The Effect of Content Feeding on Electronic Word of Mouth (E-WOM) Activities Mediated by Information Dissemination

This research was conducted to examine the mediating role of information dissemination (seeding) in the relation between marketing campaigns (content feeding) and electronic word of mouth (eWOM). The research object is three marketing campaigns from three online marketplaces in Indonesia, named: Shopee, Tokopedia, and Bukalapak. A model was built to test and evaluate and test the mediation effect existence. The number of participants was involved in this research is 200 participants, and they answer the specific question related to the research. We found that all regression path in the models is significant, that indicate direct and indirect effect happened simultaneously. Our conclusion is information dissemination (seeding activity) playing an important role as a partial mediating variable. Other implications and future research discussed. study proposed an examination of mediating role between marketing campaigns (feeding activity) and eWOM. Using a sample of 200 internet users who had direct experience in an online marketplace, this study provides insight into the moderating effect of eWoM. From the outcomes of this study, it is obvious that the study has provided several important and useful contributions to the consumer behavior from the of marketplace users. The study has successfully captured the mediating effect and contributed to the subsequent practical and theoretical implications. The result shows there is a direct and indirect effect in a model we see table and we conclude that partial mediation has become an important issue in this research. them”


129
According to the data, the Ministry of Communication and Information of Indonesia (2018) states that the number of internet users in 2017 has reached 143.26 million people or equivalent to 54.68 percent of the total population of Indonesia. This number shows an increase of 10.56 million people from the survey results in 2016. The increased connection between consumers through online means increased strength from the online review, and the ability to track these connections online to encourage interest in the e-WOM process (Dost et al., 2018). When connectivity between users increased, the tendency for someone to share information also increases. Sharing information can be in various forms such as likelihood to content sharing, giving likes, or writing and share an online review (Gao et al., 2010).
Digital platforms then transformed become an infrastructure for the flow of interactive communication. Traditionally, information flows from producers to consumers or in vertical ways, but recent information technology has transformed it becomes horizontal. Consumers have more options for an information source and more various points of view. This shows that more connectedness between consumers will change the business landscape, from a producer-consumer to a consumer-consumer. In the past 10 years, technological innovations allow stores not only in physical form but also in virtual form. Online marketplace provides consumer needs and desires in realtime through technology. Research from Google Temasek states that e-commerce sales in Indonesia this year will reach USD 10.9 billion, or around Rp 1, 477 trillion (Medcom.id, 2018). But, behind the rapid growth of electronic WoM, the question is "what drives people doing online interaction?" Desire for social interaction, desire for economic incentives, concern for other consumers, and the potential to enhance their own self-worth are the primary factors leading to eWOM behavior (Hennig-Thurau et al., 2004). Internet users expecting an enjoyment experience in interaction with other people because of ease of accessibility to the Internet and the capability to generate online content. Internet growth makes mobile shopping applications have become a popular channel to reach the omnipresent consumer of today. The impact is market space has become highly competitive, and as a consequence, cultivating repurchase behavior has become pivotal to retain market share and boost sustainability (Fernandes & Barfknecht, 2020).
Electronic word of mouth (eWoM) can have positive or negative implications for the firm. When positive conversations spread quickly, they can lead to virtually free advertising for the firm, growing brand recognition, increased sales, and so on. But, Negative eWoM, on the other hand, can cause costly or even irreparable damage (Kietzmann & Canhoto, 2013). As we stated previously when the business landscape changes from vertical to horizontal, the company cannot control the "reality" of information.
