Next Article in Journal
Multi-Objective Optimization for Order Assignment in Food Delivery Industry with Human Factor Considerations
Previous Article in Journal
Neighbourhood Environment and Cognitive Vulnerability—A Survey Investigation of Variations Across the Lifespan and Urbanity Levels
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 12
Issue 19
10.3390/su12197952
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Using the Social Influence of Electronic Word-of-Mouth for Predicting Product Sales: The Moderating Effect of Review or Reviewer Helpfulness and Product Type

by
Sangjae Lee
1,* and
Joon Yeon Choeh
2
1
College of Business Administration, Sejong University, Seoul 05006, Korea
2
Department of Software, Sejong University, Seoul 05006, Korea
*
Author to whom correspondence should be addressed.
Sustainability 2020, 12(19), 7952; https://doi.org/10.3390/su12197952
Submission received: 13 August 2020 / Revised: 11 September 2020 / Accepted: 24 September 2020 / Published: 25 September 2020
(This article belongs to the Section Economic and Business Aspects of Sustainability)
Browse Figures
Review Reports Versions Notes

Abstract

:
The social engagement of eWOM (electronic word-of-mouth) can reduce the threat of adverse selection in e-commerce. As studies that examine the social influence of eWOM are rare, the present work suggests the moderating effect of review or reviewer helpfulness and product type (experience or search goods) on the relationship between eWOM and product sales. The volume of eWOM, which is defined as the multiplication of the average length by the number of reviews, is shown to be moderated by review and reviewer helpfulness and search goods to affect product sales. Review ratings are moderated by reviewer helpfulness, and review extremity is positively (negatively) moderated by search (experience) goods and review helpfulness to affect product sales. As previous studies of differentiated sampling strategies that consider review helpfulness for predicting product sales using eWOM are lacking, this study compares the prediction power of business intelligence methods for different subsamples of products created according to high or low review and reviewer helpfulness levels. The subsample with high review or reviewer helpfulness demonstrates greater prediction performance than the subsample with low review or reviewer helpfulness when eWOM variables are used as predictors of product sales. Hence, preliminary filtering data preprocessing should consider review or reviewer helpfulness as a crucial criterion of the data quality. This will contribute to the sampling or preprocessing strategy used to predict product sales using eWOM.

1. Introduction

Consumers’ electronic word-of-mouth (eWOM) is represented as evaluations of products on review websites and is considered to exert an impact on sales [1,2]. From adverse selection and quality uncertainty frameworks [3,4,5], this paper utilizes consumers’ reliance on eWOM as a reasonable behavior to reduce information asymmetry and uncertainty in e-commerce. On the basis of social engagement and social risk frameworks, eWOM variables such as valence (review rating) and volume can be understood as crucial factors to reduce social risk and information asymmetry affecting product sales [6], purchasing intentions or decisions [7,8], product imagery [9] or brand attitudes [10]. Further, the usefulness, or helpfulness, of online reviews has been considered to be important, as helpfulness can affect the relationship between eWOM and product sales [6,11]. Helpful reviews provide review diagnosticity and a source of differentiation, resulting in greater potential value to customers [12]. It is beneficial for sales for a website manager to highlight helpful reviews to maximize their attractiveness to increase the number of visits from new customers [13]. Helpful reviews increase such sites’ business value, and providing helpful reviews tends to attract consumers who seek information and to provide greater potential value to customers. Consequently, eWOM factors such as volume, valence, and helpfulness help to reduce social risk in purchasing through e-commerce by highlighting social engagement and the quality of recommendations for each product [14]. The combination of transparency by representing eWOM factors on an e-commerce site and the social engagement of eWOM can reduce the threat of adverse selection in e-commerce [3].
Thus, this study has several objectives. First, stemming from the social influence of eWOM based on social exchange theory, our study intends to test the moderating effect of review or reviewer helpfulness and product type (i.e., experience vs. search goods) with regard to the effect of eWOM (volume, valence, review extremity) on product sales. Previous studies using the moderating effect of review or reviewer helpfulness and product type for the effect of eWOM are lacking. Recently, researchers have indicated that customers who contribute online reviews are under social influence from others [3,15]. Customers in social exchange relationships are likely to have faith in the online review system, and by facilitating individuals to share their interests, social engagement promotes social exchanges [16].
eWOM communications through online reviews serves as a type of social influence [17]. Social influence represents a process by which customers modify their thoughts or behaviors based on interaction with others who are considered as having similar buying experiences or based on experts’ opinions of a product. Consumers who are especially satisfied or dissatisfied with a brand will post their opinions as acknowledged by other consumers through eWOM interaction, intending to provide benefits from their experience. Other consumers then read eWOM as a means of information gathering to become informed about the product, and such social interaction is associated with having empathy and is considered to be equal to close discussions in the form of these interactions through eWOM. As product referral votes represent reviewers’ judgments, which can be considered to lower the uncertainty of a social exchange (Lee et al., 2015), the volume, valence, and helpfulness of eWOM can be used to support customers’ decisions to purchase, facilitating sales of the product. The willingness of consumers to depend on the brand to provide its intended quality is greatly affected by eWOM, which results in greater sales.
Based on previous studies of online reviews that suggested factors affecting perceived helpfulness [12,18,19] and moderators for eWOM [3,20,21,22], our study attempts to present the moderating effect of review or reviewer helpfulness and product type on eWOM, as these moderating effects are less studied in previous eWOM studies despite the existence of studies on other moderators between eWOM and sales. For instance, Lee et al. [3] showed that Facebook likes exert a contingent effect on product sales via deal and product traits. Consumers examine the average statistics of reviews and usefulness or helpfulness ratings as a measure of quality and preference for a specific product [23]. As review quality, as represented by helpfulness, strongly affects sales and purchasing intentions [14,24], our study intends to show that review or reviewer helpfulness can contribute as strong moderators between eWOM and sales. Instead of introducing additional direct effects, our study focuses, rather, on the moderating effects of helpfulness because helpfulness can have a synergistic effect with volume and valence, which have been much investigated as affecting sales.
Among eWOM variables, the effects of volume and valence on sales have been most commonly investigated [25,26]. Our study attempts to extend this line of eWOM studies of volume and valence by suggesting moderators of review or reviewer helpfulness, which represent review quality and thus affect sales. In order to analyze review or reviewer helpfulness, our study uses the average helpfulness for each product and treats each product as a unit of analysis so that the moderating effects of helpfulness and product type on product sales can be examined. Our study suggests that review volume is better understood as the multiplication of the average number of reviews by the average review depth (average number of reviews × average review depth review), showing that review volume has an interaction effect with review or reviewer helpfulness.
Further, there is a lack of studies on product type moderating the relationship between eWOM and product sales. The impact of an enormous amount of review data on product sales can be elusive if the interaction effects of both the review volume with product type are not considered. Further, products that provide various review depth and extremity levels in ratings have different effects on helpfulness according to the product type, such as experience and search goods [12,13,27]. Experience goods (coded as 1) (toys, DVDs, video games) indicate those that require sampling or a purchase to evaluate product quality, while search goods (coded as 0) (e.g., computers, books, PC accessories) represent products for which consumers can obtain information about product quality before purchasing [12]. In order to extend and utilize these studies of the interaction of review depth and extremity with product type on helpfulness, our study also suggests that the product type for experience and search goods exerts a moderating effect on the volume of reviews (which replaces depth in previous studies that use individual reviews as an unit of analysis) and the extremity and level of review or reviewer helpfulness for product sales.
Second, as business intelligence (BI) applications of eWOM using a differentiated sampling strategy (subsamples with high and low helpfulness) that consider the helpfulness of the sample are lacking, the study intends to propose a prediction method for product sales that uses neural networks, the k-nearest neighbor method, and the decision tree method with a sample containing high and low review or reviewer levels. Also lacking are applications of these methods to the prediction of sales in an effort to study the impact of review or reviewer helpfulness on sales prediction performance using a differentiated sampling strategy. Our study attempts to overcome this gap in the literature by contributing a sampling or preprocessing strategy for predictions of product sales using eWOM.
This study uses these BI methods to improve our understanding of specific predictors that promote product sales. Business analytics techniques are utilized in various areas of review analyses. Diverse BI methods have been utilized, such as the random forest method [1], support vector regression [28], and the decision tree method [11]. To extract helpful eWOM information to predict sales, the application of a rigorous data mining method is necessary [29,30]. This is the goal of the present study.
These three methods are selected as they are either parametric (neural networks) or non-parametric (k-nearest neighbor and decision tree methods). Neural networks are effective for estimating complex nonlinear relationships in eWOM, and the k-nearest neighbor and decision tree methods do not assume a prior probability distribution in the problem domain. Our study develops a predictive model using these BI methods that offers better prediction performance with the use of eWOM variables. The prediction performance of BI methods is contrasted between high and low review or reviewer helpfulness samples. Our results are likely to help practitioners of online businesses to identify the relationship between eWOM and product sales. The second objective of this paper is an extension of the first, as the sampling strategy according to helpfulness should be based on an understanding of the moderating role of helpfulness. Thus, after determining the moderating role of review and reviewer helpfulness, we can then ascertain the rationale for dividing the entire sample into high and low helpfulness subsamples to improve the prediction performance and thus enhance product sales.

2. Social Influence of eWOM

The relationship between eWOM and social exchange theory can be understood if factors of the social influence of eWOM, such as volume, valence, and review quality (or helpfulness), on sales are considered. Based on social exchange theory, social engagement through online reviews improves social exchange [3,16]. In relation to this, eWOM can be utilized as a source of social influence based on social exchange theory and ultimately as a socially created reputation signal for predicting product sales. Consumers present their thoughts through eWOM, as they are greatly encouraged by their concern for others and by the potential to increase their own self-worth through the posting of their opinions [31]. Because customers alternately serve as information providers and information seekers, social interaction tends to be sustained over time through ongoing posting threads. Consumers who are satisfied with a particular brand will post positive comments and let others know that they could benefit from their shopping experience, which supports the contention that positive eWOM results in favorable attitudes toward the brand [16]. eWOM created for a brand through reviews of products can be considered ads highlighting the brand’s reputation and can consequently be associated with product sales.
eWOM compiled globally in an online market can provide a crucial source of social capital capable of explaining customers’ purchasing behaviors and product sales. The statistics of eWOM that may contribute to social exchange include valence, volume, and review extremity [32,33]. For instance, reviews’ ratings exert a positive effect on book sales on online sites [34]. The total number of reviews affects sales [26,35,36]. eWOM offers a crucial market signal of reputation, and ratings are regarded widely to deliver positive or negative reputations, as eWOM factors such as volume, valence, and helpfulness represent social engagement and the quality of recommendations, and help to levy inherent social risks while ultimately positively affecting the sales of products in e-commerce, which is associated with information asymmetry between buyers and sellers.
Additional variables for eWOM and how it affects product sales are also variables that affect review helpfulness, such as review extremity, under specific conditions. For example, the average depth (and average volume, defined as the product of the average review depth and the average number of reviews in this study) of eWOM for search goods can affect product sales, as the factual nature of search reviews shows the incremental content value of those reviews regarding how the product is utilized and how it compares to alternative goods [12]. Thus, as either average review depth or average number of reviews for each product increases for search goods, product sales can increase because the review quality based on helpfulness improves [14], and this greatly influences sales and purchasing intentions [24]. Further, extreme reviews for search goods are more influential on review helpfulness than they are for experience goods. Consumers can discount extreme ratings, as they appear to show simply a different taste, while they are more understanding about moderate ratings for experience goods, as they appear to represent a more objective evaluation [12]. Thus, for search goods, more extreme reviews can have an effect on product sales because the review quality strongly influences sales and purchasing intentions.

3. Research Model

Based on social exchange theory, social engagement through online reviews promotes social exchange [16]. eWOM communications through online customer reviews provide different types of media for social influence [17]. eWOM provides signals of social proof, by which customers depend on the collaboratively shared experiences of others to determine their purchasing decisions [37]. Social influence represents the process by which consumers change their feelings or behaviors based on interactions with other consumers. WOM is a crucial market signal of product reputation, and ratings can be utilized greatly to deliver either positive or negative reputations. Thus, as structured eWOM represents social capital and supports the delivery of product knowledge, eWOM can explain changes in sales by affecting customers [38].
Using previous studies of online reviews regarding the factors influencing perceived helpfulness [12,18,19,39], and moderators of eWOM [3,20,21,22], our study attempts to discern the moderating effects of review or reviewer helpfulness and product type on the relation between eWOM and product sales, an area not thoroughly examined in previous eWOM literature. Previous studies examined other moderators of the effects of eWOM on sales or purchase behaviors. For example, Zhu and Zhang [22] showed that game popularity and player experience exert moderating effects on the role of eWOM regarding the decision to purchase a game. Wu et al. [20] suggested that consumer’ risk attitudes moderate the effects of online user reviews on consumers’ willingness to pay. Hu et al. [21] indicated that review ratings play an indirect role in sales via sentiments. In addition, Lee et al. [3] indicated that Facebook likes have a contingent effect on product sales through product and deal characteristics.
As the effects of volume and valence on sales have been much investigated [25,26], this study intends to extend this line of eWOM studies on volume and valence by indicating moderators of review or reviewer helpfulness, which are assumed to show review quality and thus affect sales. In addition, products having various review depth and extremity levels in ratings have different effects on helpfulness according to the product type, such as experience and search goods [12,13]. Based on these studies of the effects of the review depth, extremity, and product type on helpfulness, our study additionally suggests that the product type of experience and search goods exerts a moderating effect on review volume (average number of reviews × average review depth review), which was substituted for depth in previous studies to use individual products as the unit of analysis, and on the extremity of reviews or reviewer helpfulness for product sales.
The research model (Figure 1) describes how eWOM and the moderating effects of review or reviewer helpfulness and product type affect product sales, as these factors serve as market signals that offer information about the quality of a product and other unseen aspects beyond the product description. Consumer-posted information provides usefulness in that it facilitates purchasing behavior by offering indirect product usage experiences [24]. Thus, reviews perceived as more helpful exert a greater influence on the likelihood of purchase behaviors [40]. Our model specifically includes review or reviewer helpfulness, posted to exert a moderating effect on eWOM factors such as valence and the average number of reviews. Further, we suggest that the product type, i.e., experience or search goods, can influence the relationship between eWOM and helpfulness to affect product sales, as the usefulness of eWOM is different for experience and search goods [12]. In order to describe the actual “volume” of eWOM more effectively, this study uses the multiplication of the average number of words in a review by number of reviews as the “volume” of eWOM.

