Conference PaperPDF Available

The Value of Online Customer Reviews

Authors:
Conference Paper

The Value of Online Customer Reviews

Abstract and Figures

We study the effect of the volume of consumer reviews on the purchase likelihood (conversion rate) of users browsing a product page. We propose using the exponential learning curve model to study how conversion rates change with the number of reviews. We call the difference in conversion rate between having no reviews and an infinite number \textit{the value of reviews}. We find that, on average, the conversion rate of a product can increase by 142% as it accumulates reviews. To address the problem of simultaneity of increase of reviews and conversion rate, we explore the natural temporal trends throughout a product's lifecycle. We perform further controls by using user sessions where the reviews were not displayed. We also find diminishing marginal value as a product accumulates reviews, with the first five reviews driving the bulk of the aforementioned increase. Within categories, we find that the value of reviews is highest for Electronics (increase of 317%) followed by Home Living (increase of 182%) and Apparel (increase of 138%). We infer that the existence of reviews provides valuable signals to the customers, increasing their propensity to purchase. We also infer that users usually don't pay attention to the entire set of reviews, especially if there are a lot of them, but instead they focus on the first few available. Our approach can be extended and applied in a variety of settings to gain further insights.
Content may be subject to copyright.
The Value of Online Customer Reviews
Georgios Askalidis
Northwestern University
Evanston, IL
gask@u.northwestern.edu
Edward C. Malthouse
Northwestern University
Evanston, IL
ecm@northwestern.com
ABSTRACT
We study the effect of the volume of consumer reviews on
the purchase likelihood (conversion rate) of users browsing
a product page. We propose using the exponential learning
curve model to study how conversion rates change with the
number of reviews. We call the difference in conversion rate
between having no reviews and an infinite number the value
of reviews. We find that, on average, the conversion rate of a
product can increase by as much as 270% as it accumulates
reviews, amongst the users that choose to display them. We
also find diminishing marginal value as a product accumu-
lates reviews, with the first five reviews driving the bulk of
the aforementioned increase. To address the problem of si-
multaneity of increase of reviews and conversion rate, we
use customer sessions in which reviews were not displayed
as a control for trends that would have happened regardless
of the increase in the review volume. Using our framework,
we further find that high priced items have a higher value
for reviews than lower priced items. High priced items can
see their conversion rate increase by as much as 380% as
they accumulate reviews compared to 190% for low priced
items.We infer that the existence of reviews provides valu-
able signals to the customers, increasing their propensity to
purchase. We also infer that users usually don’t pay atten-
tion to the entire set of reviews, especially if there are a lot
of them, but instead they focus on the first few available.
Our approach can be extended and applied in a variety of
settings to gain further insights.
Keywords
Marketing, Online Reviews, Word of Mouth
1. INTRODUCTION
Electronic Word of Mouth (eWOM) in the form of on-
line customer reviews, is omnipresent and part of many cus-
tomers’ purchase journey. Reviews are being collected, ag-
gregated and displayed to consumers in an easy-to-digest
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full cita-
tion on the first page. Copyrights for components of this work owned by others than
ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-
publish, to post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from permissions@acm.org.
RecSys ’16, September 15-19, 2016, Boston , MA, USA
c
2016 ACM. ISBN 978-1-4503-4035-9/16/09. . . $15.00
DOI: http://dx.doi.org/10.1145/2959100.2959181
format in all types of settings: all of the top-10 U.S. on-
line retailers (as well as most of the biggest retailers in the
rest of the world) collect and display user reviews for their
products. The same is true for all the major digital stores.
Furthermore, companies like Yelp, Facebook, Google, IMDb
and Rotten Tomatoes provide platforms for users to sub-
mit reviews that are, in-turn, aggregated and displayed to
other users. User reviews are also being used to build trust
between customers in decentralized marketplaces like eBay,
Airbnb and Uber.
For online shoppers, reviews are not just an option any-
more but an expectation. A recent survey1found that 30%
of shoppers under the age of 45 consult reviews for every pur-
chase they make, while 86% say that reviews are essential in
making purchase decisions. In fact, after price, reviews are
the factor with the most impact on purchases. An exten-
sive literature has also showcased the economic importance
of positive reviews for restaurants [14, 1], books [2], movies
[6], mobile apps [8], and consumer goods [15].
In this work we aim to gain insights on a slightly more
fundamental question. We ask, should retailers seek reviews
for their products? Can the existence of reviews (without
any assumptions on their valence or other characteristics)
increase the likelihood of purchase? Our analysis suggests
positive answers. We further explore the value of reviews for
the three largest product categories of the studied retailer:
electronics, apparel and home living. Within categories, we
explore the value of reviews for high- and low-priced items.
2. THE EFFECT OF REVIEW VOLUME ON
PURCHASE
Prior research on review volume has produced mixed re-
sults. Volume is believed to exert positive effects by increas-
ing review credibility and/or signaling product popularity.
Volume has been shown to have a positive effect on box of-
fice sales [7, 13] and sales rank of electronic products [5, 11].
Moreover, it influences other outcomes such as consumer at-
tention [9, 13], product evaluation [4], product popularity
[18] and purchase intention [17]. However, [10] finds that
volume does not impact the sales of cell phones, and [3]
shows that volume has no significant impact on box office
performance. There is no evidence for a negative effect.
In order to understand the effect of volume on purchase
we propose the following assumptions:
Assumption 1: Each product has an inherent con-
1http://www.powerreviews.com/blog/survey-confirms-the-
value-of-reviews
version rate, in the absence of any reviews.
Assumption 2: Each product has an upper bound
on its conversion rate that is less than 1.
Assumption 3: As a product accumulates more re-
views, their marginal value decreases.
Consider a product (or service). When the product is first
introduced it will not have any reviews, and there will be an
initial probability of purchase, which we will call the inter-
cept. Over time the product will attract reviews. The accu-
mulating reviews should cause the probability of purchase
to increase and approach an asymptote. This asymptote can
theoretically be 1, i.e., given an infinite number of reviews
every product view can lead to purchase, but in reality we
expect it to be much lower. The exact location of the asymp-
tote will depend on other factors, such as the price, product
category, quality, and the level of marketing support. We
furthermore assume that as a product accumulates reviews,
the value of each additional review will decrease, i.e., the
marginal value of the 5th review will be higher than that of
the 50th review. If these assumptions are accepted then one
possible function for the value of reviews is the exponential
learning curve [16, equation 13.9 and §13.5]:
π(x) = γ0γ1eγ2x, x 0, γ2>0,(1)
where 0 π(x)1 is the probability that a product with x
reviews is purchased. Parameter γ0specifies the asymptote
because limx→∞ π(x) = γ0. The intercept is π(0) = γ0γ1,
and so γ1gives the distance between the intercept and the
asymptote. Thus, γ1gives the value of reviews, since it is
the difference between the conversion rate with no reviews
and when there are infinite reviews. We do not constraint
γ10, and γ1<0 would indicate that the conversion rate
decreases as reviews are added. The other parameter, γ2,
characterizes the steepness of the function, where large val-
ues indicate π(x) approaches γ0quickly and smaller values
indicate a more gradual approach. This model has been used
extensively for modeling learning curves and memory [12].
3. METHODS
We have data for the entire year 2015 from an online re-
tailer that sells high-end specialty gifts. The data record ev-
ery visit to the site for each user id, including the date and
time, what products were viewed, whether the user viewed
reviews and whether the user purchased the product. We
also know the number of reviews that the user was exposed
to, where the number increases over the year. Thus, we can
study the conversion rate as a function of the number of
reviews.
Our dataset consists of around 15.5 million page views for
1800 unique products, from 7.8 million users over the course
of one year (January 4, 2015 – January 2, 2016). For each
product, we track the number of its page views and sales as
it accumulates reviews. Hence, for each product and num-
ber of reviews, n, we calculate the product’s conversion rate
while it had nreviews. We then average the conversion rate
over all products for each n, to calculate the average conver-
sion rate as a function of the number of reviews. When we
want to study the value of reviews for products of a specific
category or characteristic (e.g., price) we repeat the above
approach but restrict our attention to the relevant products.
We then estimate the parameters of Equation 1 by using the
curve fit function of the scipy module for Python2.
Our approach has a built-in control group, since we are
tracking the conversion rate of a fixed set of products while
they accumulate more reviews. Since we are comparing the
conversion rate of the products to their past selves, many
effects due to the products’ characteristics are controlled for
within our model. What our model doesn’t inherently con-
trol for are temporal trends related to exogenous time re-
lated variables. To address this issue we use a control set of
observations. Observations regarding page views in which
the user did not display reviews. We use these sessions as
a control for time and other exogenous trends. We adjust
our exponential function to incorporate these distinction be-
tween page views where the user did and did not display the
reviews.
π(x) = (γ0+γ1disp) (γ2+γ3disp)e(γ4+γ5disp)x,
x0, γ2>0(2)
where disp is a binary indicator of whether a display of re-
views happened. By fitting Equation 1 on our data, we
isolate the effect of the display of reviews while controlling
for all other exogenous trends. To further remove any noise
from our data, we focus only on products whose first page
view was observed after March 1, 2015. This is to ensure
that our data cover the beginning of a product’s life cycle,
and we don’t use data for products that have been on the
market already for some time in the beginning of 2015.
4. RESULTS
Overall, the conversion rate of a product increases by
270% as it accumulates reviews. This increase is calculated
as the difference between the asymptote (γ1) and the inter-
cept (γ0), divided by the intercept. Moreover, over the bulk
of that effect happens within the first 10 reviews the product
receives. Figure 1 displays the fitted curve. Note that this
effect is isolated for when the display of reviews happened,
i.e., when disp equals 1 in Equation 1.
Figure 2 compares the value of reviews for high- and low-
priced items. Products were categorized into high and low
price depending whether their price was above or below the
median price amongst the studied products. The median
price was $79.99. We notice that low-priced items have con-
sistently higher conversion rates than the high-priced prod-
ucts. The value of reviews is around 190% for low-priced
items but much higher, at 380% for high-priced items. It
seems that users gain confidence in their purchase when
they see other users have bought a high-priced item. For
low-priced items, the monetary risk is lower and perhaps
that’s why the value of reviews is lower.
Finally, we also explored the value of reviews for high- and
low-quality products. We approximate a product’s quality
by observing all the reviews it accumulated over its (ob-
served in our date) lifetime. Products with a lifetime av-
erage below 3.5 were classified as low quality and products
with a lifetime average above 3.5 stars were classified as
high quality. We notice that high-quality products have
consistently higher conversion rate than lower-quality prod-
ucts. But, perhaps surprisingly, low-quality products have a
higher value for reviews, at 324% compared to high-quality
2Python Documentation, http://bit.ly/1XDwA76
products, at 135%. This could be because of the lower base-
line for low-quality products as well as that social signals are
more important and valuable when the product in question
is of lower quality.
Figure 1: Effect of Number of Displayed Reviews on the
Conversion Rate
5. CONCLUSION AND FUTURE WORK
Our work provides strong evidence for a positive value of
reviews, i.e., causal effect of the existence of reviews towards
the purchase likelihood of a browsing customer. This is with
no assumptions on the characteristics of the reviews. Practi-
tioners should take this to mean that it’s beneficial for their
platform to solicit reviews, even if they have no control over
the reviews they will get (e.g., their valence).
We explore the value of reviews for various product char-
acteristics such as price and category, but our results are
still confined under the specific retailer we study. The exact
value may change depending the specific characteristics of
the retailer. We believe our approach, and model, can help
guide future studies for other settings and retailers.
Many interesting directions are available for future work,
and our dataset can be utilized to help explore them. First,
better understanding of the value of reviews for various types
of products as well as users can be important. Are reviews
more important for returning or new users? How about users
that have submitted a review themselves? Insights here,
can help a retailer identify products and users for whom the
value of reviews is high and act accordingly, e.g., by more
prominent displays.
Moreover, the effect of valence can be better studied and
understood. What is the value of higher ratings versus lower
ratings? Studying the interaction between volume and va-
lence can also help with understanding the way that cus-
tomers interpret consumer reviews. For example, all else
being equal, is it better for a product to have a 4.5 star av-
erage rating based on 5 reviews or a 4 star average based on
50 reviews?
6. ACKNOWLEDGMENTS
We thank the Spiegel Center for Digital and Database
Marketing at Northwestern University for support.
Figure 2: Effect of Number of Displayed Reviews on Various
Price
References
[1] M. Anderson and J. Magruder. Learning from the
crowd: Regression discontinuity estimates of the effects
of an online review database. The Economic Journal,
122(563):957–989, 2012.
[2] J. A. Chevalier and D. Mayzlin. The effect of word
of mouth on sales: Online book reviews. Journal of
Marketing Research, 43(3):345–354, 2006.
[3] P. K. Chintagunta, S. Gopinath, and S. Venkataraman.
The effects of online user reviews on movie box of-
fice performance: Accounting for sequential rollout and
aggregation across local markets. Marketing Science,
29(5):944–957, 2010.
[4] J.-S. Chiou and C. Cheng. Should a company have
message boards on its web sites? Journal of Interactive
Marketing, 17(3):50–61, 2003.
[5] G. Cui, H.-K. Lui, and X. Guo. The effect of online
consumer reviews on new product sales. International
Journal of Electronic Commerce, 17(1):39–58, 2012.
[6] C. Dellarocas, N. Awad, and M. Zhang. Using
online ratings as a proxy of word-of-mouth in mo-
tion picture revenue forecasting. Available at SSRN:
http://ssrn.com/abstract=620821, 2005.
[7] W. Duan, B. Gu, and A. B. Whinston. Do online re-
views matter??an empirical investigation of panel data.
Decision support systems, 45(4):1007–1016, 2008.
[8] P. Engstrom and E. Forsell. Demand effects of con-
sumers’ stated and revealed preferences. Available at
SSRN 2253859, 2014.
[9] D. Godes and D. Mayzlin. Using online conversations
to study word-of-mouth communication. Marketing sci-
ence, 23(4):545–560, 2004.
[10] S. Gopinath, J. S. Thomas, and L. Krishnamurthi.
Investigating the relationship between the content of
online word of mouth, advertising, and brand perfor-
mance. Marketing Science, 33(2):241–258, 2014.
Table 1: Learning Curve Parameter Estimates
Category ˆγ0ˆγ1ˆγ2ˆγ3ˆγ4ˆγ5
Overall 0.07 0.053 0.042 0.03 0.078 0.27
(0.008) (0.008) (0.008) (0.015) (0.04) (0.7)
High Price 0.073 0.063 0.06 0.05 0.046 0.3
(0.0025) (0.0064) (0.01) (0.014) (3.6·106) (3.6·106)
Low Price 0.059 0.037 0.0028 0.017 13.1 -12.79
(0.002) (0.0038) (0.011) (0.014) (3.2·105) (3.2·105)
High Rated 0.051 0.03 0.001 0.013 13.58 13.37
(0.002) (0.006) (0.01) (0.014) (3.6·106) (3.6·106)
Low Rated 0.036 0.013 0.03 0.047 0.33 -0.29
(0.006) (0.16) (0.014) (0.15) (0.35) (0.64)
Values in parentheses are standard errors.
[11] B. Gu, J. Park, and P. Konana. Research note-the im-
pact of external word-of-mouth sources on retailer sales
of high-involvement products. Information Systems Re-
search, 23(1):182–196, 2012.
[12] N. Leibowitz, B. Baum, G. Enden, and A. Karniel. The
exponential learning equation as a function of success-
ful trials results in sigmoid performance. Journal of
Mathematical Psychology, 54(3):338–340, 2010.
[13] Y. Liu. Word of mouth for movies: Its dynamics and
impact on box office revenue. Journal of marketing,
70(3):74–89, 2006.
[14] M. Luca. Reviews, reputation, and revenue: The
case of Yelp.com. Harvard Business School NOM
Unit Working Paper No. 12-016. Available at SSRN:
http://ssrn.com/abstract=1928601, 2011.
[15] E. Maslowska, E. C. Malthouse, and S. Bernritter. Too
good to be true: The role of online reviews’ features in
probability to buy. International Journal of Advertis-
ing, 2016.
[16] J. Neter, M. H. Kutner, C. J. Nachtsheim, and
W. Wasserman. Applied linear statistical models, vol-
ume 4. Irwin Chicago, 1996.
[17] H. Park, J. Jeon, and W. Kwak. The influence of
perceived quality and vmd fitness of fashion brand on
brand attitude and purchase intention. Journal of Mar-
keting Management Research, 12(1):55–70, 2007.
[18] Z. Zhang, Q. Ye, R. Law, and Y. Li. The impact
of e-word-of-mouth on the online popularity of restau-
rants: A comparison of consumer reviews and editor
reviews. International Journal of Hospitality Manage-
ment, 29(4):694–700, 2010.
... Chen et al. [4] mentioned a Wall Street Journal report (2006) that 71% of online U.S. adults use consumer reviews for their purchases and 42% trust such a source. Askalidis and Malthouse [5] found that purchase likelihood increases by as much as 190% for low priced items and 380% for high priced items when product reviews are displayed. They also found that in the high availability of product reviews, users usually focus only on the first few reviews. ...
... M i (Movie) Figure 3. Reviewer's weight (Expertise_score) as a function of movieweight, deviation from the true rating, and the number of reviews. 5. Calculate the score of the reviewers. ...
Article
Full-text available
Consumer reviews have emerged as one of the most influential factors in a person’s purchase behavior. The existing open-source approaches for detecting expert reviewers and determining product ratings suffer from limitations and are susceptible to manipulation. In this work, we addressed these limitations by developing two algorithms and evaluated them on three datasets from amazon.com (the largest dataset contains nearly eight million reviews). In the first algorithm, we used a combination of the existing open-source approaches such as filtering by volume of contribution, helpfulness ratio, volume of helpfulness, and deviation from the estimated actual rating to detect the experts. The second algorithm is based on link analytic mutual iterative reinforcement of product ratings and reviewers’ weights. In the second algorithm, both reviewers and products carry weights reflecting their relative importance. The reviewers influence the product rating according to their weight. Similarly, the reviewers’ weights are impacted by their amount of deviation from the estimated actual product rating and the product’s weight. Our evaluation using three datasets from amazon.com found the second algorithm superior to the other algorithms in detecting experts and deriving product ratings, significantly reducing the avg. error and avg. Mean Squared Error of the experts over the best of the other algorithms even after maintaining similar product coverage and quantity of reviews.
... While talking about perceived trustworthiness, factors consumers' reviews or, in more specific terms, Word of Mouth (WOM) is omnipresent and integral to customers' online shopping journey. Precisely speaking, consider a service or product which is just introduced but does not contain any reviews, with the initial probability of its purchase (Askalidis & Malthouse, 2016). Over time the product/service will attract customers. ...
... Here, trustworthiness should be kept in mind as a dynamic mechanism and underlying objective of online retailers. As online retailers provide products in a virtual environment, building trust to increase positive reviews through product quality improves their brand image and motivates future consumers (Askalidis & Malthouse, 2016). Thus, the relationship between Positive Word of Mouth and Product Quality remained strongly significant (p> 0.000), indicating the potential compatibility of the study conducted by (Virvilaite et al., 2015) as they also found a causal relationship between these two variables in the Lithuanian context. ...
Article
Full-text available
Word of Mouth also works as a primary determinant of people's positive attitude towards online shopping and retailers. Notably, people are more likely to spend online shopping if they trust online retailers. This study also focuses on the online retail industry in the United Arab Emirates. The Item Response Theory primarily supports the conceptual model of the current research. We employed a cross-sectional design and selected a sample of n= 304 online consumers. Results revealed that the relationships between Positive Word of Mouth, Brand Image, and Causal Attributions are strongly validated (p> 0.000). Besides, the relationships between. Product Quality, Brand Image, and Causal Attributions are also affirmed with the path value at 1.014 and significance value at p> 0.000. Moreover, we also found a potentially significant relationship between Positive Word of Mouth and Product Quality p> 0.000). We also affirmed that the relationship between Brand Image and Causal Attribution is also validated with the significance values at p> 0.000. Lastly, the proposed relationships between we proposed a significant relationship are also validated (p> 0.000). Thus, we conclude that, today, when competition is increasing day by day, it is crucial to examine the consumer psychology that may highlight our several factors as done by the Item Response Theory. Further, we have discussed the study limitations and contributions accordingly.
... Prior research has already discussed how consumers seek online reviews and how the reviews support the customers while deciding to make a purchase decision (Bansal & Voyer, 2000;Cheung & Lee, 2012;King et al., 2014;Ruiz-Mafe et al., 2018;Tran & Strutton, 2019). Consumers often do not read all the reviews that might be present on a particular platform, but it is pretty repetitive that the latest or the first few reviews were always read before making a purchase decision (Askalidis & Malthouse, 2016). The discussion will now focus on identifying and segregating the aspects/the variables from the existing literature that strengthened the consumers' choice online and then weave them together to create a framework. ...
Article
Full-text available
The modern consumer looks at visiting restaurants as a much broader experience than just consuming food. They are aware and carefully form their decisions based on available information. Therefore, foodservice businesses should also ensure that they are being promoted positively among their peers to continue gaining profits in the long run. This study aims to explore and understand the impact of online reviews and the nature of reviews on consumer’s purchase intention on the platform of Zomato. The research uses a mixed-method approach using Sequential Explanatory Research design. In the quantitative segment, data from 208 participants were collected through a survey. In the qualitative segment, three Focus Group Discussions were conducted to gain a critical understanding of the constructs identified in the proposed framework. The results confirm that online reviews and star ratings available on Zomato significantly impact the willingness to purchase. However, the research established moderators as choice indicators that entail information seeking, risk reduction, seeking approval to visit a restaurant, reassurance to visit a restaurant, and indulging in its services. Together as choice indicators, these variables significantly moderate the perception of the reviews, thus generating purchase augmenting business. The variables are well supported as moderating variables in the findings of the study using a mixed-method approach. This research recognized and authenticated these variables considered in the study as an auxiliary variable that affects consumers’ willingness to pay for particular services, mainly food services.
Article
Purpose The number of children using and owning mobile devices has grown significantly in the last decade. By applying a uses and gratifications approach, this paper aims to explore what customers of a tablet computer for children report about the use and expectations of these devices from an information behavior perspective. Design/methodology/approach For this study, 1,185 online customer reviews published for two different versions of the Amazon Fire Tablet Kids Edition on the German Amazon website between June 16, 2019, and June 15, 2020, were analyzed. A content analysis of the reviews was conducted using different inductive coding methods. Findings Findings indicate that customers describe different aspects of children's use and families' expectations of tablets within their reviews. The expressed gratifications mostly relate to the aim of entertainment. Intentional information seeking activities were hardly mentioned within the reviews, but many customers emphasize learning as an important activity with the devices. Overall, the customer reviews reveal a mix of gratifications that differ from reported motivations for adults' tablet use. Research limitations/implications The possibility of manipulated online customer reviews must be considered. It should also be viewed critically that the children's perspectives are only indirectly included in the data. Originality/value Families' expectations of tablets as a device for children have not been a focus of research to date. This study uses an innovative research design by applying a uses and gratifications approach to online customer reviews for children's tablets. The findings add to previous research on children's use and families' expectations of tablets and contribute to our understanding of children's information behavior in connection with mobile devices.
Article
Today, reviews are the advertising medium par excellence through which companies are able to influence customers’ spending decisions. Although the initial purpose of reviews was to provide companies with a feedback tool to improve products and services based on customer needs, they soon became a way to climb the sales rankings, often illegally. In fact, deceptive and fake reviews have managed to evade the often non-existent means of validation of online platforms, proliferating a new business. To combat this phenomenon, several classification methods have been developed to train automated tools in the arduous task of distinguishing between genuine and misleading reviews, the most recent based on machine and deep learning techniques. This paper proposes a multi-label classification methodology based on the Google BERT neural language model to build a deceptive review detector aided by its sentiment awareness: improved modeling of the link between sentiment polarity and deceptiveness during the fine-tuning phase by exploiting the Binary Cross Entropy with Logits loss function adds to the advantages provided by pre-trained contextual models, which are able to capture word polysemy through word embeddings and benefit from pre-training on huge corpora. Tests were performed on the Deceptive Opinion Spam Corpus and Yelp New York City datasets, providing a quantitative and qualitative analysis of the results which, when compared with the state of the art available in the literature, showed an encouraging increase in performance.
Article
The task of detecting fraudulent reviewers is of great importance to E-commerce platforms. Existing research has invested much effort into developing comprehensive features and advanced techniques to detect fraudulent reviewers. However, most of these studies have ignored the data imbalance problem inherent in fraudulent reviewer detection: non-fraudulent reviewers are the majority, while fraudulent reviewers are the minority in real practice. To fill this gap, we propose a novel approach called ImDetector to detect fraudulent reviewers while handling data imbalance based on weighted latent Dirichlet allocation (LDA) and Kullback–Leibler (KL) divergence. Specifically, we develop a weighted LDA model to extract the latent topics of reviewers distributed on the review features. Asymmetric KL divergence is adopted to make the similarity measure between reviewers biased toward the fraudulent minority when using the K-nearest-neighbor for classification. By mapping the reviewers to the latent topics of features derived from the weighted LDA model and measuring the similarities between reviewers using asymmetric KL divergence, the data imbalance problem in fraudulent reviewer detection is alleviated. Extensive experiments on the Yelp.com dataset demonstrate that the proposed ImDetector approach is superior to the state-of-the-art techniques used for fraudulent reviewer detection. We also explain the experimental results and present the managerial implications of this paper.
Article
Existing work in the literature have shown that the number and quality of product ratings and reviews have a direct correlation with the product purchase rates in online e-commerce portals. However, the majority of the products on e-commerce portals do not have any ratings or reviews and are known as cold products (∼90% of products on Amazon are cold). As such, there has been growing interest in generating reviews for cold products by selectively transferring reviews from other similar yet warm products. Our work in this paper focuses on this specific problem and generates reviews for cold products through review selection. Similar to existing work in the literature, our work assumes a relationship between product attribute-values and the reviews that products receive. However, unlike the literature, our method (1) is not restricted to the exact surface form of a product attribute name; and, (2) can distinguish between the same attribute expressed in different forms. We achieve these two important characteristics by proposing methods to learn neural product representations that capture the semantics of product attribute-values as they relate to user reviews. More specifically, our work offers (i) an approach to learn neural representations of product attribute-values within a shared embedding space as product reviews; (ii) a weighted composition strategy to develop product representations from the representation of its attributes; and, (iii) a review selection method that selects relevant reviews for the composed product representation within the neural embedding space. We show through our extensive experiments on five datasets consisting of products from CNET.com and movies from rottentomatoes.com that our method is able to show stronger performance compared to several baselines on ROUGE-2 metrics.
Chapter
With the rise of dependency of online shopping and service providers, consumer ratings and reviews help users decide between good and bad options. Prior studies have already shown that the layout and visual cues provided with a rating scale can affect the users’ responses. This paper aims to explore: 1) users’ reaction to certain visual cues in rating scales, and 2) users’ preference in rating scale designs and how it influences the rating scores. A survey (n = 187) was conducted to collect user ratings of popular products with six different rating scale designs, using two types of visual icons (stars and emojis) and colour-schemes (using a warm-cool and a traffic-light metaphors). Statistical analysis from the survey shows that users prefer the scale with most visually informative design (traffic-light metaphor colours with emoji icons). It also shows that users tend to give their true ratings on scales they like most, rather than the scale design they are most familiar with. Based on these results, it can be concluded that user involvement is desirable in selecting the rating scale designs, and that visual cues with cognitive metaphors can ensure more accurate (truthful) rating scores from users. Our approach has novelty because we elicited the users’ own opinion on what their accurate or “true" rating is rather than only relying on analysing the data received from the rating scores. Our work can offer insights for online rating scales designs to improve the rating decision quality of users and help online business platforms provide more credible ratings to their customers.
Article
Big data defines the state where the size, speed and kind of data go beyond a memory or executing capabilities for precise and timely decision-making. Big data analytics is integrated with ML and statistical methods for processing big data and recognizes the important data. At present times, the generation of online product reviews has exponentially increased at each and every second. These applications have resulted in developing the volumes of data which can be used for prediction and classification for decision making process. Compared with other models, various techniques are applied in solving the big data problem, feature selection (FS) is known to be an efficient method. FS operations could be exploring with the application of a subset of features which is related to the topic of précised definition of the existing datasets. Deplorably, search using this type of sub-sets results in the problems of combinatorial as well as maximum time consuming. The meta-heuristic approaches are typically employed to facilitate the choice of features. This paper presents an optimal extreme learning machine (ELM) based binary particle swarm optimization to precede the FS process. The proposed method develops a Fitness Function (FF) by applying ELM. And the best solution of the FF has been explored under the application of BPSO technique. For instance, the dataset of product review which are derived from Amazon including synthetic data, which is comprised with total of 235,000 positive and 147,000 negative review records is used. The experimental result implied that the ELM-BPSO technique is comparably best
Article
Helpfulness prediction techniques have been widely incorporated into online decision support systems to identify high-quality reviews. Most current studies on helpfulness prediction assume that a review's helpfulness only relies on information from itself. In practice, however, consumers hardly process reviews independently because reviews are displayed in sequence; a review is more likely to be affected by its adjacent neighbors in the sequence, which is largely understudied. In this paper, we proposed the first end-to-end neural architecture to capture the missing interaction between reviews and their neighbors. Our model allows for a total of 12 (three selection × four aggregation) schemes that contextualize a review into the context clues learned from its neighbors. We evaluated our model on six domains of real-world online reviews against a series of state-of-the-art baselines. Experimental results confirm the influence of sequential neighbors on reviews and show that our model significantly outperforms the baselines by 1% to 5%. We further revealed how reviews are influenced by their neighbors during helpfulness perception via extensive analysis. The results and findings of our work provide theoretical contributions to the field of review helpfulness prediction and offer insights into practical decision support system design.
Article
Full-text available
Online consumer reviews are broadly believed to be a necessary and powerful marketing tool, and as such they have attracted considerable attention from both marketers and academics. However, previous research has not sufficiently focused on the effects of various review features on sales but rather used proxy measures such as consumers’ purchase intention or perceived helpfulness of reviews. Hence, the aim of this study was to investigate the effect of review valence and volume on purchase behavior. We use data from three different e-commerce websites and study light bulbs, women’s athletic shoes, natural hair care products, and herbal vitamins. The results show that, contrary to popular belief, more positive ratings do not simply result in higher sales. We find that the effect can be nonlinear, where the probability of purchase increases with rating to about 4.2-4.5 stars, but then decreases. Also, although the majority of extant research suggests that larger numbers of reviews bring more positive outcomes, we show that it is not always the case.
Article
Full-text available
We study the relative importance of online word of mouth and advertising on firm performance over time since product introduction. The current research separates the volume of consumer-generated online word of mouth (OWOM) from its valence, which has three dimensions-attribute, emotion, and recommendation oriented. Firm-initiated advertising content is also classified as attribute or emotion advertising. We also shed light on the role played by advertising content on generating the different types of OWOM conversations. We use a dynamic hierarchical linear model (DHLM) for our analysis. The proposed model is compared with a dynamic linear model, vector autoregressive/system of equations model, and a generalized Bass model. Our estimation accounts for potential endogeneity in the key measures. Among the different OWOM measures, only the valence of recommendation OWOM is found to have a direct impact on sales; i.e., not all OWOM is the same. This impact increases over time. In contrast, the impact of attribute advertising and emotion advertising decreases over time. Also, consistent with prior research, we observe that rational messages (i.e., attributeoriented advertising) wears out a bit faster than emotion-oriented advertising. Moreover, the volume of OWOM does not have a significant impact on sales. This suggests that, in our data, "what people say" is more important than "how much people say." Next, we find that recommendation OWOM valence is driven primarily by the valence of attribute OWOM when the product is new and driven by the valence of emotion OWOM when the product is more mature. Our brand-level results help us classify brands as consumer driven or firm driven, depending on the relative importance of the OWOM and advertising measures, respectively.
Article
Full-text available
Online word-of-mouth (WOM) such as consumer opinions, user experiences, and product reviews has become a major information source in consumer purchase decisions. Prior research on online WOM effect has focused mostly on low-involvement products such as books or CDs. For these products, retailer-hosted (internal) WOM is shown to influence sales overwhelmingly. Numerous surveys, however, suggest consumers often conduct pre-purchase searches for high-involvement products (e.g., digital cameras) and visit external WOM websites during the search process. In this study, we analyze the relative impact of external and internal WOMs on retailer sales for high-involvement products using a panel of sales and WOM data for 148 digital cameras from Amazon.com and three external WOM websites (Cnet, DpReview, and Epinions) over a four-month period. The results suggest that a retailer's internal WOM has a limited influence on its sales of high-involvement products, while external WOM sources have a significant impact on the retailer's sales. The findings imply that external WOM sources play an important role in the information search process.
Article
Full-text available
This study examines the effect of online reviews on new product sales for consumer electronics and video games. Analyses of panel data of 332 new products from Amazon.com over nine months reveal that the valence of reviews and the volume of page views have a stronger effect on search products, whereas the volume of reviews is more important for experience products. The results also show that the volume of reviews has a significant effect on new product sales in the early period and such effect decreases over time. Moreover, the percentage of negative reviews has a greater effect than that of positive reviews, confirming the negativity bias. Thus, marketers need to consider the distinctive influences of various aspects of online reviews when launching new products and devising e-marketing strategies.
Article
Full-text available
Internet review forums increasingly supplement expert opinion and social networks in informing consumers about product quality. However, limited empirical evidence links digital word‐of‐mouth to purchasing decisions. We implement a regression discontinuity design to estimate the effect of positive Yelp.com ratings on restaurant reservation availability. An extra half‐star rating causes restaurants to sell out 19 percentage points (49%) more frequently, with larger impacts when alternate information is more scarce. These returns suggest that restaurateurs face incentives to leave fake reviews but a rich set of robustness checks confirm that restaurants do not manipulate ratings in a confounding, discontinuous manner.
Article
This article uses actual word-of-mouth (WOM) information to examine the dynamic patterns of WOM and how it helps explain box office revenue. The WOM data were collected from the Yahoo Movies Web site. The results show that 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. More important, WOM information offers significant explanatory power for both aggregate and weekly box office revenue, especially in the early weeks after a movie opens. However, most of this explanatory power comes from the volume of WOM and not from its valence, as measured by the percentages of positive and negative messages.
Article
Knowledge of how consumers react to different quality signals is fundamental for understanding how markets work. The modern electronic marketplace has revolutionized the possibilities for consumers to gather detailed information about products and services before purchase. Specifically, a consumer can easily -- through a host of online forums and evaluation sites -- estimate a product's popularity based on either i) what other users say about the product (stated preferences) or ii) how many other users that have bought the product (revealed preferences). In this paper we compare the causal effects on demand from these two quality related signals. We study the online marketplace for Android apps on Google play. The specific way that Google play presents download and average rating information to users allows for identification of the causal effects.Our data consists of daily information, for 42 consecutive days, of more than 500 000 apps from the US version of Google play. Our main result is that consumers are much more responsive to other consumers' revealed preferences, compared to others' stated preferences. A 10 percentile increase in displayed average rating only increases downloads by about 3 percent, while a 10 percentile increase in displayed number of downloads increases downloads by about 20 percent.
Article
While the exponential learning equation, indicating a gradually diminishing improvement, is one of the standard equations to describe learning, a sigmoid behavior with initially increasing then decreasing improvement has also been suggested. Here we show that the sigmoid behavior is mathematically derived from the standard exponential equation when the independent variable of the equation is restricted to the successful trials alone. It is suggested that for tasks promoting success-based learning, performance is better described by the derived sigmoid curve.
Article
This study examined the impact of messages on a discussion forum regarding a consumer's brand evaluation and attitude toward the Web owner. Four hypotheses were developed based on consumer behavior theories. An experiment using a 2 × 3 × 2 between participants factorial design was conducted to test the hypotheses. The results showed that message favorableness and message number on an Internet discussion forum can influence a consumer's brand evaluation and attitude toward the Web owner. However, the effects are different under different levels of preexisting brand image. Negative messages hurt mainly low-image brands, whereas message repetition helps only high-image brands.