Unlike the traditional WoM which normally involves opinions expressed by acquaintances such as friends, family members or colleagues, online comments in eWoM are generated by unidentifiable sources. eWoM is significantly different from the traditional WoM due to its convenience, speed of interactions, anonymity, manyto-many communications, lack of time and space restrictions, scope and source (Leong et al., 2019). Given the distinct characteristics of internet communication (e.g., directed to multiple individuals, available to other consumers for an indefinite period of time, and anonymous), eWoM deserves the serious attention (Hennig-Thurau et al., 2004).
But, learning what motivation that drives people to doing eWoM becomes important. A recent challenge in marketing communication in the digital era is competing with tons of other information. A marketer must make their message appear as often as possible, and sometimes it can be costly to maintain those circumstances. Company can use eWoM strategy to get continually consumers' attention, but a marketer must create an extraordinary marketing stimulant first. However, the question "it is possible to make a message that has its appeal, so it can spread itself without interference from the company?". The answer is "Yes" if they can create the extraordinary ones, consumers will respond with either positive or negative feedback. A marketer can use a an extra ordinary marketing campaign to gain user attention and their involvement.
A company must create their extra ordinary marketing campaign as a feeding, then dissemination process will follow. Dissemination of information is expected to influence others, one of which is in forming a strong bond between the company and the customer. Moreover, technology has now leveled communication lines from vertical to horizontal. Communication is now not only two-way but horizontal (B2C to C2C), from the beginning of the company to the consumer, to the consumer to the consumer. Especially when assisted by social media (Hewett et al., 2016;Kozinets et al., 2010;Trusov et al., 2008).
There is a gap between the marketing campaign (feeding activity) and the expected outcome (eWoM). They must examine the intervening process between the feeding and the outcome, in other word the stimulant (such as marketing campaing), must through a seeding process before it becomes an eWoM (Hinz et al., 2011a). Its must 130 has their own appeal to deserve spread massively. A company providing a feeding in form of intractive marketing campaign to trigger consumer involvement. Positive response from a lot of user becomes a sign that stimulant deserve to voluntary sharing from user. This process called seeding (or information dissemination), and when it happen in massive then becomes viral marketing. Hence, to manage 'viral marketing', it is critical to understand what motivates consumers to share their consumption experiences. Likewise, it is important to recognize that diverse types of consumption-based content are distributed differently by often highly socially networked individuals. It follows that, the importance of eWoM as a spark for potentially viral content should not be underestimated (Kietzmann & Canhoto, 2013).
With this rapid growth, it is very important to know how the behavior of their users, especially when given a stimulant in the form of content feeding. In an industry online marketplace, given the fierce competition, one cannot rely solely on vertical communication between companies and customers. Through Web-based consumer opinion platforms (e.g., epinions.com), the Internet enables customers to share their opinions on, and experiences with, goods and services with a multitude of other consumers; that is, to engage in electronic word-of-mouth (eWoM) communication (Hennig-Thurau et al., 2004).
Marketing campaign activity related to online platform facing a several problems. At least there are three, such as related to referral programs (which encourage and incentivize current customers to contribute to customer acquisition by helping to acquire new customers); online recommendation programs (which encourage individuals to spread the word to their close social network or a broader network such as in an online review website); and seeding programs (which aim to get products into the hands of some individuals (seeds) in the hope that the consequent social influence will help accelerate and expand the growth process (Dost et al., 2018).
In this research, we focused on examine the moderating role of seeding activity, that determine the impact of feeding on eWoM activities. Seeding activity might enable information to diffuse throughout parts of the network and prevent a viral message from simply circulating in an already "infected" (Hinz et al., 2011b). Some marketing campaigns use WOM as the main vehicle to drive new product growth and seeding campaigns as a main tool to initiate the process (Dost et al., 2018). We use industries related to the development of information technology and the online marketplace as a contextual element. Research model will test with three different marketing campaign from three different online marketplace in Indonesia.

