Fang et al., 2018), which induces the formation of replying ties. Third, in the context of question-and-answer forums in OTCs, most posts can be regarded as location-oriented documents that provide representative information about the destinations. Thus, users are more likely to trust and adopt information generated by those who show the same residence location with them, because the experience shared would be more applicable for the users’ travel planning and choices (Kunz & Seshadri, 2015). Therefore, by perceiving the above-mentioned potential benefits, users are willing to engage in user interactions in the OTC by replying to others’ questions. Therefore, the following hypotheses are proposed:

H1. Users with similar user-specific characteristics, that it, (a) status, (b) preference in writing travelogues, and (c) residence locations, are more likely to form replying ties in an OTC.

User Interactions at Structure Level

To model user interactions in the OTC, we propose the following hypotheses following Lusher and Robins’ (2013) network structures typology and associated structural dependencies. First, reciprocity captures the tendency of tie formation through reciprocating an earlier interaction between actors in the network (Kim et al., 2016). In the present study context, reciprocity refers to user A’s tendency to reply to user B, who has already replied to user A. Research suggests that users are likely to reply to posts perceived as beneficial, as they have a strong sense of obligation to reciprocate (Belanche et al., 2018). On the one hand, prior interactions provide social and/or technical supports to help users address their difficulties and are more likely to evoke an affective response, such as friendship, intimacy, and empathy in OTCs (Choi et al., 2019). On the other hand, the mutual benefits potentially encourage users to contribute to the reciprocal relationship by replying to others’ posts (Belanche et al., 2018). Thereupon, the increasing reciprocal tendency exists when it comes to user replying ties in the OTCs.

Second, activity spread captures the tendency toward variation in the degree to which user A replies to multiple users in the study context, as configured in Table 1(I). In OTCs, users with higher out-degree centrality (i.e., reply to other users more often in the OTCs) have more opportunities to learn from peers and tend to receive more subsequent replying ties in return (Fang et al., 2018). Their active supporting behaviors enable them to become eminent contributors and perhaps even the leader of the OTC in the future (Liu et al., 2018). Other users may ask for favors or feel responsible for answering question actively in the OTC, which results in higher frequency of replying behaviors (González et al., 2021). It is more likely to generate activity spread at the structural level when forming replying ties in OTCs.

Third, generalized transitive closure captures the likelihood that when user A has replied to user B, and user B has replied to user C, user A is more apt to reply to user C, as shown in Table 1(II; Lusher & Robins., 2013). In other words, user B, as the shared user, creates opportunities to indirectly link user A with user C. Thus, direct ties could be formed between users A and C when user A finds it more efficient to directly reply to user C, leading to the close triad of users A, B, and C (Kim et al., 2016). In the OTC, users in a triad are always driven by certain common interests, which could foster users to form a bonding subgroup by asking and answering questions (Lee et al., 2011) and leading to the closure of a triad (Kunz & Seshadri, 2015). It is more likely to generate generalized transitive closure at the structural level when forming replying ties in OTCs.

Finally, multiple connectivity captures the tendency of tie formation through multiple two-path connected users, as shown in Table 1(III; Lusher & Robins, 2013), which highlights the role of the intermediary in building an indirect replying relationship. Specifically, although acknowledging the potential benefits of an intermediary, user B (as the intermediary) is less willing to make referrals between two indirectly connected users A and C, leaving the triad of users A, B, and C open (Kim et al., 2016). For users who are not familiar with others in the OTC, it is less costly to form and maintain indirect relationships, which provide a wide range of heterogeneous information (Horng & Wu, 2020). Therefore, there is an increasing tendency of multiple connectivity when forming replying ties in OTCs. The following hypotheses are proposed accordingly:

H2. The likelihood that the formation of user replying ties exhibits structural dependencies in terms of (a) reciprocity, (b) activity spread, (c) generalized transitive closure, and (d) multiple connectivity will significantly increase in an OTC.

Methodology

Data Collection

The dataset was collected from the users’ question-and-answer forum on Mafengwo (https://www.mafengwo.cn), a Chinese OTC that offers and shares travel information and resources for independent travelers. Mafengwo is a leading travel service platform in China, providing hundreds of millions of users with travel information and booking services for over 60,000 destinations worldwide5. Through the online community, users interact and exchange travel information with posts (asking questions) and replies (answering questions), thus providing an appropriate context for the present study.

As Mafengwo is a publicly accessible platform, its users largely keep anonymous or pseudonymous so that informed consent was not required to access and use postings (Mkono & Tribe, 2017). To further protect anonymity and confidentiality, we randomly recoded each username as NODE1, NODE2, and so forth. Moreover, this study focused on users’ interaction patterns, which do not involve any personal or private information, and no researcher interaction or intervention occurred during the data collection. Therefore, ethical clearance was not required (Hookway, 2008).

We performed the following procedures to obtain a dataset with appropriate sample size. First, through Python programming we set a time window of 4 years and crawled all the posts and corresponding replies from January 1, 2018 to December 31, 2021 in the hot topics section on the Mafengwo website to create the raw dataset. As a result, we obtained 43,867 posts in total, including 17,584 posts in 2018, 20,877 posts in 2019, 2,687 posts in 2020, and 2,719 posts in 2021. Second, considering the unbalanced distribution of the number of posts each year, we applied the method of proportionate stratified random sampling for the purpose of optimizing the representativeness of the collected data (Koffler et al., 2017). We extracted 4,000 posts from the raw dataset based on the proportion of posts in each year mentioned above, which involved 38,331 replies and 12,501 users. Third, we excluded posts or replies involving anonymous users and users with missing attributes in order to retrieve each user’s attributes. Specifically, a user’s attributes included the user ID, status, geo (geo_1/geo_2), foci, credit, travelogue, and adoption.¹ Finally, we obtained a valid dataset of 2,926 posts with 25,854 replies and 9,712 users.

