1
GAZING AT THE STARS IS NOT ENOUGH, LOOK AT THE WORD CONSTRUALS TOO
ABSTRACT
Online reviews affect purchase decisions by providing information posted online by
previous consumers. Past research on review helpfulness focused on its numerical star
ratings and on the number of words included in its textual portion. This study proposes
also including the qualitative information in those reviews to achieve a more
comprehensive understanding of the factors that impact review helpfulness. The paper
does so by looking at novel versus established products. To enable this addition the study
proposes a method to classify the words/terms in online reviews into low- and high-level
construal words by applying Latent Semantic Analysis (LSA). Low-level construal terms
are words dealing with the experience of other consumers. High-level construal terms are
words dealing with characteristics and features of the product itself. Results show that for
established products, numerical ratings, such as the number of stars, and low-level
construal strongly increase review helpfulness. Adding high-level construal terms
actually reduced helpfulness. For novel products, the number of stars, the number of
previous assessments of review helpfulness, and the verification that the reviewer bought
the product add to perceived helpfulness. That is not the case of high- and low-level
construal types of information. Online consumers may lack the knowledge to evaluate
what textual information is relevant for novel products. Both academics and practitioners
can benefit from a more comprehensive understanding of what increases the perceived
helpfulness of online product reviews.
KEYWORDS
Online consumer reviews, e-WOM, e-commerce, review helpfulness, latent semantic
analysis, construal terms
2
INTRODUCTION
Online consumer reviews are an increasingly important source of information in purchase
decisions, complementing and sometimes substituting other forms of business-to-consumer and
consumer-to-consumer information (Chevalier and Mayzlin 2006). Online reviews provide
information based on the personal experience of previous buyers of a specific product working as
free "sales assistants" to help other consumers identify the product that best matches their needs
and tastes (Chen and Xie 2008). The development of websites whose primary focus is to provide
purchase information through consumer reviews – e.g. Consumeraffairs.com or Angieslist.com –
exemplifies their growing popularity as relevant product information providers (Anderson and
Simester 2014). In fact, a 2012 survey conducted by Nielsen among 28,000 Internet users from
56 countries reported that 70% of respondents considered online consumer reviews a trustworthy
source of information, ranking them as the second most trusted form of advertising among 19
different choices (Nielsen 2012). Online product reviews impact sales (Chevalier and Mayzlin
2006; Chintagunta et al. 2010; Floyd et al. 2014; Liu 2006; Zhu and Zhang 2010), especially
helpful reviews (Chen et al. 2008). There is, therefore, a growing interest in understanding the
importance of online product reviews, as evidenced by the abundance of available literature (see
Appendix B for a list of some of the most relevant papers on online consumer reviews). This
study contributes to the expanding literature on what makes an online review helpful by
including in the analysis the high- and low-level construal terms in the written portion of the
reviews. By doing so, this paper presents an integrative approach that includes what people talk
about in the textual part of each of those online reviews.
3
The degree of helpfulness that potential customers assign to a review is a common
measure to assess the value of the information included in it (Mudambi and Schuff 2010). On
many sites, this helpfulness measure is the summation of the positive and negative votes of the
specific review by potential buyers. Review helpfulness can also represent a public endorsement
of a particular review message by a peer consumer (Metzger et al. 2010). More ‘helpful’ product
reviews can provide a competitive advantage due to their value as a differentiation tool and can
increase the time spent on an e-commerce site (Mudambi and Schuff 2010). The importance of
review helpfulness ratings has raised practical and academic interest in identifying the elements
that increase the perceived helpfulness of these product reviews.
Past research into what makes online consumers reviews helpful has concentrated on the
easily accessible numeric characteristics, such as the number of stars and on counting the number
of words included in the textual portion (e.g. Mudambi and Schuff 2010; Pan and Zhang 2011).
This study introduces a new method in marketing that allows for an assessment of the qualitative
content of those reviews too. The paper automatically integrates the construal meaning of words
in the product review to determine how the content of those reviews, in addition to the variables
employed in previous research, affects subsequent helpfulness evaluations. This study proposes a
methodology to do so based on applying Latent Semantic Analysis (LSA) and Construal Level
Theory (CLT) to identify low- and high-level construal words in the reviews. Subsequently,
these two variables are tested along with other variables employed in previous literature to assess
their importance in determining the helpfulness of the review. Rather than identify what specific
words potential customers seek for a specific product, this method has the potential to classify
the words in any review of any product into low- and high-level construal words applying CLT
and then suggest when low- or high-level construal words, or both, also contribute in forming
4
helpfulness assessments. This paper tests its hypotheses and the revised theoretical model over
product reviews relating to 16 types of light bulbs (LEDs and incandescent light bulb) available
for purchase at Amazon.com.
The contributions of this study are in: (1) proposing a new methodology based on LSA to
identify key terms in online reviews, (2) showing how CLT can be applied to classify those key
terms into low- and high-level construal terms, (3) developing a framework to assess the impact
of the informational elements available at the individual review level on its assessed helpfulness
and (4) showing that the relative importance of the variables differs depending on whether the
product is novel or established. The textual part of consumer reviews has been the focus of
limited research in marketing, possibly because of a lack of an appropriate methodology
(Chevalier and Mayzlin 2006), leaving previous attempts to process text with noisy results
(Godes and Mayzlin 2004). Adding a textual analysis of the kind proposed in this study is
important because research suggests that the textual part of product reviews may allow for richer
descriptions of the interaction of buyers with the product (King et al. 2014).
The remainder of the paper is organized as follows. In the second section, we develop the
theoretical background, leading to the research hypotheses. The third section introduces the
conceptual model. The fourth section explains the methodology. The fifth section is the data
analysis. The sixth section discusses the findings, limitations, and avenues for future research.
THEORETICAL DEVELOPMENT
The Context. Electronic Word-of-Mouth and Review Helpfulness
5
Online consumer reviews are a form of word-of-mouth (WOM). WOM is a term that
includes different types of informal communications among consumers about products, services,
or companies (Liu 2006). WOM is a source of purchasing information that influences potential
customers’ behavior (Brown and Reingen 1987; McFadden and Train 1996). It is often
considered by consumers to be a reliable and valid source of information that can reduce
purchase risks (Bickart and Schindler 2001; Godes and Mayzlin 2004; Mayzlin 2006). Possibly,
this is because of its non-commercial nature and because consumers generally trust other peer
consumers more than they trust marketers or advertisers (Sen and Lerman 2007). The electronic
counterpart of WOM (e-WOM) is defined as “any positive or negative statement made by
potential, actual or former consumers about a product or company, which is made available to a
multitude of people and institutions via the Internet” (p.39, Hennig-Thurau et al. 2004).
Typically, e-WOM interactions take place among people who have no previous connection or
relationship. This constitutes a major difference compared to traditional WOM (Dellarocas
2003). e-WOM includes a broad range of elements in addition to online reviews such as blogs,
social networking sites, online forums, and electronic bulletin boards (Floyd et al. 2014).
Although there is no standard e-WOM structure, online consumer reviews usually include
several informational elements. These elements include: (1) review valence (frequently referred
to as ‘number of stars’); (2) review text (where reviewers provide further qualitative
information); (3) the verification that the reviewer actually purchased the product or service; and
(4) an overall summary helpfulness score of the review. In the case of Amazon.com this
helpfulness score is implemented by answering the question “was this review helpful to you?”
with a dichotomous “yes” or “no” answer (thumb up/thumb down). The helpfulness assessment
6
is reported by people other than the person who posted elements (1) and (2). This helpfulness
evaluation is the dependent variable in the conceptual model of this study.
Review helpfulness is “the extent to which consumers perceive the product review as
being capable of facilitating judgment or purchase decisions” (p. 103, Li et al. 2013). Helpful
product reviews provide potential buyers with pertinent information from previous consumers
with the purpose of informing the purchasing decision process (Mudambi and Schuff 2010).
Review helpfulness can be understood as the principal tool to assess how potential customers
evaluate the quality of the information provided in each individual review. Information quality
can reduce purchase uncertainty, which can be crucial within the e-commerce context (Mudambi
and Schuff 2010). This is important because helpfulness ratings can impact sales (Chen et al.
2008; Chen and Xie 2008; Ghose et al. 2012), presumably by certifying to the value of the
information they provide (Ghose and Ipeirotis 2011). As a prominent example of the relevance
of the helpfulness ratings, Spool (2009) estimated that Amazon.com added $2.7 billion to its
annual revenue by requesting that potential customers assess the perceived helpfulness of the
review, and, through these evaluations, making the more ‘helpful’ reviews more visible to
potential customers. Additionally, helpfulness ratings allow consumers to filter the
overwhelming amount of information available on popular e-commerce sites. This may allow
potential customers to make purchase decisions even in situations of information overload (Cao
et al. 2011). The availability of product reviews can also increase the time that potential buyers
spend on a particular website, thanks to the so called “stickiness” effect of product reviews
(Mudambi and Schuff 2010). This time spent on a site can be increased even more if the
information provided actually aids consumers in their decision making, i.e. if the information
provided is in fact ‘helpful.’
7
Theory Base and Hypotheses
Allowing for this evidenced importance of e-WOM, the objective of this study is to
suggest a theory base and a related methodology that also accounts for the qualitative portion of
online reviews. This proposed theory base and related methodology should be applicable in a
consistent manner across products and service types. This need for consistent applicability is
highlighted because the specific wording that makes a review of a certain product or service
helpful might not readily apply to another product or service. In an attempt to create such an
overarching theory and supporting methodology, this section introduces CLT as a classifier of eWOM information into low- and high-level construal terms and LSA as a methodology to
identify those terms.
Being able to apply such a theory and methodology, and achieve a more comprehensive
approach to the informational elements that add to helpfulness is arguably necessary because: (1)
the web economy lowers search costs for most products (Alba et al. 1997; Lynch Jr and Ariely
2000); (2) the likelihood of information overload is greater due to the higher number of
alternatives and to the lower searching costs, especially for the most popular e-commerce sites
(Alba et al. 1997; Peterson et al. 1997; Shapiro and Varian 1999); (3) the entry costs are very low
in the online marketplaces for sellers, making it very difficult for customers to differentiate
among legitimate vendors and “fly-by-night” sellers (Tan 1999); (4) the spatial difference with
the online retailer makes it infeasible to physically examine the product prior to purchase (Tan
1999); and (5) there is a greater assortment of products on e-commerce sites, including versioned
and customized possibilities (Peterson et al. 1997; Shapiro and Varian 1999).
The textual part of online product reviews can provide at least a partial solution to those
issues. However, little research has looked at how to assess the value of the textual portions,
8
even in the form of subjective helpfulness assessments. This scarce attention may be due to a
lack of appropriate methodologies to analyze large amounts of textual data (Chevalier and
Mayzlin 2006) and due to costly analyses and noisy results (Godes and Mayzlin 2004). Although
the numerical part of product reviews conveys valuable information, the textual part could
arguably allow for richer description (King et al. 2014). The text may also convey additional
helpful information that may have nothing to do with the actual interaction with the product such
as packaging and customer service. Moreover, analyzing a numeric value rather than analyzing
its related textual review ignores the possibility that the description may deal with more than one
aspect of the quality construct or the uniformity of consumer preferences (Archak et al. 2011).
The application of CLT with LSA provides such a methodology to assess the text portion of
online reviews and add it to the existing research on what makes a review helpful.
