*3.7. Reciprocity*

Reciprocity is one of the basic principles of social relations [77,78]. The concept of reciprocity derives from social exchange theory. Reciprocity is regarded as the mechanism of social exchange [79]. Constant et al. (1994) argued that when two individuals are influenced by their social and organizational environment, the social exchange relationship will be the main determinant of their attitudes [80]. Huber also argued that based on people's desire for fairness and reciprocity, members in a society believe their mutual relationships with one another will improve [81]. This implies that when members in virtual social networks have a higher level of reciprocity, their awareness and evaluation of the relationships with other members in that network are more positive. For example, Xiong et al. argued that members' reciprocity positively affects their knowledge contribution and establishment of social relationships in virtual social networks [82].

Therefore, we introduce reciprocity as another social perspective to analyze formation mechanisms of opinion leaders in consumer advice networks. We propose hypothesis H7.

**Hypothesis 7 (H7).** *In consumer advice networks, reciprocity positively affects the establishment of network relationships.*

Based on the above hypotheses, the proposed model is summarized in Figure 1.

**Figure 1.** Research model.

#### **4. Research Design**

#### *4.1. Data Source*

In order to test the hypotheses proposed in this paper, we selected the content community of the "Qunar Travel" website (http://tips.qunar.com/, accessed on 2 May 2005) as the research object [83]. This content community is a typical consumer advice network.

Qunar Travel (https://www.qunar.com/, accessed on 2 May 2005) is one of the larger Chinese online travel websites, founded in February 2005 and headquartered in Beijing. Qunar provides consumers with real-time searches for airline tickets, hotels, venues, and vacation products, as well as group purchase options for travel products and other travel information services.

The content community of "Qunar Travel" allows registered members of the site to post a variety of reviews about hotel products and service experiences anonymously in the community. Members of the site can follow other members and can be followed by other members. Following gives them priority to see the reviews posted by the followed members and these reviews help potential consumers in their purchasing decisions. For this reason, we refer to such communities as online consumer advice networks.

#### *4.2. Data Collection*

To study the mechanisms of opinion leadership formation among influencers in consumer advice networks, we selected certified members of the network who had the "sleep tester" logo (shown in Figure 2). Both certified and uncertified members exist in consumer advice networks and allow for both relationship-establishing and social interaction between them.

"Hotel sleep testers", also known as the "hotel connoisseur", require candidates to have a keen sense of observation and feeling, a love of travel, and a willingness to share what they see and hear. The creation of hotel sleep testers has its own theoretical basis and review-based marketing has become an effective marketing approach that is widely used with good results. Research shows that 90% of hotel online booking users view and refer to other people's hotel reviews. As a result, a large number of objective reviews on online travel sites can often attract consumers to book hotels through suppliers.

**Figure 2.** Schematic diagram of hotel sleep testers' homepage.

To be certified as a "Hotel Sleep Tester", one must first apply and go through a rigorous vetting process. As such, "Hotel Sleep Testers" are a group of influencers who want to influence others, have a willingness to share what they see and hear, and aim to become opinion leaders. This study will explore which "Hotel Sleep Testers" are able to become opinion leaders and why.

We needed to select some "Hotel Sleep Testers" and observe the establishment of network relationship among them. First, eight "Hotel Sleep Testers" were randomly selected; second, 1356 "Hotel Sleep Testers" were selected from the consumer advice network (tips.qunar.com, accessed on 2 May 2005) using the snowball method; and third, the relationship establishing data, behavioral activity data, and other data of these 1356 "Hotel Sleep Testers" were obtained.

The data records of the network members date back to as early as December 2009, with a time span of 11 years of data to December 2020. The time-series data with a long time span provides us with rich data points for analyzing relationship-establishing patterns and also ensures the robustness of our findings, which is crucial for studying the formation of social network relationships.

#### *4.3. Variable Construction*

Due to the long time span of the data, we set the interval between each observation to a quarter and constructed a panel data in quarterly time intervals. The dependent variable in this study is the state of relationship establishment, which takes the value of 0 if the relationship is not established at the moment of observation and 1 if the relationship is established.

For the convenience of representation, the following assumption is made: denote the network formed by members by *<sup>G</sup>*(*<sup>V</sup>*, *<sup>E</sup>*), where *V* is the set of network nodes (i.e., members) and *E* is the set of network edges (i.e., relations). Next, we considered whether a potential edge (relationship) *Eij* = *Vi*, *Vj*- is established, with *Eij* denoting *Vi* as the following node (fan) and *Vj* as the followed node (influencer). The specific independent and control variables are defined as shown in Table 2.


#### **Table 2.** Variable explanation.

Furthermore, the measurement of the variables is explained through the personal homepage of a "hotel sleep tester" (as in Figure 2).

Contribution: the number of content posts contributed by the following node *Vj*, as shown in Figure 2 "Contribution", is 1525.

Interaction: Other members will have social interaction with the followed member *Vj*, which is mainly reflected in the comments on the content posts contributed by *Vj*, as shown in the "comment" in Figure 2. The value of interaction is the total number of comments received by member *Vj*. As shown in Figure 2, the "Total number of comments" is 1717.

Helpfulness: If other members feel that the content contributed by *Vj* is helpful, they will like the content, such as through the "Like" function in Figure 2. The value of helpfulness is the total number of likes obtained by member *Vj*. As shown in Figure 2, the "Total number of likes" is 6624.

In-degree: Member *Vj* has a number of fans. As shown in Figure 2, the "In-degree" is 977. Structural equivalence: this refers to the number of members that establish relationships withbothandinwhichthethevalue,thethestructural

 *Vi Vj*, larger stronger equivalence. Homophily: the absolute value of the difference between the level of *Vi* and *Vj*, in whichthesmallerthevalue,thehigherthehomophily.

 Reciprocity: if *Vj*is a follower of *Vi*, the value is 1 and otherwise the value is 0.

Out-degree: The number of members followed by member *Vj*. As shown in Figure 2, the"Out-degree"is27.

Level: The level of member *Vj*. As shown in Figure 2, the "Level" is 7.

Survival time: the duration member *Vj* has been in the consumer advice network, expressed as the length of time between the moment of registration and December 2020, in months.

#### *4.4. Data Analysis*

In this section, we used the software IBM SPSS 21.0 to process the collected data, including descriptive statistics and a correlation analysis, as well as the regression analysis in Section 5. Table 3 shows the descriptive statistics with data sample *n* = 1356. Table 4 shows the results of the correlation analysis between variables with the number of variables v = 10.


**Table 3.** Descriptive statistics.

Notes: Sample *n* = 1356.The data analysis was conducted using IBM SPSS 21.0 software.



Notes: Sample *n* = 1356, \* *p* < 0.05, \*\* *p* < 0.01, and \*\*\* *p* < 0.001. The data analysis was conducted using IBM SPSS 21.0 software.

#### **5. Empirical Analysis**