Variables and Measures

The dependent variable was tie formation based on users’ replying behaviors, specifically from repliers to posters. We included actor-level variables as control variables, and the dyad- and structure-level variables as independent variables in the study. We defined and constructed the actor- and dyad-level variables based on user-specific characteristics, as illustrated in Table 2.

Table 2.

Actor- and Dyad-Level Variables

Variables	Types	Descriptions
Status	Continuous variable	A user’s status depends on the accumulated experience points since their first online registration; it is calculated automatically by Mafengwo’s reward system, mainly from daily check-in, daily assignment, and so forth.
Travelogue	Dummy variable	The number of travelogues a user writes and posts. A dummy variable is set to 1 if the number of travelogues ranks in the top 25%, and 0 otherwise.
Geo_1	Dummy variable	A user’s residence is registered in their Mafengwo profile. A dummy variable is set to 1 if the user’s residence is located outside of China (including mainland China, Hong Kong, Macau, and Taiwan), and 0 otherwise.
Geo_2	Categorical variable	Following the user’s residence registered in their Mafengwo profile, a group of categorical variables is valued in integers 1–35, representing 23 provinces, five autonomous regions, four province-level municipalities, and two special administrative regions in China, and other countries and regions.
Adoption	Continuous variable	The adoption rate of a user’s answers is calculated as the total number of answers adopted by question initiators divided by the total number of the user’s answers.
Credit	Dummy variable	A user’s possession of virtual currency circulating on the Mafengwo platform. A dummy variable is set to 1 if the virtual currency ranks in the top 25%, and 0 if otherwise.
Foci	Continuous variable	The number of users the focal user follows.

Specifically, for the actor-level control variables, we included status, travelogue, geo_1, adoption, credit, and foci. We specified status and foci in logarithmic form as the data were skewed, and we added one to each value of these two variables to avoid the undefined logarithm of zero (Fang et al., 2018). We included dyadic covariates for the dyad-level variables to capture homophily by user-specific characteristics, that is, homo (status, travelogue, geo_2).

The variables are operationalized as follows. For the continuous variable “status,” we generated its corresponding dyadic covariate by calculating the absolute differences. For the dummy variable “travelogue” and the categorical variable “geo_2,” we generated their corresponding dyadic covariates by matching their types. Additionally, we made the following explanations on variable selections. The variables “geo_1” and “geo_2” both capture the user characteristic of geographic locations; specifically, geo_1 represents national-level data and geo_2 refers to provincial-level data, which are more meaningful for dyad-level analysis. Since multiple structural processes are mixed and nested in networks simultaneously (Lusher & Robins, 2013), the following structural terms were included in the study’s model, as illustrated in Table 3.

Table 3

Parameters Included in the ERGMs Estimation

Parameters	Description	Statnet Term
Actor Level
Status	Tendency of a user with a higher user’s status to reply to others	nodeocov (status)
Travelogue	Tendency of a user who shares more travelogues to reply to others	nodeofactor (travelogue)
Geo_1	Tendency of a user who resides abroad to reply to others	nodeofactor (geo_1)
Adoption	Tendency of a user whose posts have a higher adoption rate to reply to others	nodeocov (adoption)
Credit	Tendency of a user with a higher level of user credit to reply to others	nodeofactor (credit)
Foci	Tendency of a user who follows/subscribes to more users to reply to others	nodeocov (foci)
Dyad Level
Homo (Status)	Tendency of users with similar status to form a replying tie	absdiff (status)
Homo (Travelogue)	Tendency of users with similar preferences on posting travelogues to form a replying tie	nodematch (travelogue)
Homo (Geo_2)	Tendency of users from the same residence location to form a replying tie	nodematch (geo_2)
Structure Level
Edge	Baseline tendency for a replying tie to occur	edges
Reciprocity	User A’s tendency to reply to user B, who has already replied to user A	mutual
Activity Spread	Tendency for user A to reply to multiple users	gwodegree
Generalized Transitive Closure	Tendency toward the closure of transitive triads (when user A has replied to user B and user B has replied to user C, user A is more likely to reply to user C)	gwesp
Multiple Connectivity	Tendency toward the formation of multiple two-paths connecting users	gwdsp

Data Analysis

As an innovative social network approach, exponential random graph models (ERGMs) have increasingly been employed to analyze interdependent network data in exploring the tie forming and evolving process (Xiong et al., 2020). In the present study, ERGMs were applied to test the hypotheses. ERGMs predict the probability of a tie between a pair of nodes based on the attributes of nodes (Khalizadeh, 2018), with “1” representing the presence of ties among actors and “0” for otherwise. The ERGMs in this study has the general form:

\Pr (X = x) = (\frac{1}{w}) e x p [\sum_{K} α_{k} β_{k} (x)]

where $X$ is the simulated network from the model, $x$ is the observed network, w stands for a normalizing quantity to constrain the probability ranges from 0 to 1, $K$ denotes a set of network configurations in an observed network, and $β_{k} (x)$ is the network statistic in the configuration $k$ involving structural effects and users’ individual attributes. Parameter $α_{k}$ indicates the occurrence frequency of configuration $β_{k} (x)$ in consideration of other various effects in the model (Lusher & Robins, 2013).

Specifically, all parameters were estimated with Markov chain Monte Carlo maximum likelihood estimation (MCMC-MLE). Goodness-of-fit tests were conducted to evaluate the extent to which the predicted networks match the observed network.

Results

Descriptive Statistical Analysis

Table 4 presented the summary statistics of variables on users’ individual attributes in our models (in total and yearly). For continuous variables such as status, adoption, and foci, and dummy variables such as travelogue, geo_1, and credit, we provided the mean and standard deviation of each variable as descriptive statistics. For the categorical variable geo_2, we calculated the number of users per region categorized in the variable definition and provided the mean and standard deviation accordingly.

Table 4

Summary Statistics

	Total	2018	2019	2020	2021
Status
Mean	2.824	2.877	2.853	3.388	3.291
Standard Deviation	0.619	0.637	0.612	0.512	0.558
Travelogue
Mean	0.250	0.316	0.265	0.707	0.555
Standard Deviation	0.433	0.465	0.441	0.456	0.498
Geo_1
Mean	0.028	0.029	0.027	0.024	0.037
Standard Deviation	0.164	0.168	0.162	0.153	0.188
Geo_2
Mean	277.486	125.629	178.086	7.781	15.000
Standard Deviation	396.122	182.780	250.072	11.984	20.616
Adoption
Mean	0.082	0.117	0.076	0.201	0.131
Standard Deviation	0.163	0.185	0.156	0.243	0.214
Credit
Mean	0.250	0.250	0.276	0.735	0.703
Standard Deviation	0.433	0.433	0.447	0.442	0.457
Foci
Mean	3.582	3.662	3.642	4.512	4.441
Standard Deviation	1.376	1.415	1.392	1.676	1.731
Number of Users	9,712	4,397	6,233	249	465
Number of Replies	25,854	9,632	15,134	406	682
Number of Posts	2,926	1,201	1,534	99	92

In the full sample, users located in China (including mainland China, Hong Kong, Macau, and Taiwan) accounted for 97.22%, whereas 2.78% of the total users were located outside of China. We further generated a heat map for users located in China (including mainland China, Hong Kong, Macau, and Taiwan) to reflect their geographical distribution,² in which the top five regions included Beijing (17.95%), Jiangsu (15.14%), Shanghai (10.50%), Guangdong (9.39%), and Tianjin (5.96%).

ERGMs Findings

Table 5 presents the results of ERGMs estimation for the full sample, as well as additional analysis of subsamples. Model 1 includes dyadic covariates along with the actor-level variables. For continuous covariates homo (status), the negative and significant coefficient (β = -0.243, p < .01) indicates that users were more likely to form replying ties between users with similarity in status, which supported the proposed H1(a). For dummy covariates homo (geo_2), the positive and significant coefficient (β = 1.375, p < .01) indicates that users from the same residence tended to form user replying ties, which supported H1(c). Homo (travelogue) did not exhibit a homophily effect (β = -0.151, p < .01) as proposed in H1(b).

Table 5

Results of ERGMs Estimations for Full Sample and Subsamples

	Full Sample		Subsample 1 (Pre COVID-19)		Subsample 2 (During COVID-19)		Subsample 3 (Hospitality & Tourism)
Parameters	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Actor Level
Status	0.049* (0.022)	0.049 (0.042)	0.011(0.028)	0.062 (0.058)	–0.167 (0.104)	0.198** (0.074)	0.008 (0.027)	0.229*** (0.048)
Travelogue	0.506*** (0.022)	0.652*** (0.046)	0.355*** (0.03)	0.683*** (0.058)	0.592*** (0.104)	1.311*** (0.061)	0.428*** (0.028)	0.721*** (0.051)
Geo_1	–0.027(0.049)	–0.109(0.08)	–0.034 (0.065)	0.137(0.127)	–0.014(0.205)	0.193* (0.079)	0.050 (0.061)	0.255* (0.118)
Adoption	0.155*** (0.045)	–0.709*** (0.097)	–0.055 (0.066)	–0.902*** (0.133)	–0.191(0.178)	–2.166*** (0.129)	–0.035 (0.058)	–0.611*** (0.110)
Credit	0.157*** (0.021)	0.150*** (0.044)	0.095*** (0.028)	0.253*** (0.053)	0.124(0.094)	–0.296*** (0.044)	0.136*** (0.027)	0.092* (0.055)
Foci	0.050*** (0.007)	0.068*** (0.014)	0.046*** (0.009)	0.018(0.015)	0.015(0.026)	–0.111*** (0.015)	0.046*** (0.009)	–0.011*** (0.016)
Dyad Level
Homo (status)	–0.243*** (0.016)	–0.270*** (0.029)	–0.213*** (0.021)	–0.109** (0.037)	–0.611*** (0.087)	–0.604*** (0.060)	–0.222*** (0.020)	–0.232*** (0.032)
Homo (Travelogue)	–0.151*** (0.018)	–0.122*** (0.033)	–0.276*** (0.024)	–0.131** (0.042)	0.058 (0.080)	0.179*** (0.035)	–0.193*** (0.023)	–0.148*** (0.030)
Homo (Geo_2)	1.375*** (0.017)	0.986*** (0.041)	1.386*** (0.022)	1.157*** (0.036)	0.990*** (0.084)	1.063*** (0.048)	1.333*** (0.022)	1.075*** (0.042)
Structure Level
Edge	–9.149*** (0.057)	–10.092*** (0.112)	–8.516*** (0.075)	–9.701*** (0.152)	–5.747*** (0.325)	–7.945*** (0.231)	–8.745*** (0.071)	–10.208*** (0.127)
Reciprocity		9.457*** (0.224)		1.551*** (0.496)		1.408*** (0.074)		7.679*** (0.248)
Activity Spread		2.545*** (0.051)		2.875*** (0.072)		2.997*** (0.202)		2.276*** (0.062)
Generalized Transitive closure		2.102*** (0.055)		2.946*** (0.079)		3.059*** (0.034)		3.066*** (0.057)
Multiple Connectivity		0.045*** (0.002)		0.032*** (0.003)		0.146*** (0.000)		0.060*** (0.002)
AIC^a	320,468	308,307	182,685	174,664	11,860	11,210	20,4023	19,5901
BIC^b	320,632	308,536	182,840	174,881	11,969	11,363	20,4180	19,6121

Note. ^aAkaike Information Criterion. ^bBayesian Information Criterion.

***

p < .01, **p < .05, *p <0.1.