CLT and the Difference between Low- and High-Level Construal Terms
A limiting issue in assessing the textual part of online product reviews is that specific
terms that might make a posting helpful about one product or service may not be as helpful with
another. For example, a term such as “courteous” may be important across services, but
“luxurious” may be more important in contexts relating to opulence. And this is where CLT as
the theory base is advantageous. CLT provides a basis for distinguishing between two different
types of information: high-level and low-level construal terms (Irmak et al. 2013; Liberman and
Trope 1998). High-level construal terms deal with the desirability of a product or service based
on its essential features (i.e. why would customers want to buy this product or service).
Considering the desirability of an object, event, or course of action, high-level construal
information puts more importance on the end results (Trope 2012). High-level construal terms
are abstract, simple, structured and coherent, decontextualized, focused on the primary core,
9
superordinate, and goal relevant (Trope and Liberman 2003). As such, in an e-commerce
context, high-level construal terms could be understood as comprising the type of information
that is typically provided by manufacturers and retailers as they describe the product itself and its
relevant features. Examples of this type of information are technical details of RAM memory or
the name of a concert performer. In the specific case of the database of this study, the technical
details of a light bulb, such as its color temperature, constitute high-level construal information.
In contrast, low-level construal terms provide information about the feasibility or specific
‘how’ nuances of the product or service. This could include details about how other customers
used a product, their experience with it, and secondary features that are not as salient as those
included in the main desirability core. Low-level construal terms are relatively more concrete,
complex, unstructured, contextualized, focused on secondary features, subordinate, and goal
irrelevant (Trope and Liberman 2003). Consumer reviews provide also such low-level construal
terms (De Maeyer 2012). Previous research suggests that potential buyers value low-level
construal information in some cases as much as they do high-levels construal information (Irmak
et al. 2013). As Trope states, high-level construal information is a way of thinking in a bigpicture way, whereas low-level construal information is more detail-oriented (Trope 2012).
Specific examples of this type of low-level construal terms might include testimonials about how
pleasant the light produced by an LED is and how it allowed the person writing the review to see
nuanced colors in paintings better. Low-level and high-level construal terms complement each
other. Low-level construal terms provide information that allow potential customers to learn
from the experiences of previous buyers. High-level construal terms reported in the textual
portion of a review can support retailers’ claims.
High- and Low-Level Construal Information
10
Economics of Information (EoI) provides a theoretical framework for the necessity of
gathering pre-purchase information by consumers. The basic premise of EoI is that there exists
an information asymmetry between buyers and sellers, i.e. potential customers are not perfectly
informed about all the available product or service choices in the market (Biswas 2004). The
classic EoI literature suggests that a potential buyer, operating within this asymmetric
information market, would acquire information about the product until the point at which the
marginal cost of further information acquisition equals or exceeds its marginal benefit (Stigler
1961). Research shows that in online settings consumers actively search for purchase
information before buying (e.g. Punj and Moore 2009) and that their decisions are affected by
the recommendations of others (Li and Zhan 2011; Moe and Trusov 2011).
Both low- and high-level construal terms embedded in the text provided in an online
consumer review could provide people reading it with such valuable information about the
experiences of others, in addition to the other sources available. Such terms may carry meaning,
regardless of the rest of the text. Indeed, some words such as “horrible”, a low-level construal
term, and “efficient”, a high-level construal term, carry strong messages when applied in the
context of light, and may do so even regardless of the rest of the sentence. It is true that meaningchanging prefixes, such as adding “not” before “good”, may confuse such an approach.
Nonetheless, such prefixes are sufficiently rare in English that analyzing individual words across
many documents is enough for LSA to successfully factor together words to identify synonyms
(Gomez et al. 2012; Islam et al. 2012; Valle-Lisboa and Mizraji 2007), answer multiple choice
exam questions as successfully as students do in an introduction to psychology course (Landauer
et al. 1998b), and score comparably to nonnative speakers on TOEFL (Landauer and Dumais
1997). In the specific data analyzed in this study, meaning-changing prefixes were very rare.
11
Before discussing the hypotheses, some background discussion of LSA is necessary to
clarify operationally what constitutes low-level and high-level construal terms in this context.
What LSA does, more on that in the next sections, is to identify through singular value
decomposition (SVD), a process somewhat equivalent to a two-way principal components
analysis (PCA), what terms (e.g. words) and what documents (in this case, the online product
reviews) factor together in a document-to-terms frequency matrix (DTM). In analyzing terms,
LSA supports both the analysis of single words as well as phrases. (Typically, and also in this
study, terms are only single words.) Terms that factor together are assumed to carry some latent
shared meaning, much as factors are assumed to do in a PCA. And, as in a PCA, only those
factors that explain high degrees of the variance are typically retained. The resulting retained
factors identify the words that carry the most weight in explaining the variance in the DTM.
In the specific analysis run in this study some of those factors contained words that were
the same words as the manufacturers’ websites used to describe the products. The words
associated with those factors were determined to be high-level construal terms. Note that LSA
drops commonly used words such as "the", so the retained words in this category are primarily
technical in nature. Words associated with the other factors were treated as low-level construal
terms. In other words, LSA allowed the classification of commonly occurring high eigenvalue
words in the online reviews as either replicating words that appear in the manufacturers’
websites or other information that the authors of those online reviews posted. The words that
repeated the manufacturers’ websites technical information were treated as high-level construal
terms. The other words were treated as low-level construal terms, relating to information other
than the technical experience of the consumers. As a result of this process, it was possible to
identify how many high-level construal terms and how many low-level construal terms appeared
12
in each online product review. Conceptually, this process could be thought of as the equivalent
of asking human raters to identify the most influential words in the online product reviews and
count the number of times each review has what seems to them subjectively as low- or as highlevel construal terms. LSA does this automatically and objectively.
The Importance of Low-Level Construal Information
Exploring product information allows consumers to make better purchase decisions
(Kohli et al. 2004). Low-level construal terms provide information about the experience of
consumers beyond what the website details. That experience can come from the actual
interaction with the product to other related experiences such as product packaging and shipping.
This low-level evidence is generally not easy to access through other means different from eWOM. Presumably, the more low-level construal words there are, the more helpful the review
should be as a response to the need of a more informed purchase decision. More terms mean
either more information, in the case of different terms, or emphasizing a point, in the case of a
repetition of a term. More low-level construal words may also be the result of adding words that
load on other factors, adding additional facets of information. Rationally, if people reading the
post are seeking information, and if, as LSA contends, terms are facets of that information, then
the more such information is provided, the more helpful the online review should be. Consistent
with this, the first hypothesis posits:
H1: The more low-level construal terms included in a review, the more its perceived helpfulness
The Importance of High-Level Construal Information
High-level construal terms add to the helpfulness of the review in an equivalent, but
different, way. Consumers lack physical cues to evaluate products in electronic marketplaces.
This lack of cues creates risk and uncertainty since customers cannot anticipate if the purchase
13
will result in the expected benefits. To reduce this uncertainty and risk, individuals seek to verify
from other sources the claims that manufacturers make (Racherla et al. 2012; Ramirez et al.
2002). Online reviews can be one such source of verification and vice versa, they may also
question claims manufacturers make (Lee et al. 2015). Research suggests that consumers largely
trust other consumers more than they trust claims made by sellers (Sen and Lerman 2007). And
so, if the terms a manufacturer uses in its website to describe the product, such as describing a
bulb as efficient or its light as warm, are repeated in the online review then it would at least
somewhat confirm what the manufacture claims. Applying the same logic as in the previous
hypothesis, the more these terms are used, the stronger the message of confirmation with the
claims the manufacturer makes should be. The helpfulness of an online review should be higher,
the more such confirmation it provides. Following this rationale, the second hypothesis of this
study is as follows:
H2: The more high-level construal terms included in a review, the more its perceived helpfulness
Other Sources of Information in Individual Consumer Reviews
Customer ratings, such as the number of stars, are known important quantitative
informational elements of e-WOM, as they can assist potential customers in learning about the
quality of a product (Filieri 2015). However, existing literature has reported contrasting results
regarding how customer ratings affect the decision-making process. Several papers found that
extremely negative ratings are more influential, while others found that the extremely positive
ones are the most relevant for this process (e.g. Hu et al. 2009; Kuan et al. 2015; Pan and Zhang
2011; Sen and Lerman 2007; Wu 2013). Under the specific conditions of consumers evaluating
e-WOM helpfulness, Pan and Zhang (2011) found that positive ratings are more influential than
negative ones and are rated as more helpful. In other words, these authors suggest that there is a
14
positive correlation between the number of stars and review helpfulness. If that is the case, the
number of stars might be considered as a surrogate recommendation (Alex and Prendergast
2009). Furthermore, from a practical standpoint, the existence of a market for artificially
inflating ratings and fake reviews (e.g. News 2013) also suggests that there should be a positive
relationship between the number of stars and perceived helpfulness. Due to the strongly
contrasting results reported in the literature and to achieve a comprehensive approach to the
informational elements available in online reviews, this paper tests the positive relationship
reported previously by Pan and Zhang (2011):
H3: The more stars a review has, the more its perceived helpfulness
Potential customers are not perfectly informed about all the available product or service
choices in the market (Biswas 2004). Previous literature suggests that this problem can be
lessened by the provision of product information by third-parties (Faulhaber and Yao 1989).
Product reviews are suitable to mitigate this lack of market knowledge if their information is
pertinent, but only if their source is believable (Forman et al. 2008). Filieri (2015) found that if
the information source in e-WOM is assessed as credible, it will be perceived as more helpful.
One such aspect of credibility is that the person posting the online product review actually
bought the product. Conversely, if that person did not buy the product, the credibility in making
claims about it may be questionable. Perhaps indicating how strong that indication can be, many
sites – such as Amazon.com – include a verification that the reviewer actually purchased the
product. Consequently, it might be expected therefore that a verification that the review writer
bought the product should increase the perceived helpfulness of the review:
H4: The confirmation that the reviewer purchased the product will increase the perceived
helpfulness of the review
15
Accounting for Product Novelty
The need for information gathering could vary substantially depending on product
novelty. Novel products are relatively unknown to potential consumers, and so consumers might
seek more information about those products to reduce their perceived risk in purchase (Dowling
and Staelin 1994; Hawkins and Mothersbaugh 2010; Wasson 1978; Wind and Robertson 1983).
Because customers seek more information in the case of novel products, it follows that when
such information is available, reviews will be more helpful. Logically, it is expected that
information about what the buyers do not know much about, as with a novel product, should be
more helpful than telling the buyers what they presumably already know, as with an established
one. Previous research agrees that this is likely the case (Berger 2013; Wu 2013). The fifth
hypothesis tests if novelty impacts positively the helpfulness evaluation of a review:
H5: The perceived helpfulness of a review will be higher among novel products
CONCEPTUAL MODEL
Summarizing the hypotheses, the conceptual model (Figure 1) expands previous research
about what determines the perceived helpfulness of a review (e.g. Mudambi and Schuff 2010;
Pan and Zhang 2011) by integrating into the model also the qualitative information available in
the review.
[Insert Figure 1 about here]
A control variable, the number of previous helpfulness evaluations (Number of
Evaluations) that each review received, was added to Figure 1. This variable is included in the
theoretical model for two main reasons. First, the algorithm employed by Amazon.com to rank
16
the most helpful reviews is based on the aggregated number of evaluations that each review
received. This feature has an “anchoring effect,” leading to more helpful reviews receiving more
votes since they are made more salient for potential buyers (Cao et al. 2011; Wan and Nakayama
2014). Second, previous literature shows that consumers rely on the recommendations of other
online consumers when making purchasing decisions (e.g. Forman et al. 2008; Li and Zhan
2011). This social persuasion can also impact the helpfulness assessment of a review. The
number of previous helpfulness assessments can have an impact on the subsequent evaluations,
as potential customers may follow the recommendations of others regarding the helpfulness
perception of the review (Cialdini and Goldstein 2004; Deutsch and Gerard 1955). Previous
literature also controlled for the number of helpfulness evaluations (Mudambi and Schuff 2010).
The model also includes a weighting criterion, the total number of words included in the
textual portion (Number of Words). Previous research (e.g. Chevalier and Mayzlin 2006)
suggests that the length of the text also affects consumers. Various measures of text length have
been employed, such as the number of characters, the number of words, and the number of
sentences (e.g. Pan and Zhang 2011; Salehan and Kim 2016). The literature on online reviews
also suggests that a positive relationship exists between the length of the textual part and review
helpfulness (Pan and Zhang 2011). This finding would call for the inclusion of this variable in
our model as an independent variable. (This was tested. Adding the number of words as a control
variable instead of as a weight did not significantly affect perceived helpfulness and was not
selected for entry in a stepwise regression. This is important, since it highlights that the high- and
low-level construal words may have a more relevant impact on review helpfulness than the raw
number of words.) In addition to this, the inclusion of the number of words as an independent
variable may bias the analysis, since some e-commerce sites such as Amazon.com do not provide
17
reviewers with the ability to leave the textual portion blank, forcing them to add at least the
number of stars in a textual format –i.e. “five stars.” Figure 2 shows the variables included in the
conceptual model as they were extracted from each of the product reviews in the database:
[Insert Figure 2 about here]
METHODOLOGY
Analyzing the text portion of the online reviews to count the number of high-level and
the number of low-level construal terms in each review was performed with LSA. LSA is a fully
automatic statistical method that determines the similarity of meanings across words (named
“terms” in LSA) and documents (in this case online consumer reviews) by analyzing large bodies
of text. Because LSA applied SVD, terms can be associated with each other even if they do not
co-occur together. It is enough for two terms to both co-occur with a third term to make the
former two terms close in meaning (Landauer et al. 1998a). LSA has been argued to approximate
some aspects of human learning from vast amounts of text, as well as human discrimination of
meaning similarity between words (Landauer et al. 2013). Those characteristics make it
especially suitable for the kind of analysis performed in this study. Another advantage of the
usage of LSA is that it allows for the analysis of a large number of documents, in this case
product reviews, with minimal human intervention. This ability to analyze large amounts of
textual data has been highlighted in previous literature as a relevant area for further research (e.g.
Archak et al. 2011; Ghose et al. 2012). The application of LSA in this study was to identify
factors that best describe high- and low-construal terms.
18
The dataset analyzed consisted of 759 product reviews relating to 16 types of light bulbs
available for purchase at Amazon.com (8 incandescent light bulb packs and 8 LED light bulbs).
Those reviews were published online from the time the product was released until February 11,
2015. The selection of this specific category is based on three main reasons: (1) lack of
seasonality; (2) lack of heavy marketing activity, e.g. advertising, which reduces potential noise
during the analysis; and (3) fewer reviews than other very popular products, which reduces the
potential impact of other variables associated with the review – such as its posting order. The
two types of products, LED and incandescent light bulbs, differ in that LEDs were relatively new
in the period of the data collection. They were introduced in the market only after 2008 (Energy
2013). All incandescent light bulbs are established, mostly functionally equivalent products.
For each of these 16 products the variables collected were product ID, review ID, review
valence, review text, number of words included in the review text, verified purchase, number of
helpfulness evaluations, and the helpfulness rating of the review. At the time of the data
collection, these 759 reviews were assessed by 1,632 potential buyers on their perceived
helpfulness. Additionally, a helpfulness score was calculated for each review by summing the
number of positive votes minus the number of negative votes. For instance, a review with “1 of 3
people found this helpful” received a score of -1 (+1 for one potential buyer assessing the review
as helpful, -2 for two potential buyers assessing the review as non-helpful). See the Review
Helpfulness clause in Figure 2.
Product description provided by the seller/manufacturer was analyzed to identify the
high-level type of information included in the textual portion of reviews (as shown in figure 3).
[Insert Figure 3 about here]
19
Stop words (such as “the” or “a”), numbers, and punctuation were removed from the
body of the text. Then, a DTM was constructed from the remaining words in the product
description through a standard LSA function in R. Instead of working with raw frequencies, LSA
employs a transformation called “weighting.” The goal of this transformation is to significantly
reduce the influence of very frequent words that carry little meaning and give more weight to
those that convey more information (Landauer et al. 2013). The default most common weighting
approach was applied (Dumais 2004). This approach entails the multiplication of the log of the
local term frequencies (localWeight) by the inverse of the entropy of the word in the text body
(globalWeight):
;
Where:
i = term number
j = document number
wordLocalFrequency = number of appearances of term i in document j
wordGlobalFrequency = total number of appearances of term i in all the documents
ncontexts = total number of documents
From this transformed DTM, a pool of potential high-level words were obtained for
selection following the criterion described in Appendix A. After applying this selection process,
a total of 33 words were ascribed to the high-level representations category. ‘High-Level
20
Construals’ in the conceptual model is the sum of the number of times that any of these 33 words
appeared in the textual portion of each review.
To identify the low-level construal terms, the same transformed DTM was calculated
over the textual portion of the reviews – as the first step of the process in Appendix A. After
applying this process, a total number of 131 terms were selected. ‘Low-Level Construals’ in the
conceptual model is the number of times that any of those 131 terms appeared in the textual
portion of the review.
The following GLM was run to test the research hypotheses:
(1)
Where:
y = Helpfulness score
x1 = Number of terms ascribed to low-level construals in the review
x2 = Number of terms ascribed to high-level construals in the review
x3 = Number of stars
x4 = Verification of purchase: yes = 1/ no = 0. This is a classification variable
x5 = Product novelty: LED = 1 / incandescent = 0. This is a classification variable
x6 = Number of Evaluations. This is a control variable
weight = Number of words in the textual portion of the review
ANALYSIS
21
Descriptive statistics are shown in Table 1. Model 1 shows analysis results for all the
data, Model 2 for incandescent bulbs only, and Model 3 for LED only. Table 1 shows that LED
reviews were more helpful, had more words, and had both more high- and low-level construal
terms, but fewer stars. LED, however, had fewer confirmed purchases.
[Insert Table 1 about here]
The research model in Figure 1 was tested with GLM, adding the number of words as a
weight and specifying that product novelty and purchase verification are classification variables
(i.e. are nominal rather than real numbers). Results are shown in Table 2 as they refer to the three
models in Table 1. To verify the analyses, the models were run in a linear regression model too,
producing equivalent results. The Durbin Watson statistic at 1.983 was insignificant. Significant
coefficients in Table 2 are emphasized in bold. The results of the linear regression with
standardized betas (to enable comparison among the independent variables) is shown as Model 4
in Table 2. Note that the standardized betas of the number of low-level construal terms
(standardized β=.09), high-level construal terms (standardized β=-.12), and the number of stars
(standardized β=.12) are close in size, suggesting that construal terms are as important as the
number of stars. (The interactions of the verification of purchase and product novelty with lowlevel construals, high-level construals, and number of stars were also tested. None of these
interactions were significant, and none were selected automatically for entry by stepwise
regression.)
[Insert Table 2 about here]
The GLM analyses shows a nuanced picture. The more low-level construal terms there
were in the review, the more helpful the review was, supporting H1. But this is the case only
thanks to the incandescent bulb category. This suggests that potential buyers reading the reviews
22
about novel products are not very interested about the experience of other buyers with those
products, but they are interested when it concerns established products. H2, that having more
high-level construal terms in the report will increase its helpfulness score, was not supported.
The coefficient is significant and very significant in the incandescent category but with a
negative rather than positive sign. Contrary to H2, the data indicates that the more the number of
high-level construal terms there are, the less helpful the review is. It seems that buyers reading
the review are not interested in other buyers confirming what the manufacturer says, especially
in the case of established products. Buyers reading the review are behaving as if they are treating
the inclusions of such high-level construal terms as a waste of their time, reducing review
helpfulness. H3, that more stars will increase helpfulness, was supported in both categories. H4,
that verification of purchase will increase helpfulness, was only supported in the LED category.
Buyers reading the review are behaving as if to say that they know people buy incandescent light
bulbs. Therefore, that verification in the case of incandescent light bulbs adds no new
information and is insignificant. Overall, reviews are more helpful if the product being reviewed
is an LED as shown in Table 1. According to Table 2, H5 is not statistically supported. As
expected, reviews with more evaluations (the control variable) were assessed as more helpful.
A post hoc analysis was then run to verify the implication of the study that it is the
number of high- and of low-level construal terms, rather than the numbers of words altogether,
that determines helpfulness. In that post hoc analysis a stepwise linear regression was run with
all the terms in Model 1 and including beyond those also the number of words. The stepwise
regression was significant (F7,751 =133.1, p-value<.0001, R2=.55). Using the default entry and
retaining values of .05, the stepwise procedure included only the number of stars (β=.44, std.
error=.08, p-value<.0001), supporting H3, the control number of evaluations (β=.63, std.
23
error=.02, p-value<.0001), and the number of low-level construal terms (β=.08, std. error=.03, pvalue=.005), supporting H1. The variable number of words was not chosen by the stepwise
procedure to be entered.
DISCUSSION
Online reviews are clearly important to consumers and their helpfulness score does
increase as their number of stars increases as Pan and Zhang (2011) suggested. Verification of
purchase is also an important contributor to making the review more helpful, but apparently in
our data only with regard to the novel category of LED bulbs. Adding beyond previous studies of
online reviews, this study shows that also the written component of the online reviews can
contribute to their helpfulness, but in a perhaps less than obvious manner. Classifying the written
portion of the reviews into low- and high-level construal terms (through terms that reflect how
other customers employed a product and their experiences with it versus terms that replicate
what the manufacturers say, respectively), shows that low-level construal terms contribute to the
assessed helpfulness of the review but only with the established product category. High-level
construal terms, in contrast, seem to be detrimental in making the review helpful, and
significantly so with the established product category. Buyers assessing the helpfulness of the
review are behaving as if online reviews that replicate terms that the manufacturers use to
describe the product is counterproductive to making the reviews helpful, possibly indicating that
such detail may even be a waste of their time.
The importance of the number of low-level construal terms, rather than the overall
number of words, as shown in the post hoc analysis, suggests that it is not the length of the
24
review that makes it helpful but the type of terms used in it. Moreover, the classification of the
written portion of the reviews into low- and high-level construal terms adds insight into what
increases helpfulness, but notably, also into what detracts from making a review helpful.
Apparently, some kinds of information, in this case identified as high-level construal terms, are
at least in these data, of no value and potentially even be detrimental to making the review
helpful. This suggests that encouraging reviewers to describe specific experiences during the
whole purchase process should be considered. People access online reviews because they want to
learn from other consumers, but, apparently, selectively. Therefore, telling people who access
online reviews about consumer experiences is being helpful. Telling them what they already
know, or at least what they can learn directly from the manufacturer, is, at least in these data,
counterproductive. It makes the review less helpful.
