Conducting product comparative analysis to outperform competitor’s product using Teardown JST Model

Abstract

Due to the massive number of products being produced every year in every industry, firms have witnessed a tremendous growth in innovation of methods to create a sustainable competitive advantage. For the past decade and with the availability of online consumer reviews, companies and researchers have developed many approaches utilizing electronic Word-of-Mouth to improve and develop products and services to outperform competitors. The purpose of this study is to construct an effective method to perform a better product comparative analysis based on online consumer reviews. We propose a novel framework called Teardown Joint Sentiment-Topic analysis model consisting of a combination of text analytical approaches incorporated with a developed method of the traditional teardown analysis product comparative approach. The proposed approach is fully unsupervised model that employs Latent Dirichlet Allocation topic modeling to form topics which are classified according to their sentiments. Topics are then analyzed against competitive products and critical topics are identified using a developed teardown method. A case study analyzing online customer reviews of competing products in two domains (i.e., mobile phones and surveillance cameras) is conducted. The identified critical topics are further analyzed in view of products’ specifications perspective. We found that the detected aspects of the selected products are indeed critical, and hence, they need to be improved in order to gain a competitive advantage. The significant result of this study shows that the proposed method is effective in conducting products comparative analysis and provides valuable insights into utilizing the consumer reviews for product development.

Keywords

text mining customer reviews competitive advantage product comparative analysis joint sentiment topic teardown analysis

Introduction

With the recent rapid growth of the competitive market, companies have been developing their best strategies to achieve a competitive advantage aiming for leading the market. The competitive advantage is a phenomenon that was discussed by researchers for several decades and it is gained when a product is superior to all the other comparable products in the market (Porter, 1985). In order for this to be accomplished, firms have been exploring emerging methods to understand the market requirements and to meet the customer expectations. Different strategies have been proposed by researchers to assist firms in improving their current services or developing a new product. A study by (Zhang et al., 2020) exploited sales data to predict customers preferred product features using a machine learning approach. The introduction of concurrent engineering as an improvement of project management in new products development represents a new strategy that enables firms to gain the competitive advantage (Rihar et al., 2020). In today’s market, the voice of customer (VOC) is not only regarded as a trustworthy source for customers’ impressions about products but also an effective way for organizations to understand customer expectations in order to lead the market. In addition, online customer reviews and social networks are believed to influence customers’ purchase decisions (Maslowska et al., 2017). Therefore, it is essential for firms to explore the online reviews to be a step ahead among competitors in the competitive market.

There has been a vast amount of research on utilizing the VOC for several research fields like product aspect ranking (Zha et al., 2014), product design (Ireland and Liu, 2018), and sales prediction (Magdum and Department, 2017). The majority of VOC studies have explored extracting important keywords from customers’ reviews without any consideration to keywords’ synonyms or how those keywords are related. To identify product features with synonyms, researchers have employed several topic-modeling techniques, most of which were Latent Dirichlet Allocation (LDA) (Blei et al., 2003). LDA is a well-known probabilistic model used by researchers to discover the latent ideas contained in the documents. A significant number of studies have utilized LDA to understand the consumer needs in many domains (Ko et al., 2017; Korfiatis et al., 2019; Wang et al., 2018); however, LDA assumes all reviews share the same set of topics, neglecting the importance of consumer sentiments. Thus, integrating topic-modeling techniques with sentiment classification provides more insights about consumer requirements. Joint Sentiment Topic (JST) model is used to extract sentiment and mixture of topics simultaneously based on topic generation models such as LDA (Lin and He, 2009). A considerable number of studies have dealt with applying JST on textual data and online reviews to extract topics along with their sentiments (Dong et al., 2018; Li et al., 2019; Sowmiya and Chandrakala, 2015).

With respect to competitive analysis, (Wang et al., 2018) have employed LDA to do topic analysis for two competitive products. Their model has two major concerns, first, the process of identifying the critical topics from a list of extracted topics has to be done manually, and second, one of the limitations of this work is that the collected reviews need to be pre-classified into two sets of positive and negative sets based on their online ratings before applying the model. In another study, Ko et al. (2017) used LDA to extract topics and chance discovery theory to identify opportunities of one product aiming at improving firm competitiveness. They as well, have completely neglected the importance of consumer-expressed sentiment, without which, an accurate interpretation of user requirements cannot be achieved. Besides, to have an intensive competitive analysis, it is important to investigate the status of other products in the same industry rather than analyzing one firm’s product. With that being said, building reliable methods that make use of online consumer expressed feelings to construct competitive analysis effective systems needs further exploring. To fill the gaps in the literature, we propose a novel framework to extract and classify consumer requirements in form of topic-sentiment pattern incorporated with a developed teardown analysis system to determine which factors of products or services are needed to be improved in order to outperform competitors. Product teardown analysis is a comparative analysis method that helps to better understate how a competitor’s product is designed (Farel and Silverman, 2008). The traditional product teardown analysis concept involves putting hands on the competitor’s product and dissembling it into small units to learn about the product design and associated costs (John et al., 2014). This approach is time-consuming and it ignores how the consumer feels about the product. In this study, however, we develop a teardown method layer that can incorporate with JST model to determine the product’s aspects need to be improved. The teardown method layer receives the topics, keywords and their sentiments of one product from sentiment-topics matrix produced by JST model and it compares them with the topics stored in the sentiments-topic matrix of the other products. Our proposed model is fully unsupervised with no prior classification or labeled data. The motivations for this detailed competitive analysis study are as follows:

• It helps firms to better understand the customers’ requirements.

• Recognize the identifications of products according to consumer perspective.

• Develop the firm’s competitive strategies in the market.

We contribute to the literature by proposing a novel product comparative analysis model called Teardown Joint Sentiment-Topic (TDJST) to improve the current product and to identify the opportunities while developing a new product. This helps to gain competitive advantage by extracting the most essential aspects in customers’ reviews, determine the relation between them by constructing topics using LDA topic modeling and simultaneously detect sentiment for each topic. This paper is structured as follows: Section 2 discusses the literature review. Section 3 describes the research approach. Section 4 illustrates the case study while Section 5 presents the case study results and discussions. Section 6 concludes the paper and discusses ideas for future research.

Literature review

Due to the huge production development in every industry, and with the growth of information accessible on web 2.0, the customer tends to look for available alternatives in the market before making the purchase decision. Considering the boost of products in the market, recent years have witnessed a rapid rise in the competitive markets. Since customer satisfaction is one of the major indicators that measure the performance of enterprise competitiveness (Prasad, 2001), companies in the same industry are attempting to achieve sustainable competitive advantage by coping with the latest markets trends to meet the consumer expectations.

Many studies have approached achieving competitive advantage through different strategies. M. Itoh (2004), for instance, has conducted a comparative analysis of the digital camera industry based on product architecture by acquiring data from product specifications, actual market prices, and industry association manufacturing statistics to analyze the product strategy changes of the market competitors. The author argues that company’s choice of product development activity affects how they create and capture value. Based on the concept of Quality Function Deployment (QFD), Prasad (1998) has combined market research data with QFD by integrating Technical Importance Ratings (TIRs) and Customer Importance Ratings (CIRs) with value engineering and value graphs. The author assumes such combination assists product designers to obtain priorities for product improvement characteristics, which in turn, contributes to maintaining a competitive edge in global marketplace. On a similar basis, Lin et al. (2006) have suggested an integrated approach for product design to evaluate the structural relationships between customer requirements and design characteristics. A questionnaire survey was used to implement a semantic differential method for customer requirements identification. Apart from analyzing product specifications, Prasad (2016) argues that the competitive advantage can be achieved by improving the way how companies carry out their development process and how their teams utilize their resources. Similarly, many other studies developed methods attempting to achieve competitive advantage through several schemes such as product innovation and market driving (Kuncoro and Suriani, 2018), knowledge management (Lee et al., 2016), and supply chain (Wu et al., 2017). These studies, however, relied on collecting data from several sources like surveys and official online information sources missing out the importance of utilizing online consumers’ opinion. With the consumer being the cornerstone of company’s competitiveness measures to improve business performance (Prasad, 2001) and firms being adapting international marketing strategies to target international competitive markets (Samiee and Chirapanda, 2019), it becomes essential for firms to construct plans that aim at exploring a wide range of consumer opinions from different platforms and geographical locations.

The online WOM has been considered by many studies so far to be a reliable source for product development (Ko et al., 2017; Tiago and Veríssimo, 2014; Zha et al., 2014), due to which many researchers have handled competitive analysis by means of text mining. Identifying the critical factors of products to evaluate the competitive market from user-generated content has become a major research interest. Nevertheless, most of the studies in this field aimed for finding the competitor’s product or extracting critical aspects. Tkachenko and Lauw (2017) and Xu et al. (2011) have focused on extracting the comparative relation between aspects of two entities (products) based on consumer reviews that contain comparative sentences. Their methodology has four key information: Whether a sentence is comparative, identifying the compared entities, the comparison direction, and determining the aspects of entities. However, the comparative reviews do not represent a high number of consumers given that comparative reviews comprise only 10% of the total reviews (Kessler and Kuhn, 2013), Moreover, most consumers tend to compare products before or after product purchase through reading the chosen products’ reviews or checking online comparison sites. In other words, the consumer does not necessarily compare products while writing product reviews. Thus, considering comparative reviews only and neglecting other common reviews that might contain valuable information is not sufficient for competitive analysis. In addition, analyzing only aspects without any consideration to the hidden issues related to the aspects might not be meaningful, for example, the aspect “Battery” in electronic items could be associated with issues like heat, charger, size, weight, and battery life. Such relations need to be discovered in order to have more insights on aspects. This helps to determine what exactly needs to be improved about an aspect in current or future products.

Furthermore, most existing online WOM analysis models focus either on sentiment classification (Martin-Domingo et al., 2019; Rathor et al., 2018) or obtaining topics (Heng et al., 2018; Korfiatis et al., 2019) missing out having in-depth analysis of online WOM by understanding the expressed sentiment for every topic in the user-generated content. To overcome that, JST model is developed (Lin and He, 2009). JST model captures sentiment expressed by consumers about product’s aspects and group them into topics using topic-modeling techniques simultaneously. This is performed by extending LDA topic modeling with a new sentiment layer inserted before the topic layer.

This has motivated a considerable number of studies to investigate the efficiency and usability of JST model for different purposes in various domains. A recent study by Li et al. (2019) performed JST on consumer reviews suggesting that online reviews have a significant impact on product sales. Another study (Dong et al., 2018) developed a JST model named Unsupervised Topic-Sentiment Joint probabilistic model (UTSJ) to detect deceptive reviews. Expanding the use of JST, (Dermouche et al., 2014) has proposed a Time-aware Topic-Sentiment (TTS) model to jointly analyze topic-sentiment evolution over time.

One issue that needs to be raised and a neglected area in the literature of JST is how to implement a suitable JST model that can emphasize a sustainable competitive advantage in the market. To the best of our knowledge, no study has attempted to handle user-generated contents in process of developing a competitive advantage by considering similar product aspects as topics and determining their sentiments along with conducting teardown topics analysis. Product Teardown analysis or reverse engineering is a traditional benchmarking competitive tool involving disassembling the competitor product into its component parts which in turn are analyzed to have insights on matters such as design and cost (Farel and Silverman, 2008; John et al., 2014). Thus, this study introduces a new product teardown analysis approach called Teardown Joint Sentiment-Topic (TDJST) consisting of an incorporated system between developed versions of JST and teardown methods to outperform competitors’ products and obtain competitive advantage using eWOM. Product teardown analysis is a comparative analysis method that helps to better understate how a competitor’s product is designed (Farel and Silverman, 2008). The traditional product teardown analysis concept involves putting hands on the competitor’s product and dissembling it into small units to learn about the product design and associated costs (John et al., 2014). This approach is time-consuming and it ignores how the consumer feels about the product. In this study, however, we develop a teardown method layer that can incorporate with JST model to determine the product’s aspects that need to be improved.

Methodology

In order to perform a good competitive strategy to gain a competitive advantage over other competitors in the industry, this approach analyzes the target product (TG) which we intend to develop against competitors’ products (CPs). The framework of this study is shown in Figure 1 consisting of mainly there phases including the following process:

In the first step, the competitive products are precisely identified. Identifying the right competitive products is an essential task to achieve a competitive advantage, there is a considerable number of studies investigating the process of determining the right competitive product from online reviews (Tkachenko and Lauw, 2017; Wang et al., 2015; Xu et al., 2011). A serious drawback with these methods is that the vast amount of online reviews do not contain any comparative sentences or names of alternative products. In this study, however, given that most customers tend to read online reviews of each product they intend to compare or visit products comparison websites to gain more insights of products differences before deciding which product to purchase, we identify the competitors’ products by two approaches. First, we observe the popular comparisons that include the same products in comparison websites and second, we inspect the best product sales in Amazon that contain high volumes of consumers’ reviews in one industry. Once the competitive products have been identified, we collect data from popular online reviews websites. We choose online reviews as a source of data for the reason that online WOM is an influential factor for customers and companies when making purchase or investment decisions (Hu et al., 2014).

After online reviews collection, non-English reviews are filtered out and Natural Language Processing (NLP) techniques are implemented considering the unstructured nature of online reviews. Preprocessing techniques including part-of-speech (POS) tagging, lemmatization, and case conversion has been applied along with filtering worthless data as numbers, stop words and punctuation marks. Final step of phase one is filtering out the highest occurrence keywords. The highest occurrence keywords are often the names of brands and products, such keywords affect the quality of the result.

In the second phase, JST is employed to extract topics and their sentiment polarities. JST model is constructed based on the three hierarchical layers of LDA, namely, document, topic, and word layers (Blei et al., 2003; Lin and He, 2009). An additional sentiment layer is added between the document and the topic layers as shown in Figure 2 forming four layers. The LDA model is a probabilistic topic model that assigns each word W with distinct probabilities in collections of documents M to mixtures of topics Z. This implementation of LDA uses Gibbs Sampling and Parallel Topic Model of the Mallet library (Yao et al., 2009) with Sparse LDA sampling scheme and data structure (Yao et al., 2009). The replication of a node is represented by a plate to indicate multiple values or mixture components. In this model, sentiment labels are associated with documents and topics are associated with sentiment labels while words are associated with both sentiment labels and topics. S and T are the numbers of distinct sentiment and topic labels respectively. For a set of D documents C = $(d_{1}, d_{2}, \dots, d_{D})$ ; each document is a sequence of $N_{d}$ words denoted by d = $(w_{1}, w_{2}, \dots, w_{N_{d}})$ , and each word in the document is an item from a vocabulary index with V distinct terms denoted by 1,2,···,V.

In our study, for the sake of topics identification, the JST model is developed in such a way that the keyword that has the highest weight in every formed topic is assigned to be the topic name. If a repetition of topic names has occurred, the second highest weight keyword for the latter topic is set as the topic name. Lastly, a matrix containing positive and negative topics is being constructed.

Teardown analysis is then carried out to determine the critical topics as described in Algorithm 1. For each product, a sentiment-topics matrix is constructed containing topic names, topics keywords, and topics sentiment polarity, that is, values 0 for negative and 1 for positive topics. For simplicity, the comparison process here involves the topics’ names and sentiment polarities. If a topic is identified as a critical topic, then the topic name, topic keywords, and topic sentiment polarity are stored in a new matrix. Assuming that we have two or more matrices identified as TP for target product and CPs, that is, $C P_{1}, C P_{2}, C P_{3}, \dots, C P_{n}$ for competitors’ products. Topic names are stored in the first column of every matrix, that is, TP[1] and CPs[1] while each topic sentiment polarity is stored in the second column. In the first stage of teardown analysis, the polarities stored in TP[2] are checked, if a negative value (0) is detected, then compare the topic name stored in the correspondence cell with the topic names stored in CPs[1], if a match is found, check the correspondence sentiment polarity stored in CPs[2], if it is found to have a positive value (1), then store the matched topic name, topic keywords, and topic sentiment polarity in a new matrix named High priority topics (HPT). Otherwise, if the value of CP[2] is found to be negative, then store the matched topic data in a new matrix named Medium Priority Topics (MPT). If any topic is found to carry positive polarity in TP and CPs then the topic data is exported to a new matrix named Low Priority Topics (LPT). Repeat the process until all the topic names in TP[1] are checked. At the output of this stage, three lists of topics and their associated keywords, namely, HPT, MPT, and LPT are constructed. List HPM contains all the negative topics of TP that are positive in product CPs, list MPT contains all the common negative topics in products TP and CPs, while list LPT contains all the common positive topics in products TP and CPs.

Case study

Data collection

For this study, we analyze the strengths and weaknesses of popular products in two different domains. By going through the most famous real-time mobile phones’ comparison websites (e.g., GadgetsNow.com and 91mobiles.com), we found out that customers on more than three websites have often compared Samsung Galaxy M30s and Xiaomi Redmi Note 8 Pro. The other popular competitive products were chosen by going through the best products sales on Amazon. We found that among the best sales of surveillance cameras department are Wyze Cam Outdoor, Blink Outdoor, and Ring Stick Up Cam. We chose to examine those particular products for having a high volume of consumers’ reviews and for being produced by different manufacturers unlike other popular products produced by the same manufacturer. Conducting the study on competitive products from different domains chosen by various approaches assists in validating the proposed model functionality. The datasets used in this study were collected from the famous online reviews websites Amazon and Google shopping. Such websites not only play an essential role in consumer purchase behavior (Mo et al., 2015) but also assist firms to improve their products (Liu et al., 2013). The total number of the collected reviews is 10,870 for the chosen smartphones and 15,997 for the chosen surveillance cameras as shown in Table 1.

Table 1.

The collected reviews.

Smartphones	Reviews	Surveillance cameras	Reviews
Samsung Galaxy M30s	5545	Wyze Cam Outdoor	5435
Xiaomi Redmi Note 8 Pro	5325	Blink Outdoor	5579
Total	10,870	Ring Stick Up Cam	4983
Total	10,870	Total	15,997

Data preprocessing

Since our study is targeting English reviews, we filtered out all the non-English documents. As the unstructured nature of consumers’ reviews, and in order to make our data suitable for further analysis, text-preprocessing techniques are employed to clean and organize the data. To do so, several NLP techniques are applied to the dataset. Numbers, stop words, and punctuation marks have also been filtered out from the dataset. Part-of-speech (POS) tagging and lemmatization were also applied to the data. All the words containing capital letters are then converted into small letters to avoid duplication of keywords. Last but not least, unnecessary words with the highest occurrence number like the names of products and brands are removed to have quality data.

Teardown JST

In this study, JST is employed to obtain topics and sentiment simultaneously from a mixture of documents. The topic extracting process is done using LDA topic modeling which has been used efficiently to identify patterns in textual data that imply emerging topics. Another essential factor of JST is determining the number of topics. We determine the optimal number of topics based on the perplexity value generated using our textual data. We set iterations to 1000, alpha to 0.1 and beta to 0.001 as those values carry the best classification accuracy (Subeno et al., 2018). We then determine the perplexity values on a range of topics from 1 to 20 on datasets as shown in Figure 3. Lower perplexity score indicates better performance of a model. On the other hand, the higher number of topics results in an over-clustering of data into many similar topics. Thus, taking into consideration the elbow method, we set the number of topics as 10 and 9 topics for smartphones and surveillance cameras datasets, respectively, and extract top five keywords of each topic. The extracted topics are classified based on the sentiments labels given by JST sentiment layer and each topic is named as its highest weighted keyword.

The teardown method layer receives the topics and their sentiments of one product from sentiment-topics matrix produced by JST model and it compares them with the topics stored in the sentiments-topic matrix of the other products.

Assuming that we have two products to compare, the target product CP which has to outperform the competitor’s product CP. The process of teardown scheme in this study operates as follows:

(1) If a negative topic of TP exists in positive topics of CP then the topic is stored in HPT matrix.

(2) If a negative topic of TP exists in negative topics of CP then the topic is stored in MPT matrix.

(3) If a positive topic of TP exists in positive topics of CP then the topic is stored in LPT matrix.

(4) If a positive topic of TP exists in negative topics of CP then the topic is ignored.

Likewise, the mechanism process has the ability to adapt multiple products comparison. Concurrent comparisons of multiple products can be performed by comparing the topics of the target products with the topics of the other products at once, for example, extracting the negative topics of the target product which are positive in at least one of the other products. In our case study, we tested our model using two competitive products of smartphones as well as multiple competitive products of surveillance cameras.

Results and discussions

The result shown in Table 2 demonstrates the critical topics along with their sentiments and keywords. Each critical topic has been assigned a priority level implying the importance degree of the topics. The high priority is given to those topics that are negative in the target product while are positive in the competitive products. Improving such critical topics is assumed essential since those topics are considered the weaknesses that need urgent improvements to be on the same status as the competitive product. Medium priority critical topics are those with negative sentiment in both products. In such a scenario, dealing with weaknesses is given priority over improving strengths. Low priority critical topics are those topics with positive sentiments in both products. Despite the satisfaction of consumer on both topics for both products, it is still important to find solutions to be a step ahead of competitors.

Table 2.

Essential topics with keywords.

	Priority	Topics	Products
—	—	—	Samsung Galaxy M30s		Xiaomi Redmi Note 8 Pro
Smartphones	High	Camera	Neg (camera, picture, sensor, zoom, video)		Pos (camera, wide, quality, pixel, awesome)
	High	NFC	Neg (nfc, usb, call, feature, connect)		Pos (nfc, support, connect, wifi, pro)
	Medium	Amazon	Neg (amazon, replace, product, custom, return)		Neg (amazon, call, issue, warranty, product)
	Medium	Ram	Neg (ram, slow, processor, game, heat)		Neg (ram, memory, sensor, store, app)
	Low	Price	Pos (price, worth, amazon, cost, seller)		Pos (price, cheap, order, value, amount)
—	—	—	Wyze Cam Outdoor	Blink Outdoor		Ring Stick Up Cam
Surveillance cameras	High	Battery	Neg (battery, life, built, month, short)		Pos (battery, long, life, impressive, charge)	—
	High	APP	Neg (app, stream, connect, slow, wifi)		—	Pos (app, live, audio, quick, mobile)
	Low	Resolution	Pos (resolution, quality, motion, video, camera)		—	Pos (resolution, night, camera, clear, view)
	Low	Storage	Pos (storage, base, cloud, microsd, unlimited)		Pos (storage, module, sync, cloud, hub)	—
	Low	Weather	Pos (weather, climates, resistance, design, heat)		Pos (weather, dust, resistance, device, rain)	—
	Low	Alexa	Pos (alexa, amazon, google, support, smart)		Pos (alexa, connect, device, amazon, voice)	—

By observing the obtained result, we find that the topics Camera, NFC, Amazon, RAM, and Price are critical topics having common keywords expressed by smartphones consumers. Analyzing keywords provides more in-depth understanding of the topic nature. Both Camera and NFC are high priority topics that need to be considered for improvement first. The Camera topic shows how the consumers receive the camera feature. Comparing the camera specs of both devices, we found that although the Samsung device front camera resolution is higher than that of Xiaomi device, it is likely that the consumer is more concerned about the main rear camera. The keywords “quality, pixel” of Redmi note 8 pro implying the consumer satisfaction about the camera resolutions. This is notable when comparing the rear camera resolutions of both products; we find that Xiaomi offers a higher camera resolution. Another interesting finding is that despite the fact that NFC connectivity feature is not included in Samsung device, the result shows that consumers with negative sentiment have discussed NFC connectivity often. It clarifies why the consumer is expressing disappointment for missing the feature in their devices while Xiaomi users showed appreciations for acquiring it.

With regard to medium priority topics, the keywords of the topic Amazon show some dissatisfaction by consumers about “warranty” issues and “replacement” procedures by Amazon. Since negative sentiment is found in both products datasets, it is wise to develop solutions with Amazon or find more suitable channel to reach to the customer. Users of both products were comparatively displeased by the provided RAM as well and often complained about how “slow” the device is when playing “games” as observed in the keywords.

While the product price is one of the most important issues of every competitive analysis, in this comparison, however, TDJST model has identified the Price topic as a low priority topic. The reason is that most consumers have expressed positive sentiment toward the price of both products. We suspect that the consumer’s price judgment is based on two factors: The satisfaction of what they have paid in return to what they have received (as the keywords “worth” and “value” imply), and the offered competitive price in comparison to other products with similar features in the market. In spite of that, it is still essential to design price reduction strategies to strengthen a company’s long-run competitive position.

Surveillance cameras on the other hand, have more low priority topics than those of smartphones. This indicates that products have a lot in common when it comes to positive topics. The result shows Battery and APP are critical topics with high priority. Consumers have expressed dissatisfaction about the in-built battery life of Wyze Cam that can last for a few months as keywords suggest. On the contrary, consumers showed positive sentiments towards Blink outdoor long battery life. In such a scenario where battery life is an influential aspect, improving the battery quality is exceedingly needed. Wyze APP is another high priority topic which consumers complained of being “slow” and has “connectivity” issues unlike Ring cam consumers who were satisfied with the provided APP and the associated services. From the keywords, we can assume that resolving the slow app performance while adding more controls would help to gain the consumer trust.

With respect to low priority topics, the Resolution topic is received well by consumers of Wyze Cam and Ring Cam. The keywords demonstrate how the consumers praised the quality of night vision and motion detection feature. Both cameras have similar field of view but the offered field of view by Ring cam captured the consumers’ attention as the keyword “view” justifies. In such instance, increasing the field of view would be an optimal solution. The Storage topic carries positive sentiment in Wyze Cam and Blink Outdoor; this is due to the acceptable storage provided by both of the firms. The keywords reveal that offering better “cloud” storage options in the current product or increasing local storage capacity in the next new product would contribute to creating a sustainable competitive advantage. The Weather topic indicates the product quality in term of holding up against weather conditions. Consumers praised the high degree of weather resistance of Wyze Cam and Blink Outdoor, yet Wyze Cam holds an advantage of having lower cost at the similar product quality. Another low priority topic discussed by consumer is Alexa. Alexa is a voice interaction device in which the consumer can control several smart home devices including surveillance cameras. The keywords show that both products’ consumers are satisfied with the offered compatibility of Amazon Alexa. Wyze cam, however, is one-step ahead for offering the supportiveness of Google voice assistant.

Additionally, we can also observe that, unlike the smartphone industry comparison, the “Price” topic was not recognized by TDJST in this comparison. It is most likely that the consumers of the target product (i.e., Wyze Cam) were convinced about the product price, therefore the model has assigned positive sentiment to the topic, whereas in the competitor’s products (i.e., Blink outdoor and Ring Cam) consumers were not pleased with the price which is either comparatively high or very high compared with most products in the market in term of unit price or cloud storage subscription cost.

The findings show that the proposed model effectively analyzed the WOM to determine the critical topics. Compared with other Teardown competitive analysis studies (Farel and Silverman, 2008; John et al., 2014), instead of manually setting the product factors that need to be analyzed, this study has developed a teardown method that can systematically identify the critical factors and assign priority levels taking into account the expressed feelings of consumer about the product factors. This is important since analyzing wide range of consumers’ opinions contributes to more fulfilling of consumer needs.

Moreover, our approach has the advantages over other competitive analysis studies that use eWOM to achieve competitive advantage (Chen et al., 2015; Ko et al., 2017; Wang et al., 2018), rather than analyzing only aspects or topics, this study has developed a framework that contain a teardown layer incorporating with joint sentiment-topic model to group aspects and determine their sentiment polarity simultaneously.

Conclusions

The strategies for establishing a competitive advantage vary depending on the nature of the market and the ability to gain the consumer satisfaction based on consumer feedback. In this study, our objective is to build a reliable method to analyze one product against its competitors to achieve the competitive advantage. Based on the JST model, this study proposes a novel approach to conduct in-depth competitive analysis by constructing a teardown method layer incorporating with JST model to perform comparative analysis. Compared with other studies, instead of analyzing only product aspects, this study goes beyond inspecting online reviews in sentiment-topics form to proposing a method to find essential topics that are needed to be evaluated in order to outperform the competitor products.

We collected online reviews of two competitive products of smartphones and three competitive products of surveillance cameras. We employed LDA topic modeling to obtain topics and had the topics classified based on its sentiment. A designed Teardown method is implemented to identify the essentials topics. Some of the findings are justified in accordance with product specifications.

Overall, this work has few limitations. First, although the proposed model contributes to decision making in providing insights to improve the current product and provide product opportunities for new product development, it is still based on already launched products in the market for a period of time as this approach requires a high volume of user-generated content. Second, the proposed JST teardown analysis model is only applicable to products with many aspects. Further work might involve comparing products that have few aspects. Third, the reviews used in this study are extracted from the e-commerce websites Amazon and Google shopping. The fact that many consumers expressed sentiment about online retailers like Amazon services instead of the product itself might lead to losing other important topics. In future, other data sources like consumer reviews on social media could be used. Fourth, considering that gaining a competitive advantage is achieved by collaboration of strategies (Zeqiri and Nedelea, 2011), this study deals only with the information technology side of the problem and it is expected to achieve a better result if other strategies are also incorporated.

Algorithm 1

Teardown model algorithm.

Set n= number of topics, m= number of entries;

For i = 1 to n do

Set y = 0;

For k = 2 to m do \\ CP count

If TP[i][2] = 0 then

For j = 1 to n do

If CP[j][1] = TP[i][1] && CP[j][2] = 1 then

HPT [i] [y++] = TP [i][1];

HPT [i] [y++] = k;

Else If CP[j][1] = TP[i][1] && CP[j][2] = 0 then

MPT [i] [y++] = TP[i][1];

MPT [i] [y++] = k;

End if

End for

Else If TP[i][2] = 1 then

For j = 1 to n do

If CP[j][1] = TP[i][1] && CP[j][2] = 1 then

LPT [i] [y++] = TP[i][1];

LPT [i] [y++] = k;

End if

End for

End if

End for

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Abdullah Alqadhi

Mahmood Almesbahi

References

Blei

Jordan

(2003) Latent Dirichlet allocation. Journal of Machine Learning Research 3(4–5): 993–1022. DOI: 10.1016/b978-0-12-411519-4.00006-9.

Chen

Kou

Shang

, et al. (2015) Visualizing market structure through online product reviews: Integrate topic modeling, TOPSIS, and multi-dimensional scaling approaches. Electronic Commerce Research and Applications 14(1): 58–74. DOI: 10.1016/j.elerap.2014.11.004.

Dermouche

Velcin

Khouas

, et al. (2014) A joint model for topic-sentiment evolution over time. In: Proceedings - IEEE international conference on data mining, ICDM 2015-Janua, 773–778. IEEE. DOI: 10.1109/ICDM.2014.82.

Dong

S-j

Zhang

C-j

, et al. (2018) An unsupervised topic-sentiment joint probabilistic model for detecting deceptive reviews. Expert Systems with Applications 114: 210–223. DOI: 10.1016/j.eswa.2018.07.005.

Farel

Silverman

(2008) Using competitive analysis to get the competitive advantage. In: Proceedings - annual reliability and maintainability symposium, pp. 1–4. DOI: 10.1109/RAMS.2008.4925790.

Heng

Gao

Jiang

, et al. (2018) Exploring hidden factors behind online food shopping from Amazon reviews: a topic mining approach. Journal of Retailing and Consumer Services 42: 161–168. DOI: 10.1016/j.jretconser.2018.02.006.

Koh

Reddy

(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(1): 42–53. DOI: 10.1016/j.dss.2013.07.009.

Ireland

Liu

(2018) Application of data analytics for product design: sentiment analysis of online product reviews. CIRP Journal of Manufacturing Science and Technology 23: 128–144. DOI: 10.1016/j.cirpj.2018.06.003.

John

State

Jayeola

(2014) Tear Down Analysis Technique in Small and Medium Scale Enterprises in Lagos Industrial Metropolis. (January).

10.

Kessler

Kuhn

(2013) Detection of product comparisons - How far does an out-of-the-box semantic role labeling system take you? In: EMNLP 2013 - 2013 conference on empirical methods in natural language processing, proceedings of the conference, October 1892–1897.

11.

Jeong

Choi

, et al. (2017) Identifying product opportunities using social media mining: application of topic modeling and chance discovery theory. IEEE Access 6: 1680–1693. DOI: 10.1109/ACCESS.2017.2780046.

12.

Korfiatis

Stamolampros

Kourouthanassis

, et al. (2019) Measuring service quality from unstructured data: a topic modeling application on airline passengers’ online reviews. Expert Systems with Applications 116: 472–486. DOI: 10.1016/j.eswa.2018.09.037.

13.

Kuncoro

Suriani

(2018) Achieving sustainable competitive advantage through product innovation and market driving. Asia Pacific Management Review 23(3): 186–192. DOI: 10.1016/j.apmrv.2017.07.006.

14.

Lee

Foo

ATL

Leong

, et al. (2016) Can competitive advantage be achieved through knowledge management? A case study on SMEs. Expert Systems with Applications 65: 136–151. DOI: 10.1016/j.eswa.2016.08.042.

15.

Mai

(2019) The effect of online reviews on product sales: a joint sentiment-topic analysis. Information and Management 56(2): 172–184. DOI: 10.1016/j.im.2018.04.007.

16.

Lin

(2009) Joint sentiment/topic model for sentiment analysis. In: Cikm, pp. 375–384.

17.

Lin

Wang

Chen

(2006) A strategy for managing customer-oriented product design. Concurrent Engineering Research and Applications 14(3): 231–244. DOI: 10.1177/1063293X06068390.

18.

Liu

Jin

, et al. (2013) Identifying helpful online reviews: a product designer’s perspective. CAD Computer Aided Design 45(2): 180–194. DOI: 10.1016/j.cad.2012.07.008.

19.

Itoh

(2004) Product competitive advantage and product architecture: 263–268.

20.

Magdum

Department

JVM

(2017) Mining online reviews and tweets for predicting sales performance and success of movies. In: International conference on intelligent computing and control systems ICICCS 2017, pp. 334–339.

21.

Martin-Domingo

Martín

Mandsberg

(2019) Social media as a resource for sentiment analysis of Airport Service Quality (ASQ). Journal of Air Transport Management 78: 106–115. DOI: 10.1016/j.jairtraman.2019.01.004.

22.

Maslowska

Malthouse

Viswanathan

(2017) Do customer reviews drive purchase decisions? The moderating roles of review exposure and price. Decision Support Systems 98: 1–9. DOI: 10.1016/j.dss.2017.03.010.

23.

Y-F

Fan

(2015) Effect of online reviews on consumer purchase behavior. Journal of Service Science and Management 8(3): 419–424. DOI: 10.4236/jssm.2015.83043.

24.

Porter

(1985) Competitive Advantage: Creating and Sustaining Superior Performance: With a New Introduction. The Free Press.

25.

Prasad

(1998) Synthesis of market research data through a combined effort of QFD, value engineering, and value graph techniques. Qualitative Market Research: An International Journal 1(3): 156–172. DOI: 10.1108/13522759810235250.

26.

Prasad

(2001) Towards balancing multiple competitiveness measures for improving business performance in manufacturing. International Journal of Manufacturing Technology and Management 3(6): 550–569. DOI: 10.1504/IJMTM.2001.001429.

27.

Prasad

(2016) On mapping tasks during product development. Concurrent Engineering Research and Applications 24(2): 105–112. DOI: 10.1177/1063293X15625098.

28.

Rathor

Agarwal

Dimri

(2018) Comparative study of machine learning approaches for amazon reviews. Procedia Computer Science 132: 1552–1561. DOI: 10.1016/j.procs.2018.05.119.

29.

Rihar

Žužek

Kušar

(2020) How to successfully introduce concurrent engineering into new product development? Concurrent Engineering Research and Applications. DOI: 10.1177/1063293X20967929.

30.

Samiee

Chirapanda

(2019) International marketing strategy in emerging-market exporting firms. Journal of International Marketing 27(1): 20–37. DOI: 10.1177/1069031X18812731.

31.

Sowmiya

Chandrakala

(2015) Joint sentiment/topic extraction from text. Proceedings of 2014 IEEE International Conference on Advanced Communication, Control and Computing Technologies, ICACCCT 2014 (978). IEEE: 611–615. DOI: 10.1109/ICACCCT.2014.7019160.

32.

Subeno

Kusumaningrum

Farikhin (2018) Optimisation towards Latent Dirichlet Allocation: Its Topic Number and Collapsed Gibbs Sampling Inference Process. International Journal of Electrical and Computer Engineering 8(5): 3204–3213. DOI: 10.11591/IJECE.V8I5.PP3204-3213.

33.

Tiago

MTPMB

Veríssimo

JMC

(2014) Digital marketing and social media: why bother? Business Horizons 57(6): 703–708. DOI: 10.1016/j.bushor.2014.07.002.

34.

Tkachenko

Lauw

(2017) Comparative relation generative model. IEEE Transactions on Knowledge and Data Engineering 29(4): 771–783. DOI: 10.1109/TKDE.2016.2640281.

35.

Wang

Zhao

Xin

, et al. (2015) Exploiting machine learning for comparative sentences extraction. International Journal of Hybrid Information Technology 8(3): 347–354. DOI: 10.14257/ijhit.2015.8.3.31.

36.

Wang

Feng

Dai

(2018) Topic analysis of online reviews for two competitive products using latent Dirichlet allocation. Electronic Commerce Research and Applications 29: 142–156. DOI: 10.1016/j.elerap.2018.04.003.

37.

Tseng

Chiu

ASF

, et al. (2017) Achieving competitive advantage through supply chain agility under uncertainty: a novel multi-criteria decision-making structure. International Journal of Production Economics 190: 96–107. DOI: 10.1016/j.ijpe.2016.08.027.

38.

Liao

, et al. (2011) Mining comparative opinions from customer reviews for competitive intelligence. Decision Support Systems 50(4): 743–754. DOI: 10.1016/j.dss.2010.08.021.

39.

Yao

Mimno

McCallum

(2009) Efficient methods for topic model inference on streaming document collections. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining: 937–945. DOI: 10.1145/1557019.1557121.

40.

Zeqiri

Nedelea

(2011) Strategies for Achieving Competitive Advantage. The Annals of The ‘Ştefan cel Mare’ University of Suceava. Fascicle of The Faculty of Economics and Public Administration, 11 (March 2014).

41.

Zha

Tang

, et al. (2014) Product aspect ranking and its applications. IEEE Transactions on Knowledge and Data Engineering 26(5): 1211–1224. DOI: 10.1109/TKDE.2013.136.

42.

Zhang

Chu

Simeone

, et al. (2020) Machine learning-based design features decision support tool via customers purchasing data analysis. Concurrent Engineering Research and Applications. DOI: 10.1177/1063293X20963313.