Model 2 demonstrates a full model, including a variety of structural terms in the structure level, along with actor-level user-specific characteristics and dyad-level covariates. At the structure level, the edge term in ERGMs is equivalent to an intercept term in traditional regression models. Specifically, the coefficient of reciprocity (β = 9.457, p < .01) in Model 2 was positive and significant, indicating the tendency of a tie being reciprocated from user A to user B when user B had an existing tie to user A in the OTC. In addition, the positive coefficient of activity spread (β = 2.545, p < .01) demonstrated the increasing tendency to initiate a replying tie for users who often reply to others. It suggests that user A’s willingness to reply to multiple users in the OTC enhances when user A maintains high out-degree centrality in the network. Furthermore, the positive coefficient of generalized transitive closure (β = 2.102, p < .01) showed that transitivity effects existed in triad structures in the network. Thus, the likelihood of user A forming a replying tie to user C increased when both user A and user C had existing ties with a common third user (user B). The positive coefficient of multiple connectivity (β = 0.045, p < .01) simultaneously showed that more users act as the “bridge” in the network. Overall, these results suggest that the structural dependencies underlying the formation of user replying ties exhibit increasing tendencies in reciprocity, active spread, generalized transitive closure, and multiple connectivity (H2 a, b, c, and d).

Furthermore, the comparison of Model 1 and Model 2 revealed that the coefficients associated with status, adoption, credit, and foci in the actor-level were highly significant in all models (p < .01). However, the significance level of status, geo_1 decreased when structural terms were added in Model 2. Similarly, the magnitude of the coefficient associated with homo (status) dropped from -0.243 in Model 1 to -0.270 in Model 2, homo (geo_2) dropped from 1.375 in Model 1 to 0.986 in Model 2, whereas homo (travelogue) increased from -0.151 in Model 1 to -0.122 in Model 2. These results indicated that dyad-level results may be biased if the influence of endogenous structural processes was not included in examining the tie formation, which was regarded as one of the main advantages of ERGMs over conventional statistical methods (Hernández et al., 2021). The results also suggested that dyadic interactions in OTCs might be volatile in triads when network structural effects were considered.

The following two evaluation methods were adopted so as to verify the model fit of simulated networks. First, according to Akaike (1998), the values of AIC and BIC in Model 2 were smaller, indicating that Model 2, with structural terms, demonstrated a better fit. The finding in Model 2 suggests that our examination with endogenous structural effects in Model 2 provided a more comprehensive understanding of user interactions in the network context. Second, the graphical test of goodness-of-fit was employed to identify the match between the predicted and observed networks. Similar to the results of AIC and BIC values, the ROC curve of Model 2 also exhibited a good match between the predicted and observed networks.³

Additional Analysis

We conducted the following two additional analyses for robustness concerns. First, our 4-year dataset covered the ongoing COVID-19 pandemic period. We noticed that most recent literature highlighted that the pandemic could have exerted substantial influences on travel behavior (Rasoolimanesh et al., 2021). Thus, we performed additional analysis (I) with subsamples of pre-COVID-19 and during the COVID-19 pandemic. Second, considering that users’ replying behaviors may vary with the topics of posts in OTCs (Fang et al., 2018), we selected topics directly related to hospitality and tourism as another subsample and performed additional analysis (II).

Additional Analysis I

According to the World Health Organization (WHO), the outbreak date of COVID-19 was officially confirmed as December 31, 2019.⁴ For the purpose of comparing user interactions in OTCs pre and during COVID-19, we set December 31, 2019, as the cut-off value, and divided the dataset into Subsample 1 (covering posts from January 1, 2019 to December 31, 2019) and Subsample 2 (covering posts from January 1, 2020 to December 31, 2021).

Models 3–6 in Table 5 presents the results of ERGMs estimations for additional Analysis I. In the comparisons between Subsample 1 and Subsample 2, the results of key variables were consistent with the results of ERGMs estimations in the full sample. We also noticed that the coefficient of homo (travelogue) remained significant but positive in Subsample 2 (β = 0.179, p < .01), which suggested an increasing tendency for users with similar preferences in writing travelogues to form replying ties. One possible reason is the dramatically declined number of user interactions on OTCs during the ongoing COVID-19 pandemic, which led to the relatively limited subsample size.

Additional Analysis II

We identified the topics of each post in the full sample using text analysis. Specifically, following Lee et al. (2011), Xiang et al. (2017), and Hou et al. (2019), we employed keywords such as “restaurant,” “hotel,” “gourmet,” and “accommodation” as tags for hospitality-related posts, and keywords such as “tourism,” “scenery,” “tour,” “travel,” “attraction,” and “destination” as tags for tourism-related posts. Then, for further robustness tests, we filtered 2,014 from 2,926 posts as Subsample 3 of hospitality- and tourism-related posts (586 hospitality-related posts and 1,428 tourism-related posts), including 7,087 users and 16,201 replies. As shown in Table 5, the coefficients and significance levels of the key variables in Models 7 and 8 remained consistent with the results in the full sample (Models 1 and 2), which indicated the robustness of our key findings.

Conclusions and Discussion

Conclusions

Today’s tourists commonly use OTCs to seek and share travel-related information between peers by asking and answering questions online. However, little is known about the mechanism of how and why user interactions occur in OTCs (El-Manstrly et al., 2020). Rather than focusing on single dyads, this study advanced prior studies (Kim & Kim, 2021; Lin et al., 2019) on dyadic user interactions by examining the more complex and realistic triadic relationships in the network context. From the social network perspective, the study focused on the phenomenon of users’ replying ties in question-and-answer forums in OTCs and aimed to explore the mechanism of tie formation by modeling underlying structural dependencies of user interactions.

First, consistent with Horng and Wu (2019), the empirical results verified homophily and heterogeneity as the basis of dyadic interactions in OTCs, indicating users with similar or dissimilar attributes as the antecedents of tie formation. Specifically, in our research context, higher-status OTC users were more likely to form replying ties with higher-status users; users from the same residence location were more likely to form replying ties. Furthermore, the study results highlighted that user interactions at a dyad level may be less stable. Dyadic user interactions declined or increased in the magnitude of coefficients when structural terms in the structure level were included. This finding makes significant contributions to clarify the inconsistent understanding in existing literature regarding the influence of user attributes on replying tie formation in OTCs (Lv et al., 2021).