A. Technological Developments, DSMM, And Marketing Communication
Digital Social Media and Mobile (DSMM) marketing have evolved to offer marketing in reaching, informing, binding, selling, learning places, and providing services to consumers (Lamberton & Stephen, 2016). Internet things produce new ways of communicating, some important things that can be noted are: 1) As an expression facilitator, 2) Helping as a decision-making tool, 3) As a source of marketing intelligence. Simultaneous developments in technology, individual equipment, and infrastructure enable companies to explore the potential of these things to increase as commercial excellence (Yu et al., 2017).
In the past, the limitation of facilities to make communication that occurred was top-down between the company (firm) with consumers (consumers), or what is called F2C. However, with this available, it allows communication between consumers to consumers (C2C), so that the form of communication that occurs is not only vertical but horizontal (Batra & Keller, 2016;de Vries et al., 2017). Consumers can more easily communicate with other consumers, so they can use it to express emotions, help in making purchasing decisions, and share information with other consumers. Forms of communication between consumers to other consumers can be in the form of Word-of-Mouth Communication or WoM.

B. Word Of Mouth Communication (WoM) And E-Wom
Lie (2016), summarizes some opinions to define Word-of-Mouth communication, that is, communication made orally, person-to-person between recipient and sender, who receives non-commercial messages about brands, products or services. WoM is the result of an event (or part of a company's marketing activity) that is retold by consumers to other consumers. WoM is more of a reaction caused by a stimulant. In the initial definition, Word of Mouth Communication (WoM) is a form of Face-to-Face communication, where consumers communicate directly to other consumers, then with the development of information technology, interaction can 131 be done through many media so that the term Electronic Word of Mouth Communication appears or is popular with the term E-WoM. Important elements in WoM are the presence of feeding or stimulants and Seeding or spreading reactions. WoM, which is part of the marketing communication strategy, will be designed and prepared as part of supporting promotional activities. Various forms of stimulants can cause WoM, such as Video clips or clips (Liu et al., 2018), Stories on the platform content (Kanuri et al., 2018), or very well-known examples of Farmville and Cityville games (Schulze et al., 2013). With massive distribution, excellence can be achieved in Fast, Easy and Cheap. However, like a double-edged sword, if not careful then eWoM can be a negative e-WoM. It can damage the reputation of a company that was built for years in a matter of hours (Hewett et al., 2016), as happened in #deleteUBER, United Airlines and the most recent example in Indonesia is #UninstallBukaLapak (Herhausen et al., 2019;Prasetia, 2019).

C. The Marketing Campaign (Content Feeding)
Electronic Word-of-Mouth will not succeed if the content feeding (or feed) is not powerful. Electronic WoM will produce a communication called viral marketing. An effective feed can be in the form of a reward as a consumer effort to help companies in promoting something (Orsingher & Wirtz, 2018), but the best bait can generate consumer participation. Viral is a condition in which a message that spreads reciprocally is like an epidemic driven by

D. The Diffusion Process (Content Seeding)
consumers (Hinz et al., 2011b). When positive messages spread massively, it is very helpful for companies, especially in conveying messages and cost-effectiveness (Pescher et al., 2014). Firm (company) must be able to design creative feeding that can move the audience, such as: making video content, photos, or benefits offered to consumers. An example of feeding on a successful WoM activity is the advertisement of the Tipp Ex product "Hunter shoots a bear" which reached 10 million clicks in four weeks (Hinz et al., 2011b).
Scholars have paid increasing attention to firms' emerging efforts to create programs that generate amplified WOM through tools such as seeding programs, influencer marketing, and referral reward programs. Seeding campaigns may be better considered as substitutes for advertising, rather than complements, at least for certain consumers. (Dost et al., 2018). The development of social media such as Facebook and Twitter has increased interest in WoM and Viral Marketing to optimize the business through online channels (Berger & Milkman, 2012;Yu-Jia, 2012). Richer content allows messages to be made in various variations. Firm (company) must be able to design content that can make the message spread as much as possible, this can be an advantage especially in terms of cost efficiency (Pescher et al., 2014).
In general, viral marketing campaigns are initiated by companies by providing certain stimuli and actively disseminated by consumers selected or not. There are two important things, namely: the effect of viral marketing on consumers and research related to what factors drive participation in viral marketing campaigns (Pescher et al., 2014). Channel selection is also important, Hinz et. al., (2011a), states that an effective seeding strategy can be done via E-mail, e-mail over the web, interface; direct messages on social network sites (Yang, 2018).
Engaging with social media daily makes people more involve with decision-making, such as Generation Y consumers. They should be more likely to share their opinions, and engage in eWOM, whether positive or negative. Consequently, if they experience good service, they engage typically in positive eWOM within their online communities via either social media or other methods of communication. On the other hand, if they are experiencing service failure, they will disseminate negative eWOM to their peers or families to express their complaints or emotional reactions, such as anger, unhappiness, and dissatisfaction (Zhang et al., 2017). Generation Y consumers are likely to disseminate positive eWOM about good service, negative eWOM about service failure and positive eWOM following satisfactory service recovery because they can engage in eWOM with ease using multiple, technology-supported media sources (Zhang et al., 2017).