3.1. Moderating Effect of Helpfulness for eWOM

Perceived product value or quality is affected by eWOM, which has normative as well as informational effects on consumers’ conformity and beliefs [41] and purchasing behavior or intention [42,43]. Specifically, the volume of reviews positively affects sales of experience products, while page views positively influence search products’ sales [44]. eWOM volume levels are positively related to higher box-office performance outcomes [26,36]. The volume of eWOM represents the distribution of eWOM, promoting consumer awareness or consumer evaluations of products and expediting their consequent purchase decisions [36]. eWOM’s awareness effect increases eWOM dissemination to help create buzz, which in turn leads to even greater product sales.
Review valence has been posited to affect sales. Review ratings are positively associated with the sales ranks of books [1]. Exposure to positive or negative reviews, showing the average product rating, influences consumers’ purchase intentions [45]. Seller reputation tends to facilitate consumer interest in bidding and increases the prices of products [46]. Regarding box-office revenue, review ratings are a more crucial determinant than the volume of eWOM [25]. In a similar vein with regard to eWOM volume, the effect of eWOM on product evaluation results in a positive feedback mechanism in that increases in eWOM valence could lead to higher sales. Thus, product sales will be higher if many reviews and high review ratings exist.
While the average review volume and ratings are related to sales, this relationship becomes greater if the helpfulness of reviews is greater because helpfulness can help to strategically attract the attention of consumers [47]. Because revenue significantly increases via much helpful reviews, online review sites have begun to establish a voting mechanism to attract helpful reviews [48]. Review helpfulness combined with high review volume and valence levels can be considered as more enabling of the promotion of purchasing behavior [13]. The intention of consumers to online purchase is greatly affected by perceived review usefulness [49] as well as by the review volume and valence levels. The usefulness of eWOM information has a significant influence on customer trust [49]. Consumers will have greater trust in review contents if reviews are valuable such that they satisfy their information needs [50]. We suggest that the effect of the review volume and ratings on sales can be increased if reviews are helpful because review helpfulness represents the quality of these reviews.
Hypothesis 1 (H1).
Review helpfulness has a positive moderating effect on the relationship between the volume of reviews (average number of reviews × the average review depth) and product sales.
Hypothesis 2 (H2).
Review helpfulness has a positive moderating effect on the relationship between the average review rating and product sales.

3.2. Moderating Effect of Product Type

The average length of a review or the review depth is strongly associated with the helpfulness of the review, which is more true for search goods [3] as the number of arguments in a message positively represents diagnosticity and more details about where and how the product is usable [51]. As quality information exerts a positive influence on purchase behavior [52], review helpfulness exerts a great moderating effect on the effect of the average review depth on box-office revenue [39]. Our study intends to test whether the product of the average number of reviews and the review depth can be moderated by helpfulness for sales, which is more facilitated in the case of search goods.
Given a large volume of reviews, if such reviews are helpful, they can have a higher influence on the purchasing behavior of potential consumers as compared to less helpful reviews [13]. Customers who read high-quality reviews are likely to evaluate search products more positively, resulting in a greater attitude toward purchasing [53]. Because the review depth is more related to review helpfulness in the case of search goods [3], product sales will become greater if a large amount of reviews are available for search goods and at the same time these reviews are helpful. Thus, the average review volume affects product sales more when the helpfulness of reviews becomes higher and the product type is search goods.
Hypothesis 3 (H3).
Review helpfulness has a positive moderating effect on the relationship between the volume of reviews (average number of reviews × the average review depth) and product sales, and in addition, search (experience) goods has a positive (negative) moderating effect on the same relationship.
Either a very low or high rating is indicative of an extreme review (positive or negative, respectively). Review extremity is defined in terms of the deviation from the average review rating. Product types such as experience (coded as 1) and search goods (coded as 0) have been considered as a moderator of the impact of review extremity on helpfulness [3]. Establishing a balanced review with both positive and negative comments is positively related to the helpfulness for experience goods [27]. For experience goods, consumers are not convinced when reading about the tastes and subjective evaluations of others, though they are often greatly assuring of their own extreme views. Consumers may underestimate extreme ratings if they represent a simple difference in taste. A consumer who has an initially positive evaluation of a product can agree with an extremely positive review but is not likely to perceive that an extreme review will help the purchase decision process.
Extreme arguments when reviewing search goods, however, can be considered as reliable because it is rather easy to verify objective assertions about tangible aspects. Consumers consider extreme ratings of search goods as useful, as they may indicate a more persuasive and convincing assessment. Thus, extreme comments for search goods can provide more of a clue of logical thinking than extreme arguments in reviews of experience goods. In contrast, Chen (2016) posits that for experience (search) goods, customers perceive one-sided reviews as more (less) helpful than two-sided reviews that are posted by expert reviewers. Consequently, it is necessary to test whether review extremity is moderated by product type to affect review helpfulness, and here we test whether review extremity positively (negatively) interacts with search (experience) goods to have an effect on product sales, as customers are believed to expect more balanced opinions for reviews of experience goods, which are mostly subjective in nature, compared to reviews for search goods.
Review helpfulness represents review quality and the convincing trait of the argument [54], having an impact on customers’ attitudes and purchase decisions [14,42]. Review contents delivering facts rather than engaging in story-telling promote customers’ attitudes along with their purchase decisions [55]. Review helpfulness means that consumers agree with the quality of the reviews such that it influences their purchasing decision. Thus, as the average review extremity interacts with the product type to affect helpfulness, this influences product sales more if review helpfulness is in fact greater as a result of a match between review extremity and product type.
Hypothesis 4 (H4).
Review helpfulness has a positive moderating effect on the relationship between average review extremity and product sales, and in addition, search (experience) goods has a positive (negative) moderating effect on the same relationship.

3.3. Moderating Effect of Reviewer Helpfulness for eWOM

Reviewer credibility and qualification comprise issues of review quality [56], and reviews providing high ratings can turn out to be more helpful if these are created by highly reputable reviewers receiving numerous helpfulness votes. This study considers reviewer helpfulness separately from review helpfulness as a moderating factor for product sales because reviews created by more helpful reviewers tend to deliver more persuasive and credible messages that affect consumers’ purchasing decisions.
Reviewer helpfulness can roughly represent the user’s social influence because the more followers he/she has, the greater the impact of his/her eWOM through the people he/she can reach. Customers tend to consider the target product according to the quality review, with such reviews likely to be posted by helpful reviewers. Reviews posted by “top reviewers” create stronger trustworthiness of the source than reviews posted by laypeople [57]. Thus, eWOM from a certain reviewer group could provide a strong effect on sales [33].
Top online reviewers can greatly contribute to attracting consumers’ attention, and financial benefits may be possible from their reputation [58]. In order to lower customers’ concerns regarding reviewer credibility of reviews, review websites have begun to offer helpfulness votes of reviewers [57]. As reviewers vary, according to their helpfulness, review quality and quantity levels, and sometimes hide their intent [59,60], the reviewer’s qualification can be represented by the cumulative helpfulness of the reviewer [61]. Past helpfulness records of reviewers are likely to determine future helpfulness ratings [62]. Consequently, reviewer helpfulness will much likely determine average review helpfulness for each product. Thus, similar hypotheses regarding reviewer helpfulness can be suggested, closely related to the previously suggested hypotheses for review helpfulness in this section.
Hypothesis 5 (H5).
Reviewer helpfulness has a positive moderating effect on the relationship between the volume of reviews (average number of reviews × the average review depth) and product sales.
Hypothesis 6 (H6).
Reviewer helpfulness has a positive moderating effect on the relationship between the average review rating and product sales.
Hypothesis 7 (H7).
Reviewer helpfulness has a positive moderating effect on the relationship between the volume of reviews (average number of reviews × the average review depth) and product sales, and in addition, search (experience) goods has a positive (negative) moderating effect on the same relationship.
Hypothesis 8 (H8).
Reviewer helpfulness has a positive moderating effect on the relationship between average review extremity and product sales, and in addition, search (experience) goods has a positive (negative) moderating effect on the same relationship.
Figure 1 presents the procedures of the data analysis in this study, and Figure 2 presents the research model. We classified subsamples of products according to review and reviewer helpfulness levels and applied business analysis (BI) methods to compare the prediction power for each subsample of products according to review and reviewer helpfulness. A regression analysis was conducted to determine the explanatory power in each subsample of products according to review and reviewer helpfulness. We used neural networks (manual, automatic) and the k-nearest neighbor and decision tree methods to investigate the prediction performance. Our study uses multiple regression analysis with interaction terms to test the effects of eWOM and helpfulness on sales.