Second, the study uncovered the consequences of homophily or heterogeneity, which provides answers to the research question of how homophilous or heterogeneous relationships evolve across multiple levels (actor, dyad, triad, and network) and their different types of outcomes in terms of network structure. The research findings revealed that users’ replying ties in OTCs exhibited significantly positive structural dependencies in terms of reciprocity, activity spread, generalized transitive closure, and multiple connectivity. Specifically, dyadic interactions between users were reciprocity-oriented (Belanche et al., 2018), which means user A is more likely to reciprocate to user B, who has already formed a replying tie with user A in the OTC. In terms of triadic interactions, activity spread indicated that the tendency of star-like structures increased when users initiated a replying tie to multiple users in the OTC. The results on generalized transitive closure suggested that users were more likely to form replying ties with a third user; however, open triads tended to become triad closures. In other words, when user A replied to user B and user B replied to user C, user A was more likely to reply to user C. The results on multiple connectivity indicated that certain users could enjoy brokerage benefits (Lusher & Robins, 2013) in the network.

In addition, the study employed the methodology of ERGMs, which are more appropriate than conventional statistical methods like regressions in network analysis due to their advantages in accommodating the interdependence of network data and endogeneity. The full model with structural terms provided a better fit with observed data compared to Models 1 and 2, indicating that dyadic interactions and triadic interactions exist simultaneously. In recent years, ERGMs have been applied to limited studies in tourism and hospitality research, such as destination marketing organization network (Williams & Hristov, 2018), destination management (Farmaki et al., 2019; Khalilzadeh, 2018), tourism flows (Lozano & Gutierrez, 2018) and attraction network (Hernández et al., 2021). In this study, we applied ERGMs for directed network in OTCs, which overcame the limitations of traditional statistical methods by incorporating network data interdependency and endogeneity (Kim et al., 2016) and exerted unique explanatory power in the context of OTCs.

Theoretical Contributions

Very limited research has examined dyadic user interactions by considering the network context in which they are embedded. This study emphasizes the significant distinctions between single dyads and triads in networks by modeling user interactions in OTCs. The findings confirm that variations in dyadic user interactions may occur if endogenous structural effects are included and provide unique explanations regarding inconsistent findings of how user-specific characteristics influence user interactions in the previous literature (e.g., Choi et al., 2019; Fang et al., 2018; Liu & Park, 2015; Shen et al., 2020). Specifically, the study presents the following theoretical implications.

First, the study discussed how user interactions occur and their outcomes across levels by revealing the underlying network mechanism in OTCs. Prior research mainly focused on dyadic interactions between users (Chan et al., 2017; Lee et al., 2014; Lv et al., 2021), but neglected the real, complex network context in which dyads are embedded (Lomi et al., 2014). Therefore, the study specifically focused on triadic rather than traditional dyadic user interactions, which provides a new understanding and horizon to the existing hospitality and tourism network literature by modeling structural dependencies underlying user interactions in OTCs. Therefore, the study makes significant contributions to the body of knowledge by incorporating structure-level effects with actor- and dyad-level effects for a comprehensive and integrated understanding of network mechanisms in the hospitality and tourism sector.

Second, while extant literature considers the principle of homophily as the antecedents of dyadic interaction (Kunz & Seshadri, 2015), the study extends the social network theory by focusing on the consequences derived from homophily across levels in OTCs (Ertug et al., 2022). Specifically, the study investigates “how homophilous ties among users evolve after their formation” instead of “how users with similar attributes form ties” in terms of network structure.

Third, the study applied ERGMs in examining the formation of user replying ties in OTCs, which demonstrates distinct advantages of ERGMs over conventional statistical methodologies in dealing with the interdependence of network data. Specifically, using the directed network of OTC as the research context, this study included both exogenous effects (i.e., actor-level effects and dyad-level effects) and endogenous effects (i.e., structure-level effects). More importantly, the study provided meaningful explanations for the inconsistent findings on the co-occurrence of ties by fully considering structural dependencies (e.g., the propensity to visit attractions with similar ratings in Hernández et al. (2021) versus the propensity to visit attractions with dissimilar ratings in Liu et al. (2017). Therefore, the study makes significant methodological contributions to hospitality and tourism network research by offering a new pathway to understand network mechanisms.

Furthermore, recent literature has emphasized the negative impacts of the ongoing COVID-19 pandemic on tourist behaviors (Hao et al., 2022) and on the hospitality and tourism industry (Ntounis et al., 2022). This study specifically focused on peer-to-peer user interactions in OTCs and compared user interactions between pre COVID-19 and during COVID-19, which extended the current discussions on network analysis and offered new insights into theorizing the complicated reflections of tourist behaviors in different contexts.

Practical Implications

This study examined user interactions in an OTC over a 4-year period, which delivers the following practical implications. First, interactivity is regarded as a significant driver of value co-creation in OTCs (Shen et al., 2020). Our findings assist OTC practitioners to recognize the user-specific characteristics that drive users to form replying ties, which helps in forecasting patterns, structures, and tendencies of user interactions in today’s online hospitality and tourism context. In addition to user interactions in single dyads, user interactions may occur via a third party in triads or through multiple third parties in the network context. Considering the fast-growing popularity of OTCs, current OTCs encompass millions of users and vast amounts of information (Fang et al., 2018; Zhou, et al., 2021), it would be helpful to adopt artificial intelligence technologies (e.g., automated recommender systems on the internet) when tracing and predicting users’ replying behavior and interaction patterns. For example, based on users’ characteristics, replying records, and patterns, OTC managers can automatically match users with certain questions and recommend them to reply, leading to more user engagement in question-and-answer forums and better user interaction efficiency. These practical implications can also be generalized to other online communities to better understand users’ replying ties and further promote user interactions.