III. RESEARCH METHODOLOGY
Research design is a blueprint for fulfilling research objectives. Our research objective is to examine the moderating role of seeding activity in the relation between feeding activity and eWoM. We use a marketing campaign in online marketplace provider as a contextual element to examine the relationship (for details see table  132 1). There are three issues we concern about in this research design, namely: (1) The sampling design, (2) The data collection design, and (3) The measurement instrument(s) (Schindler, 2019). In sampling design, we focused to identify the sources of any information we want to examine. First, for who or what should be measured? and how do we access or recruit that source? In this research, our subject is any internet user that ever interacts with the marketing program (such as they ever know or are involved with the campaign). Data collection design involves many decisions related to how, when, how often, and where data will be collected. In this research, we use crosssectional as a time dimension, which means data only collect in one period of time. A measurement instrument is a tool for collecting data on a study's variables. As a research tool, we use survey questionnaires to collect responses with various types of questions (categorical and interval data type questions). Note: 1 Marketing campaign from Shopee named "Goyang shopee" or "Shopee's shakes". This marketing campaign asks the audience to shake their smartphone to participate in their interactive games. Their member can get a lot of kind of rewards. 2 Marketing campaign from Tokopedia named "Tokopoints". This marketing campaign offers its members to collects point rewards and redeems them. 3 Marketing campaign from Bukalapak named "Nego Cincai" or free translation as "Easy bid". This marketing campaign provides a "big opportunity" from participating in their marketing campaign.

A. Sampling design, sampling method and data collection
In this subprocess of research design, we will attempt to answers the question: From whom or what (target population) does the data need to be collected, and how and from how many (cases)? The population is the whole object that has the same characteristics studied, while the sample is part of the population taken for generalization. We define our population as every internet user that ever interacts with the marketing program (such as they ever know or are involved with the campaign). Sample statistics are descriptors of those same relevant variables computed from sample data. In this study, our sample (or respondents) are those people who had been exposed to marketing activity through electronic media. There are three online marketplaces are being studied, namely: Shopee, Tokopedia, and Bukalapak. The selection of the online marketplace is due to the availability of various advertisements and creative marketing campaigns aimed at becoming e-WoM (viral). More specifically, there are three online marketplace provider as an object in this research, each of them have unique marketing campaign (see table 1). For the first object, all versions of "Goyang Shopee" from Shopee, the second object is the "Tokopoints program from Tokopedia", and the last is the "Cinta Nego from Bukalapak (for all versions)".
Before data collection process, we have decided to determine the proper sampling method. There are two kind of sampling method named probability and non-probability sampling method, they are different in term of statistic probability procedures. Probability procedures is based on the concept of random selection (a controlled procedure) that assures that each case is given a known nonzero chance of selection. These procedures usually conduct when we exactly know the number of populations. In this research the exact number of populations is unknown, giving this circumstance, we decide using nonprobability sampling with purposive sampling using judgement sampling. Judgment sampling occurs when a researcher selects sample members to conform to some 133 criterion. Our basic criterion is user at least one of online marketplace or having an experience with it (shopee, tokopedia and bukalapak).
To define the number of sample, in case of sampling method with criterion (judgement sampling), we refers to Hair (2014) to determine a minimum sample to conduct quantitative research. According to minimum sample size, at least the number of participant is 100 respondents or 150 participant to models with seven constructs or less, modest communalities (0.5), and no under identified constructs (Hair et al., 2014, p. 574). A total of 200 participants was participated in this study conducted from June to August 2019, and each participant has direct experience in all three marketing campaigns. Before the participant fill the questionnaire, we asked them to watch again the marketing campaign to recall the memory, and please the participant to interact with each marketing campaign. Sampling distribution in this research test with Bayesian Estimation, all of them shows a normal distribution (see exhibit 1).