4. Data Collection

Our study utilizes crawled data from Amazon for the research purpose here to develop the metrics of eWOM and product-related variables. The final sample includes 1834 products. The eWOM and product related data was created between 2011 and 2016. The summaries of the product- and eWOM-related variables are shown in Table 1. Books, DVDs, music, and videos are the major products in the sample, and experience and search goods account for 32.9% and 67.1% of all products, respectively. Descriptions of the measures for the research variables are presented in Table 2, and the corresponding basic statistics are presented in Table 3. For instance, review helpfulness is defined as the proportion of the number of helpful votes among the total number of votes that evaluated if the review was helpful or not. Reviewer helpfulness is described as the proportion of the number of helpful votes among the total number of votes that were received by the specific reviewer to evaluate if the review by that reviewer was helpful or not. We suggest “the average number of reviews × the average number of words in a review” as the “volume” of eWOM in this study.
As the study objective is to test the effect of a differentiated sampling strategy considering the helpfulness of a sample on the prediction performance, the 1834 products cases used here are divided into two subsamples according to the average of review or reviewer helpfulness (0.733 and 0.755) in order to compare the prediction accuracy levels between the two subsamples. For each subset, the number of records is modified (by randomly excluding surplus records) to be equal to the smaller number between the two subsamples (i.e., 767, 731) by excluding randomly the records in the larger sample. Subsequently, each subsample is divided into the test (15 or 20 records) and the training sample.

5. Prediction Methods

One of the objectives of this study is to compare the prediction performance outcomes between the high review(er) helpfulness and low review(er) helpfulness subsamples, and this study utilizes neural networks with manually or automatically determined architectures along with the k-nearest neighbor with the best k method, and decision trees with pruning. The procedure of the data analysis is presented in Figure 2. This study applies these BI methods because the intent here is to compare both parametric (neural networks) and non-parametric BI methods (k-nearest neighbor method and decision trees) to investigate the effect of a differentiated sampling strategy by comparing the differences in the prediction performance outcomes between two subsamples in terms of review or reviewer helpfulness.
With neural network models, it is not necessary to specify the correct form of the relationship between input and output variables. Multilayer backpropagation networks are the most widely used method in neural networks [63]. In this study, the architecture, i.e., the details of the hidden layer and corresponding number of nodes, are determined manually or automatically from the BI solution. The number of epochs, the error tolerance, the weight decay, the gradient descent step size, and the weight change momentum are 30, 0.01, 0, 0.1, and 0.6, respectively. The logistic function is chosen as the transfer (activation) function for the hidden layer and the output layer.
kNN or case-based reasoning (CBR) relies on the use of knowledge from the case base by which to make predictions for new situations. This method builds a knowledgebase to suggest a reasoning approach while adapting existing knowledge retrieved from the case base, applying the knowledge and updating the predictions for new situations. The k closest product of the training dataset is suggested for each product in the validation dataset. The value of the best k is found to be that which minimizes the prediction error in the validation dataset.
The decision tree method involves the training of a tree that consists of branches, paths (a set of branches), and leaf nodes to predict new cases. A path of branches refers to the conditions of the predictor’s value and the leaves are the states of the dependent variable. Our study applies pruning to the trained tree to minimize prediction errors in the validation set.
We used six eWOM variables (review depth, review volume, average review rating, review extremity, reviewer rank, and average number of elapsed days after posting until parsing) and two product-related variables (price and product type) as predictors for the outcome variable, i.e., ln (sales rank). We used the Analytical Solver Platform [63] for the prediction experiments. The prediction error is presented in terms of the RMSR (root mean squared residual). We selected review and reviewer helpfulness as the criteria for the sampling strategy in the comparison of the prediction performance outcomes, excluding product type because the numbers of search and experiences goods are rather unbalanced. Instead, we focused on the quality aspects of reviews as a criterion of the sampling strategy for prediction performance.

6. Results and Discussions

Table 4 presents the regression analysis results, where ln (product rank) is the dependent variable. Price, the average number of days after review posting until review parsing, and the average reviewer rank are used as control variables in an effort to control for time-varying product factors or product-level heterogeneity. Table 4 shows the two most widely examined variables for direct effects, i.e., the average number of reviews and the average review rating.
The multiple regression analysis with interaction terms in Table 4 is used to test the moderating effects of review or reviewer helpfulness and product type. As a higher value for the sales rank is linked to lower sales, a negative sign of the estimated coefficient for interaction terms indicates a positive impact of eWOM on sales. For instance, the estimated coefficient for the average number of reviews × the average number of words in a review × review helpfulness is −0.402 (p-value < 0.05). Thus, this shows that review helpfulness is positively moderating the relationship between eWOM volume and product sales. Table 4 shows that hypotheses H1 (review volume with helpfulness), H2 (review volume with helpfulness and search goods), H4 (review extremity with helpfulness and search goods), H5 (volume with reviewer helpfulness), and H7 (review rating with reviewer helpfulness) are accepted. This shows that the social influence of eWOM based on social exchange theory can be applied to the effect of eWOM on product sales such that shoppers check the average amount of reviews, ratings from other customers, the helpfulness of reviews and reviewers, and the product type as sources of accurate and unbiased information regarding a particular product before they are involved in the final social exchange transaction process, which is consistent with the social influence of eWOM [16].
The significant coefficient of the average number of reviews rather than the review rating shows that the number of reviews for a product has more of an effect on product sales than the review rating, an outcome consistent with those in previous studies, such as that of Duan et al. [26]. The level of review activity of a product can be used to explain the usefulness of reviews of the product [64], as consumers are more apt to post reviews for products about which many other persons have contributed [65]. Thus, with an increase in the number of reviews, the usefulness of the reviews may increase, having an ultimate effect on product sales.
The volume of eWOM, determined by multiplying the review number by the review depth (length), exerts a significant moderating effect on sales with high review helpfulness and search goods. With reviewer helpfulness, the eWOM volume also has a moderating effect on product sales. This indicates that greater eWOM volumes lead to greater product sales, especially when review or reviewer helpfulness is greater. When there exists a larger amount of eWOM providing higher helpfulness, product sales increase.
These significant moderating effects of eWOM indicate that both the average review number and depth are important indicators of the usefulness of eWOM, thus affecting product sales. The information depth of the content could enhance diagnosticity, increasing review value, as they more likely provide detailed arguments about the features of products [66]. As the word count has been found to be a crucial factor for the review helpfulness of reputable reviewers’ reviews [61], the length of reviews should be moderated by reviewer helpfulness to have an impact on sales.
Review ratings are moderated by reviewer helpfulness and not review helpfulness, having an influence on sales. This shows that review ratings are a more convincing indicator of sales when more average reviewer helpfulness for a product is higher. The average review helpfulness for a product fails to moderate the relationship between the average review rating and product sales, indicating that among the three determinants (volume, rating, extremity) of product sales, review ratings are not always a crucial variable moderated by review helpfulness for product sales. This shows that regarding the effect of review ratings on sales, source credibility as represented by reviewer helpfulness is a more crucial factor than the helpfulness of review contents.
Review extremity is positively (negatively) moderated by search (experience) goods (having a value of 0) and review helpfulness to affect product sales. This shows that for search (experience) goods, more (less) extreme reviews and greater helpfulness of reviews collectively contribute to product sales to a greater extent. This indicates that for the effect of review extremity on sales, source credibility as represented by reviewer helpfulness is a less important factor than the helpfulness of reviews in contents.
The strong moderating effect of reviewer helpfulness with the volumes and ratings of the three eWOM variables show that the average helpfulness of a reviewer greatly affects the helpfulness of his/her reviews and significantly provides the conditions in which eWOM variables have a greater influence on product sales. Reviewers contribute to online platforms in order to maintain their reputations [67], and consumers are likely to search for reputable online reviewers, helping them determine trustful reviews [68]. Review authorship promotes corresponding information persuasiveness [13], which directly influences purchasing intention. The percentage of helpfulness votes among all votes represents the level of reviewer qualification [61], and previous information about reviewers plays a crucial role in product sales through its interaction with eWOM. Our results pertaining to the significant moderating roles of review and reviewer helpfulness and product type extend those of previous studies of online reviews that indicated moderators for eWOM [3,20,21,22] by showing the significant moderating roles of moderators that were relatively less studied in these works.
This study examined the effect of the differentiated sampling strategy by considering the helpfulness of the sample for predicting product sales using eWOM. We investigate the prediction performance of neural networks (manual, automatic) and the k-nearest neighbor and decision tree methods, where 767 movies in subsamples of high and low review helpfulness are divided into a training and a validation set. There are 20 or 30 records in the validation set, with the remaining records comprising the training set. Thus, 32 pairs constituting the training and validation sets are prepared. The prediction errors in the validation set in terms of the root mean square residual are compared using a t-test between the high and low subsamples across the 32 validation sets. For 731 movies in the high and low reviewer helpfulness subsample, the 31 pairs making up the training and validation set are similarly composed. The prediction errors are similarly tested by means of a t-test between the high and low reviewer helpfulness subsamples.
Table 5 indicates that the high review or reviewer subsample shows better prediction performance when neural networks (manual and automatic) are applied to predict box-office revenue. When either the k-nearest neighbor or the decision tree method is used, there is no significant difference in the prediction error. With a nonlinear parametric estimation method, such as a neural network, high review or reviewer helpfulness is a better condition for predicting box-office revenue using eWOM variables, whereas for non-parametric estimation methods, such as the k-nearest neighbor or decision tree methods, the moderating influence of review or reviewer helpfulness is not significant. The moderating role of review or reviewer helpfulness with regard to predictions of box-office revenue becomes clear when a nonlinear parametric estimation method is used, indicating that the relationship between eWOM variables and box-office revenue can be better understood when using a nonlinear estimation method. For predictions using neural networks, eWOM with more helpfulness shows more prediction power for product sales, and preliminary filtering as a data preprocessing means can utilize review or reviewer helpfulness as a crucial criterion of data quality.