Second, with the ongoing COVID-19 pandemic worldwide, travel has become more challenging and requires additional health screening, temperature checking, quarantining, and seemingly constant regulation change in destinations. Travel and tourism may see the “new normal” with divergent destination experiences and tourist behaviors. Our findings indicated that there is an increasing tendency for users of a certain status to reply to others of a same status. In contrast, we found a decreasing tendency of users in the same locations to reply to others in the during COVID-19 subsample compared with the subsample of pre COVID-19. Besides, the tendencies of activity spread, generalized transitive closure, and multiple connectivity increase in the subsample of during COVID-19. These slight changes suggest that the number of active users in the OTC is declining, which is a reflection of the negative impacts of the COVID-19 pandemic on the hospitality and tourism industry worldwide (Fotiadis et al., 2021). Consequently, the OTC serves as an essential and effective channel for users to fulfill their increasing needs to exchange and share first-hand, real-time, and in-depth travel information to mitigate the potential risks and uncertainties under the new hospitality and tourism context (El-Manstrly et al., 2020). Therefore, understanding the formation and evolving structure of user interactions in OTCs will enable online platform marketers and managers to better serve online users in their travel.

Limitations and Future Research

This study has the following limitations, which suggests a need for future research. First, taking users’ question-and-answer forums in Mafengwo as an example, the study used the dataset collected from a single Chinese online community and formed a directed network with the size of 9,712 nodes and 25,854 edges as our focal network, where nodes referred to users and ties referred to sending and replying relationships. Although Mafengwo encompasses a large number of online users and active interactions, a single business case may have limited the generalizability of the findings. Future studies should examine the data from different countries and cultural backgrounds for comparison and provide more extensive understandings of user interactions in OTCs or other contexts.

Second, the study collected longitudinal data over a 4-year period, which provided a large and comprehensive sample for OTCs research. However, due to the severe impact of the COVID-19 pandemic on the hospitality and tourism industry (Ntounis et al., 2022), OTCs also experienced a decline in user participation, which was also reflected in the unbalanced distribution of the number of posts across 2018 to 2021. Future research is suggested when the hospitality and tourism industry fully recovers from the recession by performing a simulation with large-scale data or even the entire population data, so as to provide a full picture of the complex user interactions in OTCs and draw conclusive remarks on replying tie formation and evolution.

Supplemental Material

sj-docx-1-jht-10.1177_10963480221141616 – Supplemental material for User Interactions in Online Travel Communities: A Social Network Perspective

Supplemental material, sj-docx-1-jht-10.1177_10963480221141616 for User Interactions in Online Travel Communities: A Social Network Perspective by Bing Liu, Fang Meng, Chaoliang Luo and Hui Jiang in Journal of Hospitality & Tourism Research

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Humanities and Social Sciences Foundation of the Ministry of Education of China (grant number 22YJA630048) and the National Natural Science Foundation of China (grant number 71902049).

ORCID iDs

Bing Liu

Fang Meng

Chaoliang Luo

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Bing Liu, PhD (email: liub33@mail.sysu.edu.cn), is an associate professor at Business School, Sun Yat-sen University, Shenzhen, China.

Fang Meng, PhD (email: fmeng@hrsm.sc.edu), is a professor at School of Hotel, Restaurant and Tourism Management, University of South Carolina, Columbia, SC, USA.

Chaoliang Luo, PhD (email: luochl3@mail2.sysu.edu.cn), is an associate professor at School of Business, Guizhou University of Finance and economics, Guiyang, China.

Hui Jiang (email: hj17@email.sc.edu), is a Ph.D. student at School of Hotel, Restaurant and Tourism Management, University of South Carolina, Columbia, SC, USA.

References

Chi

O. H.

Ouyang

(2019). Categorizing peer-to-peer review site features and examining their impacts on room sales. Journal of Hospitality Marketing & Management, 28(7), 862–881. https://doi.org/10.1080/19368623.2019.1568341

Akaike

(1998). Information theory and an extension of the maximum likelihood principle. Springer. https://doi.org/10.1007/978-1-4612-1694-0_15

Akhtar

Kim

W. G.

Ahmad

Akhtar

M. N.

Siddiqi

U. I.

Riaz

(2019). Mechanisms of consumers’ trust development in reviewers’ supplementary reviews: A reviewer-reader similarity perspective. Tourism Management Perspectives, 31, 95–108. https://doi.org/10.1016/j.tmp.2019.04.001

Belanche

Casaló

L. V.

Flavián

Guinalíu

(2018). Reciprocity and commitment in online travel communities. Industrial Management & Data Systems, 119(2), 397–411. https://doi.org/10.1108/IMDS-03-2018-0098

Chan

I. C. C.

Lam

L. W.

Chow

C. W.

Fong

L. H. N.

Law

(2017). The effect of online reviews on hotel booking intention: The role of reader-reviewer similarity. International Journal of Hospitality Management, 66, 54–65. https://doi.org/10.1016/j.ijhm.2017.06.007

Chang

Hou

R. J.

Wang

Cui

A. P.

Zhang

C. B.

(2020). Effects of intrinsic and extrinsic motivation on social loafing in online travel communities. Computers in Human Behavior, 109, Article 106360. https://doi.org/10.1016/j.chb.2020.106360

Choi

Kim

Cha

Suh

Y. K.

Kim

K. H.

(2019). Travelers’ parasocial interactions in online travel communities. Journal of Travel & Tourism Marketing, 36(8), 888–904. https://doi.org/10.1080/10548408.2019.1657053

Colladon

A. F.

Guardabascio

Innarella

(2019). Using social network and semantic analysis to analyze online travel forums and forecast tourism demand. Decision Support Systems, 123, Article 113075. https://doi.org/10.1016/j.dss.2019.113075

Currarini

Matheson

Vega-Redondo

(2016). A simple model of homophily in social networks. European Economic Review, 90, 18–39. https://doi.org/10.1016/j.euroecorev.2016.03.011

10.

Duffy

(2019). If I say you’re authentic, then you’re authentic: Power and privilege revealed in travel blogs. Tourist Studies, 19(4), 569–584. https://doi.org/10.1177/1468797619865387

11.