B. Variables
Before conducting the data analysis, we need to clarify the research variables, the definitions, indicators, and measurements. A variable is a measurable symbol of an event, act, characteristic, trait, or attribute. To meet our research objective, one or more variables are used as a substitute for a concept or construct. Three various type of variable employed in this research, named marketing campaign or feeding activity as an independent variable (we symbolize with "X"), information diffusion or seeding activity as a mediating variable (we symbolize with "M"), and electronic word of mouth (eWoM) as a dependent variable (we symbolize with "Y"). Variables definition and indicators present in detail at table 2. To collect data we use questionnaire as a research instruments. The measurement scale used is interval with a Likert type scale. Scale type range between 1 to 6 points of scale, ranging from 1 (one) for strongly disagree, and 6 (six) for strongly agree. Following the research design where the sample was taken is online marketplace users, such as Shopee, Bukalapak, and Tokopedia, the enumerator will screen respondents who meet these qualifications. Our sample is users of online marketplace, namely Shopee, Tokopedia and Bukalapak. Before starting to fill the questionnaire, we refreshed the participants' memories by playing a video contains the marketing campaign for each online marketplace.

134
To examine the moderating role of seeding activity on the relationship between marketing campaign (feeding activity) and eWoM, we adopted Baron and Kenny (1986) mediating-moderating variable concept of measurement. Traditionally, the X  Y relationship is tested before mediation to determine whether there is an effect to mediate, and it is also tested after introducing a potential mediator to determine whether that mediator fully or partially accounts for the effect (Rucker et al., 2011). After the first step, followed with to test the effect of Me variable as an intervening the relationship between X and Y. If there is no relationship between them after Me variable intervene the relationship, it means a full mediating relationship existed. But, if the X and Y relationship still significant after the intervening variable joined, and the relationship between X  Me (see equation 3) and Me  Y (see equation 3) is also significant, it means partial mediating were exist. To describe the moderating variable test as a present in equation 1-3 ( Baron & Kenny, 1986).

C. Data Analysis 1) Profiles and Behavior
Profiles and behaviors reflect the characteristics of the object under study. After recapitulation of data based on the results of data collection, then it can be seen how the results of the profile and behavior of respondents. Based on the results, the first thing discussed was the respondent profile. The profile of respondents is a description of the characteristics of respondents in general associated with research. In this respondent profile, there are several variables, namely: Gender and type of social media users. The results showed that the majority of respondents were female (75%) and used Instagram-based social media (69%). Respondent behavior is a depiction of how respondents behaved in the past, in this case, related to their activities in the online marketplace, such as knowledge of marketing activities of each online marketplace. There are several questions related to this behavior, namely: the level of awareness, following the latest info, and referring to other people (referral activity). The results of data processing will analyze the behavior of respondents separately based on each online marketplace. The recapitulation of the respondent's behavior is reflected in figures 1-3.
The first description starts with figure 1, namely the awareness level. Based on figure 1, it appears that Shopee's marketing campaign, "Goyang Shopee from Shopee" has the highest level of awareness, which is 92%. This means that of the total respondents, 92% knew this program. It can be said that this marketing campaign is the most popular among others. Next is Tokopoints from Tokopedia which only has an awareness level of 17.4% and Cinta Nego from Bukalapak at 6.3%. While the remaining 5.3% of respondents did not know all three. The result shows "there are 5.3% respondents who did not know all three-marketing campaign", we guess it is happen because of a lack of awareness to the marketing campaigns. This small segment likely not aware of the marketing campaigns that appear every time they access one of these online marketplaces. As we mentioned before, we are doing an awareness procedure previously, this procedure is also intended so that each participant has direct experience of the marketing campaign. After all, we still consider to include this segment because of they have an experience using the online marketplace.
The level of awareness (awareness level) will generally be followed by a desire to find out more, then the description of figure 2 is related to this. Based on figure 2, the Shake Shopee Campaign remains at the highest interest level compared to other online marketplaces. This campaign has a percentage of 66.8%, which means that several respondents have their interests in the form of the following information from this campaign. For the Tokopoints campaign from Tokopedia it has a percentage of 10%, and Cinta Nego from Bukalapak at 2.1%. But what stands out is as much as 27.4% of respondents are not at all interested in all of these marketing campaigns.
After having an interest, usually, people will convey interest to other people or commonly referred to as referral activity. Based on figure 3, in the Goyang Shopee ad campaign, there were 56.8% of respondents were willing to conduct referral activity, while for Tokopoints in Tokopedia there were 5.3% respondents, while for the Cinta Nego ad campaign from Bukalapak only 0.5% were willing. The interesting thing is that there are 39.5% who are not interested in spreading one of the three ad campaigns.

2) Reliability and Validity Test
To ensure the main variable showing their performance, we conduct reliability and validity test. Reliability is the degree to which an assessment tool produces stable and consistent results, while validity refers to how well a test measures what it is purported to measure. To determine the instrument is reliable, it's value must more than 136 0.6, and the value of Corrected Item Total Correlation (CITC) must more than r table (0.1388). We run data using SPSS to get reliability and validity output. There are 11 items we use in this research to explain the research question. According to table 3, all the α value is > 0.6 it means all of the items are reliable or all of the instrument produces a stable and consistent result. Whereas the result of the validity test shows all of the items have more than r table (0.1388), it means the instrument items measure what it is purported to measure. Note: 200 participants participated in this study, each participant has direct experience in all of three marketing campaign.

3) Analysis of Main Variables
The main variables in this study are marketing campaign or feeding activity (X), information diffusion or seeding (Me), and electronic word of mouth or eWoM (Y). In each object there is one model that is tested After being processed and inputted on a statistical program, then the data is processed with SEM modeling. The results of data processing are listed in table 4. Structured equation modeling (SEM) was performed to test each model to examine the mediation effect using regression analysis. Table 4 shows the result of regression analysis for each research object: Shopee, Tokopedia, and Bukalapak. Each object evaluates with three models of mediating variable test as mentioned in equations 1, 2, and 3. Full mediation occurs when the existence of mediating variable makes the relationship between X and Y becomes insignificant. On another side, partial mediation occurs when the relationship between X and Y significant, even in the existence of a significant mediating relationship occurs. Table 4 confirm the existence of mediation variable due to all coefficients are significant. Table 5 contains some important information about direct and indirect effect to testing the type of mediation role. Because all of model was significant according to table 4, it means there is no full mediation exist. Another type mediation, occurs if the direct effect between X and Y significant, then the effect of mediating variable to dependent variable and independent variable to mediating variable is significant. To test the partial mediation, we compare the indirect effect to direct effect. Table 6 present a resume of mediation test of each model. According to table 6, all type of mediation is partial mediation. This is mean, the mediation variable still exists even the relation between feeding activity is significant to eWOM. Partial mediation indicates the existence of another potential mediating variable(s).  Note: Indirect effect is interaction between coefficient in path X Me and Me  Y This research examines the mediation role of seeding (information dissemination) in the relation between marketing campaign (feeding activity) and eWoM. In this study, we found that different object of an online marketplace (Shopee, Tokopedia, and Bukalapak) has the same mediation effect. This indicates that respondents tend to find it difficult to distinguish between each type of feeding activity (marketing campaign), even for the popular ones. To deepen the discussion, we present a set of mediation tests proposed by Baron and Kenny (1986) in detail, and the result of partial mediation is discussed in the following section.

137
Three sets of mediation models presented in table 4 show indicate a significant level in every path. First we discuss the set of mediating models on Shopee. The first model between feeding and seeding shows the probability of getting a critical ratio as large as 7.288 in absolute value is less than 0.001. In other words, the regression weight for feeding in the prediction of Seeding is significantly different from zero at the 0.001 level (two-tailed). Then, second model show the probability of getting a critical ratio as large as 4.392 in absolute value is less than 0.001. the regression weight for Feeding in the prediction of eWOM is significantly different from zero at the 0.001 level (two-tailed). And the third model between seeding and eWOM show the probability of getting a critical ratio as large as 7.705 in absolute value is less than 0.001. The regression weight for Seeding in the prediction of eWOM is significantly different from zero at the 0.001 level (two-tailed).
Second, we discuss the set of mediating models on Tokopedia. The first model between feeding and seeding the probability of getting a critical ratio as large as 11.603 in absolute value is less than 0.001. The regression weight for Feeding in the prediction of Seeding is significantly different from zero at the 0.001 level (two-tailed). Then, second model between feeding and eWoM the probability of getting a critical ratio as large as 9.19 in absolute value is less than 0.001. The regression weight for Feeding in the prediction of eWOM is significantly different from zero at the 0.001 level (two-tailed). And the third model between seeding and eWoM the probability of getting a critical ratio as large as 7.644 in absolute value is less than 0.001. The regression weight for Seeding in the prediction of eWOM is significantly different from zero at the 0.001 level (two-tailed).
Last, we discuss the set of mediating models on Bukalapak. The first model between feeding and seeding the probability of getting a critical ratio as large as 12.659 in absolute value is less than 0.001. In other words, the regression weight for Feeding in the prediction of Seeding is significantly different from zero at the 0.001 level (two-tailed). Then, in the second model between feeding and eWoM, the probability of getting a critical ratio as large as 4.409 in absolute value is less than 0.001. The regression weight for Feeding in the prediction of eWOM is significantly different from zero at the 0.001 level (two-tailed). And the third model between seeding and eWoM The probability of getting a critical ratio as large as 8.741 in absolute value is less than 0.001. The regression weight for Seeding in the prediction of eWOM is significantly different from zero at the 0.001 level (two-tailed).
All of the mediation models show a partial mediation is exists. The significant value for all equations is below 0.05, which means all regression equations are significant (see table 4). This indicates that eWOM requires seeding steps after the feeding step is given. Refers to the context (i.e., marketing campaigns conducted by Shopee, Tokopedia, and Bukalapak), then marketing campaign activities require a stage of information dissemination so that the interesting content offered becomes eWOM. But how important is the role of seeding (information dissemination) as a mediating variable.
The distinction between full and partial mediation has been influential in theory testing and the development of social psychological knowledge (Rucker et al., 2011). In a simple mediation model with one mediator, full mediation suggests that a researcher has completely explained the process by which X influences Y and there is no need to test for further indirect effects. In the case of partial mediation, there is a clear implication that other indirect effects could (and probably should) be examined and tested empirically.
We try to compare the direct effect between the three models. According to table 5, the highest coefficient of direct effect was found at Tokopedia with 0.546. This direct coefficient followed by Shopee with 0.416, and the lowest is Bukalapak with 0.310. But, different results come from comparing the indirect effect of all models. Mediating models of Bukalapak precisely have the highest coefficient with 0.478, followed by Shoppe (0.440) and Tokopedia (0.302). It happens because of the interaction between coefficients in equations 1, 2, and 3. Interaction determines the magnitude of the indirect effect coefficient.

V. CONCLUSION
This study proposed an examination of mediating role between marketing campaigns (feeding activity) and eWOM. Using a sample of 200 internet users who had direct experience in an online marketplace, this study provides insight into the moderating effect of eWoM. From the outcomes of this study, it is obvious that the study has provided several important and useful contributions to the consumer behavior literature especially from the context of online marketplace users. The study has successfully captured the mediating effect and contributed to the subsequent practical and theoretical implications. The result shows there is a direct and indirect effect in a model (as we see in table 4), and we conclude that partial mediation has become an important issue in this research.
Another finding from our study is that opposite trend in descriptive of user behavior and "none of them" criterion. We ask three questions about their response to the marketing campaign, using yes or no questions. The first question is to measure awareness level, while the second question is to measure the interest in the marketing campaign, and the last question is to measure the referral activity. All questions are related to one and another and have hierarchical order. Awareness level makes people realize that marketing campaign exists, following the latest information (Interest) indicate that the user involves in the marketing campaign, and referral activity indicates the user shares their thoughts without temporal or spatial constraints.
Remind the hierarchical order of question, we found an opposite direction from awareness to referral activity. At the question of awareness level, Shoppe succeeds gain 92% user awareness, followed by Tokopedia with 17.4%, then Bukalapak 6.3%, and None of them 5.3%. At the question about interest level, Shopee succeeds to gain interest 66.8% user, Tokopedia 10%, then Bukalapak 2.1%, and None of them 27.4%. And the last question related to referral activity, are they willing to share the information with others? 56.8% users willing to share information about Shoppe, 5.3% user willing to share information about Tokopedia, then 0.5% user willing to share about Bukalapak, and 39.5% of users willing to share none of them. We found a downward trend in awareness level, then interest stages, and referral activity. Conversely, the "none of them" criterion shows an upward trend. It is a kind of new threat to a marketing campaign in the real world that we cannot avoid.
These results answer that the activity of feeding content does not always encourage people to doing eWOM. The company must evaluate its engagement program, increase sharing participation of consumers, lead them to choose a channel and encourage them to a willingness to share. The availability of communication tools that can disseminate information quickly, precisely and cheaply, made the company start to consider conducting ad campaigns that can produce eWOM. Feeding requires the seeding stage to produce eWOM. Based on the previous discussion, seeding becomes an important step that is passed after the feeding stage. The success of this stage will produce eWOM, where people will be willing to disseminate the information.

VI. LIMITATION AND FUTURE RESEARCH DIRECTION
We realize there is an opportunity to improvement for the next research. First, our research limited to people who has an experience to use certain online marketplace. In the future comparison between group contains of active user must conduct. This research only focus on within subject, each respondent given their opinion for all of online marketplace. Between subject provides oppportunity to explore the behavior variation making comparison between group. Second, our research variable very focused on the process of feeding in marketing campaign, and information dissemination, future research can start to improving a motivational aspect as a predictor of eWoM. To making the marketing campaing goes viral naturally, a message must share massively from a lot of user with voluntarily. What drives them to "help" the company's marketing campaign must be examine in the future.
Third, the most frequent issue in this research is the attempt to examine the effect of partial mediation. In a simple mediation model with one mediator, full mediation suggests that a researcher has completely explained the process by which feeding activity influences electronic word of mouth. If it happens, there is no need to test for further indirect effects. In the case of partial mediation, there is a clear implication that other indirect effects could (and probably should) be examined and tested empirically. We see this become an opportunity to explore and examine another mediating variable.
We initiate a few suggestions and future research direction that may help other scholars to replicate our study. Here is some of improving suggestion: (1) Improvement in research variable(s) with the attitudinal construct (cognitive, affective, or conative construct) is very important to consider (Rageh Ismail & Spinelli, 2012), (2) using wide time dimension or longitudinal research (Leong et al., 2019), (3) using control variable to examine the 140 effect of a different group of the respondent (Zhang et al., 2017), and (4) examine the antecedent variable that has potentially affected to eWoM (Srivastava & Sharma, 2017).

DECLARATION OF CONFLICTING INTEREST
Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Model
Model 1