7. Conclusions and Implications

Our study presents the moderating role of review or reviewer helpfulness in predicting product sales using eWOM. The results indicate that review or reviewer helpfulness moderates the effect of volume or valence on product sales. Review extremity is positively (negatively) moderated by the search (experience) goods and review helpfulness. When using nonlinear parametric estimation methods, such as neural networks with the architecture determined in a manual or automated approach, the high review or reviewer helpfulness subsample provides greater prediction performance than the low review or reviewer helpfulness subsample in predicting product sales, supporting the moderating role of review or reviewer helpfulness in predicting product sales using eWOM.

7.1. Implications For Researchers

Our study has several implications for researchers. First, stemming from the social influence of eWOM based on social exchange theory, our study focuses on the moderating effect of review or reviewer helpfulness with eWOM to exert an effect on sales. While numerous studies have examined the role of WOM for sales, studies of the interaction effect between eWOM and the helpfulness of eWOM using the social influence of eWOM based on social exchange theory are lacking. Our study shows that based on social exchange theory, social engagement through online reviews facilitates social exchange, as a shopper will search for reviews from friends as a signal of nonjudgmental and emotional support guidance during the purchasing decision process, as influenced by the average volume, ratings, helpfulness of eWOM, and the product type. Consumers consider the average statistics of reviews and ratings from others as a signal of the quality of and preference for a specific product. Product referral votes indicate reviewers’ evaluations, which can be utilized to decrease the uncertainty of social exchanges [3], and the interaction between the volume and valence with helpfulness of eWOM can be harnessed to help researchers explain purchasing decisions and the sales of products.
Based on previous studies of online reviews that suggested moderators of eWOM [3,20,21,22], our study determines the moderating effects of review or reviewer helpfulness and product type on the effect of eWOM on sales, as these moderating effects are relatively less investigated in previous eWOM studies despite the importance of review quality levels on product sales [14,24]. Specifically, review volume, represented by the multiplication of the average number of reviews and the average review depth, is moderated by review or reviewer helpfulness. Further, our study suggests that product type for experience or search goods can interact with review volume and the extremity of reviews, as well as with review helpfulness for product sales. This study provides insights in the eWOM literature in that the helpfulness of eWOM can contribute to product sales through its moderating effect of review helpfulness and product type.
Second, in order to investigate the effect of a differentiated sampling strategy by considering the helpfulness of a sample to predict product sales performance using eWOM, the BI methods of neural networks and the k-nearest neighbor and decision trees methods were applied here to compare prediction performance outcomes between high and low review or reviewer helpfulness subsamples. Our study attempts to accomplish the second objective as an extension of the first, as after we understand the moderating role of helpfulness, we can proceed with a sampling strategy according to helpfulness.
While previous studies investigated how eWOM affects helpfulness or sales performance, studies that apply BI methods to investigate the effect of a differentiated sampling strategy by comparing differences in prediction performance outcomes according to certain sample characteristics are lacking. Our study makes contributions to the eWOM literature on BI by suggesting review or reviewer helpfulness as a moderating factor in predictions of product sales.
Our study has limitations in that the most common (95.2%) product classes are music, DVDs, and videos. This may reduce the generalizability of the research results. Accordingly, more diverse categories of products can be included in future studies. Further, our study used sales rank as a proxy for sales, and it may be necessary to use actual product sales data to test the moderating effect of eWOM and review (reviewer) helpfulness on product sales. Further, the future can employ more recent data to suggest more insights on the relationships between eWOM and product sales. Our research may be exploratory in nature as there are a lack of studies on the moderators between eWOM and sales, and there may be other moderators or mediators. For instance, the effect of volume and valence on helpfulness and the effect of helpfulness on sales (which mean that helpfulness is a mediator not moderator) will be considered in future research issues.

7.2. Implications for Practitioners

Our study also has several implications for those in business. First, the important eWOM variables posited to be interacting with helpfulness and thus to affect product sales should be considered in the design of websites to enhance product sales. The results of this study can delineate design policies to facilitate more sales based on created value for customers through eWOM. It is better to exhibit the tally of eWOM in terms of the average volume, valence, and helpfulness (of the review and reviewer) for an array of products when customers are viewing products of interest so as to offer more complete summarized information about eWOM and to trace the growth of eWOM for a given product. As customers recognize greater helpfulness from eWOM, companies will realize greater levels of business outcome through eWOM. Thus, guidelines for the design of review systems can be devised such that both eWOM and helpfulness as related to eWOM should be encouraged as crucial aspects of the strategy of many online retailers. Websites could encourage reviewers to create helpful reviews by offering incentives and rewards, which can contribute to the ultimate level of product sales through eWOM.
Second, from the standpoint of BI analysts, the selection strategy for samples can be improved by considering the aspects of the review or reviewer helpfulness of eWOM. This will contribute to a sampling or preprocessing strategy for predicting product sales using eWOM. The prediction performance can be improved by focusing on the sample with high review or reviewer helpfulness. This shows that eWOM with greater levels of helpfulness can provide more prediction power for product sales and that preliminary filtering as a means of data preprocessing should consider review or reviewer helpfulness as a crucial criterion of data quality.

Author Contributions

Conceptualization, S.L.; methodology, S.L.; software, S.L. and J.Y.C.; validation, S.L.; formal analysis: S.L.; investigation, S.L.; resources, J.Y.C.; data curation, S.L.; writing—original draft preparation, S.L.; writing—review and editing, S.L.; supervision, S.L.; funding acquisition, S.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Ghose, A.; Ipeirotis, P.G. Estimating the helpfulness and economic impact of product reviews: Mining text and reviewer characteristics. IEEE Trans. Knowl. Data Eng. 2010, 23, 1498–1512. [Google Scholar] [CrossRef] [Green Version]
  2. Mauri, A.G.; Minazzi, R. Web reviews influence on expectations and purchasing intentions of hotel potential customers. Int. J. Hosp. Manag. 2013, 34, 99–107. [Google Scholar] [CrossRef]
  3. Lee, K.; Lee, B.; Oh, W. Thumbs up, sales up? The contingent effect of Facebook likes on sales performance in social commerce. J. Manag. Inf. Syst. 2015, 32, 109–143. [Google Scholar] [CrossRef]
  4. Li, X.; Hitt, L.M. Price effects in online product reviews: An analytical model and empirical analysis. MIS Q. 2010, 34, 809–831. [Google Scholar] [CrossRef]
  5. Ma, X.; Khansa, L.; Deng, Y.; Kim, S.S. Impact of prior reviews on the subsequent review process in reputation systems. J. Manag. Inf. Syst. 2013, 30, 279–310. [Google Scholar] [CrossRef]
  6. Lee, S.; Choeh, J.Y. The interactive impact of online word-of-mouth and review helpfulness on box office revenue. Manag. Decis. 2018, 56, 849–866. [Google Scholar] [CrossRef]
  7. Kunja, S.R.; GVRK, A. Examining the effect of eWOM on the customer purchase intention through value co-creation (VCC) in social networking sites (SNSs): A study of select Facebook fan pages of smartphone brands in India. Manag. Res. Rev. 2018. [Google Scholar] [CrossRef]
  8. Prasad, S.; Gupta, I.C.; Totala, N.K. Social media usage, electronic word of mouth and purchase-decision involvement. Asia Pac. J. Bus. Adm. 2017, 9, 134–145. [Google Scholar] [CrossRef]
  9. Jalilvand, M.R.; Heidari, A. Comparing face-to-face and electronic word-of-mouth in destination image formation: The case of Iran. Inf. Technol. People 2017, 30, 710–735. [Google Scholar] [CrossRef] [Green Version]
  10. Kudeshia, C.; Kumar, A. Social eWOM: Does it affect the brand attitude and purchase intention of brands? Manag. Res. Rev. 2017, 40, 310–330. [Google Scholar] [CrossRef]
  11. Lee, S.; Choeh, J.Y. Exploring the determinants of and predicting the helpfulness of online user reviews using decision trees. Manag. Decis. 2017, 55, 681–700. [Google Scholar] [CrossRef]
  12. Mudambi, S.M.; Schuff, D. What makes a helpful online review? A study of customer reviews on Amazon.com. MIS Q. 2010, 34, 185–200. [Google Scholar] [CrossRef] [Green Version]
  13. Li, M.; Huang, L.; Tan, C.H.; Wei, K.K. Helpfulness of online product reviews as seen by consumers: Source and content features. Int. J. Electron. Commer. 2013, 17, 101–136. [Google Scholar] [CrossRef] [Green Version]
  14. Chen, C.C.; Tseng, Y.-D. Quality evaluation of product reviews using an information quality framework. Decis. Support Syst. 2011, 50, 755–768. [Google Scholar] [CrossRef]
  15. Wang, A.; Zhang, M.; Hann, I.H. Socially nudged: A quasi-experimental study of friends’ social influence in online product ratings. Inf. Syst. Res. 2018, 29, 641–655. [Google Scholar] [CrossRef] [Green Version]
  16. Amblee, N.; Bui, T. Harnessing the influence of social proof in online shopping: The effect of electronic word of mouth on sales of digital microproducts. Int. J. Electron. Commer. 2011, 16, 91–114. [Google Scholar] [CrossRef]
  17. Park, D.; Lee, J. eWOM overload and its effect on consumer behavioral intention depending on consumer involvement. Electron. Commer. Res. Appl. 2009, 7, 386–398. [Google Scholar] [CrossRef]
  18. Karimi, S.; Wang, F. Online review helpfulness: Impact of reviewer profile image. Decis. Support Syst. 2017, 96, 39–48. [Google Scholar] [CrossRef]
  19. Zhou, S.; Guo, B. The order effect on online review helpfulness: A social influence perspective. Decis. Support Syst. 2017, 93, 77–87. [Google Scholar] [CrossRef]
  20. Wu, J.; Wu, Y.; Sun, J.; Yang, Z. User reviews and uncertainty assessment: A two stage model of consumers’ willingness-to-pay in online markets. Decis. Support Syst. 2013, 55, 175–185. [Google Scholar] [CrossRef]
  21. Hu, N.; Koh, N.S.; Reddy, S.K. Ratings lead you to the product, reviews help you clinch it? The mediating role of online review sentiments on product sales. Decis. Support Syst. 2014, 57, 42–53. [Google Scholar] [CrossRef]
  22. Zhu, F.; Zhang, X. Impact of online consumer reviews on sales: The moderating role of product and consumer characteristics. J. Mark. 2010, 74, 133–148. [Google Scholar] [CrossRef]
  23. Teng, S.; Khong, K.W.; Goh, W.W.; Chong, A.Y.L. Examining the antecedents of persuasive eWOM messages in social media. Online Inf. Rev. 2014, 38, 746–768. [Google Scholar] [CrossRef]
  24. Park, D.-H.; Lee, J.; Han, I. The effect of on-line consumer reviews on consumer purchasing intention: The moderating role of involvement. Int. J. Electron. Commer. 2007, 11, 125–148. [Google Scholar] [CrossRef]
  25. Chintagunta, P.K.; Gopinath, S.; Venkataraman, S. The effects of online user reviews on movie box office performance: Accounting for sequential rollout and aggregation across local markets. Mark. Sci. 2010, 29, 944–957. [Google Scholar] [CrossRef]
  26. Duan, W.; Gu, B.; Whinston, A.B. The dynamics of online word-of-mouth and product sales: An empirical investigation of the movie industry. J. Retail. 2008, 84, 233–242. [Google Scholar] [CrossRef]
  27. Weathers, D.; Swain, S.D.; Grover, V. Can online product reviews be more helpful? Examining characteristics of information content by product type. Decis. Support Syst. 2015, 79, 12–23. [Google Scholar] [CrossRef]
  28. Ngo-Ye, T.L.; Sinha, A.P. The influence of reviewer engagement characteristics on online review helpfulness: A text regression model. Decis. Support Syst. 2014, 61, 47–58. [Google Scholar] [CrossRef]
  29. Archak, N.; Ghose, A.; Ipeirotis, P.G. Deriving the pricing power of product features by mining consumer reviews. Manag. Sci. 2011, 57, 1485–1509. [Google Scholar] [CrossRef] [Green Version]
  30. Li, X.; Wu, C.; Mai, F. The effect of online reviews on product sales: A joint sentiment-topic analysis. Inf. Manag. 2019, 56, 172–184. [Google Scholar] [CrossRef]
  31. Huang, C.C.; Lin, T.C.; Lin, K.J. Factors affecting pass-along email intentions (PAEIs): Integrating the social capital and social cognition theories. Electron. Commer. Res. Appl. 2009, 8, 160–169. [Google Scholar] [CrossRef]
  32. Du, J.; Xu, H.; Huang, X. Box office prediction based on microblog. Expert Syst. Appl. 2014, 41, 1680–1689. [Google Scholar] [CrossRef]
  33. Rui, H.; Liu, Y.; Whinston, A. Whose and what chatter matters? The effect of tweets on movie sales. Decis. Support Syst. 2013, 55, 863–870. [Google Scholar] [CrossRef] [Green Version]