El-Manstrly

Ali

Steedman

(2020). Virtual travel community members’ stickiness behaviour: How and when it develops. International Journal of Hospitality Management, 88, Article 102535. https://doi.org/10.1016/j.ijhm.2020.102535

12.

Ertug

Brennecke

Kovács

Zou

(2022). What does homophily do? A review of the consequences of homophily. Academy of Management Annals, 16(1), 38–69. https://doi.org/10.5465/annals.2020.0230

13.

Fang

Chen

Wang

George

(2018). Not all posts are treated equal: An empirical investigation of post replying behavior in an online travel community. Information & Management, 55(7), 890–900. https://doi.org/10.1016/j.im.2018.04.003

14.

Farmaki

Khalilzadeh

Altinay

(2019). Travel motivation and demotivation within politically unstable nations. Tourism Management Perspectives, 29, 118–130. https://doi.org/10.1016/j.tmp.2018.11.004

15.

Filieri

McLeay

(2014). E-WOM and accommodation: An analysis of the factors that influence travelers’ adoption of information from online reviews. Journal of Travel Research, 53(1), 44–57. https://doi.org/10.1177/0047287513481274

16.

Fotiadis

Polyzos

Huan

T. C. T.

(2021). The good, the bad and the ugly on COVID-19 tourism recovery. Annals of Tourism Research, 87, Article 103117. https://doi.org/10.1016/j.annals.2020.103117

17.

Ghahramani

Khalilzadeh

Birendra

K. C.

(2018). Tour guides’ communication ecosystems: An inferential social network analysis approach. Information Technology & Tourism, 20(1–4), 103–130. https://doi.org/10.1007/s40558-018-0114-y

18.

González

I. O.

Camarero

Cabezudo

R. S. J.

(2021). SOS to my followers! The role of marketing communications in reinforcing online travel community value during times of crisis. Tourism Management Perspectives, 39, Article 100843. https://doi.org/10.1016/j.tmp.2021.100843

19.

Hao

Park

Chon

(2022). Social media and disaster risk reduction and management: How have reddit travel communities experienced the COVID-19 pandemic? Journal of Hospitality & Tourism Research, 0(0). https://doi.org/10.1177/10963480221081115.

20.

Hernández

J. M.

Santana-Jiménez

González-Martel

(2021). Factors influencing the co-occurrence of visits to attractions: The case of Madrid, Spain. Tourism Management, 83, Article 104236. https://doi.org/10.1016/j.tourman.2020.104236

21.

Hookway

(2008). Entering the blogosphere’: Some strategies for using blogs in social research. Qualitative Research, 8(1), 91–113. https://doi.org/10.1177/1468794107085298

22.

Horng

S. M.

C. L.

(2020). How behaviors on social network sites and online social capital influence social commerce intentions. Information & Management, 57(2), Article 103176. https://doi.org/10.1016/j.im.2019.103176

23.

Hou

Cui

Meng

Lian

(2019). Opinion mining from online travel reviews: A comparative analysis of Chinese major OTAs using semantic association analysis. Tourism Management, 74, 276–289. https://doi.org/10.1016/j.tourman.2019.03.009

24.

Jin

Zhong

Zhai

(2015). Why users contribute knowledge to online communities: An empirical study of an online social Q&A community. Information & Management, 52(7), 840–849. https://doi.org/10.1016/j.im.2015.07.005

25.

Khalilzadeh

(2018). Demonstration of exponential random graph models in tourism studies: Is tourism a means of global peace or the bottom line? Annals of Tourism Research, 69, 31–41. https://doi.org/10.1016/j.annals.2017.12.007

26.

Kim

J. J.

(2021). Emotional attachment, age and online travel community behaviour: The role of parasocial interaction. Current Issues in Tourism, 24(24), 3466–3488. https://doi.org/10.1080/13683500.2021.1952942

27.

Kim

J. Y.

Howard

Cox Pahnke

Boeker

(2016). Understanding network formation in strategy research: Exponential random graph models. Strategic Management Journal, 37(1), 22–44. https://doi.org/10.1002/smj.2454

28.

Koffler

Shonfield

Vickers

(2017). Beyond pedigree—Optimizing and measuring representativeness in large-scale LCAs. The International Journal of Life Cycle Assessment, 22(7), 1065–1077. https://doi.org/10.1007/s11367-016-1223-5

29.

Kunz

Seshadri

(2015). From virtual travelers to real friends: Relationship-building insights from an online travel community. Journal of Business Research, 68(9), 1822–1828. https://doi.org/10.1016/j.jbusres.2015.01.009.

30.

Lawrence

B. S.

Shah

N. P.

(2020). Homophily: Measures and meaning. Academy of Management Annals, 14(2), 513–597.

31.

Lee

H. A.

Law

Murphy

(2011). Helpful reviewers in TripAdvisor, an online travel community. Journal of Travel & Tourism Marketing, 28(7), 675–688. https://doi.org/10.1080/10548408.2011.611739

32.

Lee

Reid

Kim

W. G.

(2014). Understanding knowledge sharing in online travel communities: Antecedents and the moderating effects of interaction modes. Journal of Hospitality & Tourism Research, 38 (2), 222–242. https://doi.org/10.1177/1096348012451454

33.

Lin

Zhang

Gursoy

(2019). Impact of tourist-to-tourist interaction on tourism experience: The mediating role of cohesion and intimacy. Annals of Tourism Research, 76(C), 153–167. https://doi.org/10.1016/j.annals.2019.03.009

34.

Liu

Huang

S. S.

(2017). An application of network analysis on tourist attractions: The case of Xinjiang, China. Tourism Management, 58, 132–141. https://doi.org/10.1016/j.tourman.2016.10.009

35.

Liu

Park

(2015). What makes a useful online review? Implication for travel product websites. Tourism Management, 47, 140–151. https://doi.org/10.1016/j.tourman.2014.09.020

36.

Liu

Schuckert

Law

(2018). Utilitarianism and knowledge growth during status seeking: Evidence from text mining of online reviews. Tourism Management, 66, 38–46. https://doi.org/10.1016/j.tourman.2017.11.005

37.

Lomi

Lusher

Pattison

P. E.

Robins

(2014). The focused organization of advice relations: A study in boundary crossing. Organization Science, 25(2), 438–457. https://doi.org/10.1287/orsc.2013.0850

38.

Lozano

Gutiérrez

(2018). A complex network analysis of global tourism flows. International Journal of Tourism Research, 20(5), 588–604. https://doi.org/10.1002/jtr.2208

39.

Lusher

Robins

(2013). Formation of social network structure. In Lusher

Koskinen

Robins

(Eds), Exponential random graph models for social networks: Theory, methods and applications (pp. 16–28). Cambridge University Press. https://doi.org/10.1017/CBO9780511894701.004

40.

Zhang

(2021). Value co-destruction: The influence of failed interactions on members’ behaviors in online travel communities. Computers in Human Behavior, 122, Article 106829. https://doi.org/10.1016/j.chb.2021.106829

41.

Melamed

Sweitzer

Simpson

Abernathy

J. Z.

Harrell

Munn

C. W.

(2020). Homophily and segregation in cooperative networks. American Journal of Sociology, 125(4), 1084–1127. https://doi.org/10.1086/708142

42.

Mkono

Tribe

(2017). Beyond reviewing: Uncovering the multiple roles of tourism social media users. Journal of Travel Research, 56(3), 287–298. https://doi.org/10.1177/0047287516636236

43.

Moon

Miao

Hanks

Line

N. D.

(2019). Peer-to-peer interactions: Perspectives of Airbnb guests and hosts. International Journal of Hospitality Management, 77, 405–414. https://doi.org/10.1016/j.ijhm.2018.08.004

44.

Munn

(2020). Angry by design: Toxic communication and technical architectures. Humanities and Social Sciences Communications, 7(1), 1–11. https://doi.org/10.1057/s41599-020-00550-7

45.

Nguyen

Q. N.

Sidorova

Torres

(2022). User interactions with chatbot interfaces vs. Menu-based interfaces: An empirical study. Computers in Human Behavior, 128, Article 107093. https://doi.org/10.1016/j.chb.2021.107093

46.

Ntounis

Parker

Skinner

Steadman

Warnaby

(2022). Tourism and hospitality industry resilience during the COVID-19 pandemic: Evidence from England. Current Issues in Tourism, 25(1), 46–59. https://doi.org/10.1080/13683500.2021.1883556

47.

Opper

Nee

Brehm

(2015). Homophily in the career mobility of China’s political elite. Social Science Research, 54, 332–352. https://doi.org/10.1016/j.ssresearch.2015.08.007

48.

Pantano

Priporas

C. V.

Stylos

(2017). You will like it! Using open data to predict tourists’ response to a tourist attraction. Tourism Management, 60, 430–438. https://doi.org/10.1016/j.tourman.2016.12.020

49.

Pantano

Priporas

C. V.

Stylos

Dennis

(2019). Facilitating tourists’ decision making through open data analyses: A novel recommender system. Tourism Management Perspectives, 31, 323–331. https://doi.org/10.1016/j.tmp.2019.06.003

50.

Podolny

J. M.

(2005). Status signals: A sociological study of market competition. Princeton University Press.

51.

Rasoolimanesh

S. M.

Seyfi

Rastegar

Hall

C. M.

(2021). Destination image during the COVID-19 pandemic and future travel behavior: The moderating role of past experience. Journal of Destination Marketing & Management, 21, Article 100620. https://doi.org/10.1016/j.jdmm.2021.100620

52.

Sharifi

(2019). Examining the impacts of positive and negative online consumer reviews on behavioral intentions: Role of need for cognitive closure and satisfaction guarantees. Journal of Hospitality Marketing & Management, 28(4), 397–426. https://doi.org/10.1080/19368623.2019.1531804

53.

Shen

Xue

(2020). The effect of online interaction and trust on consumers’ value co-creation behavior in the online travel community. Journal of Travel & Tourism Marketing, 37(4), 418–428. https://doi.org/10.1080/10548408.2018.1553749

54.

Williams

N. L.

Hristov

(2018). An examination of DMO network identity using Exponential Random Graph Models. Tourism Management, 68, 177–186. http://doi.org/10.1016/j.tourman.2018.03.014

55.

Xiang

Fan

(2017). A comparative analysis of major online review platforms: Implications for social media analytics in hospitality and tourism. Tourism Management, 58, 51–65. https://doi.org/10.1016/j.tourman.2016.10.001

56.

Xiong

Feng

Tang

(2020). Understanding user-to-user interaction on government microblogs: An exponential random graph model with the homophily and emotional effect. Information Processing & Management, 57(4), Article 102229. https://doi.org/10.1016/j.ipm.2020.102229

57.

Chen

Pearce

Mohammadi

Pearce

P. L.

(2021). Reaching audiences through travel vlogs: The perspective of involvement. Tourism Management, 86, Article 104326. https://doi.org/10.1016/j.tourman.2021.104326

58.

Zhang

(2021). Online expression as well-be(com)ing: A study of travel blogs on Nepal by Chinese female tourists. Tourism Management, 83, Article 104224. https://doi.org/10.1016/j.tourman.2020.104224

59.

Xue

Wang

Gursoy

Song

(2021). Effects of customer-to-customer social interactions in virtual travel communities on brand attachment: The mediating role of social well-being. Tourism Management Perspectives, 38, Article 100790. https://doi.org/10.1016/j.tmp.2021.100790

60.

Zhou

Kim

W. G.

Okumus

Cobanoglu

(2021). Understanding online travel communities: A literature review and future research directions in hospitality and tourism. Journal of Travel & Tourism Marketing, 38(2), 194–212. https://doi.org/10.1080/10548408.2021.1887052

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

3.19 MB