Contribution to Theory
Online reviews are clearly helpful. Discovering what exactly makes them helpful is the
topic of this study. Specifically, what is the contribution of high- and of low-level construal
terms embedded in the review to make it helpful, and presumably how do they influence the
decision-making process of other interested customers? And, relatively, how important are the
high- and low-level construal terms in the textual part of the review relative to its numerical stars
rating? The approach this study took, analyzing low- and high-level construal terms rather than
number of words as done by previous research, sheds new light on these questions. Putting this
approach into context, past research that looked at the written portion of online reviews
examined text length (Pan and Zhang 2011), the impact of linguistic styles (Ludwig et al. 2013),
the role of emotions (e.g. Felbermayr and Nanopoulos 2016; Li and Zhan 2011), and the
influence of very specific features of the product included in the textual part of the review
25
(Archak et al. 2011). Also splitting review texts into a consumer experience category and a
product features category has been done before (e.g. Li et al. 2013). What is new in the approach
taken in this study is that this categorization is centered on construal terms, and it is objective
and numeric. This duality of the textual portion fits with CLT. While the application of CLT is
not novel in marketing – especially within consumer behavior research (e.g. Irmak et al. 2013;
Kim et al. 2009; Trope et al. 2007) – this theory could possibly by applied to shed light also to ecommerce research. Indeed, among the online review studies listed in the Appendix B, only Jin
et al. (2014) included CLT in their theoretical framework. Using CLT to classify terms into lowlevel construal terms (related to the usage experience) and into high-level construal terms
(related to the features of the product), and assessing essentially equivalent functionality over
novel and established products, adds new insight into what makes reviews helpful. For example,
that, at least in these data, novel products have almost double the number of low-level construal
terms than the established product category (see Table 1) and almost triple the number of highlevel construal terms. The even higher ratio of high-level construal terms in novel products may
be because information about features and characteristics may be easier to convey with fewer
terms, due to their more objective nature (such as the color temperature of the light being
'2700K'). Information on how previous buyers used the product, the description of the usage
context, and other additional experiences, such as how long it took the package to be delivered,
will be more varied and hence demand more words.
This new approach to assessing review helpfulness expands previous research on the
topic. The results suggest that is not only the “raw” number of words, i.e. how much is written,
that may increase the helpfulness perception of the review – an approach common in the early
literature (e.g. Chevalier and Mayzlin 2006) – but, rather, what is written as reflected by the
26
construal terms. It is not that previous research did not recognize the importance of what versus
how much, but rather that apparently previous research did not have the tools to objectively do
so. Indeed, Filieri (2015) suggested that potential buyers are more influenced by the quality of
the information (information quality being defined by information depth and breadth, relevance,
credibility, and factuality) than by information quantity. Mudambi and Schuff (2010) also
suggested that information depth is an important predictor of review helpfulness. Cao et al.
(2011) found that semantic characteristics (related to the ‘substance’ of the text) are the most
important in determining helpfulness votes. The addition of high- and low-level construal terms
allows the analysis of the text portion through perhaps more nuanced lenses. This new
perspective might also apply to the emerging literature in marketing that diverts from employing
just review ratings – such as the number of stars – in the analyses (Wang et al. 2015; Wu et al.
2015). Review ratings have been widely employed in previous literature – for a meta-analysis on
the impact of the number of stars and the volume of reviews on sales see Floyd et al. (2014).
Nevertheless, recent literature reports important shortcomings of relying solely on these reviews
ratings (De Langhe et al. 2016).
The analyses presented suggest that an intrinsic attribute of the product, its novelty, does
not, by itself, increase the assessed helpfulness of the review as suggested. The novelty of the
product, however, does determine what specific informational elements online buyers rely on
when forming an idea of the performance and quality of a product. For established products,
low-level construal terms and the number of stars increase helpfulness. For novel products, the
number of stars, the verification of purchase, and the number of previous evaluations are more
indicative than the remaining variables. Possibly, consumers of novel products are less certain
what to look for in the written section of online reviews.
27
Previous research recognized the importance of providing product information for novel
products (e.g. Wasson 1978; Wind and Robertson 1983) in order to overcome the perceived risk
that is due to the lack of knowledge about the product (Hawkins and Mothersbaugh 2010).
Research also acknowledges that consumers’ interest switches from different aspects of
information depending on whether the product is a novel or an established one (Park and Kim
2008). To the best of our knowledge, this is the first attempt to address the inquiry from King et
al. (2014) on how the textual content relates to numerical ratings. In our data, for established
products, more stars and more terms that describe the experiences of previous customers
increases the helpfulness ratings of the review. Both variables move in the same direction.
However, more stars and fewer terms describing features and characteristics significantly
increase helpfulness evaluations. These two variables move in opposing directions. This is an
interesting finding. High-level type of information has been traditionally provided by vendors
and manufacturers through advertising or other forms of business-to-consumers communications.
The inclusion of these terms, referring essential features and characteristics, may not be helpful
in corroborating claims made by manufacturers as suggested. Conversely, the inclusion of this
high-level type of information may signal some kind of manipulation of the review from the
vendor or manufacturer (Hu et al. 2011a; Hu et al. 2011b). Potential customers may expect
reviewers to express their personal experiences with the product (Hu et al. 2012) rather than
report features, as vendors and manufacturers usually do. In the case of novel products, neither
low-level nor high-level construal terms contributed to increase review helpfulness. A potential
explanation of this finding is that the lack of knowledge about the product may hinder the ability
to assess what information is helpful in making purchase decisions. Potential buyers are exposed
to concepts and episodes through product reviews. These concepts and episodes gain depth of
28
meaning by becoming associated with additional concepts and episodes. The process of building
knowledge structures around a product depends on the exposure of the reader to enough
information. Therefore, acquiring the ability to assess the helpfulness of the information may
require them to build enough knowledge about the product (Hawkins and Mothersbaugh 2010;
Piaget 2001). As Table 2 suggests, in the case of novel products, potential customers rely more
on numerical information such as stars or the number of previous helpfulness evaluations. In the
case of established products, online customers value more complex information, such as lowlevel construals probably due to a better knowledge of what messages can be considered
relevant.
The novelty of the product is also relevant when assessing the impact of other
informational elements of individual reviews, such as the verification that the reviewer
purchased the product. The literature on e-WOM has found contrasting results regarding the
influence of different aspects of source credibility. Filieri (2015) suggested that the review will
be perceived as more helpful if the information source is considered credible. Willemsen et al.
(2011) found a weak positive relationship between the perception of helpfulness of the
information and the claims of expertise of reviewers. Other research found no impact of source
credibility over perceived information helpfulness (Cheung et al. 2008). The inclusion of
novelty, may help explain those contrasting results. The analysis section shows that the
helpfulness scores of the novel category improved significantly with the verification that the
reviewer purchased the product. Potential customers may rely more heavily on the information
provided by verified reviewers. They probably assess such information as more credible.
Contribution to Methodology
29
LSA is a natural language processing tool that can factor statistically influential terms
across large bodies of text. CLT is a popular theory in the social sciences that classifies
information into high- and low-construal terms. Adding CLT to LSA can, as done in this study,
guide the classification of terms into terms dealing with the product and terms dealing with other
experience. This methodological framework might have potential beyond the context of online
consumer reviews and e-commerce. Social scientists may be able to apply this methodology to
study considerably larger bodies of text than the traditional qualitative approaches, especially as
it requires no human coders or raters.
The application of LSA as well as other natural language processing techniques for text
analysis offers new avenues of research within online contexts. A similar methodology as the
one presented in this paper could be applied to analyze other sources of online data, such as
social media or specialized blogs data. Likewise, LSA can provide new insights about the
learning process from textual information online. LSA has been argued to approximate how
people learn from text and how they discern the similarity of meaning among words (Landauer et
al. 2013). LSA may be adequate to simulate ‘what’ potential buyers can learn from the textual
data available online, e.g. from product reviews. This research avenue differs from previous
studies in marketing that focused more on ‘how’ consumers learn from online reviews (Zhao et
al. 2013). As a potential application of this simulation, LSA could provide additional insights on
how to develop better ways to overcome information overload in online settings.
Limitations and Avenues for Future Research
The data analyzed contained 16 products from two different categories that were
deliberately chosen because they are essentially functionally equivalent: both product categories
are used for the same purpose of producing artificial light. As research suggests that product type
30
may play an important role in determining the perceived helpfulness of online reviews (e.g.
Moore 2015; Mudambi and Schuff 2010; Pan and Zhang 2011), replicating the study with a
broader scope of products may reveal new insight. This may be especially relevant in the case of
low-level construal terms, which are highly contextualized (Trope and Liberman 2003). What is
perceived as helpful within a specific context may not be perceived as such within a different
one. In this sense, marketers may be interested in characterizing and making more accessible
those reviews that are considered more helpful. Marketers could also be interested in making less
helpful reviews less accessible or even deleting them. A potential way to achieve this goal is
through an extensive characterization of how the informational elements impact the perceived
helpfulness of a review across different product categories.
Conclusions
New methodological approaches are needed to achieve a comprehensive understanding of
the role of the different informational elements available online, especially in the case of textual
data that, nowadays, is ubiquitous for online users. Textual data is readily available to influence
the purchase intentions of potential customers anytime. In the U.S. just mobile connections to the
Internet generate more than 18 million megabytes of data every minute (James 2016). This
research proposes and illustrates how LSA, in connection with a theoretical framework, provided
in this case by CLT, can advance the current marketing methodological approaches to achieve
that comprehensive approach. The method proposed in this study can identify some aspects of
what information from reviews is considered important and when it is important. This could
open new avenues into the study of online reviews, determining what information is important
and when.
31
REFERENCES
Alba, J., J. Lynch, B. Weitz, C. Janiszewski, and et al. (1997), "Interactive home shopping: Consumer,
retailer, and manufacturer incentives to participate in electronic marketplaces," Journal of Marketing,
61 (3), 38-53.
Alex, S.L.T. and G. Prendergast (2009), "Is a "star" worth a thousand words?," European Journal of
Marketing, 43 (11/12), 1269-80.
Anderson, E. T. and D. I. Simester (2014), "Reviews Without a Purchase: Low Ratings, Loyal Customers,
and Deception," Journal of Marketing Research, 51 (3), 249-69.
Archak, N., A. Ghose, and P.G. Ipeirotis (2011), "Deriving the Pricing Power of Product Features by
Mining Consumer Reviews," Management Science, 57 (8), 1485-509.
Berger, J. (2013), Contagious. New York, NY: Simon & Schuster.
Bickart, B. and R.M. Schindler (2001), "Internet forums as influential sources of consumer information,"
Journal of Interactive Marketing, 15 (3), 31-40.
Biswas, D. (2004), "Economics of information in the Web economy: Towards a new theory?," Journal of
Business Research, 57 (7), 724-33.
Brown, J.J. and P.H. Reingen (1987), "Social Ties and Word-of-Mouth Referral Behavior," Journal of
Consumer Research, 14 (3), 350-62.
Cao, Q., W. Duan, and Q. Gan (2011), "Exploring determinants of voting for the “helpfulness” of online
user reviews: A text mining approach," Decision Support Systems, 50 (2), 511-21.
Chen, P.-Y., S. Dhanasobhon, and M.D. Smith (2008), "All reviews are not created equal: The
disaggregate impact of reviews and reviewers at amazon. com," Available at SSRN:
http://ssrn.com/abstract=918083.
Chen, Y. and J. Xie (2008), "Online Consumer Review: Word-of-Mouth as a New Element of Marketing
Communication Mix," Management Science, 54 (3), 477-91.
Cheung, C. M. K., M. K.O. Lee, and N. Rabjohn (2008), "The impact of electronic word‐of‐mouth: The
adoption of online opinions in online customer communities," Internet Research, 18 (3), 229-47.
32
Chevalier, J.A. and D. Mayzlin (2006), "The Effect of Word of Mouth on Sales: Online Book Reviews,"
Journal of Marketing Research, 43 (3), 345-54.
Chintagunta, Pradeep K., Shyam Gopinath, and Sriram Venkataraman (2010), "The effects of online user
reviews on movie box office performance: accounting for sequential rollout and aggregation across local
markets," Marketing Science, 29, 944+.
Cialdini, R.B. and N.J. Goldstein (2004), "Social influence: Compliance and conformity," Annu. Rev.
Psychol., 55, 591-621.
De Langhe, B., P. M. Fernbach, and D. R. Lichtenstein (2016), "Navigating by the Stars: Investigating the
Actual and Perceived Validity of Online User Ratings," Journal of Consumer Research, 42 (6), 817-33.
De Maeyer, P. (2012), "Impact of online consumer reviews on sales and price strategies: a review and
directions for future research," Journal of Product & Brand Management, 21 (2), 132-39.
Dellarocas, C. (2003), "The Digitization of Word of Mouth: Promise and Challenges of Online Feedback
Mechanisms," Management Science, 49 (10), 1407-24.
Deutsch, M. and H.B. Gerard (1955), "A study of normative and informational social influences upon
individual judgment," The journal of abnormal and social psychology, 51 (3), 629.
Dowling, G.R. and R. Staelin (1994), "A Model of Perceived Risk and Intended Risk-Handling Activity,"
Journal of Consumer Research, 21 (1), 119-34.
Dumais, S.T. (2004), "Latent semantic analysis," Annual Review of Information Science and Technology,
38 (1), 188-230.
Energy, US Department of (2013), "The History of the Light Bulb."
Faulhaber, G.R. and D.A. Yao (1989), ""Fly-by-Night" Firms and the Market for Product Reviews," The
Journal of Industrial Economics, 38 (1), 65-77.
Felbermayr, A. and A. Nanopoulos (2016), "The Role of Emotions for the Perceived Usefulness in Online
Customer Reviews," Journal of Interactive Marketing, 36, 60-76.
Filieri, R. (2015), "What makes online reviews helpful? A diagnosticity-adoption framework to explain
informational and normative influences in e-WOM," Journal of Business Research, 68 (6), 1261-70.
33
Floyd, K., R. Freling, S. Alhoqail, H. Y. Cho, and T. Freling (2014), "How Online Product Reviews Affect
Retail Sales: A Meta-analysis," Journal of Retailing, 90 (2), 217-32.
Forman, C., A. Ghose, and B. Wiesenfeld (2008), "Examining the Relationship Between Reviews and
Sales: The Role of Reviewer Identity Disclosure in Electronic Markets," Information Systems Research, 19
(3), 291-313,93-95.
Ghose, A. and P.G. Ipeirotis (2011), "Estimating the Helpfulness and Economic Impact of Product
Reviews: Mining Text and Reviewer Characteristics," IEEE Transactions on Knowledge and Data
Engineering 23 (10), 1498-512.
Ghose, A., P.G. Ipeirotis, and B. Li (2012), "Designing Ranking Systems for Hotels on Travel Search
Engines by Mining User-Generated and Crowdsourced Content," Marketing Science, 31 (3), 493-520.
Godes, D. and D. Mayzlin (2004), "Using online conversations to study word-of-mouth communication,"
Marketing Science, 23, 545+.
Gomez, Juan Carlos, Erik Boiy, and Marie-Francine Moens (2012), "Highly Discriminative Statistical
Features for Email Classification," Knowlege Information Systems, 31, 23-53.
Hawkins, D.I. and D.L. Mothersbaugh (2010), Consumer Behavior. Building Marketing Strategy (11th
ed.). New York, NY: McGraw-Hill/Irwin.
Hennig-Thurau, T., K.P. Gwinner, G. Walsh, and D.D. Gremler (2004), "Electronic word-of-mouth via
consumer-opinion platforms: What motivates consumers to articulate themselves on the Internet?,"
Journal of Interactive Marketing, 18 (1), 38.
Hu, N., I. Bose, Y. Gao, and L. Liu (2011a), "Manipulation in digital word-of-mouth: A reality check for
book reviews," Vol. 50: Elsevier B.V.
Hu, N., I. Bose, N.S. Koh, and L. Liu (2012), "Manipulation of online reviews: An analysis of ratings,
readability, and sentiments," Decision Support Systems, 52 (3), 674-84.
Hu, N., L. Liu, and V. Sambamurthy (2011b), "Fraud detection in online consumer reviews," Decision
Support Systems, 50 (3), 614-26.
Hu, N., J. Zhang, and P.A. Pavlou (2009), "Overcoming the J-shaped distribution of product reviews,"
Commun. ACM, 52 (10), 144-47.
Irmak, C., C.J. Wakslak, and Y. Trope (2013), "Selling the Forest, Buying the Trees: The Effect of Construal
Level on Seller- Buyer Price Discrepancy," Journal of Consumer Research, 40 (2), 284-97.
34
Islam, Aminul, Evangelos Milios, and Vlado Keselj (2012), "Text Similarity using Google Tri-grams," in
25th Canadian Conference on Artificial Intelligence, L. Kosseim and D. Inkpen (Eds.) Vol. LNAI 7310.
Toronto, Canada: Springer.
James, J. (2016), "Data Never Sleeps 4.0," (accessed 09/05/2016, 2016), [available at
https://www.domo.com/blog/2016/06/data-never-sleeps-4-0/].
Jin, Liyin, Bingyan Hu, and Yanqun He (2014), "The Recent versus The Out-Dated: An Experimental
Examination of the Time-Variant Effects of Online Consumer Reviews," Journal of Retailing, 90 (4), 55266.
Kim, Y.-J., J. Park, and R. Wyer (2009), "Effects of Temporal Distance and Memory on Consumer
Judgments," Journal of Consumer Research, 36 (4), 634-45.
King, R.A., P. Racherla, and V.D. Bush (2014), "What We Know and Don't Know About Online Word-ofMouth: A Review and Synthesis of the Literature," Journal of Interactive Marketing, 28 (3), 167-83.
Kohli, R., S. Devaraj, and M. A. Mahmood (2004), "Understanding Determinants of Online Consumer
Satisfaction: A Decision Process Perspective," Journal of Management Information Systems, 21 (1), 11535.
Kuan, K.K.Y., K.-L. Hui, P. Prasarnphanich, and H.-Y. Lai (2015), "What Makes a Review Voted? An
Empirical Investigation of Review Voting in Online Review Systems," Journal of the Association for
Information Systems, 16 (1), 48-71.
Landauer, T.K., P.W. Foltz, and D. Laham (1998a), "An introduction to latent semantic analysis,"
Discourse Proceses, 25 (2-3), 259-84.
Landauer, T.K., D.S. McNamara, S. Dennis, and W. Kintsch (2013), Handbook of latent semantic analysis.
New York, NY: Psychology Press.
Landauer, Thomas K. and Susan T. Dumais (1997), "A Solution to Plato’s Problem: The Latent Semantic
Analysis Theory of Acquisition, Induction, and Representation of Knowledge," Psychological Review, 104
(2), 211-40.
Landauer, Thomas K., Peter W. Foltz, and D. Laham (1998b), "An Introduction to Latent Semantic
Analysis," Discourse Processes, 25 (2&3), 259-84.
Lee, J., J.-N. Lee, and B.C.Y. Tan (2015), "Antecedents of cognitive trust and affective distrust and their
mediating roles in building customer loyalty," Information Systems Frontiers, 17 (1), 159-75.
35
Li, J. and L. Zhan (2011), "Online Persuasion: How the Written Word Drives WOM," Journal of Advertising
Research, 51 (1), 239-57.
Li, M., L. Huang, C-H Tan, and K-K Wei (2013), "Helpfulness of Online Product Reviews as Seen by
Consumers: Source and Content Features," International Journal of Electronic Commerce, 17 (4), 101-36.
Liberman, N. and Y. Trope (1998), "The role of feasibility and desirability considerations in near and
distant future decisions: A test of temporal construal theory," Journal of Personality and Social
Psychology, 75 (1), 5-18.
Liu, Y. (2006), "Word of Mouth for Movies: Its Dynamics and Impact on Box Office Revenue," Journal of
Marketing, 70 (3), 74-89.
Ludwig, S., K. de Ruyter, M. Friedman, E. C. Brüggen, M. Wetzels, and G. Pfann (2013), "More Than
Words: The Influence of Affective Content and Linguistic Style Matches in Online Reviews on Conversion
Rates," Journal of Marketing, 77 (1), 87-103.
Lynch Jr, J.G. and D. Ariely (2000), "Wine online: Search costs affect competition on price, quality, and
distribution," Marketing Science, 19 (1), 83-103.
Mayzlin, D. (2006), "Promotional Chat on the Internet," Marketing Science, 25 (2), 155-63,201.
McFadden, D.L. and K.E. Train (1996), "Consumers' Evaluation of New Products: Learning from Self and
Others," Journal of Political Economy, 104 (4), 683-703.
Metzger, M.J., A.J. Flanagin, and R.B. Medders (2010), "Social and Heuristic Approaches to Credibility
Evaluation Online," Journal of Communication, 60 (3), 413-39.
Moe, W. W. and M. Trusov (2011), "The Value of Social Dynamics in Online Product Ratings Forums,"
Journal of Marketing Research 48 (3), 444-56.
Moore, S. G. (2015), "Attitude Predictability and Helpfulness in Online Reviews: The Role of Explained
Actions and Reactions," Journal of Consumer Research, 42 (1), 30-44.
Mudambi, S.M. and D. Schuff (2010), "What Makes a Helpful Online Review? A Study of Customer
Reviews on Amazon.com," MIS Quarterly, 34 (1), 185-200.
News, BBC (2013), "Samsung probed in Taiwan over 'fake web reviews'," in BBC News Technology.
36
Nielsen (2012), "Consumer Trust in Online, Social and Mobile Advertising Grows," Nielsen.
Pan, Y. and J.Q. Zhang (2011), "Born Unequal: A Study of the Helpfulness of User-Generated Product
Reviews," Journal of Retailing, 87 (4), 598-612.
Park, D.-H. and S. Kim (2008), "The effects of consumer knowledge on message processing of electronic
word-of-mouth via online consumer reviews," Electronic Commerce Research and Applications, 7 (4),
399-410.
Peterson, R.A. , S. Balasubramanian, and B.J. Bronnenberg (1997), "Exploring the Implications of the
Internet for Consumer Marketing " Journal of the Academy of Marketing Science, 25 (4), 329-46.
Piaget, J. (2001), Studies in Reflecting Abstraction. New York, NY: Psychology Press.
Punj, G. and R. Moore (2009), "Information search and consideration set formation in a web-based store
environment," Journal of Business Research, 62 (6), 644-50.
Racherla, P., M. Mandviwalla, and D.J. Connolly (2012), "Factors affecting consumers' trust in online
product reviews," Journal of Consumer Behaviour, 11 (2), 94-104.
Ramirez, A., J.B. Walther, J.K. Burgoon, and M. Sunnafrank (2002), "Information Seeking Strategies,
Uncertainty, and Computer-Mediated Communication. Toward a Conceptual Model," Human
Communication Research, 28 (2), 213-28.
Salehan, M. and D.J. Kim (2016), "Predicting the performance of online consumer reviews: A sentiment
mining approach to big data analytics," Decision Support Systems, 81, 30-40.
Sen, S. and D. Lerman (2007), "Why are you telling me this? An examination into negative consumer
reviews on the Web," Journal of Interactive Marketing, 21 (4), 76-94.
Shapiro, C. and H.R. Varian (1999), Information Rules. Boston, MA: Harvard Business School Publications.
Spool, J.M. (2009), "The Magic Behind Amazon's 2.7 Billion Dollar Question," in User Interface
Engineering: User Interface Engineering.
Stigler, G.J. (1961), "The Economics of Information," Journal of Political Economy, 69 (3), 213-25.
Tan, S.J. (1999), "Strategies for reducing consumers’ risk aversion in Internet shopping," Journal of
Consumer Marketing, 16 (2), 163-80.
37
Trope, Y. and N. Liberman (2003), "Temporal construal," Psychological Review, 110 (3), 403-21.
Trope, Y., N. Liberman, and C. J. Wakslak (2007), "Construal Levels and Psychological Distance: Effects on
Representation, Prediction, Evaluation, and Behavior," Journal of Consumer Psychology, 17 (2), 83-95.
Trope, Y.L. (2012), "Construal Level Theory," in Handbook of Theories of Social Psychology, P. K. Van
Lange, ed. Washington DC: Sage Publications Ltd. .
Valle-Lisboa, Juan C. and Eduardo Mizraji (2007), "The Uncovering of Hidden Structures by Latent
Semantic Analysis," Information Sciences, 177 (19), 4122–47.
Wan, Y. and M. Nakayama (2014), "The reliability of online review helpfulness," Journal of Electronic
Commerce Research, 15 (3), 179.
Wang, F., X. Liu, and E. Fang (2015), "User Reviews Variance, Critic Reviews Variance, and Product Sales:
An Exploration of Customer Breadth and Depth Effects," Journal of Retailing, 91 (3), 372-89.
Wasson, C.R. (1978), Dynamic competitive strategy & product life cycles: Austin Press.
Willemsen, L. M., P. C. Neijens, F. Bronner, and J. A. de Ridder (2011), "“Highly Recommended!” The
Content Characteristics and Perceived Usefulness of Online Consumer Reviews," Journal of ComputerMediated Communication, 17 (1), 19-38.
Wind, Y. and T.S Robertson (1983), "Marketing strategy: New directions for theory and research," The
Journal of Marketing, 12-25.
Wu, C., H. Che, T.Y. Chan, and X. Lu (2015), "The Economic Value of Online Reviews," Marketing Science,
34 (5), 739-54.
Wu, P.F. (2013), "In Search of Negativity Bias: An Empirical Study of Perceived Helpfulness of Online
Reviews," Psychology & Marketing, 30 (11), 971-84.
Zhao, Y., S. Yang, V. Narayan, and Y. Zhao (2013), "Modeling Consumer Learning from Online Product
Reviews," Marketing Science, 32 (1), 153-69.
Zhu, F. and X. Zhang (2010), "Impact of Online Consumer Reviews on Sales: The Moderating Role of
Product and Consumer Characteristics," Journal of Marketing, 74 (2), 133-48.
38
TABLES
Mean
Std
Helpfulness Score
N
Mean
Std
Number of Words
N
Mean
Number of Stars in
Std
the Review
N
Mean
Number of HighStd
Level Construal
Terms in the Review
N
Mean
Number of LowLevel Construal
Std
Terms in the Review
N
Verification of
Purchase
Mean
Std
N
Review is about
Model 1
Model 2
Established
Both Established
Product
and Novel
Category
Product
(Incandescent
Categories
bulbs) only
1.28
0.57
4.51
3.31
759
431
55.18
33.57
77.8
28.08
759
431
4.09
4.26
1.39
1.28
759
431
7.11
4.13
9.92
3.43
759
431
11.57
8.32
12.6
5.19
759
431
0.85
0.35
759
0.96
0.21
431
Model 3
Novel
Product
Category
(LED)
only
2.2
5.58
328
83.59
107.56
328
3.87
1.5
328
11.02
13.63
328
15.84
17.33
328
0.72
0.45
328
Significance levels *** at .001, ** at .01, * at .05, and # at .1
Table 1. Descriptive statistics of the complete dataset
T test
comparing
Incandescent
bulbs with
LED
-4.70***
-8.21***
3.74***
-8.93***
-7.61***
8.93***
39
Model 1
(all the data)
Hypothesis
H1
H2
H3
H4
H5
Contr
ol
Intercept
Number of LowLevel Construal
Terms
Number of
High-Level
Construal Terms
Number of Stars
Verification of
Purchase
(Bought = 1)
Product Novelty
(LED = 1)
Number of
Evaluations
R square
-2.24(.41)***
Model 2
Model 3
(only
(only LED
incandescent
bulbs)
bulbs)
Estimate (Std. Err)
-11.02(3.66)**
-2.70(.86)**
Model 4
Standardized
Linear
Regression
.02(.01)*
1.25(.13)***
.01(.03)
.09*
-.04(.01)**
-1.03(.24)***
-.05(.04)
-.12**
.57(.08) ***
3.33(.52)***
1.21(.20)***
.12***
.68(.28)*
4.34(3.21)
2.85(.65)***
.05*
.40(.26)
.03
.77(.01)***
-.16(.07)*
.82(.03)***
.91**
.83
.55
.87
.83
Significance levels *** at .001, ** at .01, * at .05, and # at .1
Table 2. GLM and Standardized Linear Regression Analyses Results
40
FIGURES
Figure 1. Conceptual model
41
Figure 2. Variables extracted from reviews included in the conceptual model
42
Figure 3. Operationalization of high-level construal terms
43
APPENDIX A
The high- and low-level construal word selection criterion is based on a tradeoff between
the number of words selected and the average importance of those words. The construal words
selection follows an 11-step process.
Step (1), in addition to the transformed DTM (DTM1) already constructed to identify
potential high-level words, a new transformed DTM (DTM2) was constructed over the textual
portion of the 759 reviews following the same methodology as in the case of the potential highlevel construal words. Those words that are found in both matrices (DTM1 and DTM2) become
the pool of potential high-level words. All others in the review DTM2 are the pool of low-level
terms. Step (2), a reduced-rank SVD was calculated over DTM2. SVD retains the k-largest
singular values of this matrix and the remainders are set to 0. This results in a reduced-dimension
SVD representation, which is the best k-dimensional approximation to the original matrix
employing least-squares. After applying SVD, each document and each term is represented as a
k-dimensional vector in a semantic space (specifically in this research, as a vector of 149
dimensions).
Step (3), similarities and associations were then computed on this reduced-dimension
semantic space. Since both documents and terms are represented as vectors in this space,
document-document, term-term, and document-term similarities and associations can be
calculated (Dumais 2004). Step (4), from this semantic space, cosine measures of “closeness in
meaning” were computed by collecting the 20 terms most close to every one of the original 759
reviews. 1,497 terms were obtained – with their corresponding appearance frequencies in each
44
review. Step (5), the total frequency for each word i is calculated over each document j for a total
of r documents:
Where:
x = frequency of word i in document j
Step (6), the i rows are ordered in descending order by Fi. Step (7), to obtain the average
importance of each word across the 759 reviews R and in each of the 16 product descriptions D,
we calculated the average importance of word k in matrix M as:
Where:
y = estimate of how many times a word would occur in each of an infinite sample of reviews
(Landauer et al. 1998a) (each cell value in the semantic space)
k = word in the semantic space (row in the semantic space)
l = documents in the semantic space (column in the semantic space). They vary between 1 and t
t = total number of documents in the semantic space (r in M = R, d in M = D)
M = semantic space matrix (R = reviews, D = descriptions)
Choosing words for inclusion in the construal words set is based on the average
cumulative importance of these words, following the frequency ordering obtained in step (6).
Step (8), to choose a cutoff point for the average of words to retain given the importance of the
words, we inspect a graph of the cumulative average importance over the full set of potential
construal words.
45
Cumulative Average Importance
Cumulative Average Term Importance
2
1.5
1
0.5
0
0
200
400
600
800
1000
1200
1400
1600
Number of Terms in Construal Set
Upon inspection of the previous figure, a cutoff point of .5 is chosen. We seek to include
only those words with an average cumulative importance greater than .5. Step (9), maintaining
the ordering found in step (6), we calculate the average cumulative importance at each unique
frequency level in the set of potential construal terms. Step (10), for each frequency level, we
calculate the cumulative number of words included in the construal set and the cumulative
average importance for the two matrices. Step (11), we take the average importance at each
unique frequency level.
Our inclusion pool is based on the average of the cumulative average importance for the
unique frequency levels from matrix R and the cumulative average importance for the matching
words found in matrix D. All words that have a unique frequency level where this average is
greater than or equal to .5 are included. In this case, we find that the minimum frequency of the
included review words is 25, the resulting number of words is 164 and the breakdown of highlevel and low-level construal words is 33 and 131 respectively.
46
APPENDIX B
Authors
(year)
Vol.
(Issue)
Key Findings
Sample
Journal of
Marketing
Research
51 (3)
Deceptive reviews may not be limited to the strategic actions of firms. Instead, the phenomenon
may be far more prevalent, extending to individual customers who have no financial incentive to
influence product ratings
330,975 reviews from a
prominent retailer
Management
Science
57 (8)
This paper demonstrates how textual data can be used to learn consumers’ relative preferences for
different product features and also how text can be used for predictive modeling of future changes in
sales
Sales data and review data
for digital cameras and
camcorders from Amazon
Establishing Trust in
Electronic Commerce through
Online Word of Mouth: An
Examination across Genders
Finding disseminators via
electronic word of mouth
message for effective
marketing communications
Exploring determinants of
voting for the “helpfulness”
of online user reviews: A text
mining approach
Journal of
Management
Information
Systems
24 (4)
The effect of trust on intention to shop online is stronger for women than for men. Men value their
ability to post content online, whereas women value the responsive participation of other consumers
to the content they have posted. Online word-of-mouth quality affects online trust differently across
genders
Decision
Support
Systems
67
The authors identify three types of opinion leaders through e-WOM using a message-based
approach. They demonstrate that e-WOM of opinion leaders drives product sales due to their
product experience and knowledge background
Decision
Support
Systems
50 (2)
The findings show that the semantic characteristics are more influential than other characteristics in
affecting how many helpfulness votes reviews receive. The findings also suggest that reviews with
extreme opinions receive more helpfulness votes than those with mixed or neutral opinions
Chen, C. C., &
Tseng, Y.-D.
(2011)
Quality evaluation of product
reviews using an information
quality framework
Decision
Support
Systems
50 (4)
The authors treat the evaluation of review quality as a classification problem and employ an
effective information quality framework to extract representative review features
Chen, Y., & Xie,
J. (2008)
Online Consumer Review:
Word-of-Mouth as a New
Element of Marketing
Communication Mix
Management
Science
54 (3)
The results reveal that if the review information is sufficiently informative, the seller-created
product attribute information and buyer-created review information will interact with each other
Survey focused on
asynchronous online
forums, 1,561 usable
responses
Amazon review dataset.
Sample of 350,122 book,
music, video and DVD
titles
Data from CNET
Download.com for
software programs in
Enterprise Computing
The first 150 reviews were
collected for ten digital
cameras and ten mp3
players on Amazon
Third-party reviews from
CNET.com, consumer
reviews from
Amazon.com, product
attributes from CNET.com
Decision
Support
Systems
53 (1)
The authors focus on the factors that drive consumers to spread positive eWOM in online consumeropinion platforms. They identified a number of key motives of consumers' eWOM intention and
developed an associated model
203 members of a
consumer review
community
Decision
Support
Systems
54 (1)
The authors used the social communication framework to summarize and classify prior eWOM
studies. They identified key factors related to the major elements of the social communication
literature and built an integrative framework explaining the impact of eWOM communication on
consumer behavior
No empirical test
Journal of
Marketing
Research
43 (3)
Reviews are overwhelmingly positive; an improvement in a book's reviews leads to an increase in
relative sales at that site; the impact of one-star reviews is greater than the impact of five-star
reviews; and customers read review text rather than relying only on summary statistics
Book characteristics and
user reviews from
Amazon.com and bn.com
Anderson, E. T.,
& Simester, D. I.
(2014)
Archak, N.,
Ghose, A., &
Ipeirotis, P. G.
(2011)
Awad, N. F., &
Ragowsky, A.
(2008)
Bao, T., &
Chang, T. S.
(2014)
Cao, Q., Duan,
W., & Gan, Q.
(2011)
Cheung, C. M.
K., & Lee, M. K.
O. (2012)
Cheung, C. M.
K., & Thadani,
D. R. (2012)
Chevalier, J. A.,
& Mayzlin, D.
(2006)
Title
Journal
Reviews Without a Purchase:
Low Ratings, Loyal
Customers, and Deception
Deriving the Pricing Power of
Product Features by Mining
Consumer Reviews
What drives consumers to
spread electronic word of
mouth in online consumeropinion platforms
The impact of electronic
word-of-mouth
communication: A literature
analysis and integrative
model
The Effect of Word of Mouth
on Sales: Online Book
Reviews
47
Chintagunta, P.
K., Gopinath, S.,
&
Venkataraman,
S. (2010)
Clemons, E. K.,
Gao, G. G., &
Hitt, L. M.
(2006)
De Langhe, B.,
Fernbach, P. M.,
& Lichtenstein,
D. R. (2016)
de Matos, C., &
Rossi, C. (2008)
Dellarocas, C.
(2003)
Dellarocas, C.
(2006)
Dellarocas, C.,
Gao, G., &
Narayan, R.
(2010)
Duan, W., Gu,
B., Whinston,
A.B. (2008)
Duhan, D. F.,
Johnson, S. D.,
Wilcox, J. B., &
Harrell, G. D.
(1997)
Floyd, K.,
Freling, R.,
Alhoqail, S.,
Cho, H. Y., &
Freling, T.
(2014)
Ghose, A.,
Ipeirotis, P. G.,
& Li, B. (2012)
The effects of online user
reviews on movie box office
performance: accounting for
sequential rollout and
aggregation across local
markets
Marketing
Science
When Online Reviews Meet
Hyperdifferentiation: A Study
of the Craft Beer Industry
Navigating by the Stars:
Investigating the Actual and
Perceived Validity of Online
User Ratings
Word-of-mouth
communications in
marketing: a meta-analytic
review of the antecedents and
moderators
The Digitization of Word of
Mouth: Promise and
Challenges of Online
Feedback Mechanisms
Strategic Manipulation of
Internet Opinion Forums:
Implications for Consumers
and Firms
Are Consumers More Likely
to Contribute Online Reviews
for Hit or Niche Products?
Do online reviews matter? —
An empirical investigation of
panel data
29 (5)
The authors find that it is the valence of reviews that seems to matter and not the volume. When the
authors carry out the analysis with aggregated national data, they obtain the same results as previous
studies that volume matters but not the valence. Using various market-level controls in the national
data model, the authors attempt to identify the source of this difference
Daily box office ticket
sales data for 148 movies
along with user review
data from Yahoo! Movies
Journal of
Management
Information
Systems
23 (2)
They find that the variance of ratings and the strength of the most positive quartile of reviews play a
significant role in determining which new products grow fastest in the market-place
281,868 ratings for 1,159
U.S. craft brewers from
6,212 reviewers from
Ratebeer.com
Journal of
Consumer
Research
42 (6)
Journal of the
Academy of
Marketing
Science
36 (4)
Management
Science
49 (10)
Management
Science
52 (10)
Journal of
Management
Information
Systems
Decision
Support
Systems
Average user ratings (1) lack convergence with Consumer Reports scores, (2) are often based on
insufficient sample sizes which limits their informativeness, (3) do not predict resale prices in the
used-product marketplace, and (4) are higher for more expensive products and premium brands,
controlling for Consumer Reports scores
WOM valence is a significant moderator, cross-sectional studies show a stronger influence of
satisfaction and loyalty on WOM activity than longitudinal studies, studies of WOM behavior show
a weaker link between loyalty and WOM activity than studies of WOM intentions, and satisfaction
has a stronger relationship with positive WOM than loyalty, whereas (dis)loyalty has a stronger
relationship with negative WOM than does (dis)satisfaction
The paper provides an overview of relevant work in game theory and economics on the topic of
reputation. It discusses how this body of work is being extended and combined with insights from
computer science, management science, sociology, and psychology to take into consideration the
special properties of online environments
The authors' result implies the existence of settings where online forum manipulation benefits
consumers. If the precision of honest consumer opinions that firms manipulate is sufficiently high,
firms of all types, as well as society, would be strictly better off if manipulation of online forums
was not possible
27 (2)
The authors' findings suggest that online forum designers who wish to increase the contribution of
user reviews for lesser-known products should make information about the volume of previously
posted reviews a less-prominent feature of their sites
45 (4)
The result shows that online user reviews have little persuasive effect on consumer purchase
decisions. Nevertheless, the authors find that box office sales are significantly influenced by the
volume of online posting, suggesting the importance of awareness effect
1272 products across 120
vertically differentiated
product categories
127 studies, which
produced 162 independent
samples, 348 effect sizes,
and a cumulated N of
64,364 subjects
eBay's Case Study
Econometric Model
Reviews posted on
Yahoo! Movies,
production and weekly
box office
Data from Yahoo!
Movies, Variety.com, and
BoxOfficeMojo.com
Influences on consumer use
of word-of-mouth
recommendation sources
Journal of the
Academy of
Marketing
Science
25 (4)
The authors present a model that proposes two routes of influence on the choice of recommendation
sources. The model proposes that the prior knowledge level of the consumer, the perceived level of
task difficulty, and the type of evaluative cues sought by consumers influence their choice of
recommendation sources
245 questionnaires
How Online Product Reviews
Affect Retail Sales: A Metaanalysis
Journal of
Retailing
90 (2)
Online product reviews have a significant impact on sales elasticity. Our research also provides
interesting insights about important variables which augment or diminish the influence that online
product reviews exert on retailer performance
Meta-analysis of 26
empirical studies yielding
443 sales elasticities
Designing Ranking Systems
for Hotels on Travel Search
Engines by Mining UserGenerated and Crowdsourced
Content
Marketing
Science
31 (3)
The authors propose to generate a ranking system that recommends products that provide, on
average, the best value for the consumer’s money
U.S. hotel reservations
through Travelocity,
supplemented with data
from social media sources
48
Godes, D., &
Mayzlin, D.
(2004)
Using online conversations to
study word-of-mouth
communication
Marketing
Science
23 (4)
Godes, D., &
Silva, J. C.
(2012)
Sequential and Temporal
Dynamics of Online Opinion
Marketing
Science
31 (3)
Decision
Support
Systems
Hu, N., Bose, I.,
Koh, N. S., &
Liu, L. (2012)
Hu, N., Koh, N.
S., & Reddy, S.
K. (2014)
Huang, L., Tan,
C.-H., Ke, W., &
Wei, K.-K.
(2013)
Jin, L., Hu, B., &
He, Y.
(2014)
Khare, A.,
Labrecque, L. I.,
& Asare, A. K.
(2011)
Manipulation of online
reviews: An analysis of
ratings, readability, and
sentiments
Ratings lead you to the
product, reviews help you
clinch it? The mediating role
of online review sentiments
on product sales
Comprehension and
Assessment of Product
Reviews: A Review-Product
Congruity Proposition
The Recent versus The OutDated: An Experimental
Examination of the TimeVariant Effects of Online
Consumer Reviews
The Assimilative and
Contrastive Effects of Wordof-Mouth Volume: An
Experimental Examination of
Online Consumer Ratings
Online conversations may offer an easy and cost-effective opportunity to measure word of mouth.
The authors show that a measure of the dispersion of conversations across communities has
explanatory power in a dynamic model of TV ratings
The authors find that, once controlling for calendar date, the residual average temporal pattern is
increasing. With respect to sequential dynamics, they find that
ratings decrease: the nth rating is, on average, lower than the n−1th when controlling for time,
reviewer effects, and book effects
44 TV US shows, data
from Nielsen ratings, and
Usenet newsgroup
Book review data gathered
from Amazon.com.
Reviews for each of the
top 350 selling titles
52 (3)
The authors discover that around 10.3% of the products are subject to online reviews manipulation.
Consumers are only able to detect manipulation taking place through ratings, but not through
sentiments
The dataset consisted of
4490 books, with 610,713
online reviews
Decision
Support
Systems
57
Ratings do not have a significant direct impact on sales but have an indirect impact through
sentiments. Sentiments, however, have a direct significant impact on sales. The two most accessible
types of reviews – most helpful and most recent – play a significant role in determining sales
Panel of data on over 4000
books from Amazon.com
Journal of
Management
Information
Systems
30 (3)
The two matching conditions, (1) attribute-based reviews describing a search product and (2)
experience-based reviews describing an experience product, could lead consumers to perceive
higher review helpfulness and lower cognitive effort to comprehend the reviews
Experimental settings
Journal of
Retailing
90 (4)
Although recent online reviews are more influential in shifting consumer preferences towards nearfuture consumption decisions, the relative influence of out-dated online reviews in shifting
consumer preferences increases when consumers are making distant-future consumption decisions
Experimental Settings
Journal of
Retailing
87 (1)
High volume accentuates or assimilates perceptions of positivity or negativity of WOM targets.
Depending upon how high volume interacts with WOM consensus and consumer decision
precommitment, it can contrast preference away from the valence of a target also. Consumers differ
in their susceptibility to the influence of high volume
Experimental Settings
Data from 83 blogs,
representing
approximately 1,376,000
words and 6722 postings
Kozinets, R. V.,
de Valck, K.,
Wojnicki, A. C.,
& Wilner, S. J. S.
(2010)
Networked Narratives:
Understanding Word-ofMouth Marketing in Online
Communities
Journal of
Marketing
74 (2)
The findings indicate that the network of communications offers four social media communication
strategies—evaluation, embracing, endorsement, and explanation. Each is influenced by character
narrative, communications forum, communal norms, and the nature of the marketing promotion
Liu, Y. (2006)
Word of Mouth for Movies:
Its Dynamics and Impact on
Box Office Revenue
Journal of
Marketing
70 (3)
WOM activities are the most active during a movie's prerelease and opening week and that movie
audiences tend to hold relatively high expectations before release but become more critical in the
opening week. WOM information offers significant explanatory power for box office revenue, most
comes from the volume of WOM and not from its valence
WOM data collected from
the Yahoo Movies
message board
Ludwig, S., de
Ruyter, K.,
Friedman, M.,
Brüggen, E. C.,
Wetzels, M., &
Pfann, G. (2013)
More Than Words: The
Influence of Affective
Content and Linguistic Style
Matches in Online Reviews
on Conversion Rates
Journal of
Marketing
77 (1)
The influence of positive affective content on conversion rates is asymmetrical. No such taperingoff effect occurs for changes in negative affective content in reviews. Furthermore, positive changes
in affective cues and increasing congruence with the product interest group’s typical linguistic style
directly and conjointly increase conversion rates
Sample of 591 books
available for sale on
Amazon.com with 18,682
customer reviews
Ma, X., Khansa,
L., Deng, Y., &
Kim, S. S. (2013)
Impact of Prior Reviews on
the Subsequent Review
Process in Reputation
Systems
Journal of
Management
Information
Systems
30 (3)
Male reviewers or those who lack experience, geographic mobility, or social connectedness are
more prone to being influenced by prior reviews. The authors also found that longer and more
frequent reviews can reduce online reviews’ biases
Panel data set of 744
individual consumers
collected from Yelp
49
The authors find a unique equilibrium where online word of mouth is persuasive despite the
promotional chat activity by competing firms. In this equilibrium, firms spend more resources
promoting inferior products, in striking contrast to existing advertising literature
Across individuals, the results show that positive ratings environments increase posting incidence,
whereas negative ratings environments discourage posting. The authors' results also indicate
important differences across individuals in how they respond to previously posted ratings. These
dynamics affect the evolution of online product opinions
The results show that although ratings behavior is significantly influenced by previously posted
ratings and can directly improve sales, the effects are relatively short lived once indirect effects are
considered
Mayzlin, D.
(2006)
Promotional Chat on the
Internet
Marketing
Science
25 (2)
Moe, W. W., &
Schweidel, D. A.
(2012)
Online Product Opinions:
Incidence, Evaluation, and
Evolution
Marketing
Science
31 (3)
Moe, W. W., &
Trusov, M.
(2011)
The Value of Social
Dynamics in Online Product
Ratings Forums
Attitude Predictability and
Helpfulness in Online
Reviews: The Role of
Explained Actions and
Reactions
What Makes a Helpful Online
Review? A Study of
Customer Reviews on
Amazon.com
Born Unequal: A Study of the
Helpfulness of UserGenerated Product Reviews
Predicting the performance of
online consumer reviews: A
sentiment mining approach to
big data analytics
The influence of online
product recommendations on
consumers’ online choices
How Does the Variance of
Product Ratings Matter?
From valence to emotions:
Exploring the distribution of
emotions in online product
reviews
The Impact of MarketingInduced versus Word-ofMouth Customer Acquisition
on Customer Equity Growth
User Reviews Variance,
Critic Reviews Variance, and
Product Sales: An
Exploration of Customer
Breadth and Depth Effects
Saliency effects of online
reviews embedded in the
description on sales:
Moderating role of reputation
Can online product reviews
be more helpful? Examining
Journal of
Marketing
Research
48 (3)
Journal of
Consumer
Research
42(1)
Results show that review writers explain their actions more than their reactions for utilitarian
products, but they explain their reactions more than their actions for hedonic products
Experimental Settings
MIS Quarterly
34 (1)
Review extremity, review depth, and product type affect the perceived helpfulness of the review.
Product type moderates the effect of review extremity on the helpfulness of the review
1,587 reviews from
Amazon.com across six
products
Journal of
Retailing
87 (4)
Decision
Support
Systems
81
Journal of
Retailing
80 (2)
Results indicate that subjects who consulted product recommendations selected recommended
products twice as often as subjects who did not consult recommendations
Experimental Settings
Management
Science
58 (4)
A high average rating indicates a high product quality, whereas a high variance of ratings is
associated with a niche product, one that some consumers love and others hate
667 books available from
Amazon and BN
Decision
Support
Systems
81
The authors find that there is a difference in the emotional content of reviews across search and
experience goods in the early stages of product launch. However, these differences disappear over
time as the addition of reviews reduces the information asymmetry gap
15,849 online customer
reviews from
Amazon.com
45 (1)
Marketing-induced customers add more short-term value, but word-of-mouth customers add nearly
twice as much long-term value to the firm.
Internet firm provided free
Web hosting to registered
users during a 70-week
observation period
Journal of
Retailing
91 (3)
After recognizing a high variance in user reviews, many potential buyers may simply exclude the
focal product from their consideration sets for fear that it does not match their needs and
preferences. High user reviews variance, in combination with high critic reviews variance, can elicit
a sense of uniqueness and enhance purchase intentions of some consumers. Quality signals can
strengthen the positive customer depth effect
Secondary data and
experimental settings
combined
Decision
Support
Systems
87
This study explores the effects of online reviews embedded in the product description (OED) on
sales. Results indicate that OED has a positive effect on sales, and a high reputation strengthens the
impact of OED on sales
Secondary data and
experimental settings
combined
Decision
Support
79
The authors find that, while review content affects helpfulness in complex ways, these effects are
well explained by the proposed framework. The data suggest that review writers who explicitly
Over 8000 helpfulness
ratings from reviews
Moore, S. G.
(2015)
Mudambi, S. M.,
& Schuff, D.
(2010)
Pan, Y., &
Zhang, J. Q.
(2011)
Salehan, M., &
Kim, D. J. (2016)
Senecal, S., &
Nantel, J. (2004)
Sun, M. (2012)
Ullah, R.,
Amblee, N.,
Kim, W., & Lee,
H. (2016)
Villanueva, J.,
Yoo, S., &
Hanssens, D. M.
(2008)
Wang, F., Liu,
X., & Fang, E.
(2015)
Wang, Z., Li, H.,
Ye, Q., & Law,
R. (2016)
Weathers, D.,
Swain, S. D., &
Journal of
Marketing
Research
Both review valence and length have positive effects on review helpfulness, but the product type
moderates these effects. A curvilinear relationship exists between expressed reviewer
innovativeness and review helpfulness
The authors' findings show that reviews with higher levels of positive sentiment in the title receive
more readerships. Sentimental reviews with neutral polarity in the text are also perceived to be more
helpful. The length and longevity of a review positively influence both its readership and
helpfulness
Econometric Model
Data from BazaarVoice.
10,460 ratings across
1,811 products
Sample of 500 beauty
products with 3801 ratings
posted
Experimental Settings
35,000 online reviews of
20 different products from
Amazon.com
50
Grover, V.
(2015)
characteristics of information
content by product type
Systems
Wu, C., Che, H.,
Chan, T. Y., &
Lu, X. (2015)
The Economic Value of
Online Reviews
Marketing
Science
Xu, K., Liao, S.
S., Li, J., &
Song, Y. (2011)
Mining comparative opinions
from customer reviews for
Competitive Intelligence
Decision
Support
Systems
Ya, Y.,
Vadakkepatt, G.
G., & Joshi, A.
M. (2015)
A Meta-Analysis of
Electronic Word-of-Mouth
Elasticity
Journal of
Marketing
Yang, J.,
Sarathy, R., &
Lee, J. (2016)
Yin, D., Bond, S.
D., & Zhang, H.
(2014)
The effect of product review
balance and volume on online
Shoppers' risk perception and
purchase intention
Anxious or Angry? Effects of
Discrete Emotions on the
Perceived Helpfulness of
Online Reviews
34 (5)
attempt to enhance review diagnosticity or credibility are often ineffective or systematically
unhelpful
The authors conduct a series of counterfactual experiments and find that the value from Dianping is
about 7 CNY for each user, and about 8.6 CNY from each user for the reviewed restaurants. The
majority of the value comes from reviews on restaurant quality, and contextual comments are more
valuable than numerical ratings in reviews
posted on Amazon.com
User browsing and user
dining choices collected
from Dianping.com
50 (4)
The authors proposed a novel graphical model to extract and visualize comparative relations
between products from customer reviews, with the interdependencies among relations taken into
consideration, to help enterprises discover potential risks and further design new products and
marketing strategies
Data from: 1) online
shopping sites; 2) review
sites; 3) blogs; 4) social
network sites; and 5)
emails in CRM
79 (2)
The analysis reveals that electronic word-of-mouth volume (valence) elasticity is .236 (.417).
Volume and valence elasticities are studied under different contexts
51 studies
Decision
Support
Systems
89
The four proposed risk concerns are good predictors of online shoppers' overall risk in e-commerce;
perceived risk is a major determinant of online shoppers' attitude toward purchasing, which in turn
determines their purchase intention. No significant causal effect between perceived uncertainty and
purchase intention was found
302 questionnaires
MIS Quarterly
38 (2)
The findings demonstrate the importance of examining discrete emotions in online word-of-mouth,
and they carry important practical implications for consumers and online retailers
Experimental Settings
72,916 customers'
numerical ratings for 1628
different movies from
EachMovie
Ying, Y.,
Feinberg, F., &
Wedel, M.
(2006)
Leveraging Missing Ratings
to Improve Online
Recommendation Systems
Journal of
Marketing
Research
43 (3)
Recommendation quality is improved substantially by jointly modeling "selection" and "ratings,"
both whether and how an item is rated. Accounting for missing ratings and various sources of
heterogeneity offers a rich portrait of which items are rated well, which are rated at all, and how
these processes are intertwined
Yubo, C., &
Jinhong, X.
(2008)
Online Consumer Review:
Word-of-Mouth as a New
Element of Marketing
Communication Mix
Management
Science
54 (3)
The authors show when and how the seller should adjust its own marketing communication strategy
in response to consumer reviews. The authors also identify product/market conditions under which
the seller benefits from facilitating such buyer-created information
The sample includes 120
digital camera models
reviewed by CNET.com
Examining the influence of
online reviews on consumers'
decision-making: A heuristic–
systematic model
Decision
Support
Systems
67
Argument quality of online reviews, which is characterized by perceived informativeness and
perceived persuasiveness, has a significant effect on consumers' purchase intention. In addition, the
authors find that source credibility and perceived quantity of reviews have direct impacts on
purchase intention
The model is empirically
tested with 191 users of an
existing online review site
Modeling Consumer Learning
from Online Product Reviews
Marketing
Science
32 (1)
Consumers learn more from online reviews of book titles than from their own experience with other
books of the same genre
Panel data set of 1,919
book purchases by 243
consumers
Impact of Online Consumer
Reviews on Sales: The
Moderating Role of Product
and Consumer Characteristics
Journal of
Marketing
74 (2)
Online reviews are more influential for less popular games and games whose players have greater
Internet experience. The article shows differential impact of consumer reviews across products in
the same product category and suggests that firms' online marketing strategies should be contingent
on product and consumer characteristics
Data on console sales and
game sales from NPD.
Review data from
GameSpot.com
Zhang, K. Z. K.,
Zhao, S. J.,
Cheung, C. M.
K., & Lee, M. K.
O. (2014)
Zhao, Y., Yang,
S., Narayan, V.,
& Zhao, Y.
(2013)
Zhu, F., &
Zhang, X. (2010)
Table 6. Most relevant literature on online consumer reviews (key findings as reported by the authors)