Conference PaperPDF Available

HOW HELPFUL ARE COMPARATIVE REVIEWS FOR PREDICTING PRODUCT DEMAND?

Authors:

Abstract and Figures

The increased Internet usage has driven a rapid growth of e-commerce transactions. One of the key determinants of the increased online transactions is the influence of electronic word-of-mouth (eWOM) in the form of online reviews. In particular, comparative reviews that compare similar products provide valuable information for consumers to evaluate multiple products and play a pivotal role in driving consumer purchase decisions. By constructing a product network based on products connected by comparative reviews, we develop several new network centrality measures and empirically examine the impact of eWOM through these new centrality measures and the semantic similarity of the comparative reviews. We find that the comparative reviews are key eWOM measures that influence the product's sales within a product network. Our findings also demonstrate that the text semantic similarity is a better measure of the strength of tie in a comparative product network than the review sentiment. Our study contributes to the eWOM literature by utilizing text review semantic similarity to capturing review strength based on the latent product features, and to the network graph theory through the new centrality measures we have developed. Overall, our findings provide important insights for e-commerce platform operators and vendors to leverage the impact of eWOM and help consumers compare products in a more effective manner.
Content may be subject to copyright.
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 1
HOW HELPFUL ARE COMPARATIVE REVIEWS FOR
PREDICTING PRODUCT DEMAND?
Research paper
Vemprala, Naga, The University of Texas, San Antonio, USA, Naga.Vemprala@gmail.com
Liu, Charles, The University of Texas, San Antonio, USA, Charles.Liu@utsa.edu
Choo, Kim-Kwang Raymond, The University of Texas, San Antonio, USA,
Raymond.Choo@fulbrightmail.org
Abstract
The increased Internet usage has driven a rapid growth of e-commerce transactions. One of the key
determinants of the increased online transactions is the influence of electronic word-of-mouth (eWOM) in
the form of online reviews. In particular, comparative reviews that compare similar products provide
valuable information for consumers to evaluate multiple products and play a pivotal role in driving
consumer purchase decisions. By constructing a product network based on products connected by
comparative reviews, we develop several new network centrality measures and empirically examine the
impact of eWOM through these new centrality measures and the semantic similarity of the comparative
reviews. We find that the comparative reviews are key eWOM measures that influence the products sales
within a product network. Our findings also demonstrate that the text semantic similarity is a better measure
of the strength of tie in a comparative product network than the review sentiment. Our study contributes
to the eWOM literature by utilizing text review semantic similarity to capturing review strength based on
the latent product features, and to the network graph theory through the new centrality measures we have
developed. Overall, our findings provide important insights for e-commerce platform operators and
vendors to leverage the impact of eWOM and help consumers compare products in a more effective manner.
Keywords: eWOM, WOM, network graph theory, strength of ties, centrality, customer reviews, text
analytics, text mining
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 2
1 Introduction
With the rapid expansion of e-commerce over the last decade, there is an increasing amount of consumer
opinions being exchanged about the products in the form of online customer reviews, also commonly
known as electronic word-of-mouth (eWOM). eWOM greatly influences the way consumers evaluate
products and has led to extensive research on how consumers process and derive useful information from
customer reviews. One particular research trend is the extraction and analysis of information from customer
reviews of rival products to evaluate the impact of product comparison on sales (Chen et al., 2017; Church
et al., 2015; Kim et al., 2018). Most e-commerce platforms encourage their users to post detailed reviews
about products based on their user experience, which often results in direct comparison across similar
competing products. Such comparisons often generate heated discussions among consumers, further
extending the influence of eWOM. Prior literature has shown that comparative reviews can influence review
rating and product demand (Chen et al., 2017; Z. Zhang et al., 2013). However, there is a lack of
understanding of how the text semantics of these comparative reviews influence consumer’s perception of
comparison outcome and subsequent product sales.
E-commerce platforms have traditionally adopted numeric metrics, such as review ratings and the
helpfulness of the review, to assist consumers to evaluate products. Although these metrics are useful,
studies have shown that the numeric ratings suffer under-reporting bias on various factors (Trenz et al.,
2013). In contrast, the review text embeds rich description on product features and quality. In particular,
comparative reviews provide useful information about what makes a product better relative to the other
products being compared. Such observations have motivated researchers to utilize text sentiment analysis
to explore the impact of comparative reviews on product sales (Ghose et al., 2010; Hu et al., 2014; Salehan
et al., 2016). However, most of these analyses are conducted at the individual product review level and
focus on review characteristics such as review length, readability, and subjectivity (Ghose et al., 2010;
Mudambi et al., 2010). While the above characteristics may influence a consumer’s evaluation of the
product being reviewed, consumers often have several candidate products in mind as they go through the
product reviews. Therefore, consumers often form their perception about the superiority of a product based
on various comparisons across multiple products. Moreover, when evaluating a product based on review
text, consumers with a preferred product specification are very likely to be highly sensitive to specific
keywords and text and heavily influenced by the text semantics, rather than relying on the overall sentiment
of the review text.
Social discussion via eWOM about specific features of a product comparing it with a competitor products
acts as a collective signal and a significant driver of purchase influence. The customers writing text reviews
provides their opinion about a product feature based on their experience generates a signal for other users
on the eCommerce platform reading the reviews. Extracting such context sensitive sentiments from online
customer reviews is not only helpful for customers but also provide helpful information to product designers
to understand customer opinions at a fine grained level (Jin, J., Ji, P., & Kwong, C. K., 2016). Hence,
examining the impact of comparative reviews requires a much-expanded framework that measures the
influence of eWOM with respect to all competing products, and more importantly, considers the text
reviews semantic orientation rather than overall text review sentiment. To achieve this objective, we
construct a network graph connected by various products involved in comparative reviews, and apply text
mining techniques to review text to examine how the characteristics of the comparative reviews and
properties of the resulting product network can influence consumer perception of the product quality and
market dynamics.
The product network we develop consists of all products in a given product category (Chen et al., 2017; Z.
Zhang et al., 2013) as comparative reviews rarely involve products from different categories. Within this
product network, the key components are network nodes and ties. A tie is formed when two products
(nodes) are compared in a product review. In this case, the relationship between the two connected products
can be measured by tie strength. Prior studies generally measure the strength of the comparisons based on
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 3
the sentiment of the reviews or the review comments, with the assumption that the sentiment is captured by
the frequency of positive (or negative) words. However, a review can evaluate a product based on multiple
features and consumers generally have vastly different tastes on these features. For example, a positive
review on the product’s price may be less useful for quality- or brand-conscious customers. Therefore, in
our current study, we propose to measure the strength of tie based on semantic similarity with comparative
review text. Semantic similarity is a measure of closeness between texts based on their syntactical
representation. It considers the meaning of the text rather than the exact presence of words (Deerwester et
al., 1990). Building such a network enables researchers to focus on the structure of the product network
that addresses the influence of eWOM through the ties (i.e. comparison between two products) and strength
of the comparisons. A product network constructed based on the ties of the products that have been directly
compared allows us to aggregate and leverage the micro-level effects of individual reviews into a much
stronger network-level macro effect, thus better explaining the influence of the dyadic nature of tie in terms
of the node, and tie properties.
In addition, the product networks constructed in prior studies primarily consider only the tie level influence,
which use the number of comparisons to calculate the product influence on the demand, and ignore the
strength of the ties while considering the centrality measures. A number of studies from a wide range of
fields have begun to explore the issue of node influence considering the number of ties the node makes, the
distance of the connection and considered both ties and distance while measuring the strength of ties. In a
social network, the strength of ties is a function of trust, frequency of contact, and emotional intensity
(Granovetter, 1977). In the biological disease networks, the influence of an endemic disease being spread
is measured based on the contact or establishing a tie, and the number of frequent contacts together (Chu et
al., 2011). On a similar analogy, even though the comparative reviews are influential in product sales, the
importance or the strength of comparative reviews should also be considered along with the comparisons
made on the product to measure the influence on product sales. Opsahl et al., (2010) provided alternative
measures to calculate the network properties utilizing both the ties and tie strengths (Opsahl et al., 2010).
We adopt the measures considering both the number of ties and the weights to compute the product
influence in this research, and seek to address the below research questions.
RQ1: Is text semantic similarity a better measure of the impact of eWOM on product sales than text review
sentiment?
RQ2: How are product position in network and the strength of comparative review associated with the
eWOM influence?
RQ3: Is there any difference in the strength of comparative reviews for the products with a high rating and
those with a low rating?
To answer these questions, we construct two predictive models using several comparative network
centrality measures. We then compare these two predictive models based on the review sentiment and
semantic similarity measures. Our results show that several centrality measures are strong predictors of sale
and the model using semantic similarity produces a more significant predictive power than the model based
on sentiment analysis. To the best of our knowledge, our study is among the first few studies (e.g. Chen et
al., 2017; Z. Zhang et al., 2013) that model products as nodes in the network, and is the first study that
applies the network centralities based on both the network ties and the semantic similarity of text to discover
the relationships among competing products in electronic markets. Due to this unique approach, the results
of our study benefit product manufacturers and market operators (e.g. Amazon) in identifying the relative
strength of eWOM and cater to customers who seek to compare multiple products. Such information can
also help vendors develop more effective marketing strategies to better promote their products.
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 4
2 Prior Literature
2.1 eWOM and Product Sales Prediction
The original WOM influence is defined as “the change in attitude and/or behavioural [purchase] intention
resulting from an interpersonal information exchange” (Gilly et al., 1998, p. 84). Extending this concept to
e-commerce, eWOM refers to “any positive or negative statement made by potential, actual, or former
customers about a product or company, which is made available to a multitude of people and institutions
via the internet” (Hennig-Thurau et al., 2004, p. 39). There are two streams of eWOM influence studies in
the extant literature. One investigates how the content of the review influences customer purchase decisions
or market outcomes (J. A. Chevalier et al., 2006; Hu et al., 2014), and the other resolves around examining
how the valence of the review (i.e. the number of reviews or the helpfulness of the reviews) influences
review credibility and facilitates product purchase (Mudambi et al., 2010; Weathers et al., 2015). Among
these studies, both sales volume and total sales have been used to measure customer purchase intention.
However, sales rank emerges as the most widely used outcome variable as sales volume is difficult to obtain
and sales rank dollar amount is believed to be the best proxy for a product’s sales performance (Chen et al.,
2017; J. Chevalier et al., 2003; Hu et al., 2014; Z. Zhang et al., 2013). On Amazon, for example, Amazon
best sellers (ABS) rank is widely accepted as a prime indicator of product sales as it follows a Pareto
distribution (J. Chevalier et al., 2003). In the second stream of eWOM studies, the focus is on the
helpfulness and readability of the reviews and how they help to stimulate purchase intention and decision
(Ghose et al., 2010; Hu et al., 2008; Mudambi et al., 2010; Salehan et al., 2016). Recent studies utilizing
the textual characteristics of the reviews posit that the readability of the reviews and the review depth,
which is the number of words used in a review and a measure of cognitive effort that a user invests, influence
product sales (Ghose et al., 2010). Using a two stage least squares model, Ghose et al. (2010) find that text
reviews and review characteristics, such as review rating, review sentiment, and readability change the
average product rating, and consequently product sales.
2.2 Product Comparisons, Competitive Intelligence, and Sales Prediction
Prior research has shown that product comparison increases the likelihood of product purchase, and
experienced buyers will explicitly look for comparative reviews as they evaluate their choices (Dhar et al.,
2004). Therefore, text mining of comparative reviews has gained popularity in recent years, and various
studies have found evidence that demonstrates its potential to increase sales (Vemprala et al., 2019; Xu et
al., 2011). Building on a product network formed using the comparative reviews, Z. Zhang et al. (2013)
estimate how the sentiment polarity in the review affects product sales based on the sentiment scores
assigned to the review text. Their results show that the higher the sentiment of comparative reviews, the
lower the sales rank. The comparative network uses directed links based on the polarity of review sentiment
in their research. Specifically, one creates a link towards the product if the review text polarity is positive,
and if the review polarity is negative then we create a link outwards from the product to the compared
product. The more the number of such positive reviews, the more influence these eWOM carries towards
sales. The follow-on studies based on their research and other similar research posit that the position of a
product within a network and the number of connections has minimal to no influence on the product sales.
2.3 Strength of Ties
Social networks are one of the most prominent forms of networks that help in the diffusion of product
related information among consumers. For example, this effect has been observed in book sales where the
book readers trust the opinions of other readers who read similar books in the past (J. A. Chevalier et al.,
2006). Similar kind of effect is validated and tested in studies that examine product recommendations and
identify competitors through eWOM sentiments analysis (Leem et al., 2014). They find that competing
products resemble entities in a social network due to their competitive relationship and the fact that
customers are often directly comparing the products. Recent studies on e-commerce sales prediction and
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 5
product demand have emphasized the network perspective of ties among products that promotes product
sales (Liu et al., 2018; Z. Zhang et al., 2013). Online consumers are often actively looking for a direct
comparison of the product that they are interested in. If the comparisons favour a given product in a single
review, then such comparison will significantly influence consumer’s preference and consequently the
likelihood of purchase (Z. Zhang et al., 2013). In addition, user’s perception will be enhanced if the
comparison outcome is consistent over several similar comparisons across multiple products. One measure
to capture the influence of the review is the text review sentiment. In a product network, the review text
compares two products based on the product features. Zhang et al., (2013) capture the comparative review
sentiment of product and a competitor product through a sentiment score, which is used as a weight of the
network and compared with multiple directed networks. Another measure is the semantic relationship
embedded in the text, which is argued to be a more suitable approach for measuring eWOM influence
(Aggarwal, Vaidyanathan, & Venkatesh, 2009).
3 Hypotheses
In the existing eWOM literature, due to the increasing abundance of online reviews, there has been a recent
shift of focus from using numerical predictors (star rating, price) in their statistical models to understanding
the impact of review text (Z. Zhang et al., 2013). Hence, rather than depending on the numerical predictors,
recent studies start to explore other forms of text measures such as review sentiment (Z. Zhang et al., 2013).
Even though the review sentiment is a good measure of WOM influence, the sentiment is based on the
polarity of text and the use of language in the review, which leads to two potential problems. First, the
sentiment is specific to the set of words used in the review but not specific to the product features that the
customer is interested in. There are various factors (e.g. product quality, wait-time in receiving the product,
overall closure of transactions, and package) which may cause these sentiments. The closer the review is to
the customer expectations, the higher is the influence that the review generates towards purchase decision.
The degree of satisfaction on these features vary from customer to customer. Text sentiment considers the
language usage but not the relevance of text. Semantic relationship is a more suitable approach for
measuring eWOM influence (Aggarwal, Vaidyanathan, & Venkatesh, 2009). Second, the range of values
for the overall sentiment of a review does not vary much as it is specific to the individual text words used.
One cannot rely on the sentiment of one review and link it to other review to measure the influence based
on the strength of this value. For example, let us consider two separate comparisons, between products P1,
P2 and between P2, P3, respectively. The review sentiment of P1-P2 comparison has no relation with the
review sentiment of P2-P3. Hence, calculating the product influence based on the comparative review
sentiment would not generalize to other product networks.
In contrast, semantic similarity of review text emphasizes the relevance of text used in comparison to all
the text reviews, which has gained traction due to the capability of extracting meaning from the text than
providing the polarity of text in the case of sentiment analysis. It is measured as a semantic distance between
review text and the hidden features mentioned in the text reviews. Hence, it would allow us to better assess
the impact of comparative review on product sales. Accordingly, we hypothesize that:
H1: Review text semantic similarity in a product comparison network is more strongly associated with
product sales than text sentiment.
Prior literature on eWOM influence examines the influence of product eWOM from a single-entity
orientation, which ignores the direct comparison of competing products by customers. Z. Zhang et al.,
(2013) propose comparison networks to study the influence of one product comparing it to all the possible
products within a network. They propose various centrality measures on the comparison networks
considering both the directional and unidirectional characteristics of the network. Comparison such as
comparative text reviews are directional in nature. For these directional comparisons, they consider the in-
degree and out-degree centrality measures. Comparison between product A and B creates an in-degree link
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 6
between the network nodes A and B in the product network with direction of the link going from B to A if
the review text sentiment is positive for product A. On the contrary, an out-degree link from A to B is
formed when the review sentiment favours product B. Given that incoming links represent the extent of
positive recommendation over the focal product, we would expect the in-degree centrality to have a positive
influence on the sales of the product. Hence, we predict that:
H2: The in-degree centrality of a product in a comparison network positively affects the product sales.
In addition to these directional centrality measures, there exists a different group of centrality measures
such as closeness centrality, Betweenness centrality, eigenvector, and PageRank, which capture the
influence of the node without considering the directionality. Closeness centrality focuses on how close a
node is to all the other nodes in the network beyond those it is directly connected to. A node is considered
to be central if it can quickly reach all the others. Betweenness centrality measures the extent to which a
particular node lies between other nodes in a network. Eigenvector and PageRank measure how close a
node is to influential nodes in the network, which presumably allows a node to share some of the influences
carried by those influential nodes. In general, these undirectional centrality measures are based on position
of nodes in a network and primarily focus on the number and type of links a node connects with in the
network or the strength of the link, which is the sentiment score in their comparison network. In the previous
studies, there has been inconclusive findings on whether these undirectional centrality measures have not
been shown to have a significant impact on predicting the eWOM influence (Leem et al., 2014; Z. Zhang
et al., 2013).
Using Z. Zhang et al. (2013)’s study as the baseline, Chen et al. (2017) consider intercommunication
network on top of the product comparison network to predict the eWOM influence. They also use
comparative reviews and review’s comments to derive the product influence. However, the weights of
network ties are the sentiment of the comparative review or comment text, and do not consider the semantic
relationship between reviews and comments. The product influence measure should capture two important
factors. First, multiple usersinterests about the similar feature measured based on semantic similarity, and
second, it should consider the network ties that a user can reach from a specific product (Bakshy et al.,
2012). Network centrality measures, namely, degree centrality, betweenness centrality, and closeness
centrality are shown to capture the node influence in a network (Chu et al., 2011; Leem et al., 2014; Shi et
al., 2007). Closeness centrality and betweenness centrality consider the distance between two nodes or
products. For example, if a customer reads a comparative review of Samsung S9 and Xiaomi 7, and another
comparative review of Xiaomi 7 and Huawei P30, then there is an intermediate tie. This leads to a higher
search cost. The more these number of ties, the more is the distance and less is the influence it creates on
the first product sales. Closeness centrality relies on the length of the paths from a node to all other nodes
in the network, and is defined as the inverse total length. Betweenness centrality relies on the identification
of shortest paths, and measures the number of them that passes through a node. These measures are
consistently proven to be significant in predicting product demand (Leem et al., 2014). However, they do
not consider the importance of node position. Recent studies addressed the issue and provided separate
measures considering equal importance to the ties and their strength (Opsahl et al., 2010). Hence, we
consider both the semantic similarity as weights, and the product position within a comparison network to
calculate the centrality measures and frame our hypothesis as below:
H3: The betweenness centrality of a product in a comparison network positively affects the product sales.
H4: The closeness centrality of a product in a comparison network positively affects the product sales.
Prior literature on product promotion in E-commerce posits that the product sales of search products
depends on the product rating (L. Zhang et al., 2013). When it comes to examining the impact of the
comparative reviews, we argue that if a highly-rated product was involve in a comparison, it tends to have
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 7
a stronger persuading effect than if a product with a relatively low rating was being used in the comparison.
Accordingly, product sales are stronger for highly rated products with higher semantic similarity and the
sales decreases with an increase in the number of similar comparisons. If more number of comparisons on
similar feature are available, then the comparative review may not be helpful, even when the product is
highly rated. This is because as the customer has more options (products) to choose from. In such a scenario,
the influence of WOM semantic similarity will be minimal. However, the reverse might not be true, as the
low rating product has relative hidden advantages like price. If the customer requirement is satisfied in the
comparative review, the low rating product benefits through long chain of comparisons on similar feature.
H5: The product rating moderates the relationship between closeness centrality and product sales such
that the high rated product receives lesser sales with higher closeness centrality.
H6: The product rating moderates the relationship between betweenness centrality and product sales such
that the high rated product receives lesser sales with higher betweenness centrality.
4 Research Methodology
4.1 Data collection
To provide an outline of our methodology, we summarize the steps involved in our research methodology
in Figure 1. In step 1, in order to build and analyze the impact of comparative text reviews semantic
similarity, we collected customer review data from Amazon.com, a major e-commerce platform, using an
automated software script. Our data collection focused on the customer reviews for mobile phones due to
their popularity on Amazon and the large number of reviews. The final dataset contains over 4 million
customer reviews (about 45,000 text reviews per day) for 236 mobile phones over a 90-day period. The
descriptive statistics of the key variables and their correlations are presented in Tables 1 and 2.
Figure 1. Research framework.
Key Variables
Count
Min
Max
Mean
Std. dev
Reviews
45,416
5
2665
213
26
Comparative Reviews
6188
2
326
24
8
Price (Pr)
6188
46.94
1089.95
296.82
216.55
Review depth (Rd)
6188
78.86514
782.3379
226.4906
118.8241
AverageRating (avgR)
6188
1
5
3.3
.4160978
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 8
In-degree centrality (In_C)
6188
0
56.53377
2.121413
4.390354
Out-degree centrality (Out_C)
6188
0
44.02748
3.54457
5.547812
Closeness Centrality (CC)
6188
0
3.0946
1.21899
.4707519
Betweenness Centrality (BC)
6188
1.029024
.0278096
.072319
Sales Rank (SR)
6188
1
348
-
-
Table 1. Descriptive Statistics
Pr
Rd
avgR
In_C
Out_C
CC
BC
SR
Pr
1
Rd
0.1819
1
avgR
0.3658
0.2551
1
In_C
-0.0339
-0.0423
-0.0816
1
Out_C
0.2145
0.3033
0.1744
0.1199
1
CC
0.0496
0.2182
0.0068
0.2683
0.3802
1
BC
-0.0034
0.1828
0.1338
0.1956
0.4118
0.3477
1
SR
0.0703
-0.0892
0.0394
-0.0518
0.1278
-0.0743
-0.0911
1
Table 2. Correlation Matrix
4.2 Extracting comparative reviews
The second step in our research methodology is to identify the comparative reviews. Our approach is similar
to that of Z. Zhang et al. (2013). We first extracted the product names from the text reviews we obtained.
Then, we implemented pre-processing logic to remove the stop words (a, an, the, and other commonly used
words) and created a dictionary of product tree structure. The tree structure contains information such as
product brand, product series and model (e.g. BLU Grand M2, and Samsung Galaxy Note 10+). We also
applied other set of heuristics such as removing the product feature specifications (e.g. 32 GB, carrier
specifications such as AT&T), colour and other commonly used keywords that are not unique for the
product. We then calculated bi-grams of the product name. We split the review text into bi-grams after
implementing the same text pre-processing logic that we implemented to create product names bi-grams.
These bi-grams of review text and the bi-grams of all the products are compared to extract the common
match of the bi-grams. Based on the common match of the bi-grams in the review text and the products
dictionary, we identified the presence of comparative reviews.
4.3 Latent Semantic Analysis
The next step in our methodology is to use the comparative reviews and identify the latent features that are
used in the text to compare the products. The latent features are extracted based on the latent semantic
analysis (LSA). LSA is a topic model algorithm for grouping text documents, and can be used to recognize
different words by their context instead of their spelling. In addition, the same word can have two different
meanings. Therefore, two same words with different meanings are separated into two different topics based
on their hidden meanings using LSA. Other forms of topic models group similar words together based on
the dictionary meaning. LSA is an unsupervised machine learning model, where no labels are required for
the training data. LSA returns the hidden topics from the comparative reviews.
LSA requires the user to provide the number of topics to be extracted. Users compare products on many
features. However, as the product type is restricted to mobile phones, the number of topics is also limited.
Hence, we chose the top five topics from the comparative reviews. Using the topic keywords, we measure
the semantic similarity between the comparative review and each topic. The topic mentioned in the
comparative review is the topic that has greater value of similarity to the comparative review.
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 9
4.4 Product Network and Centrality
We constructed a product network using the comparative reviews. The products that are compared in the
reviews are the network nodes, and the links represent a comparison in the network. We used both directed
network centrality measures and undirected network centrality measures to estimate the product’s eWOM.
The product’s eWOM influence depends not only on the strength of the comparison, the semantic similarity
to the features mentioned in the review, but also on the network position of a product. For example, new
products may have attractive features, but they are compared only with few products on the key topics.
However, as the number of connections of this new product is limited, so does its eWOM influence is
limited too. Hence, we propose modified centrality measures for both the directed and undirected networks,
and use these new measures in the directed and undirected measures to capture product influence. Finally,
we statistically validate the significance of these network measures on product eWOM.
4.4.1 Undirected Graph Centrality Measures
In network theory, the node influence is a characteristic that specifies the importance of a node in the
network. A central node is the most influential node that enjoys some privileges either through the number
of connections or due to the strength of these connections. In the case of the product comparison network,
we posit that there are three parameters that decide the influence of a product. These three parameters are:
the semantic similarity of comparative review text, the number of comparisons made, and the sentiment of
the comparative text. These three parameters drive a customer’s product search cost. If the comparative
review is related to the product feature customer is expecting, and if it is positive and strong, then the
customer decides to purchase the product. However, the customer is also likely to look for one or more
alternate products. This is referred to as search cost. The alternative product is usually a competitor product
that is mentioned in the comparative review. The search cost increases if the comparative review is not
related to the feature that the customer is expecting or if the sentiment of the review is negative. The extent
of searching for alternatives is limited. It depends on the number of product comparisons, the strength of
comparisons, and the sentiment of comparisons. Existing centrality measures closeness centrality and
betweenness centrality is based only on the number of connections or on strength, but not both. Degree
centrality is based on the number of connections. In a social network, the influence of one node to reach
many other nodes to access the information quickly is not determined by how closely the node is connected
to all other nodes, but due to the position of the connections within the network (Opsahl et al., 2010). To
capture this feature of connectedness through node position, closeness centrality is defined as the inverse
sum of the shortest distances to all other nodes from a focal node. Another measure of a node’s influence
to reach other nodes to access the required information is based on the shortest path that connects the nodes
in a network with strong connections. Betweenness centrality (CB) is designed to capture this feature. CB
measures the degree to which a node lies on the shortest path between two other nodes. None of the
centrality measures consider both. Hence, we adopted the centrality measures from Opsahl et al. (2010) as
below:
Degree Centrality:
 

--- (1)
Closeness Centrality:
 
  --- (2)
Betweenness Centrality:
  


 --- (3)
In the above equations, w denotes the strength of the connection, α is the tuning parameter, which
determines the importance of the number of ties compared to tie strength, i is the product (node) for which
influence is measuredand j is the product (node) that is compared in the review text (both i and j form a
connection), gjk is the number of binary shortest paths between two nodes and gjk(i) is the number of those
paths go through node ‘i’, N is the total number of products in the comparative product network, S is the
sum of the strength for all the connections that product i has, and k is the number of connections. d is the
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 10
shortest distance between connections product i and j considering, strength w * path length between i and
j. It is measured as below:
 
   
 --- (4)
In Equation (4), h is the intermediate product between product comparison path i and j.
For an undirected network, the directions of connections do not matter and only the strength of the
connections is important. We measure the strength of the connections consolidating all the comparative
reviews semantic similarity. However, we account for the polarity of comparative text review sentiment
while summing up the overall similarity value. For example, there are two comparative reviews between
P1 and P2. The first review is for product P1, and it is a positive of product P1 (and compared with product
P2). The second review is a negative review for product P2, in comparison to product P1. In this scenario,
we will add the semantic similarity values of both text reviews. This process ensures that the semantic
similarity is consistent across the products of the network.
4.4.2 Directed Graph Centrality Measures
We built a directed network to measure the product’s eWOM based on the relevance of the comparative
review text and product sentiment, and calculated the overall review text sentiment based on the
SentiWordNet lexicon (Esuli et al., 2006). If the review for a product P1 has a comparison of product P1
with product P2, and the sentiment of the comparative text is positive, then the comparison is said to cast a
positive vote on P1. Hence, the direction of comparison is towards P1, which is a directed edge from P2 to
P1. If the review sentiment is negative, then a directed edge from P1 to P2 is created. The strength of the
edge is measured based on the semantic similarity and the product position within the network. We
hypothesize that the in-degree centrality influences product sales. We adopted the in-degree centrality
measure based on the strength of the comparison and the number of in-degree comparisons as below:


  

 --- (5)
In the above equation, kin is the number of connections for product i’ which compare product i’ through a
positive review sentiment. Specifically, 
 is the in-degree centrality considering both the number of
positive sentiment comparative reviews towards product i and the strength of the semantic similarity of
comparative text.
4.5 Panel Data Regression Analysis
Our study consists of two parts that closely relate to the two research streams on text mining of customer
reviews. The first part is to investigate the impact of semantic relationship when compared to the
comparative review sentiment on the product’s eWOM influence. Second, we introduce two important
measures in the context of product comparison network to measure the product influence by considering
both the strength of the semantic relationships in a network and the product position within the network.
Thus, we built a set of predicting models to compare the relationships and carried out regression analysis
based on the panel data built using daily cross sectional data from the product network. The dependent
variable we use in all these models is the best sellers rank. On Amazon, ABS rank is widely accepted as a
prime indicator of product sales as it follows a Pareto distribution (J. Chevalier et al., 2003). ABS rank is a
numerical value given to each of the products sold on amazon relative to the other products. As our data
collected is for the unlocked mobile devices, we used ABS rank within the unlocked mobiles category.
Also, our predictors in the model do not have any derived variables. Hence, we used fixed effect model for
our panel data regression.
The following are the predictive models we built in this study:
Model 1:          
  
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 11
In the above model, i = 1, …, N denotes the index of products, and t is the time component representing
per day observations of our panel data.
Model 2: Model with the same number of coefficients using centrality measures calculated based on the
sentiment of comparative review sentiment as the strength.
Model 3: Model with the same number of coefficients using centrality measures calculated based on the
semantic similarity of comparative review as the strength alone.
Model 4: Model with the same number of coefficients using centrality measures calculated based on the
product position alone in the network.
Models 5 and 6 are the predictive models that test the significance of predictors without interaction terms.
Centrality measures in model 5 use semantic similarity, while the measures in model 6 use review sentiment
as tie strength. Both models consider the product position and semantic similarity in the centrality measures.
5 Results and Discussions
We conducted a fixed-effect estimation for the unbalanced panel data. Table 3 shows the results for the six
models.
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Pr
.000618
(.000196)***
.000624
(.000637)
.000586
(.000294)**
.000312
(.000128)***
.000564
(.000257)**
.000631
(.000307)**
Rd
.434152
(.614625)
.201196
(.109198)
.158431
(.166321)
.127671
(.185231)
-.000109
(.000236)
-
avgR
.0162641
(.0931126)
.0181176
(.0132456)
-.009176
(.011217)
-.203218
(.056196)***
-.316452
(.056191) ***
-.312167
(.054326)***
In_C
-.0135343
(.005153)***
-.000106
(.000046)**
-.019916
(.009814)**
.000128
(.000196)
-.0113216
(.000182)***
-.019812
(.008136)**
Out_C
.021636
(.106523) **
.008121
(.001056)***
.078126
(.057424)
.025786
(.012312)**
.0109923
(.004981)**
.017643
(.007581)**
CC
-1.20116
(.118307)***
1.105131
(.534126)**
1.091256
(.188321)***
.021652
(.023116)
-.109786
(.000109)***
0.000126
(.000098)
BC
1.127412
(1.165127)
-2.234158
(1.86142)
-2.564391
(2.324879)
.009121
(.872106)
-.451326
(.209321)*
-.482144
(.245108)*
avgR
*CC
-.362316
(.181006)**
-.313516
(.184652)*
-1.10656
(.986517)
-.003166
(.003451)
-
-
avgR
*BC
-.6149134
(.169126)***
.368512
(.453514)
-.261437
(.045716) ***
-.010926
(1.076142)
-
-
R2
0.36
0.33
0.35
0.24
0.32
0.21
Table 3. Regression Results (***p<0.01, **p<0.05, *p<0.10)
Consistent with prior literature, we used price as a control variable in all our models. The coefficients of
the price are positive across all the models, which suggest that the increase in the product price (increase in
sales rank is considered as decreasing in sales) has a negative impact on product sales. We compared models
1 and 2, 5 and 6 based on the F-statistics (567.90, 396.85 for 1, 2 models and 374.25, 283.6 for 5, 6
respectively), to suggest that semantic similarity provides a higher explanatory power. We constructed the
predicators for models 1 and 5 using semantic similarity, models 2 and 6 using text sentiment, and models
5 and 6 with no interaction terms. We observed from the results presented in Table 3 that the predictors for
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 12
models 1 and 5 are significant at the 0.05 level, and have a higher explanatory power (than models 2 and
6). This provides support for our hypothesis H1, in the sense that review text semantic similarity in a product
comparison network provides better eWOM influence measures than text sentiment. From both model 1
and 5, in-degree centrality is significant and negative, supporting our hypothesis h2 such that with an
increase in the number of positive comparative reviews for a given product, its sales increases. Based on
the sign of betweenness centrality and closeness centrality measures of model 5, we can support our claim
for hypotheses H3 H4, such that the betweenness and closeness centrality measures calculated based on
the semantic similarity of text reviews and the product position within a comparison network positively
influence product sales. However, the significance of betweenness centrality is at the 0.1 level.
Model 3 uses the centrality measures captured based on the semantic relationship. The coefficient of
closeness centrality is positive, which implies that the product sales increases when the review provides
meaningful information. Also, half of the predictors are not significant. This suggests that the model is
incorrect. The coefficients for models 1 and 5 are both consistent. This provides support for our hypotheses
H3-H4, such that products with both favourable semantic similarity distance and network positions within
a comparative network secure better sales.
The coefficients of closeness and betweenness centrality measures are negative in models 1 and 4. This
suggests that the eWOM influence has a positive impact on sales, and the value of sales rank (better rank)
decreases when the centrality measures are high and positive with close semantic relationship. From the
regression results, models 1 and 5 have consistent results with the model 1 showing interaction between an
average product rating and the closeness centrality and betweenness centrality, and without interactions in
model 5. The interaction term in our model is also significant, and thus providing support for hypotheses 5
and 6. From the interaction plot, we observe that an increase in closeness centrality for highly rated products
resulted in increased sales rank.
Figure 2. Interaction Plot between closeness centrality and average rating.
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 13
Figure 3. Interaction Plot between, betweenness centrality and average rating.
Additionally, to check for the possible multi-collinearity between our predictors, we conducted a VIF
(Variance Inflation Factor) test. As shown in Table 4, we do not observe any significant collinearity (VIF
> 10) between the variable.
Variable
VIF
1/VIF
Price (Pr)
1.28
0.78
Review depth (Rd)
2.19
0.46
AverageRating (avgR)
2.46
0.41
In-degree centrality (In_C)
1.22
0.82
Out-degree centrality (Out_C)
1.55
0.65
Closeness Centrality (CC)
1.09
0.92
Betweenness Centrality (BC)
1.18
0.85
Sales Rank (SR)
1.46
0.68
Table 4. Collinearity Statistics
6 Findings, Contributions, and Limitations
Our study confirms that the comparative reviews are key eWOM measures that influence the products sales
within a product network constructed from similar products competing in the same product category. Our
findings demonstrate that the text semantic similarity is a better measure of strength for a comparative
product network than the review sentiment. We have provided empirical evidence that the centrality
measures based out of product position and semantic similarity explain the relationship between product
sales and comparative reviews better than the centrality measures considering either the product position
or the semantic similarity alone. We have also shown that the in-degree centrality is a significant measure
of the product influence, a finding that is consistent with prior studies (Leem et al., 2014).
Our research makes several theoretical contributions towards the literature on eWOM, the network theory,
and the theory of the strength of ties through the new measures we have developed for the product
comparison networks. Current literature typically examines the impact of the strength of ties and the
positions of nodes in the network separately. Integrating these two important network properties, our
empirical validation of centrality measures based on both the strength and positions provides new insights
into the existing literature. Our research also provides practical implications for decision making on what
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 14
features of a product manufacturers should focus when they seek to improve their product quality and
identifies the significance of comparative reviews that influences product eWOM.
Like many other empirical studies, our study is not without limitations. Even though we have demonstrated
that the semantic relationship of comparative reviews provides better insights in answering eWOM
influence on product sales, the findings were demonstrated in only one product category. These limitations
present opportunities for future research. As an ongoing effort to extend our study, we are continuing our
data collection and expanding it to include more product categories on multiple e-commerce platforms, so
that we can examine if the findings can be generalized to more product categories and across e-commerce
platforms.
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 15
References
Aggarwal, P., Vaidyanathan, R., & Venkatesh, A. (2009). Using lexical semantic analysis to derive online
brand positions: An application to retail marketing research. Journal of Retailing, 85(2), 145-158.
Bakshy, E., Eckles, D., Yan, R., & Rosenn, I. (2012). Social influence in social advertising: evidence from
field experiments. Paper presented at the Proceedings of the 13th ACM conference on electronic
commerce.
Chen, K., Luo, P., & Wang, H. J. I. (2017). An influence framework on product word-of-mouth (WoM)
measurement. 54(2), 228-240.
Chevalier, J., & Goolsbee, A. (2003). Measuring prices and price competition online: Amazon. com and
BarnesandNoble. com. Quantitative marketing and Economics, 1(2), 203-222.
Chevalier, J. A., & Mayzlin, D. (2006). The effect of word of mouth on sales: Online book reviews. Journal
of marketing research, 43(3), 345-354.
Chu, X., Zhang, Z., Guan, J., & Zhou, S. (2011). Epidemic spreading with nonlinear infectivity in weighted
scale-free networks. Physica A: Statistical Mechanics and its Applications, 390(3), 471-481.
Church, E. M., Iyer, L. S., & Zhao, X. J. I. S. M. (2015). Using interest graphs to predict rich-media
diffusion in content-based online social networks. 32(3), 210-219.
Deerwester, S., Dumais, S. T., Furnas, G. W., Landauer, T. K., & Harshman, R. (1990). Indexing by latent
semantic analysis. Journal of the American society for information science, 41(6), 391-407.
Dhar, R., & Nowlis, S. M. (2004). To buy or not to buy: Response mode effects on consumer choice.
Journal of marketing research, 41(4), 423-432.
Esuli, A., & Sebastiani, F. (2006). Sentiwordnet: A publicly available lexical resource for opinion mining.
Paper presented at the LREC.
Ghose, A., & Ipeirotis, P. G. (2010). Estimating the helpfulness and economic impact of product reviews:
Mining text and reviewer characteristics. IEEE Transactions on Knowledge and Data Engineering,
23(10), 1498-1512.
Gilly, M. C., Graham, J. L., Wolfinbarger, M. F., & Yale, L. J. (1998). A dyadic study of interpersonal
information search. Journal of the academy of marketing science, 26(2), 83-100.
Granovetter, M. S. (1977). The strength of weak ties Social networks (pp. 347-367): Elsevier.
Hennig-Thurau, T., Gwinner, K. P., Walsh, G., & Gremler, D. D. (2004). Electronic word-of-mouth via
consumer-opinion platforms: what motivates consumers to articulate themselves on the internet?
Journal of interactive marketing, 18(1), 38-52.
Hu, N., Koh, N. S., & Reddy, S. K. (2014). Ratings lead you to the product, reviews help you clinch it? The
mediating role of online review sentiments on product sales. Decision Support Systems, 57, 42-53.
Hu, N., Liu, L., & Zhang, J. J. (2008). Do online reviews affect product sales? The role of reviewer
characteristics and temporal effects. Information Technology and management, 9(3), 201-214.
Jin, J., Ji, P., & Kwong, C. K. (2016). What makes consumers unsatisfied with your products: Review
analysis at a fine-grained level. Engineering Applications of Artificial Intelligence, 47, 38-48.
Kim, S., Kandampully, J., & Bilgihan, A. J. C. i. H. B. (2018). The influence of eWOM communications:
An application of online social network framework. 80, 243-254.
Leem, B., & Chun, H. (2014). An impact of online recommendation network on demand. Expert Systems
with Applications, 41(4), 1723-1729.
Vemprala et al. /How Helpful are Comparative Reviews?
Twenty-Eigth European Conference on Information Systems (ECIS2020), Marrakesh, Morocco. 16
Liu, C., Jozani, M. M., & Choo, R. (2018). Canalization or increased diffusion? An empirical analysis on
the impact of the recommendation system in the mobile app market. Proceedings of the 51st Hawaii
International Conference on System Sciences.
Mudambi, S. M., & Schuff, D. (2010). What makes a helpful review? A study of customer reviews on
Amazon. com. MIS quarterly, 34(1), 185-200.
Opsahl, T., Agneessens, F., & Skvoretz, J. (2010). Node centrality in weighted networks: Generalizing
degree and shortest paths. Social networks, 32(3), 245-251.
Salehan, M., & Kim, D. J. (2016). Predicting the performance of online consumer reviews: A sentiment
mining approach to big data analytics. Decision Support Systems, 81, 30-40.
Shi, X., Adamic, L. A., & Strauss, M. J. (2007). Networks of strong ties. Physica A: Statistical Mechanics
and its Applications, 378(1), 33-47.
Trenz, M., & Berger, B. (2013). Analyzing Online Customer Reviews-An Interdisciplinary Literature
Review And Research Agenda. Paper presented at the ECIS.
Vemprala, N., Xiong, R. R., Liu, C. Z., & Choo, K.-K. R. (2019). Where Does My Product Stand? A Social
Network Perspective on Online Product Reviews. Paper presented at the Proceedings of the 52nd
Hawaii International Conference on System Sciences.
Weathers, D., Swain, S. D., & Grover, V. (2015). Can online product reviews be more helpful? Examining
characteristics of information content by product type. Decision Support Systems, 79, 12-23.
Xu, K., Liao, S. S., Li, J., & Song, Y. (2011). Mining comparative opinions from customer reviews for
competitive intelligence. Decision Support Systems, 50(4), 743-754.
Zhang, L., Ma, B., & Cartwright, D. K. (2013). The impact of online user reviews on cameras sales.
European Journal of Marketing, 47(7), 1115-1128.
Zhang, Z., Guo, C., & Goes, P. (2013). Product comparison networks for competitive analysis of online
word-of-mouth. ACM Transactions on Management Information Systems (TMIS), 3(4), 20.
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
Online retailers have increasingly adopted product recommendation systems as an effective tool to improve product visibility and promote sales. This study examines the impact of the recommendation system in the popular Google Play mobile app store. By analyzing a 60-day panel dataset with 235,638 observations from 9,735 apps, we investigate how the characteristics of the recommended apps relative to those of the focal apps affect the adoption of mobile apps in this volatile market. Our results show that the relative strength of the recommended apps over the focal app plays a key role in influencing the outcome of recommendations. Moreover, the heterogeneity of the recommendations as represented by the diversity of the popularity of the recommended apps is positively associated with a more even distribution of revenue in the market. These findings provide insights for mobile app market operators to enhance the design of their recommendation systems.
Article
Full-text available
Many online retailers and other product-oriented websites allow people to post product reviews for use by shoppers. While research indicates that these reviews influence consumers' shopping attitudes and behaviors, questions remain about how consumers evaluate the product reviews themselves. With the current research, we introduce a new methodology for identifying the review factors that shoppers use to evaluate review helpfulness, and we integrate prior literature to provide a framework that explains how these factors reflect readers' general concerns about the diagnosticity (uncertainty and equivocality) and credibility (trust and expertise) of electronic word-of-mouth. Based on this framework, we offer predictions about how the relative importance of diagnosticity and credibility should vary systematically across search and experience product types. By analyzing secondary data consisting of over 8000 helpfulness ratings from product reviews posted by shoppers on Amazon.com, we find that, while review content affects helpfulness in complex ways, these effects are well explained by the proposed framework. Interestingly, the data suggest that review writers who explicitly attempt to enhance review diagnosticity or credibility are often ineffective or systematically unhelpful. Our findings have implications for both IS developers and retailers for designing online decision support systems to optimize communication practices and better manage consumer-generated content and interactions among consumers.
Conference Paper
Full-text available
Online customer reviews increasingly exert influence on customers’ purchase decisions when shopping online and give new importance to the concept of word-of-mouth. This is reflected in a growing body of academic literature across varying disciplines that draws on online customer reviews as source of information. However, these studies apply varying methods and obtain contradicting results. We conduct a systematic and interdisciplinary literature review to examine how online customer review data is used in academic research and which insights these studies provide. Analyzing 49 journal articles we find that the most studies investigate online customer reviews’ effect on sales, review helpfulness or review manipulation. Furthermore, the variety of product categories and review websites these studies receive their information from is rather limited. The results reveal that previous research can only provide an imperfect understanding of the impact of electronic word-of-mouth on business. We therefore develop a twofold research agenda with regards to (a) studies investigating customer reviews and its effects and (b) issues with respect to the use of customer reviews as a data source in academic research.
Article
With the development of e-commerce platforms, online customer reviews have become an important instrument for providing product word-of-mouth (WoM) information. Analyzing and measuring WoM is quite valuable in product design, sales prediction, marketing strategy, and other decision-making tasks. In contrast to previous studies that analyze product WoM focusing on a single product, we propose an influence framework to measure WoM from a market perspective. In this framework, we combine product competition relationships and customer intercommunication relationships to construct a two-layer network and calculate the node influence effects in the network. To compare different product WoM measures, we use product sales as a predictor and build a series of predictive models. In the experiments conducted based on Amazon.com data, we find that, first, textual sentiment analysis produces a better summary of customer opinions than rating scores. Second, product-comparative relationships provide additional information on measuring product WoM. Third, the customer intercommunication feature in social media is useful for measuring the collective opinions about a product. The influence framework and experimental findings have both theoretical and managerial implications.
Article
Although online consumer reviews (OCRs) have helped consumers to know about the strengths and weaknesses of different products and find the ones that best suit their needs, they introduce a challenge for businesses to analyze them because of their volume, variety, velocity and veracity. This research investigates the predictors of readership and helpfulness of OCR using a sentiment mining approach for big data analytics. Our 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. Because the current methods used for sorting OCR may bias both their readership and helpfulness, the approach used in this study can be adopted by online vendors to develop scalable automated systems for sorting and classification of big OCR data which will benefit both vendors and consumers.
Article
Online product reviews contain valuable information regarding customer requirements (CRs). Intelligent analysis of a large volume of online CRs attracts interest from researchers in various fields. However, many research studies only concern sentiment polarity in the product feature level. With these results, designers still need to read a list of reviews to absorb comprehensive CRs. In this research, online reviews are analyzed at a fine-grained level. In particular, aspects of product features and detailed reasons of consumers are extracted from online reviews to inform designers regarding what leads to unsatisfied opinions. This research starts from the identification of product features and the sentiment analysis with the help of pros and cons reviews. Next, the approach of conditional random fields is employed to detect aspects of product features and detailed reasons from online reviews jointly. In addition, a co-clustering algorithm is devised to group similar aspects and reasons to provide a concise description about CRs. Finally, utilizing customer reviews of six mobiles in Amazon.com, a case study is presented to illustrate how the proposed approaches benefit product designers in the elicitation of CRs by the analysis of online opinion data.
Article
Rich-media, pictures and videos, is becoming an increasingly important aspect of online social networks. Unlike social networks, where users are connected primarily because of being friends, peers, or coworkers, content-based networks (CBN) build connections between individuals founded on a shared interest in rich-media content. In this study, we examine “interest-graphs” comprised of these content-based connections. As we show, interest graph analysis provides important advantages over traditional social network analysis (SNA) to identify valuable network members and predicting rich-media diffusion.
Article
Purpose The purpose of this research is to help better understand the impact of online user reviews on sales of search goods. Design/methodology/approach This research is based on digital camera sales data collected from amazon.com and two studies are included in this research. The first study is based on a static model and sample data from one time stamp. The second study is based on two sample data collected from two different time stamps, and a dynamic model is proposed. Findings The results from the first study reveal that the average online customer review, the number of online reviews, the price and the camera's physical properties such as the number of pixels and the optimal zoom number (but not LCD screen size) have significant influence on digital camera sales. The results from the second study show that the sales from the previous period are an important indicator for future sales. In addition, change in price, change in average online review rating and change in the total number of online reviews are all significantly associated with future sales. Research limitations/implications The research reveals that there is a significant relationship between the online user review and sales of search goods, and the influence of online user reviews on search goods sales is different from that on experience goods. It also recognizes that the product specifications influence the sales of search goods. In addition, the research on search goods shows that price at the specific time and price changes are significant factors affecting sales. Practical implications The research indicates that retailers should provide channels for, and encourage, customer online reviews for search goods to improve sales. It is also beneficial for online retailers to provide detailed product attributes to help their customers make the purchase decision. Carefully designed and executed price promotions could also be effective ways to improve sales of searchable goods. Originality/value This study is one of the first attempts to investigate the impact of online user reviews on sales of search goods.