- Email: [email protected]
Mining customer knowledge for exploring online group buying behavior
Expert Systems with Applications 39 (2012) 3708–3716
Contents lists available at SciVerse ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
Mining customer knowledge for exploring online group buying behavior Shu-hsien Liao a,⇑, Pei-hui Chu b, Yin-ju Chen a, Chia-Chen Chang a a b
Department of Management Sciences, Tamkang University, No. 151, Yingjuan Rd., Danshuei Dist., New Taipei City 251, Taiwan, ROC Department of Information Management, National Taipei College of Business, No. 321, Sec. 1, Jinan Rd., Zhongzheng District, Taipei City 10051, Taiwan, ROC
a r t i c l e
i n f o
Keywords: Data mining Association rules Cluster analysis Online group buying Online group buying behavior
a b s t r a c t Online group buying is an effective marketing method. By using online group buying, customers get unbelievable discounts on premium products and services. This not only meets customer demand, but also helps sellers to find new ways to sell products sales and open up new business models, all parties benefit in these transactions. During these bleak economic times, group buying has become extremely popular. Therefore, this study proposes a data mining approach for exploring online group buying behavior in Taiwan. Thus, this study uses the Apriori algorithm as an association rules approach, and clustering analysis for data mining, which is implemented for mining customer knowledge among online group buying customers in Taiwan. The results of knowledge extraction from data mining are illustrated as knowledge patterns, rules, and knowledge maps in order to propose suggestions and solutions to online group buying firms for future development. Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction In these tough economic times, it is not only important to find new sources of income but also to cut down on expenditures. By using online group buying, it is easy to find more people in a short period of time to share freight costs and to buy in bulk so as to lower prices. It is also easier to get bigger discounts when more people take part in a group purchase. On the other hand, online group buying is a model in which multiple buyers cooperate and buy the same good/ service in order to bargain with the proprietor, if there are enough buyers, they may aggregate buyer power to get volume discounts. All parties benefit in these transactions. In the past, group buyers were limited to members of companies, friends, families or communities. They filled out the types of goods and quantity of items to be purchased on the flyers to order goods. As the internet develops, it is becoming an increasingly prosperous network for many types of commerce. In addition, the large shopping sites are familiar with online shopping markets and the follow-up group buying platforms are learning this market. They have created group buying discount zones and provided some incentives and discounts to attract more users to visit and buy. The more visitors that come, the more goods are exhibited and sold. Thus, the consumption patterns of group buying have become more active in online shopping markets. According to pollster online survey of group buying, the most preferred goods are foodstuffs, making up to 33%. The group buying shares of other items are: clothing and accessories 11%, cosmetics 8%, articles of daily use 5%, others 3% and home appliances 2%. How⇑ Corresponding author. E-mail address: [email protected] (S.-h. Liao). 0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.09.066
ever, 36% of people in Taiwan have yet to experience group buying (Market Intelligence & Consulting Institute in Taiwan, 2010). As online group buying (OGB) market increases and a great variety commodity stats become available, consumer demand is changing fast, implicitly shortening the product life cycle. Clear assessment of the overall sales strategies of internet group buying has a positive effect. However over-reliance can result in the cart coming before the horse. In 2009, as many as 50.3% of online stores did not break even, while 5.3% increased sales compared to last year in Taiwan. The reason is that business operation was impacted by the overall economy. More and more intense price wars led to descending overall profits (Market Intelligence & Consulting Institute in Taiwan, 2010). In addition, in electronic marketplaces, group buying is seen as an effective form of electronic commerce and a promising field. For example, Tokuro and Takayuki (2004) proposed using decision support systems for buyers in group buying. Their system supports buyers’ decision making by using the Analytic Hierarchy Process with three methods for group integration. First, buyers trade in simple group buying. Second, all buyers are integrated. Third, some buyers are integrated. Thus, buyers’ multi-attribute utilities are effectively expressed in group integration and buyers can purchase goods at a lower price. Buyers’ payments are decided based on their degree of compromise. Software agents can be useful in forming buyers’ groups since humans have considerable difficulties in finding Pareto-optimal deals (no buyer can be better without another being worse) in negotiation situations (Frederick & Brahim, 2006). The above study developed a negotiation protocol for software agents, which evaluated whether or not the problem is difficult on average and why. This protocol is probably able to find
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
a Pareto-optimal solution and, furthermore, minimize the worst distance to the ideal among all software agents given strict preference ordering. In addition, Miguel and María (2009) explored the circumstances under which the retailers’ use of the buying group’s brand name may benefit them. Their research findings show that the retailer’s use of the buying group’s brand name is more capable of improving the retailer’s economic satisfaction with the buying group when differentiation is perceived to be a source of competitive advantage, when the environment is perceived as more dynamic and when the retailer is strategically integrated in the relationship with the buying group. However, only a few studies have explored online group buying behavior patterns and segments from customers. On the other hand, customers play an important role as business assets. Most of the parties involved in sales, such as the commercial web sites, retailers and channels, are aware of the need for businesses to acquire better customer knowledge. However, this is easier said than done since customers’ knowledge is concealed within the customers. It is available but not accessible, and there is little possibility of exploring the full volume of data that should be collected for its potential value. Inefficient utilization renders the data collected useless, causing databases to become ‘data dumps’ (Keim, Pansea, Sipsa, & Northb, 2004). Thus, finding ways to effectively process and use data is an artificial issue that calls for new techniques to help analyze, understand or even visualize the huge amounts of stored data gathered from business and scientific applications (Liao & Chen, 2004). Among the new techniques developed, data mining is a process of discovering significant knowledge, such as patterns, associations, changes, anomalies and significant structures from large amounts of data stored in databases, data warehouses, or other information repositories (Keim et al., 2004). Customer knowledge extracted through data mining can be integrated with products and marketing knowledge from research and can be provided to up stream suppliers as well as downstream retailers. Thus, it can serve as a reference for product development, product promotion and customer relationship management. When effectively utilized, such knowledge extraction can enable enterprises to gain a competitive edge by producing customer-oriented goods that increase consumer satisfaction (Arie & Sterling, 2006; Liao, Chen, Chieh, & Hsiao, 2009; Liao, Chen, & Dang, 2010; Liao, Chen, & Hsu, 2009; Liao, Chen, & Tseng, 2009; Liao, Ho, & Yang, 2010; Liao, Hsieh, & Huang, 2008). Accordingly, this study investigates online group buying behavior, and implements data mining approach to analyze Taiwanese customers. There are two data mining stages implemented in this study. First, this study employs the k-means algorithm to cluster the customers into potential customers and target customers, and uses the Apriori algorithm to generate association rules for each cluster. The rules are proposed to the group buying firms to help them attain possible new customers, services and sales. The rest of this paper is organized as follows. Section 2 introduces the proposed data mining system, which includes the system framework, system design, and physical database design. Section 3 introduces the data mining approach, including the association rules and cluster analysis. Section 4 presents the data mining process and the analyzed results. Section 5 describes research findings, managerial implications. Finally, a brief conclusion is presented in Section 6. 2. Data mining systems 2.1. Research framework Because online group buying customer information and product sales information are difficult to obtain, this study used a designed
3709
questionnaire to collect research data. This study uses k-means to sort customers into clusters to generate association rules for each cluster, and then proposes suggestions and solutions for online group buying to open up possible new services and sales. 2.2. System framework This study collected the past data of customers to establish a database system, We then analyze the entire database system by data mining to find the association between group buying customer behavior and commodities buying patterns, including consumer shopping preferences and demand consideration, etc. The aim of this approach is to enable businesses to further understand group buying experience/patterns, rather than the psychology and orientation of experienced online group buying customers. The propose is to make suitable marketing suggestions, so that they can really provide customers with preferred products and services, while reducing marketing costs and increasing business profits. The system framework is shown in Fig. 1. Accordingly, the system design diagram is shown in Fig. 2. It shows that this study incorporates data related to customers, customer behavior and products into the database, and analyzes the entire database by data mining for customer patterns and market segmentations to find the different types of target customers. Online firms can evaluate customer knowledge management for marketing and service, and then determine efficient means to achieve the goals of exploring online group buying behavior. In this study, the design and operation of a physical database is used to construct a relational database enter data in the table through Microsoft Access 2003. Although general database software cannot accommodate too many people online simultaneously, Microsoft SQL 2005 can satisfy this need, because the general database systems use standard structured query language (SQL). Because each type of data storage and processing is different, in order to give a programming language access to the information database system, manufacturers can design a driver for all types of language using standard SQL, and then access their database through a regional network. Microsoft’s console provides an open database link (open database connectivity; ODBC). Thus, administrators can manage a variety of ODBC drivers (Fig. 3). 2.3. Questionnaire design and data collection This study uses the questionnaire approach to collect data from general customers who have group buying or online group buying experience, and establishes the database system using collected data. The main purpose of the questionnaire design is to understand the customer motivations of the entities involved in group buying, the degree of involvement of internet group buying, and explore the factors which affect the correlation between internet group buying behavior and customer psychology. The questionnaire is divided into five parts. Part 1 focuses on customer personal information (9 items), part 2 discusses the customer psychology of individual customers engaged group buying (9 items), part 3 looks at the customer tendencies of individual online group buying (9 items), while part 4 focuses on the shopping behavior of individual online group buying (4 items) and the service mechanism of online group buying (10 items). The quantity of preliminary draft questionnaire sent and responded to, the pre-test questionnaire and formal questionnaire are shown in Table 1. Thus, the questionnaires were collected from November 1st to November 30th of 2009. Altogether, we sent 720 questionnaires and 621 were collected. Excluding omissions and incomplete answers, there
3710
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
Fig. 1. System framework.
were 550 valid responses, with an effective collection ratio of 88.6%. 3. Data mining 3.1. Association rules Discovering association rules is an important data mining problem (Agrawal, Imilienski, & Swami, 1993), and there has been considerable research on using association rules for data mining problems. The association rules algorithm is used mainly to determine the relationships between items or features that occur synchronously in databases. For instance, during a trip to the shopping center, if the people who buy item X also buy item Y as well, there exists a relationship between item X and item Y. Such informa-
tion is useful for decision makers. Therefore, the main purpose of implementing the association rules algorithm is to find synchronous relationships by analyzing random data and to use these relationships as a reference for decision-making. The association rules are defined as follows (Wang, Chuang, Hsu, & Keh, 2004): Make I = {i1, i2, . . . , im} the item set, in which each item represents a specific literal. D stands for a set of transactions in a database in which each transaction T represents an item set such that T # I. That is, each item set T is a non-empty sub-item set of I. The association rules are an implication of the form X ? Y, where T X I, Y I and X Y = U. The rule X ? Y holds in the transaction set D according to two measurement standards – support and confidence. Support (denoted as_ Sup(X, D)) represents the rate of transactions in D containing the item set X. Support is used to evaluate the statistical importance of D, and the higher its value, the
3711
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
standard called confidence (denoted as Conf(X ? Y)), representing the rate of transactions in D that contain both X and Y. That is, T Conf(X ? Y) = Sup(X Y)/Sup(X, D). In this case, Conf(X ? Y) denotes that if a transaction includes X, the chance that this transaction also contains Y is relatively high. The measure of confidence is then used to evaluate the level of confidence about the association rules X ? Y. Given a set of transactions D, the problem of mining association rules is used to generate all transaction rules that have certain levels of user-specified minimum support (called Min sup) and confidence (called Minconf) (Kouris, Makris, & Tsakalidis, 2005). According to Agrawal and Shafer (1996), the problem of mining association rules can be broken down into two steps. The first step is to detect a large item set whose support is greater than Min sup, and the second step is to generate association rules, using the large item set. Such rules must satisfy the following two conditions: 1. SupðX [ Y; DÞ P Min sup. 2. Conf ðX ! YÞ P Minconf To explore association rules, many researchers use the Apriori algorithm (Agrawal et al., 1993). In order to reduce the possible biases incurred when using these measurement standards, the simplest way to judge the standard is to use the lift judgment. Lift is defined as: Lift = Confidence(X ? Y)/Sup(Y) (Wang et al., 2004). 3.2. Cluster analysis and k-means algorithm
Fig. 2. System design.
more important the transaction set D is. Therefore, the rule X ? Y S which has support Sup(X Y, D) represents the rate of transactions S in D containing X Y. Each rule X ? Y also has another measuring
The process of partitioning a large set of patterns into disjoint and homogeneous clusters is fundamental in knowledge acquisition. It is called Clustering in most studies and it has been applied in various fields, including data mining, statistical data analysis, compression and vector quantization. The k-means is a very popular algorithm and is one of the best for implementing the clustering process. k-Means clustering proceeds in the following order.
Fig. 3. Physical database designs.
Table 1 Send and receive questionnaire quantity. Questionnaire
During
Send
Receive
Effective questionnaire
Returns-ratio
Draft Pre-test Formal
2009/9/26–2009/10/15 2009/10/21–2009/10/25 2009/11/1–2009/11/30
– 25 720
– 22 621
– 20 550
– 90.9% 88.6%
3712
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
Firstly, the k numbers of observations are randomly selected from all N number of observations according to the number of clusters, and these become centers of the initial clusters. Secondly, for each of the remaining N–k observations, the nearest cluster is found in terms of the Euclidean distance with respect to xik-means algorithm for cluster analysis as a data mining approach (Ture, Kurt, Turhan, & Ozdamar, 2005). 4. Data mining and analysis 4.1. k-Mean analysis This study employs the k-means algorithm to cluster the questionnaire survey data, and categorize data into group of potential customers (Cluster-1) and target customers (Cluster-2). 4.1.1. Customer adumbration Common features of both cluster include: most of them are 31– 40-year-old unmarried females, they go online several times a day, the majority spend between 3 h to 5 h a day online time, the majority of group buying goods are foodstuffs, most people group buy goods with colleagues because of inexpensive prices or simple necessity, most people trust the quality and safety of the products, most companies do not prohibit the staff from carrying out group buying during working hours and they even cross-departmental colleagues to join in group buying. In addition, the differing characteristics between each cluster are shown in Table 2. (1) Potential customers (Cluster-1): These customers have good economic capacity, with monthly disposable income amounting to more than 40,001 New Taiwan dollars (NTD), their online time does not exceed 3 h, they have experience with entity group buying, but no experience of online group buying and they will consider the online group buying in the future. The reasons that members of this group do not have experience with internet group buying are that they are unable to confirm product quality, the security of network transactions and individual data protection. (2) Target customers (Cluster-2): Their monthly disposable amount income is around 10,001–20,000 NTD, the reason they purchase goods by group buying is not only due to the inexpensive price, but also but also because of good word of mouth and convenience factors; they complete group buying transactions in 15 min. This group of customer has experience of online group buying from both large and small firms in the past three months the frequency of online
group buying is approximately 1–3 times; most of the purchased through group buying are no more than 1000 NTD for foodstuffs. The people they invite to join them in group buying are classmates, friends and colleagues. They know about group buying sites and group buying news. 4.1.2. Customer behavior The customer behavior analysis of the two clusters shows that there are some common features as below: (1) They find information about group buying products through introductions from relatives/ friends and newspapers/magazines reports. (2) The main reasons why they purchase goods from the same group buying websites are good product quality and low price. (3) All of them have an optimistic view of the future development and future prospects of online group buying. (4) The longest time they are willing to wait for popular commodities is 15 days. (5) The preferred payment method is cash on delivery. (6) Upon obtaining unsatisfactory products, they may return the products after home delivery service. (7) They think group buying sites should provided special lines for customer service and msn online service. (8) Expected service contents should ensure product quality and after service to make the procedures and processes of group buying clear and easy to understand. (9) Expected services functions should compare the function of price and product and strengthen after service consultation. The behaviors of each cluster are shown as below. The reasons potential customers do not shop on OGB are uncertainty about commercial quality, personal data security and transaction security considerations. For the target customers, the majority have consumed products from large group buying sites, and many purchased goods from small dealers; over the past three months, the majority of customers have engaged in online group buying 1–3 times; the majority buy foodstuff goods under 500 NTD, followed by purchases of foodstuff goods between 501 and 1000 NTD. The online group buys goods due to demand, because prices are below the market price and for the sake of convenience. Most people in this group receive information resources about group buying sets and group buying goods from schoolmates or colleagues, relatives or friends. Recommendations from friends to carry out online group buying are related to low price and good commodity quality. The main products purchased through online group buying differ between the two clusters. Potential customers mainly buy: Leisure commodities, foodstuffs, cosmetology maintenance commodities, clothing/popular commodities and 3C commodities. Target customers mainly buy: foodstuffs, cosmetology maintenance commodities, home appliances, clothing/popular commodities and toy/game commodities. The customer behavior of each cluster is shown in Table 3.
Table 2 k-Means clustering results (customer profile). Cluster
Cluster-1
Cluster-2
Naming Feature
Potential customer Mostly unmarried females, with good economic capacity for their age level. They only participate in group buying with company members 232 Female (55.36%) Male (44.64%) Mostly above 40,001 (NTD)
Target customer Mostly unmarried females with monthly disposable income. In addition to being involved in group buying with company colleagues, they also have experience with online group buying 318 Female (62.15%) Male (37.85%) Mostly during 10,001–20,000 (NTD)
Samples Sex Monthly disposable amount Information source of group buying/ collective buying The reason why they join in group (collective) buying Experience of online group buying
1. 2. 3. 1. 2.
Recommendations of colleagues and friends Colleagues invite them to buy Vendors provide samples to eat and to test Cheap price Simple demand
No More than half of them consider being involved in online group buying in the future
1. 2. 3. 1. 2. Yes
Recommendations of colleagues and friends Network introduction Vendors provide samples to eat and to test Cheap price Well-known products
3713
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716 Table 3 k-Means clustering results (customer behavior). Behavior
Cluster-1
Cluster-2
Do not shopping on OGB
–
Has consumed products at these group buying sites
1. Commercial quality uncertainty 2. Personal data security 3. Transaction security considerations –
Over the past three months, the frequency of online group buying Often buy goods online groups
– –
Movement of online group buying
–
Information source of group buying sets and goods
–
The reasons friends recommend online group buying
–
The goods most members to buy in online group and buying sets
1. 2. 3. 4. 5.
Leisure commodities Foodstuffs cosmetology maintenance commodities Clothing/popular commodities 3C commodities
4.1.3. Customer online group buying tendencies and demands This study used k-means to divide customers into potential customers (Cluster-1) and target customers (Cluster-2), and then used the Apriori algorithm to do the association analysis for each cluster. This was followed by a progressive method to identify the association rules. These association rules were used to find out customer tendencies and to explore customer demand in online group buying. These results will provide an effective reference to enhance marketing operations in related industries. The association diagrams of both clusters are shown in Figs. 4 and 5. Further, further investigation explores gender and all age levels of shopping priority, in order to find association rules between each cluster and group buying purchases. (1) Potential customers (Cluster-1): They have no experience of online group buying and there are goods which they urgently want to purchase from online group buying. If there were the opportunity to join online group buying, the first consideration would be to buy frequently purchased goods. Some of the male and 21–30 year old customers have partiality toward buying 3C products. If they were to participate
1. Large dealers 2. Small dealers 1–3 times 1. Foodstuff under 500 NTD 2. Foodstuff between 501–1000 NTD 1. Just demand 2. Below the market price 3. Convenience 1. Schoolmates or colleagues 2. Relatives or friends 1. Low price 2. Good commodity quality 1. Foodstuffs 2. Cosmetology maintenance commodities 3. Home appliances 4. Clothing/popular commodities 5. Toy/game commodities
in online group buying in the future, some of the female customers would have partiality toward buying foodstuffs, as shown in Table 4. (2) Target customers (Cluster-2): Because this population has online group buying experience, they have more ideas about online group buying. For online group buying, the order of preference for most male customers is as follows: foodstuff and 3C products. The order of preference for most female consumers is as follows: foodstuffs, cosmetic products, apparel and popular products. The order of preference for 21–30 year old customers is foodstuffs, cosmetic products, apparel and popular products, leisure goods, 3C products and game/toy products, as shown in Table 5. 5. Research findings and managerial implications Through the implementation of data mining, this study divided the customers into potential customers and target customers. The former group has potential for future online group buying. The findings of this study will be helpful in the development of potential customer markets by manufacturers. The latter group represents
Fig. 4. Association diagram of Cluster-1’s online group buying experience.
3714
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
Fig. 5. Association diagram of Cluster-2’s online group buying experience.
Table 4 The potential customer buying preferences of online group buying (Cluster-1). Rule
Support %
Confidence %
Life
Consequent
Antecedent
R1 R2 R3
11.16 11.16 21.89
88.46 42.31 72.55
1.98 1.47 1.31
Male 21–30 years old Female
Online group buying preference01 = 3C product Online group buying preference01 = 3C product Online group buying preference01 = foodstuff
the existing online group buying target customers. This is the group which sellers wish to understand and grasp. This study explores their purpose of group buying and their service mechanism demands for group buying. Some research findings and managerial implications are discussed as follows. 5.1. Customer spending tendencies in online group buying Online group buying is a highly competitive market. Companies must work diligently of they want to join the online group buying market. The mining analysis describes two clusters (potential customers and target customers) in an online group buying marketing knowledge map, as shown in Fig. 6. Based on the knowledge map’s information, this study proposed two entirely different strategies for each cluster. 5.1.1. Strengthen the quality and characteristics of the goods Network products often vary in regard to quality. Thus, the seller should enhance the authenticity information of the product content, and enhance the unique nature of the goods. If the seller reduces the cost by reducing product quality or functional requirements, they will lose customer loyalty, and also lose the opportunity to develop new customers. When the competition of homogeneous goods is too intense, sellers should try not to enter a vicious price war of the competition. Rather, they should try to improve product quality and create product differentiation in order to improve competitiveness. 5.1.2. Strengthening the network transaction safety mechanism In order to safeguard customers’ online security, each big shopping site must strengthen their safety mechanism. They may use a credit card safe transaction mechanism and strengthen advocacy by asking their customers to note their transaction security. Small dealers may use cash on delivery, in-store-pickup or cooperate with the convenience stores’ cash flow payment mechanism, so
customers can order goods on the web more conveniently and securely. 5.2. Group buying websites’ goods and services development According to the report by the Organization for Economic Cooperation and Development (OECD), the financial tsunami and the economic recession have prompted customers and businesses to find low-priced goods through the internet. Thus, the global online shopping platform is flourishing. OECD also mentioned, however, that the rise of online shopping also faces potential obstacles, including the security of personal information, dissatisfaction with products, goods delivery that does not meet customer expectations and so on. All of these problems have potential to influence customer confidence in of online group buying. In order to meet the requirements of target customers, and prompt potential customers to join, sellers should promote website security and service quality. In this way, they can provide the most ideal goods and supply complete pre- and after-sales service, to create a win–win situation with customers. Potential customers do not have experience with online group buying. If they have the opportunity to join OGB in the future to purchase goods, they have no sense of urgency and they lack ideas. Some females have only the choice of conservative group buying goods, like food commodities. Some male customers are only interested in 3C products. Among target customers, most of the males prefer products such as foodstuffs and 3C products. They are not interested in transportation and cosmetology maintenance products; most of the females prefer products like foodstuffs, cosmetology maintenance and apparel and popular products. The most uninteresting products for them are 3C products, household electrical appliances and transportation products. Moreover, this study also discovered that the most frequent online customers are 21–30 years old, and the most desired commodity for the 31–40 year old customers is in the food category. The majority of customers
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
3715
Table 5 The target customers’ buying preference in online group buying (Cluster-2). Rule
Support %
Confidence %
Life
Consequent
Antecedent
R1
12.93
92.68
2.44
Male
R2 R3
11.04 10.09
88.57 62.50
2.34 1.65
Male Male
R4
12.30
92.31
1.48
Female
R5 R6
10.73 11.99
88.24 81.58
1.42 1.31
Female Female
R8
12.62
55.00
1.45
R9
12.62
52.50
1.38
R12
11.36
50.00
1.32
R13
12.62
50.00
1.32
21–30 old 21–30 old 21–30 old 21–30 old
Online group buying maintenance Online group buying Online group buying commodities Online group buying popular articles Online group buying Online group buying commodities Online group buying commodities Online group buying
years years years years
preference10 = cosmetology
–
preference02 = 3C products preference07 = traffic
Online group buying preference01 = foodstuffs –
preference03 = clothing and
preference04 = leisure
Online group buying preference02 = cosmetology maintenance – Online group buying preference09 = home appliances Online group buying preference01 = foodstuffs
preference05 = 3C products
–
Online group buying preference05 = toy/game products Online group buying preference03 = clothing and popular articles
–
preference08 = 3C products preference10 = traffic
in these two age groups are unmarried. They are not interested in household goods, household electrical appliances and transportation category goods. The unit prices of household goods, household electrical appliances and transportation category goods are more expensive. Before purchasing these goods, customers prefer to test the items in person. Therefore, to put these goods into the group buying market, it is necessary to provide customers with the opportunity to test the products.
Online group buying preference02 = cosmetology maintenance
5.3. Enhance the competitiveness of the business services of group buying platforms What potential customers care most about is product quality. Moreover, after-sales service must allow them to return products if they are not satisfied. The thing that worries them most is the network security mechanisms. They prefer to pay cash on delivery. Word of mouth through trusted friends and family and newspaper
Foodstuff
Security trading mechanism
Recommended gifts, feedback
Fig. 6. Knowledge map of online group buying.
3716
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
and magazines reports will attract their attention. Finally, the findings of this study can serve as a reference to group buying platforms as they attempt to improve the quality of their service. (1) Group buying platforms: The target customers like to use credit cards or ATM transfers to pay bills, and they prefer websites that provide 24-h customer service hotlines to answer shopping-related questions at any time. The majority of group buying platforms currently do not provide these cash flow and service mechanisms. The target customers do not want to pay for shipping when they shop online and the prices must be lower than the market price. If group buying platforms provide a price and merchandise comparison function, they can help customers make decisions about purchases. (2) Group buying stores: The target customers who have shopped ay group buying stores are mostly between 31 and 50 years old. Their monthly disposable income is relatively high. Turning this population into loyal customers can increase profitability. This study found that the OGB motives are as follows: inexpensive price, good quality and more product types. The target customers hope that the flows of websites’ group buying processes are easy to understand. They hope that websites provide the following services: the latest group buying messages, ranking and evaluation of commodity items, messenger online customer services, 24-h advisory special service lines and online credit card or COD payment methods. The target customers also want more realistic feedback such as: cumulative consumption offset cash or expense offset by bonus points. More choices for customers would shorten the sales cycles of goods. Therefore, for the group buying stores, the fastest way to grab market opportunities is to allow more well-known providers and small dealers move into their platform. In addition to saving product development time, this may also allow customers to form unconscious routines by going to those shopping platforms with well-known brands, quality guarantees and fair prices. (3) Large dealers: This type of website service mechanism is very good already. Offering msn online customer service, a 24-h advisory special service line and online credit card or COD payment security mechanisms will greatly enhance customer’s confidence. (4) Small dealers: This study found that customers expect the website service functions of small dealers to be as complete as those of large dealers. However, it this is difficult for them to compete in regard to human resources and site management, making it difficult for this type of site to sustain operation. Well-known products and new customer promotion attract customers to visit their site. If they want to receive steady orders each month, the best way is to put their own products on the ready appropriate platform. This would increase commodities to exposure and sales opportunities. The original website can be the official image website. Combined with a simple shopping cart, this enables the customers to recognize shops. In addition, customers can choose from many different purchasing channels.
5.4. Strengthen online channels and combine group buying From the clustering results (customer features), we discover that these two categories of customers all have experience with group buying. The total group amount purchased by office workers is absolutely more than that of individuals. In addition, fewer deliver addresses may save a lot of deliver costs. Therefore, enhancing
group buying and combining online group buying channels is a strategy for future operation. 6. Conclusions There are many stores investing in the online group buying market. However, only the minority really succeed. Only by creating unique commodities, having the competitive advantage in regard to price and quality and so on, can stores meet the customers’ needs and strive for each customer to repeat buying. In this way, stores can stand out and find the key to success. Thus, this study finds some online group buying behavior patterns, including customer purchase preferences and customer purchase demands, in order to generate different online group buying marketing alternatives for merchants. These research results will provide owners with some useful references to discover potential customers, develop latent business possibilities, maintain the loyalty of target customers, and earn higher profits with online group buying. Acknowledgment This research was funded by the National Science Council, Taiwan, Republic of China, under Contract No. NSC 100-2410-H-032018–MY3. References Agrawal, R., Imilienski, T., & Swami, A. (1993). Mining association rules between sets of items in large databases. In Proceedings of the 1993 ACM SIGMOD international conference on management of data (pp. 207–216). Agrawal, R., & Shafer, J. (1996). Parallel mining of association rules. IEEE Transactions on Knowledge and Data Engineering, 8(6), 962–969. Arie, B., & Sterling, L. (2006). Generating rules from examples of human multiattribute decision making should be simple. Expert Systems with Applications, 31(2), 390–396. Frederick, A., & Brahim, C. D. (2006). Performance of software agents in nontransferable payoff group buying. Journal of Experimental & Theoretical Artificial Intelligence, 18(1), 17–48. Keim, D. A., Pansea, C., Sipsa, M., & Northb, S. C. (2004). Pixel based visual data mining of geo-spatial data. Computers & Graphics, 28, 327–344. Kouris, I. N., Makris, C. H., & Tsakalidis, A. K. (2005). Using information retrieval techniques for supporting data mining. Data & Knowledge Engineering, 52, 353–383. Liao, S. H., & Chen, Y. J. (2004). Mining customer knowledge for electronic catalog marketing. Expert Systems with Applications, 27, 521–532. Liao, S. H., Hsieh, C. L., & Huang, S. P. (2008). Mining product maps for new product development. Expert Systems with Applications, 34(1), 50–62. Liao, S. H., Chen, C. M., Chieh, C. L., & Hsiao, S. C. (2009). Mining information users’ knowledge for one-to-one marketing on information appliance. Expert Systems with Applications, 36(3), 4967–4979. Liao, S. H., Chen, Y. N., & Tseng, Y. Y. (2009). Mining demand chain knowledge of life insurance market for new product development. Expert Systems with Applications, 36(2), 9422–9437. Liao, S. H., Chen, J. L., & Hsu, T. Y. (2009). Ontology-based data mining approach implemented for sport marketing. Expert Systems with Applications, 36(8), 11045–11056. Liao, S. H., Ho, H. H., & Yang, F. C. (2010). Ontology-based data mining approach implemented on exploring product and brand spectrum. Expert Systems with Applications, 36(9), 11730–11744. Liao, S. H., Chen, Y., & Dang, M. Y. (2010). Mining customer knowledge for the development of new tourism products and customer relationship management. Expert Systems with Applications, 37(6), 4212–4223. Market Intelligence & Consulting Institute (Taiwan) (2010). 2009 Taiwan online group buying behavior report. http://www.mic.iii.org.tw/english/default.asp. Miguel, H. E., & María, Á. N. B. (2009). Accessing retailer equity through integration in retailers’ buying groups. International Journal of Retail & Distribution Management, 37(1), 43–62. Tokuro, M., & Takayuki, I. (2004). A group formation support system based on substitute goods in group buying. Systems and Computers in Japan, 35(10), 23–31. Ture, M., Kurt, I., Turhan, K. A., & Ozdamar, K. (2005). Comparing classification techniques for predicting essential hypertension. Expert Systems with Applications, 16(4), 379–384. Wang, Y. F., Chuang, Y. L., Hsu, M. H., & Keh, H. C. (2004). A personalized recommender system for the cosmetic business. Expert Systems with Applications, 26, 42.
Contents lists available at SciVerse ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
Mining customer knowledge for exploring online group buying behavior Shu-hsien Liao a,⇑, Pei-hui Chu b, Yin-ju Chen a, Chia-Chen Chang a a b
Department of Management Sciences, Tamkang University, No. 151, Yingjuan Rd., Danshuei Dist., New Taipei City 251, Taiwan, ROC Department of Information Management, National Taipei College of Business, No. 321, Sec. 1, Jinan Rd., Zhongzheng District, Taipei City 10051, Taiwan, ROC
a r t i c l e
i n f o
Keywords: Data mining Association rules Cluster analysis Online group buying Online group buying behavior
a b s t r a c t Online group buying is an effective marketing method. By using online group buying, customers get unbelievable discounts on premium products and services. This not only meets customer demand, but also helps sellers to find new ways to sell products sales and open up new business models, all parties benefit in these transactions. During these bleak economic times, group buying has become extremely popular. Therefore, this study proposes a data mining approach for exploring online group buying behavior in Taiwan. Thus, this study uses the Apriori algorithm as an association rules approach, and clustering analysis for data mining, which is implemented for mining customer knowledge among online group buying customers in Taiwan. The results of knowledge extraction from data mining are illustrated as knowledge patterns, rules, and knowledge maps in order to propose suggestions and solutions to online group buying firms for future development. Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction In these tough economic times, it is not only important to find new sources of income but also to cut down on expenditures. By using online group buying, it is easy to find more people in a short period of time to share freight costs and to buy in bulk so as to lower prices. It is also easier to get bigger discounts when more people take part in a group purchase. On the other hand, online group buying is a model in which multiple buyers cooperate and buy the same good/ service in order to bargain with the proprietor, if there are enough buyers, they may aggregate buyer power to get volume discounts. All parties benefit in these transactions. In the past, group buyers were limited to members of companies, friends, families or communities. They filled out the types of goods and quantity of items to be purchased on the flyers to order goods. As the internet develops, it is becoming an increasingly prosperous network for many types of commerce. In addition, the large shopping sites are familiar with online shopping markets and the follow-up group buying platforms are learning this market. They have created group buying discount zones and provided some incentives and discounts to attract more users to visit and buy. The more visitors that come, the more goods are exhibited and sold. Thus, the consumption patterns of group buying have become more active in online shopping markets. According to pollster online survey of group buying, the most preferred goods are foodstuffs, making up to 33%. The group buying shares of other items are: clothing and accessories 11%, cosmetics 8%, articles of daily use 5%, others 3% and home appliances 2%. How⇑ Corresponding author. E-mail address: [email protected] (S.-h. Liao). 0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.09.066
ever, 36% of people in Taiwan have yet to experience group buying (Market Intelligence & Consulting Institute in Taiwan, 2010). As online group buying (OGB) market increases and a great variety commodity stats become available, consumer demand is changing fast, implicitly shortening the product life cycle. Clear assessment of the overall sales strategies of internet group buying has a positive effect. However over-reliance can result in the cart coming before the horse. In 2009, as many as 50.3% of online stores did not break even, while 5.3% increased sales compared to last year in Taiwan. The reason is that business operation was impacted by the overall economy. More and more intense price wars led to descending overall profits (Market Intelligence & Consulting Institute in Taiwan, 2010). In addition, in electronic marketplaces, group buying is seen as an effective form of electronic commerce and a promising field. For example, Tokuro and Takayuki (2004) proposed using decision support systems for buyers in group buying. Their system supports buyers’ decision making by using the Analytic Hierarchy Process with three methods for group integration. First, buyers trade in simple group buying. Second, all buyers are integrated. Third, some buyers are integrated. Thus, buyers’ multi-attribute utilities are effectively expressed in group integration and buyers can purchase goods at a lower price. Buyers’ payments are decided based on their degree of compromise. Software agents can be useful in forming buyers’ groups since humans have considerable difficulties in finding Pareto-optimal deals (no buyer can be better without another being worse) in negotiation situations (Frederick & Brahim, 2006). The above study developed a negotiation protocol for software agents, which evaluated whether or not the problem is difficult on average and why. This protocol is probably able to find
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
a Pareto-optimal solution and, furthermore, minimize the worst distance to the ideal among all software agents given strict preference ordering. In addition, Miguel and María (2009) explored the circumstances under which the retailers’ use of the buying group’s brand name may benefit them. Their research findings show that the retailer’s use of the buying group’s brand name is more capable of improving the retailer’s economic satisfaction with the buying group when differentiation is perceived to be a source of competitive advantage, when the environment is perceived as more dynamic and when the retailer is strategically integrated in the relationship with the buying group. However, only a few studies have explored online group buying behavior patterns and segments from customers. On the other hand, customers play an important role as business assets. Most of the parties involved in sales, such as the commercial web sites, retailers and channels, are aware of the need for businesses to acquire better customer knowledge. However, this is easier said than done since customers’ knowledge is concealed within the customers. It is available but not accessible, and there is little possibility of exploring the full volume of data that should be collected for its potential value. Inefficient utilization renders the data collected useless, causing databases to become ‘data dumps’ (Keim, Pansea, Sipsa, & Northb, 2004). Thus, finding ways to effectively process and use data is an artificial issue that calls for new techniques to help analyze, understand or even visualize the huge amounts of stored data gathered from business and scientific applications (Liao & Chen, 2004). Among the new techniques developed, data mining is a process of discovering significant knowledge, such as patterns, associations, changes, anomalies and significant structures from large amounts of data stored in databases, data warehouses, or other information repositories (Keim et al., 2004). Customer knowledge extracted through data mining can be integrated with products and marketing knowledge from research and can be provided to up stream suppliers as well as downstream retailers. Thus, it can serve as a reference for product development, product promotion and customer relationship management. When effectively utilized, such knowledge extraction can enable enterprises to gain a competitive edge by producing customer-oriented goods that increase consumer satisfaction (Arie & Sterling, 2006; Liao, Chen, Chieh, & Hsiao, 2009; Liao, Chen, & Dang, 2010; Liao, Chen, & Hsu, 2009; Liao, Chen, & Tseng, 2009; Liao, Ho, & Yang, 2010; Liao, Hsieh, & Huang, 2008). Accordingly, this study investigates online group buying behavior, and implements data mining approach to analyze Taiwanese customers. There are two data mining stages implemented in this study. First, this study employs the k-means algorithm to cluster the customers into potential customers and target customers, and uses the Apriori algorithm to generate association rules for each cluster. The rules are proposed to the group buying firms to help them attain possible new customers, services and sales. The rest of this paper is organized as follows. Section 2 introduces the proposed data mining system, which includes the system framework, system design, and physical database design. Section 3 introduces the data mining approach, including the association rules and cluster analysis. Section 4 presents the data mining process and the analyzed results. Section 5 describes research findings, managerial implications. Finally, a brief conclusion is presented in Section 6. 2. Data mining systems 2.1. Research framework Because online group buying customer information and product sales information are difficult to obtain, this study used a designed
3709
questionnaire to collect research data. This study uses k-means to sort customers into clusters to generate association rules for each cluster, and then proposes suggestions and solutions for online group buying to open up possible new services and sales. 2.2. System framework This study collected the past data of customers to establish a database system, We then analyze the entire database system by data mining to find the association between group buying customer behavior and commodities buying patterns, including consumer shopping preferences and demand consideration, etc. The aim of this approach is to enable businesses to further understand group buying experience/patterns, rather than the psychology and orientation of experienced online group buying customers. The propose is to make suitable marketing suggestions, so that they can really provide customers with preferred products and services, while reducing marketing costs and increasing business profits. The system framework is shown in Fig. 1. Accordingly, the system design diagram is shown in Fig. 2. It shows that this study incorporates data related to customers, customer behavior and products into the database, and analyzes the entire database by data mining for customer patterns and market segmentations to find the different types of target customers. Online firms can evaluate customer knowledge management for marketing and service, and then determine efficient means to achieve the goals of exploring online group buying behavior. In this study, the design and operation of a physical database is used to construct a relational database enter data in the table through Microsoft Access 2003. Although general database software cannot accommodate too many people online simultaneously, Microsoft SQL 2005 can satisfy this need, because the general database systems use standard structured query language (SQL). Because each type of data storage and processing is different, in order to give a programming language access to the information database system, manufacturers can design a driver for all types of language using standard SQL, and then access their database through a regional network. Microsoft’s console provides an open database link (open database connectivity; ODBC). Thus, administrators can manage a variety of ODBC drivers (Fig. 3). 2.3. Questionnaire design and data collection This study uses the questionnaire approach to collect data from general customers who have group buying or online group buying experience, and establishes the database system using collected data. The main purpose of the questionnaire design is to understand the customer motivations of the entities involved in group buying, the degree of involvement of internet group buying, and explore the factors which affect the correlation between internet group buying behavior and customer psychology. The questionnaire is divided into five parts. Part 1 focuses on customer personal information (9 items), part 2 discusses the customer psychology of individual customers engaged group buying (9 items), part 3 looks at the customer tendencies of individual online group buying (9 items), while part 4 focuses on the shopping behavior of individual online group buying (4 items) and the service mechanism of online group buying (10 items). The quantity of preliminary draft questionnaire sent and responded to, the pre-test questionnaire and formal questionnaire are shown in Table 1. Thus, the questionnaires were collected from November 1st to November 30th of 2009. Altogether, we sent 720 questionnaires and 621 were collected. Excluding omissions and incomplete answers, there
3710
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
Fig. 1. System framework.
were 550 valid responses, with an effective collection ratio of 88.6%. 3. Data mining 3.1. Association rules Discovering association rules is an important data mining problem (Agrawal, Imilienski, & Swami, 1993), and there has been considerable research on using association rules for data mining problems. The association rules algorithm is used mainly to determine the relationships between items or features that occur synchronously in databases. For instance, during a trip to the shopping center, if the people who buy item X also buy item Y as well, there exists a relationship between item X and item Y. Such informa-
tion is useful for decision makers. Therefore, the main purpose of implementing the association rules algorithm is to find synchronous relationships by analyzing random data and to use these relationships as a reference for decision-making. The association rules are defined as follows (Wang, Chuang, Hsu, & Keh, 2004): Make I = {i1, i2, . . . , im} the item set, in which each item represents a specific literal. D stands for a set of transactions in a database in which each transaction T represents an item set such that T # I. That is, each item set T is a non-empty sub-item set of I. The association rules are an implication of the form X ? Y, where T X I, Y I and X Y = U. The rule X ? Y holds in the transaction set D according to two measurement standards – support and confidence. Support (denoted as_ Sup(X, D)) represents the rate of transactions in D containing the item set X. Support is used to evaluate the statistical importance of D, and the higher its value, the
3711
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
standard called confidence (denoted as Conf(X ? Y)), representing the rate of transactions in D that contain both X and Y. That is, T Conf(X ? Y) = Sup(X Y)/Sup(X, D). In this case, Conf(X ? Y) denotes that if a transaction includes X, the chance that this transaction also contains Y is relatively high. The measure of confidence is then used to evaluate the level of confidence about the association rules X ? Y. Given a set of transactions D, the problem of mining association rules is used to generate all transaction rules that have certain levels of user-specified minimum support (called Min sup) and confidence (called Minconf) (Kouris, Makris, & Tsakalidis, 2005). According to Agrawal and Shafer (1996), the problem of mining association rules can be broken down into two steps. The first step is to detect a large item set whose support is greater than Min sup, and the second step is to generate association rules, using the large item set. Such rules must satisfy the following two conditions: 1. SupðX [ Y; DÞ P Min sup. 2. Conf ðX ! YÞ P Minconf To explore association rules, many researchers use the Apriori algorithm (Agrawal et al., 1993). In order to reduce the possible biases incurred when using these measurement standards, the simplest way to judge the standard is to use the lift judgment. Lift is defined as: Lift = Confidence(X ? Y)/Sup(Y) (Wang et al., 2004). 3.2. Cluster analysis and k-means algorithm
Fig. 2. System design.
more important the transaction set D is. Therefore, the rule X ? Y S which has support Sup(X Y, D) represents the rate of transactions S in D containing X Y. Each rule X ? Y also has another measuring
The process of partitioning a large set of patterns into disjoint and homogeneous clusters is fundamental in knowledge acquisition. It is called Clustering in most studies and it has been applied in various fields, including data mining, statistical data analysis, compression and vector quantization. The k-means is a very popular algorithm and is one of the best for implementing the clustering process. k-Means clustering proceeds in the following order.
Fig. 3. Physical database designs.
Table 1 Send and receive questionnaire quantity. Questionnaire
During
Send
Receive
Effective questionnaire
Returns-ratio
Draft Pre-test Formal
2009/9/26–2009/10/15 2009/10/21–2009/10/25 2009/11/1–2009/11/30
– 25 720
– 22 621
– 20 550
– 90.9% 88.6%
3712
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
Firstly, the k numbers of observations are randomly selected from all N number of observations according to the number of clusters, and these become centers of the initial clusters. Secondly, for each of the remaining N–k observations, the nearest cluster is found in terms of the Euclidean distance with respect to xik-means algorithm for cluster analysis as a data mining approach (Ture, Kurt, Turhan, & Ozdamar, 2005). 4. Data mining and analysis 4.1. k-Mean analysis This study employs the k-means algorithm to cluster the questionnaire survey data, and categorize data into group of potential customers (Cluster-1) and target customers (Cluster-2). 4.1.1. Customer adumbration Common features of both cluster include: most of them are 31– 40-year-old unmarried females, they go online several times a day, the majority spend between 3 h to 5 h a day online time, the majority of group buying goods are foodstuffs, most people group buy goods with colleagues because of inexpensive prices or simple necessity, most people trust the quality and safety of the products, most companies do not prohibit the staff from carrying out group buying during working hours and they even cross-departmental colleagues to join in group buying. In addition, the differing characteristics between each cluster are shown in Table 2. (1) Potential customers (Cluster-1): These customers have good economic capacity, with monthly disposable income amounting to more than 40,001 New Taiwan dollars (NTD), their online time does not exceed 3 h, they have experience with entity group buying, but no experience of online group buying and they will consider the online group buying in the future. The reasons that members of this group do not have experience with internet group buying are that they are unable to confirm product quality, the security of network transactions and individual data protection. (2) Target customers (Cluster-2): Their monthly disposable amount income is around 10,001–20,000 NTD, the reason they purchase goods by group buying is not only due to the inexpensive price, but also but also because of good word of mouth and convenience factors; they complete group buying transactions in 15 min. This group of customer has experience of online group buying from both large and small firms in the past three months the frequency of online
group buying is approximately 1–3 times; most of the purchased through group buying are no more than 1000 NTD for foodstuffs. The people they invite to join them in group buying are classmates, friends and colleagues. They know about group buying sites and group buying news. 4.1.2. Customer behavior The customer behavior analysis of the two clusters shows that there are some common features as below: (1) They find information about group buying products through introductions from relatives/ friends and newspapers/magazines reports. (2) The main reasons why they purchase goods from the same group buying websites are good product quality and low price. (3) All of them have an optimistic view of the future development and future prospects of online group buying. (4) The longest time they are willing to wait for popular commodities is 15 days. (5) The preferred payment method is cash on delivery. (6) Upon obtaining unsatisfactory products, they may return the products after home delivery service. (7) They think group buying sites should provided special lines for customer service and msn online service. (8) Expected service contents should ensure product quality and after service to make the procedures and processes of group buying clear and easy to understand. (9) Expected services functions should compare the function of price and product and strengthen after service consultation. The behaviors of each cluster are shown as below. The reasons potential customers do not shop on OGB are uncertainty about commercial quality, personal data security and transaction security considerations. For the target customers, the majority have consumed products from large group buying sites, and many purchased goods from small dealers; over the past three months, the majority of customers have engaged in online group buying 1–3 times; the majority buy foodstuff goods under 500 NTD, followed by purchases of foodstuff goods between 501 and 1000 NTD. The online group buys goods due to demand, because prices are below the market price and for the sake of convenience. Most people in this group receive information resources about group buying sets and group buying goods from schoolmates or colleagues, relatives or friends. Recommendations from friends to carry out online group buying are related to low price and good commodity quality. The main products purchased through online group buying differ between the two clusters. Potential customers mainly buy: Leisure commodities, foodstuffs, cosmetology maintenance commodities, clothing/popular commodities and 3C commodities. Target customers mainly buy: foodstuffs, cosmetology maintenance commodities, home appliances, clothing/popular commodities and toy/game commodities. The customer behavior of each cluster is shown in Table 3.
Table 2 k-Means clustering results (customer profile). Cluster
Cluster-1
Cluster-2
Naming Feature
Potential customer Mostly unmarried females, with good economic capacity for their age level. They only participate in group buying with company members 232 Female (55.36%) Male (44.64%) Mostly above 40,001 (NTD)
Target customer Mostly unmarried females with monthly disposable income. In addition to being involved in group buying with company colleagues, they also have experience with online group buying 318 Female (62.15%) Male (37.85%) Mostly during 10,001–20,000 (NTD)
Samples Sex Monthly disposable amount Information source of group buying/ collective buying The reason why they join in group (collective) buying Experience of online group buying
1. 2. 3. 1. 2.
Recommendations of colleagues and friends Colleagues invite them to buy Vendors provide samples to eat and to test Cheap price Simple demand
No More than half of them consider being involved in online group buying in the future
1. 2. 3. 1. 2. Yes
Recommendations of colleagues and friends Network introduction Vendors provide samples to eat and to test Cheap price Well-known products
3713
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716 Table 3 k-Means clustering results (customer behavior). Behavior
Cluster-1
Cluster-2
Do not shopping on OGB
–
Has consumed products at these group buying sites
1. Commercial quality uncertainty 2. Personal data security 3. Transaction security considerations –
Over the past three months, the frequency of online group buying Often buy goods online groups
– –
Movement of online group buying
–
Information source of group buying sets and goods
–
The reasons friends recommend online group buying
–
The goods most members to buy in online group and buying sets
1. 2. 3. 4. 5.
Leisure commodities Foodstuffs cosmetology maintenance commodities Clothing/popular commodities 3C commodities
4.1.3. Customer online group buying tendencies and demands This study used k-means to divide customers into potential customers (Cluster-1) and target customers (Cluster-2), and then used the Apriori algorithm to do the association analysis for each cluster. This was followed by a progressive method to identify the association rules. These association rules were used to find out customer tendencies and to explore customer demand in online group buying. These results will provide an effective reference to enhance marketing operations in related industries. The association diagrams of both clusters are shown in Figs. 4 and 5. Further, further investigation explores gender and all age levels of shopping priority, in order to find association rules between each cluster and group buying purchases. (1) Potential customers (Cluster-1): They have no experience of online group buying and there are goods which they urgently want to purchase from online group buying. If there were the opportunity to join online group buying, the first consideration would be to buy frequently purchased goods. Some of the male and 21–30 year old customers have partiality toward buying 3C products. If they were to participate
1. Large dealers 2. Small dealers 1–3 times 1. Foodstuff under 500 NTD 2. Foodstuff between 501–1000 NTD 1. Just demand 2. Below the market price 3. Convenience 1. Schoolmates or colleagues 2. Relatives or friends 1. Low price 2. Good commodity quality 1. Foodstuffs 2. Cosmetology maintenance commodities 3. Home appliances 4. Clothing/popular commodities 5. Toy/game commodities
in online group buying in the future, some of the female customers would have partiality toward buying foodstuffs, as shown in Table 4. (2) Target customers (Cluster-2): Because this population has online group buying experience, they have more ideas about online group buying. For online group buying, the order of preference for most male customers is as follows: foodstuff and 3C products. The order of preference for most female consumers is as follows: foodstuffs, cosmetic products, apparel and popular products. The order of preference for 21–30 year old customers is foodstuffs, cosmetic products, apparel and popular products, leisure goods, 3C products and game/toy products, as shown in Table 5. 5. Research findings and managerial implications Through the implementation of data mining, this study divided the customers into potential customers and target customers. The former group has potential for future online group buying. The findings of this study will be helpful in the development of potential customer markets by manufacturers. The latter group represents
Fig. 4. Association diagram of Cluster-1’s online group buying experience.
3714
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
Fig. 5. Association diagram of Cluster-2’s online group buying experience.
Table 4 The potential customer buying preferences of online group buying (Cluster-1). Rule
Support %
Confidence %
Life
Consequent
Antecedent
R1 R2 R3
11.16 11.16 21.89
88.46 42.31 72.55
1.98 1.47 1.31
Male 21–30 years old Female
Online group buying preference01 = 3C product Online group buying preference01 = 3C product Online group buying preference01 = foodstuff
the existing online group buying target customers. This is the group which sellers wish to understand and grasp. This study explores their purpose of group buying and their service mechanism demands for group buying. Some research findings and managerial implications are discussed as follows. 5.1. Customer spending tendencies in online group buying Online group buying is a highly competitive market. Companies must work diligently of they want to join the online group buying market. The mining analysis describes two clusters (potential customers and target customers) in an online group buying marketing knowledge map, as shown in Fig. 6. Based on the knowledge map’s information, this study proposed two entirely different strategies for each cluster. 5.1.1. Strengthen the quality and characteristics of the goods Network products often vary in regard to quality. Thus, the seller should enhance the authenticity information of the product content, and enhance the unique nature of the goods. If the seller reduces the cost by reducing product quality or functional requirements, they will lose customer loyalty, and also lose the opportunity to develop new customers. When the competition of homogeneous goods is too intense, sellers should try not to enter a vicious price war of the competition. Rather, they should try to improve product quality and create product differentiation in order to improve competitiveness. 5.1.2. Strengthening the network transaction safety mechanism In order to safeguard customers’ online security, each big shopping site must strengthen their safety mechanism. They may use a credit card safe transaction mechanism and strengthen advocacy by asking their customers to note their transaction security. Small dealers may use cash on delivery, in-store-pickup or cooperate with the convenience stores’ cash flow payment mechanism, so
customers can order goods on the web more conveniently and securely. 5.2. Group buying websites’ goods and services development According to the report by the Organization for Economic Cooperation and Development (OECD), the financial tsunami and the economic recession have prompted customers and businesses to find low-priced goods through the internet. Thus, the global online shopping platform is flourishing. OECD also mentioned, however, that the rise of online shopping also faces potential obstacles, including the security of personal information, dissatisfaction with products, goods delivery that does not meet customer expectations and so on. All of these problems have potential to influence customer confidence in of online group buying. In order to meet the requirements of target customers, and prompt potential customers to join, sellers should promote website security and service quality. In this way, they can provide the most ideal goods and supply complete pre- and after-sales service, to create a win–win situation with customers. Potential customers do not have experience with online group buying. If they have the opportunity to join OGB in the future to purchase goods, they have no sense of urgency and they lack ideas. Some females have only the choice of conservative group buying goods, like food commodities. Some male customers are only interested in 3C products. Among target customers, most of the males prefer products such as foodstuffs and 3C products. They are not interested in transportation and cosmetology maintenance products; most of the females prefer products like foodstuffs, cosmetology maintenance and apparel and popular products. The most uninteresting products for them are 3C products, household electrical appliances and transportation products. Moreover, this study also discovered that the most frequent online customers are 21–30 years old, and the most desired commodity for the 31–40 year old customers is in the food category. The majority of customers
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
3715
Table 5 The target customers’ buying preference in online group buying (Cluster-2). Rule
Support %
Confidence %
Life
Consequent
Antecedent
R1
12.93
92.68
2.44
Male
R2 R3
11.04 10.09
88.57 62.50
2.34 1.65
Male Male
R4
12.30
92.31
1.48
Female
R5 R6
10.73 11.99
88.24 81.58
1.42 1.31
Female Female
R8
12.62
55.00
1.45
R9
12.62
52.50
1.38
R12
11.36
50.00
1.32
R13
12.62
50.00
1.32
21–30 old 21–30 old 21–30 old 21–30 old
Online group buying maintenance Online group buying Online group buying commodities Online group buying popular articles Online group buying Online group buying commodities Online group buying commodities Online group buying
years years years years
preference10 = cosmetology
–
preference02 = 3C products preference07 = traffic
Online group buying preference01 = foodstuffs –
preference03 = clothing and
preference04 = leisure
Online group buying preference02 = cosmetology maintenance – Online group buying preference09 = home appliances Online group buying preference01 = foodstuffs
preference05 = 3C products
–
Online group buying preference05 = toy/game products Online group buying preference03 = clothing and popular articles
–
preference08 = 3C products preference10 = traffic
in these two age groups are unmarried. They are not interested in household goods, household electrical appliances and transportation category goods. The unit prices of household goods, household electrical appliances and transportation category goods are more expensive. Before purchasing these goods, customers prefer to test the items in person. Therefore, to put these goods into the group buying market, it is necessary to provide customers with the opportunity to test the products.
Online group buying preference02 = cosmetology maintenance
5.3. Enhance the competitiveness of the business services of group buying platforms What potential customers care most about is product quality. Moreover, after-sales service must allow them to return products if they are not satisfied. The thing that worries them most is the network security mechanisms. They prefer to pay cash on delivery. Word of mouth through trusted friends and family and newspaper
Foodstuff
Security trading mechanism
Recommended gifts, feedback
Fig. 6. Knowledge map of online group buying.
3716
S.-h. Liao et al. / Expert Systems with Applications 39 (2012) 3708–3716
and magazines reports will attract their attention. Finally, the findings of this study can serve as a reference to group buying platforms as they attempt to improve the quality of their service. (1) Group buying platforms: The target customers like to use credit cards or ATM transfers to pay bills, and they prefer websites that provide 24-h customer service hotlines to answer shopping-related questions at any time. The majority of group buying platforms currently do not provide these cash flow and service mechanisms. The target customers do not want to pay for shipping when they shop online and the prices must be lower than the market price. If group buying platforms provide a price and merchandise comparison function, they can help customers make decisions about purchases. (2) Group buying stores: The target customers who have shopped ay group buying stores are mostly between 31 and 50 years old. Their monthly disposable income is relatively high. Turning this population into loyal customers can increase profitability. This study found that the OGB motives are as follows: inexpensive price, good quality and more product types. The target customers hope that the flows of websites’ group buying processes are easy to understand. They hope that websites provide the following services: the latest group buying messages, ranking and evaluation of commodity items, messenger online customer services, 24-h advisory special service lines and online credit card or COD payment methods. The target customers also want more realistic feedback such as: cumulative consumption offset cash or expense offset by bonus points. More choices for customers would shorten the sales cycles of goods. Therefore, for the group buying stores, the fastest way to grab market opportunities is to allow more well-known providers and small dealers move into their platform. In addition to saving product development time, this may also allow customers to form unconscious routines by going to those shopping platforms with well-known brands, quality guarantees and fair prices. (3) Large dealers: This type of website service mechanism is very good already. Offering msn online customer service, a 24-h advisory special service line and online credit card or COD payment security mechanisms will greatly enhance customer’s confidence. (4) Small dealers: This study found that customers expect the website service functions of small dealers to be as complete as those of large dealers. However, it this is difficult for them to compete in regard to human resources and site management, making it difficult for this type of site to sustain operation. Well-known products and new customer promotion attract customers to visit their site. If they want to receive steady orders each month, the best way is to put their own products on the ready appropriate platform. This would increase commodities to exposure and sales opportunities. The original website can be the official image website. Combined with a simple shopping cart, this enables the customers to recognize shops. In addition, customers can choose from many different purchasing channels.
5.4. Strengthen online channels and combine group buying From the clustering results (customer features), we discover that these two categories of customers all have experience with group buying. The total group amount purchased by office workers is absolutely more than that of individuals. In addition, fewer deliver addresses may save a lot of deliver costs. Therefore, enhancing
group buying and combining online group buying channels is a strategy for future operation. 6. Conclusions There are many stores investing in the online group buying market. However, only the minority really succeed. Only by creating unique commodities, having the competitive advantage in regard to price and quality and so on, can stores meet the customers’ needs and strive for each customer to repeat buying. In this way, stores can stand out and find the key to success. Thus, this study finds some online group buying behavior patterns, including customer purchase preferences and customer purchase demands, in order to generate different online group buying marketing alternatives for merchants. These research results will provide owners with some useful references to discover potential customers, develop latent business possibilities, maintain the loyalty of target customers, and earn higher profits with online group buying. Acknowledgment This research was funded by the National Science Council, Taiwan, Republic of China, under Contract No. NSC 100-2410-H-032018–MY3. References Agrawal, R., Imilienski, T., & Swami, A. (1993). Mining association rules between sets of items in large databases. In Proceedings of the 1993 ACM SIGMOD international conference on management of data (pp. 207–216). Agrawal, R., & Shafer, J. (1996). Parallel mining of association rules. IEEE Transactions on Knowledge and Data Engineering, 8(6), 962–969. Arie, B., & Sterling, L. (2006). Generating rules from examples of human multiattribute decision making should be simple. Expert Systems with Applications, 31(2), 390–396. Frederick, A., & Brahim, C. D. (2006). Performance of software agents in nontransferable payoff group buying. Journal of Experimental & Theoretical Artificial Intelligence, 18(1), 17–48. Keim, D. A., Pansea, C., Sipsa, M., & Northb, S. C. (2004). Pixel based visual data mining of geo-spatial data. Computers & Graphics, 28, 327–344. Kouris, I. N., Makris, C. H., & Tsakalidis, A. K. (2005). Using information retrieval techniques for supporting data mining. Data & Knowledge Engineering, 52, 353–383. Liao, S. H., & Chen, Y. J. (2004). Mining customer knowledge for electronic catalog marketing. Expert Systems with Applications, 27, 521–532. Liao, S. H., Hsieh, C. L., & Huang, S. P. (2008). Mining product maps for new product development. Expert Systems with Applications, 34(1), 50–62. Liao, S. H., Chen, C. M., Chieh, C. L., & Hsiao, S. C. (2009). Mining information users’ knowledge for one-to-one marketing on information appliance. Expert Systems with Applications, 36(3), 4967–4979. Liao, S. H., Chen, Y. N., & Tseng, Y. Y. (2009). Mining demand chain knowledge of life insurance market for new product development. Expert Systems with Applications, 36(2), 9422–9437. Liao, S. H., Chen, J. L., & Hsu, T. Y. (2009). Ontology-based data mining approach implemented for sport marketing. Expert Systems with Applications, 36(8), 11045–11056. Liao, S. H., Ho, H. H., & Yang, F. C. (2010). Ontology-based data mining approach implemented on exploring product and brand spectrum. Expert Systems with Applications, 36(9), 11730–11744. Liao, S. H., Chen, Y., & Dang, M. Y. (2010). Mining customer knowledge for the development of new tourism products and customer relationship management. Expert Systems with Applications, 37(6), 4212–4223. Market Intelligence & Consulting Institute (Taiwan) (2010). 2009 Taiwan online group buying behavior report. http://www.mic.iii.org.tw/english/default.asp. Miguel, H. E., & María, Á. N. B. (2009). Accessing retailer equity through integration in retailers’ buying groups. International Journal of Retail & Distribution Management, 37(1), 43–62. Tokuro, M., & Takayuki, I. (2004). A group formation support system based on substitute goods in group buying. Systems and Computers in Japan, 35(10), 23–31. Ture, M., Kurt, I., Turhan, K. A., & Ozdamar, K. (2005). Comparing classification techniques for predicting essential hypertension. Expert Systems with Applications, 16(4), 379–384. Wang, Y. F., Chuang, Y. L., Hsu, M. H., & Keh, H. C. (2004). A personalized recommender system for the cosmetic business. Expert Systems with Applications, 26, 42.