  34. Reinstein, D.; Snyder, C.M. The influence of expert reviews on consumer demand for experience goods: A case study of movie critics. J. Ind. Econ. 2005, 53, 27–51. [Google Scholar] [CrossRef]
  35. Forman, C.; Ghose, A.; Wiesenfeld, B. Examining the relationship between reviews and sales: The role of reviewer identity disclosure in electronic markets. Inf. Syst. Res. 2008, 19, 291–313. [Google Scholar] [CrossRef]
  36. Liu, Y. Word of mouth for movies: Its dynamics and impact on box office revenue. J. Mark. 2006, 70, 74–89. [Google Scholar] [CrossRef]
  37. Premzaai, K.; Castaldo, S.; Grosso, M.; Raman, P.; Brudvig, S.; Hofacker, C. Customer information sharing with e-vendors: The roles of incentives and trust. Int. J. Electron. Commer. 2010, 14, 63–74. [Google Scholar] [CrossRef]
  38. Hung, K.H.; Li, S.Y. The influence of eWOM on virtual consumer communities: Social capital, consumer learning, and behavioral outcomes. J. Advert. Res. 2007, 47, 485–495. [Google Scholar] [CrossRef]
  39. Lee, S.; Choeh, J.Y. The determinants of helpfulness of online reviews. Behav. Inf. Technol. 2016, 35, 853–863. [Google Scholar] [CrossRef]
  40. Chevalier, J.A.; Mayzlin, D. The effect of word of mouth on sales: Online book reviews. J. Mark. Res. 2006, 43, 345–354. [Google Scholar] [CrossRef] [Green Version]
  41. Cheung, M.Y.; Luo, C.; Sia, C.L.; Chen, H. Credibility of electronic word-of-mouth: Informational and normative determinants of on-line consumer recommendations. Int. J. Electron. Commer. 2009, 13, 9–38. [Google Scholar] [CrossRef]
  42. Chen, C.-H.; Nguyen, B.; Klaus, P.P.; Wu, M.-S. Exploring electronic word-of-mouth (eWOM) in the consumer purchase decision-making process: The case of online holidays—Evidence from United Kingdom (UK) consumers. J. Travel Tour. Mark. 2015, 32, 953–970. [Google Scholar] [CrossRef] [Green Version]
  43. Liang, L.J.; Choi, H.S.C.; Joppe, M. Understanding repurchase intention of Airbnb consumers: Perceived authenticity, electronic word-of-mouth, and price sensitivity. J. Travel Tour. Mark. 2016, 35, 1224750. [Google Scholar] [CrossRef]
  44. Cui, G.; Lui, H.-K.; Guo, X. The effect of online consumer reviews on new product sales. Int. J. Electron. Commer. 2012, 17, 39–58. [Google Scholar] [CrossRef]
  45. Hsu, C.-L.; Yu, L.-C.; Kuo-Chien, C. Exploring the effects of online customer reviews, regulatory focus, and product type on purchase intention: Perceived justice as a moderator. Comput. Hum. Behav. 2017, 69, 335–346. [Google Scholar] [CrossRef]
  46. Van Der Heide, B.; Johnson, B.K.; Vang, M.H. The effects of product photographs and reputation systems on consumer behavior and product cost on eBay. Comput. Hum. Behav. 2013, 29, 570–578. [Google Scholar] [CrossRef]
  47. Connors, L.; Mudambi, S.M.; Schuff, D. Is it the review or the reviewer? A multi-method approach to determine the antecedents of online review helpfulness. In Proceedings of the 44th Hawaii International Conference on System Sciences, Kauai, HI, USA, 4–7 January 2011; IEEE Computer Society: Los Alamitos, CA, USA, 2011. [Google Scholar]
  48. Yin, D.; Bond, S.; Zhang, H. Anxious or angry? Effects of discrete emotions on the perceived helpfulness of online review. MIS Q. 2014, 38, 539–560. [Google Scholar] [CrossRef]
  49. Elwalda, A.; Lü, K.; Ali, M. Perceived derived attributes of online customer reviews. Comput. Hum. Behav. 2016, 56, 306–319. [Google Scholar] [CrossRef] [Green Version]
  50. Filieri, R.; Alguezaui, S.; McLeay, F. Why do travelers trust TripAdvisor? Antecedents of trust towards consumer-generated media and its influence on recommendation adoption and word of mouth. Tour. Manag. 2015, 51, 174–185. [Google Scholar] [CrossRef] [Green Version]
  51. Baek, H.; Ahn, J.; Choi, Y. Helpfulness of online consumer reviews: Readers’ objectives and review cues. Int. J. Electron. Commer. 2012, 17, 99–126. [Google Scholar] [CrossRef]
  52. Pan, Y.; Zhang, J. Born unequal: A study of the helpfulness of user generated product reviews. J. Retail. 2011, 87, 598–612. [Google Scholar] [CrossRef]
  53. Lee, E.-J.; Shin, S.Y. When do consumers buy online product reviews? Effects of review quality, product type, and reviewer’s photo. Comput. Hum. Behav. 2014, 31, 356–366. [Google Scholar] [CrossRef]
  54. Cheung, C.M.K.; Thadani, D.R. The impact of electronic word-of-mouth communication: A literature analysis and integrative model. Decis. Support Syst. 2012, 54, 461–470. [Google Scholar] [CrossRef]
  55. Papathanassis, A.; Knolle, F. Exploring the adoption and processing of online holiday reviews: A grounded theory approach. Tour. Manag. 2011, 32, 215–224. [Google Scholar] [CrossRef]
  56. Sotiriadis, M.D.; van Zyl, C. Electronic word-of-mouth and online reviews in tourism services: The use of twitter by tourists. Electron. Commer. Res. 2013, 13, 103–124. [Google Scholar] [CrossRef]
  57. Shan, Y. How credible are online product reviews? The effects of self-generated and system generated cues on source credibility evaluation. Comput. Hum. Behav. 2016, 55, 633–641. [Google Scholar] [CrossRef]
  58. Shen, W.; Hu, Y.J.; Ulmer, J.R. Competing for attention: An empirical study of online reviewers’ strategic behavior. MIS Q. 2015, 39, 683–696. [Google Scholar] [CrossRef]
  59. Ganu, G.; Kakodkar, Y.; Marian, A. Improving the quality of predictions using textual information in online user reviews. Inf. Syst. 2013, 38, 1–15. [Google Scholar] [CrossRef]
  60. Li, J.; Zhan, L. Online persuasion: How the written word drives WOM evidence from consumer-generated product reviews. J. Advert. Res. N. Y. 2011, 51, 239–257. [Google Scholar] [CrossRef]
  61. Huang, A.H.; Chen, K.; Yen, D.C.; Tran, T.P. A study of factors that contribute to online review helpfulness. Comput. Hum. Behav. 2014, 48, 17–27. [Google Scholar] [CrossRef]
  62. Hong, H.; Xua, D.; AlanWang, G.; Fan, W. Understanding the determinants of online review helpfulness: A meta-analytic investigation. Decis. Support Syst. 2017, 102, 1–11. [Google Scholar] [CrossRef]
  63. Shmueli, G.; Patel, N.R.; Bruce, P.C. Data Mining for Business Analytics, 3rd ed.; John Wiley & Sons, Inc.: Hoboken, NJ, USA, 2018. [Google Scholar]
  64. Kwon, O.; Sung, Y. Shifting selves and product reviews: How the effects of product reviews vary depending on the self-views and self-regulatory goals of consumers. Int. J. Electron. Commer. 2012, 17, 59–82. [Google Scholar] [CrossRef]
  65. Dellarocas, C.; Gao, G.; Narayan, R. Are consumers more likely to contribute online reviews for hit or niche products? J. Manag. Inf. Syst. 2010, 27, 127–158. [Google Scholar] [CrossRef]
  66. Schindler, R.M.; Bickart, B. Perceived helpfulness of online consumer reviews: The role of message content and style. J. Consum. Behav. 2012, 11, 234–243. [Google Scholar] [CrossRef] [Green Version]
  67. Cheung, C.M.; Lee, M.K. What drives consumers to spread electronic word of mouth in online consumer-opinion platforms. Decis. Support Syst. 2012, 53, 218–225. [Google Scholar] [CrossRef]
  68. Ku, Y.-C.; Wei, C.-P.; Hsiao, H.-W. To whom should I listen? Finding reputable reviewers in opinion-sharing communities. Decis. Support Syst. 2012, 53, 534–542. [Google Scholar] [CrossRef]
Sustainability 12 07952 g001 550
Figure 1. Research Model.
Figure 1. Research Model.
Sustainability 12 07952 g001
Sustainability 12 07952 g002 550
Figure 2. Procedures of the data analysis.
Figure 2. Procedures of the data analysis.
Sustainability 12 07952 g002
Table
Table 1. The sample of the study. Product type is also shown in parenthesis.
Table 1. The sample of the study. Product type is also shown in parenthesis.
Items FrequencyPercent
Product typeExperience goods60432.9
Search goods123067.1
Product classConsumer Electronics (experience)10.1
Video (experience)1437.8
Video games (experience)40.2
DVD (experience)45324.7
Toy (experience)20.1
Software (experience)10.1
Sports (search)10.1
Single detail page miscellaneous (search)10.1
PC accessory (search)20.1
Pet products (search)20.1
Home (search)20.1
Home improvement (search)20.1
Kitchen (search)10.1
Music (search)115062.7
Baby products (search)10.1
Book (search)633.4
Wireless devices (search)50.3
Review helpfulness (according to average (0.733))High review helpfulness subsample106758.2
Low review helpfulness subsample76741.8
Reviewer helpfulness (according to average (0.755))High reviewer helpfulness subsample110360.1
Low reviewer helpfulness subsample73139.1
Total 1834100
Table
Table 2. eWOM and movie-related variables used in this study.
Table 2. eWOM and movie-related variables used in this study.
CategoryVariablesMeasures Description
eWOMReview depthThe review length in terms of words on the average
Review volume The average number of reviews × The average number of words in a review
Average review ratingThe review rating on the average
Review extremityThe absolute value of the difference between the average of all user ratings and the reviewers’ rating
Reviewer rankThe rank of a specific reviewer announced by Amazon
Average elapsed days after posting until parsingThe average elapsed days after posting until parsing for data analysis
Review helpfulnessThe proportion of the number of helpful votes among the total number of votes that evaluated if the review was helpful or not
Reviewer helpfulnessThe proportion of the number of helpful votes among the total number of votes that were received by the specific reviewer to evaluate if the review by that reviewer was helpful or not
ProductSales rank The rank of sales for a specific product (The greater sales, the smaller rank value)
PriceThe price of product
Product type Experience goods (coded as 1) (products that necessitate purchasing or sampling for the evaluation of product quality) or search goods (coded as 0) (products about which consumers may obtain information on product quality before purchasing)
Table
Table 3. Descriptive statistics of variables in the study.
Table 3. Descriptive statistics of variables in the study.
VariablesMinimumMaximumMeanStd. Deviation
Price0.01265.0012.2310.42
Sales rank13.002,199,944.0061,133.76103,943.22
Reviewer rank82.0015,811,962.004,026,059.092,803,538.89
Elapsed days after launching until posting0.003860.001674.21795.91
Elapsed days after posting until parsing52.004973.502412.03835.46
The average number of reviews1.00385.0016.2620.52
The average number of words in a review10.00467.14108.4049.58
The average number of reviews × The average number of words in a review10.0030,728.071878.492517.29
Average review rating1.005.004.370.57
Average review extremity0.002.000.580.38
Table
Table 4. Regression analysis using eWOM for ln (sales rank).
Table 4. Regression analysis using eWOM for ln (sales rank).
Standardized CoefficientsStandard ErrortSig.
Price0.119 ***0.0036.6810.000
Reviewer rank−0.042 **0.000−1.9510.026
Elapsed days after posting until parsing0.261 ***0.00013.4090.000
The average number of reviews−2.684 ***0.043−4.7370.000
Average review rating−0.0160.085−0.5090.305
The average number of reviews × The average number of words in a review × review helpfulness−0.402 **0.000−2.0170.022
The average number of reviews × The average number of words in a review × review helpfulness × product type0.248 ***0.0002.3570.010
Average review rating × review helpfulness−0.0040.039−0.1340.447
Average review extremity × review helpfulness × product type0.126 **0.4111.7930.037
The average number of reviews × The average number of words in a review × reviewer helpfulness−1.676 ***0.000−3.1360.001
The average number of reviews × The average number of words in a review × reviewer helpfulness × product type0.0160.0000.1550.439
Average review rating × reviewer helpfulness−0.052 *0.074−1.5200.065
Average review extremity × reviewer helpfulness × product type0.0180.3790.2500.402
Control variables = price, reviewer rank, elapsed days after posting until parsing. * p < 0.1 ** p < 0.05 *** p < 0.01, F = 88.59, p-value = 0.000.
Table
Table 5. Average prediction error for business intelligence (BI) methods for high and low review and reviewer helpfulness groups. Error is in RMSR (root mean squared residual).
Table 5. Average prediction error for business intelligence (BI) methods for high and low review and reviewer helpfulness groups. Error is in RMSR (root mean squared residual).
Number of Prediction Error Data Used to CompareNeural Network (Manual)Neural Network (Automatic)k-Nearest Neighbor (Best k Method)Decision Tress (with Pruning)
Review helpfulnessHigh helpfulness (767)32144,789.75138,806.9691,444.6675,725.91
Low helpfulness (767)32259,720.67246,872.1195,264.6274,494.48
t-value
(p-value)		 −9.024 (0.000)−8.275 (0.000)−0.218 (0.828)0.067 (0.947)
Reviewer helpfulnessHigh helpfulness (731)31147,865.44138,134.2397,642.9073,953.45
Low helpfulness (731)31258,334.47240,704.5448101,095.6371,586.80
t-value
(p-value)		 −9.288 (0.000)−8.223 (0.000)−0.197 (0.845)0.135 (0.893)

Share and Cite

MDPI and ACS Style

Lee, S.; Choeh, J.Y. Using the Social Influence of Electronic Word-of-Mouth for Predicting Product Sales: The Moderating Effect of Review or Reviewer Helpfulness and Product Type. Sustainability 2020, 12, 7952. https://doi.org/10.3390/su12197952

AMA Style

Lee S, Choeh JY. Using the Social Influence of Electronic Word-of-Mouth for Predicting Product Sales: The Moderating Effect of Review or Reviewer Helpfulness and Product Type. Sustainability. 2020; 12(19):7952. https://doi.org/10.3390/su12197952

Chicago/Turabian Style

Lee, Sangjae, and Joon Yeon Choeh. 2020. "Using the Social Influence of Electronic Word-of-Mouth for Predicting Product Sales: The Moderating Effect of Review or Reviewer Helpfulness and Product Type" Sustainability 12, no. 19: 7952. https://doi.org/10.3390/su12197952

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop