行政院國家科學委員會補助專題研究計畫成果報告
※※※※※※※※※※※※※※※※※※※※※※※※※※
※ ※
※
以顧客需求為主之可適性線上品類管理
※
※
An Adaptive Approach to Customer-centered
※
※
Online Category Management
※
※ ※
※※※※※※※※※※※※※※※※※※※※※※※※※※
計畫類別：
R個別型計畫 □整合型計畫
計畫編號：NSC90－2416－H－110－036
執行期間： 90 年 8 月 1 日至 91 年 7 月 31 日
計畫主持人：張德民博士 國立中山大學資訊管理系
計畫參與人員：劉江倫 國立中山大學資訊管理系
賴志明 國立中山大學資訊管理系
本成果報告包括以下應繳交之附件：
□赴國外出差或研習心得報告一份
□赴大陸地區出差或研習心得報告一份
□出席國際學術會議心得報告及發表之論文各一份
□國際合作研究計畫國外研究報告書一份
執行單位：國立中山大學資訊管理學系
中 華 民 國 91 年 10 月 18 日
行政院國家科學委員會專題研究計畫成果報告
計畫編號：NSC 90-2416-H-110-036
執行期限：90 年 8 月 1 日至 91 年 7 月 31 日
主持人：張德民博士 國立中山大學資訊管理系
計畫參與人員：劉江倫 國立中山大學資訊管理系
賴志明 國立中山大學資訊管理系
中文摘要 電子商務伴隨著網際網路的興起而蓬 勃發展，對零售商而言，電子商務為他們 帶來了新的銷售通路。和多數的線上型錄 商一樣，越來越多的零售商也透過「線上 型錄」的表現方式來進行線上產品銷售。 線上零售商在網站設計上常採用商品類 別，本研究即針對線上零售商之品類管理 提出兩階段的作法，來達到以顧客為主且 方便消費者瀏覽的目的。第一階段是區隔 產品市場，進一步建立線上型錄的網站架 構，在線上型錄越上層的產品代表越受消 費者的歡迎。第二階段，隨著消費者對於 產品的喜好的改變，我們設計一套可適性 的演算法來調整原來的網站架構。本研究 相信設計出來的線上型錄可以更符合消費 者需求，而且也可以為線上零售商創造更 大的商機。 關鍵詞：電子商務、線上品類管理、市場 區隔、網頁調適 AbstractThe Internet with growing electronic commerce provides a new selling channel for retailers. More and more retailers attempt aggressively to organize their catalogs for online sales just as online catalogers. Online retailing web sites usually employ categories in their web design. In this paper, we propose a two-stage approach to the customer-centered online category management. In the first stage, cluster-based market segmentation is employed to construct the web hierarchy with products in higher levels indicating more interesting to customers. The second stage is to dynamically adjust the hierarchy when customers’ preference
indicated by browsing patterns is changed. It is desired that the online catalog organization designed using such an approach can receive more customers’ attentions and further contribute to the growth of retailing business on the web.
Keywords : Electronic Commerce, Online Category Management, Market Segmentation, Adaptation of Web Page
1. Introduction
As the prevalence of the Internet and World Wide Web (WWW), retailing business has gone through dramatic changes with the rapid progress of electronic commerce that operates online sales literally from anywhere, anytime (Kalakota & Whinston, 1997). Electronic retailing (e-tailing) has created an unprecedented ‘format’ with high deliver value at low cost (Gerbert et al., 1999). One of the major concerns for successful e-tailing is a good design of online catalogs that are electronic counterparts of shop’s shelves, sections, and departments. Online catalog organization has thus become an important issue for e-tailing business development.
Current online catalog design, however, has its deficiency. According to Evans (1998), people who surf through the web can be classified into searchers and browsers. Most online retailing web sites provide search functions for searchers to spot desired products. By contrast, they provide little help for casual browsers, who take a more open-minded and exploratory approach to navigation (Evans, 1998). For those customers, it is important to
determine which products are most interesting to them, and present those products on top of the web hierarchy for their efficient navigation (Evans, 1998; Huizingh, 2000). It is a strategy to provide browsers limited but essential products in one web page.
To employ customer-centered category management to help browsers find limited and essential products, we propose a practical solution by employing market segmentation to segment products from customers’ viewpoints. Traditional market segmentation is to segment customers with different marketing strategies applied to different customer segments. Similar idea can be easily employed to segment products. Different segments of products indicate different degrees of importance to customers. Such information can help browsers for efficient navigation and assist e-tailers to prioritize their products in the online catalog organization.
Furthermore, static online catalog may be readily outdated because customers’ interests are changing over time. It is time-consuming and costly to apply the market segmentation every time to restructure the catalog organization. On the other hand, it is much easier to acquire customers’ browsing behavior in WWW environments. An adaptive algorithm can be developed to dynamically adjust the catalog organization once changes of customers’ browsing patterns have been detected. The result should be more fit to customers’ needs so that browsers can easily navigate the online catalog organization before they lose their patience.
2. Online Category Management Framework
In this section, we propose an adaptive approach to the customer-centered online category management for e-tailers. The framework of the proposed adaptive approach is shown in Figure 1. It mainly includes two stages. The first stage is to generate an initial catalog organization based on the cluster analysis of products while an adaptive algorithm is
applied to refine the organization in the second stage. This two-stage method consists of five steps. Detailed descriptions are addressed in the following.
Step1. Categories Determination and Purchasing Data Collection
The first thing to start online catalog is to determine the variety of categories for a specific e-tailing store. For example, Land’s End Store is selling garment online and the categories for Land’s End may include knit tops, polos & t-shirts, sweaters, tailored pants and so on. The categories for Amazon bookstore may be classified as art & architecture, history, horror and the like.
Arrangement between categories in electronic catalog can be achieved by utilizing ‘reach’ values. Reach is the percentage of households that purchase the product category in one-year period (Blattberg et al. 1995). Categories with higher reach value should be placed on more important position and vice versa. Purchasing Data Browsing Data Cluster Analysis Initial Online Catalog Adaptive Online Catalog Adaptive Algorithm Categories Determineation Align RFM Data Extracted Applied Generate Applied Generate Stage I Stage II Construct Construct Attract Attract
Figure 1 Framework of Online Category Management
After the categories are determined and arranged, one may proceed to collect customer’s purchasing data for the purpose of product segmentation. Purchasing records can reflect customers’ behaviors towards the selling goods and behavioral data are the best starting point for marketing (Ha & Park, 1998; Marcus, 1998; Saarenvirta, 1998). For retailers, it is easy to gather customers’ purchasing data from routine transaction records. Such data are then used as the bases for market segmentation. It is desired that customers’ purchasing data reflect their preferences about products so that the segmentation is meaningful with respect to customers’ needs.
Step 2. Data Extraction with RFM Variables Among many behavioral variables, RFM variables are well-known behavioral variables to both researchers and practitioners. To thoroughly understand customers, most database marketers employ RFM variables to segment markets of customers (Newell, 1997). RFM variables that represent a customer’s latest purchase, number of purchases, and average purchase amount can be derived from transaction records. Marcus (1998) pointed out the possibility that frequency is positively correlated to total purchase amount and suggested using average purchases amount to reduce their co-linearity.
It is interesting to note that RFM can also be applied to segment products to reflect different degrees of their importance. Julander (1992) suggested that analyzing transaction data in groups of products helps to study customer’s shopping behaviors. In our method, we propose to apply RFM variables to segment market of products rather than customers.
Recency, frequency and monetary are transformed into a product’s last date of being purchased, number of being purchased, and average sales amount. The last date of being purchased for a product refers to customers’ preferences. If the date is far away from today, it means that product is out of fashion. The number of being purchased represents a product’s popularity whether this product sells well.
Average sales amount stands for significance of a product.
Step 3. Cluster Analysis and Initial Online Catalog
After purchasing data with respect to RFM variables are generated, the next step is to segment products based on their RFM values. One popular segmentation technique used in marketing field is cluster analysis. Traditional RFM analysis arbitrarily segment markets into equal segments regardless of the distribution of transaction data. Cluster analysis outperforms traditional RFM method by considering data distribution. Clusters are more representative of data characteristics that indicate different degrees of product importance. We adopt the two-stage clustering algorithm introduced by Punj and Stewart (1983) that determines number of clusters using Ward’s method in the first stage and applies K-means analysis to form clusters in the second stage.
Clusters obtained from the clustering algorithm can then be utilized in the online catalog organization. Our idea is to construct a linear hierarchy of which each level corresponds to a cluster of products. Clusters of products on the upper levels indicate more popularity to customers and more fit to customers’ needs. In this manner, it allows browsers to navigate desired products more efficiently.
Criteria to determine cluster priority in the hierarchy are addressed here. Generally speaking, upper levels of clusters should possess higher values of RFM variables than lower levels. In case of conflicts existing among the three variables, several research works have implicated their relative importance. Marcus (1998) eliminated recency (R) value to structure customer value matrix to gain strategic positioning on customers. Ha and Park (1998) made use of RM values to coordinate strategic positioning of corner clusters. Therefore, it is evident that monetary (M) value is the most important criterion to consider. Furthermore, Marcus strongly disapproved of R among RFM variables while the purpose of Ha and Park’s work was to utilize RM values to facilitate RFM
coordination. In conclusion, we consider frequency (F) as the second important criterion with recency (R) as the least one.
Step 4. Browsing Data Collection
In Stage II, we consider the customers’ changing preferences toward products. No once popular products can stand for champion forever. This is why marketers always conduct market surveys to catch up the trend. The purpose of Stage II is to dynamically adjust the online catalog. That is to say, products in different layers may be swapped due to their popularity changed. We, however, try to avoid performing cluster-based segmentation method all over again because it is time-consuming and costly.
To detect customers’ changing behavior in the WWW environment, the most effective way is to record customers’ browsing patterns. Access frequency to a web page may indicate its importance to browsers. As cited by Garofalakis et al. (1999), relative access (RA) is more meaningful to absolute access (AA). RA was defined as follows:
(Eq. 1)
where a is a function determined by d (depth of layers), n (number of pages on the same layer) and r (number of links toward this page).
This method defines a multiplier of AA to render RA without sound grounds. In our opinion, each factor that affects RA should be regarded as a criterion. Each criterion is then assigned a weight to indicate the relative importance among all criteria. We therefore define the equation for RA as follows:
(Eq. 2) where c1, c2, and c3 are weights and values for d,
n and AA are normalized into f1(d), f2(n) and
f3(AA). Note that factor r is trivial to consider in
our case because each page is linked from exclusively one source (its upper level).
To determine the weights, we employ the analytical hierarchy process (AHP) method proposed by Saaty (1980), which is popular in solving multiple criteria decision problems. The main idea to derive the weights is based on paired comparisons on a nine-level scale. Details
of how to use excel to perform AHP analysis could found in the book by Eppen et al. (1998). Step 5. Adaptive Online Catalog
Since weights of each criterion are determined, we could take the next step to dynamically adjust the online catalog based on the RA value derived. To use RA values, it is intuitive that products with higher RA values should be placed on higher layer of the web hierarchy. The algorithm for online catalog adaptation starts from creating the data structure of a linked list for products. This data structure allows sorting to perform easily for later usage. Initial online catalog is then stored within this structure. RA value for each product is calculated accordingly and a sorting mechanism is performed. Finally, sorted products are organized from the top product layer to the bottom layer and the adaptive online catalog is generated.
3. Conclusions
In this research, we propose a two-stage approach for the customer-centered online catalog designs. In the first stage, cluster-based market segmentation is employed to construct the web hierarchy with products in higher levels indicating more interesting to customers. The second stage is to dynamically adjust the hierarchy when customers’ preference indicated from browsing patterns is changed. It is desired that the online catalog organization designed using such an approach can receive more customers’ attentions and further contribute to the growth of retailing business on the web. The future work of this research is to conduct several experiments based on real cases to justify its feasibility in effective online catalog designs. 4. Self Evaluation
This research work is consistent with the original idea of the proposed project, that is, to design customer-centered online catalogs based on category management for retailing business. We successfully achieve our expected goals in
choosing appropriate market segmentation variables for cluster analysis and using proper clustering algorithm to create clusters that construct the web hierarchy. We also determine suitable factors of customers’ browsing data to consider in the web adaptation stage. We, however, need further work on performance evaluation of the new design due to the time limit. Nonetheless, the research so far does make sufficient contributions to researchers and practitioners. Overall, this work is suitable for journal publication.
5. References
Blattberg, R., Fox, E., and Purk, M. “Category
Management-Complete Set”. (Washington,
DC: Food Marketing Institute, 1995)。
Eppen, G. D., Gould, F. J., Schmidt, C. P., Moore, J. H., and Weatherford, L. R. “Introductory Management Science: Decision
Modeling with Spreadsheets”. 5th edition (Upper Saddle River, NJ; Prentice Hall, Inc., 1998)
Evans, M. B. “Guidelines for Web Design”. (1998) [Online] Available: http://response.restoration.noaa.gov/webmastr/th issite.html
Garofalakis, J.; Kappos; P., and Mourloukos, D. “Web site optimization using page popularity”. IEEE Internet Computing, Vol.3-4, 1999, pp. 22 –29.
Gerbert, P., Schneider, D., and Birch A. “The
Age of E-tail”. (Mill Street, Oxford; Capstone
Publishing Ltd., 1999)
Ha, S. H. and Park, S. C. “Application of
Data Mining Tools to Hotel Data Mart on the Internet for Database Marketing”. Expert
Systems with Applications, Vol.15, 1998, pp. 1-31.
Huizingh, E. K.R.E. “The Content and Design
of Web Sites: An Empirical Study ” .
Information & Management, Vol.37, 2000, pp. 123-134.
Julander, C-R. “Basket Analysis: A New Way
of Analyzing Scanner Data”. International
Journal of Retail and Distribution Management, Vol. 20, No. 7, 1992, pp. 10-18.
Kalakota, R. and Whinston, A. B. ”Electronic
Commerce: A Manager’s Guide”. (Reading,
Massachusetts; Addison-Wesley, 1997), pp. 217-250.
Markus, C. “A Practical yet Meaningful
Approach to Customer Segmentation”.
Journal of Consumer Marketing, Vol.15, 1998, pp. 494-504.
Newell, F. “The New Rules of Marketing:
How to Use One-to-one relationship marketing to be the leader in your industry”.
(New York; McGraw-Hill Companies, Inc., 1997)
Punj, G. and Stewart, D. W. “Cluster Analysis
in Marketing Research: Review and Suggestions for Application”. Journal of
Marketing Research, Vol.20, 1983, pp. 134-148.
Saarenvirta, G. “Data Mining to Improve
Profitability”. CMA Management, Vol.72,
1998, pp. 8-12.
Saaty, T. L. “The Analytical Hierarchy
Process: Planning, Priority, Setting, Resource Allocation ”. (New York; London:
