S. YAJING et al.: CUSTOMER VALUE HIERARCHY

CUSTOMER VALUE HIERARCHY BASED CUSTOMER DEMAND ANALYSIS IN PERSONALIZED SERVICE RECOMMENDER SYSTEM SI YAJING[1], SHU HUAYING[2], QI JIAYIN[2] [1]

[2]

Economics and Management Department, Beijing Material Institute 1 Fu He Street, Tongzhou District, Beijing, China, 101149 E-mail: [email protected] Economics and Managment School, Beijing University of Posts and Telecommunications 190, 10 Xi Tu Cheng Road, Haidian District, Beijing, China, 100876

Abstract: Recommender systems are powerful tools for promoting marketing in the mobile industry. An effective recommender system can help boost the mobile service provider’s marketing by finding potential customers and recommend customers to engage additional services that are not originally engaged. Based on the background of the mobile industry, this paper proposes the framework and process of a mobile service recommendation. The customer value hierarchy-based customer demand analysis is used. Firstly, a contour model of customer value hierarchy is obtained by investigation and specific interview; secondly, the significant attributes of customer value layers are screened out; then a customer demand discrimination model is built where the customer demand objective layer is the output of the model and the customer demand attribute layer is its input. A well-formed model can dynamically identify the customer demand objectives from their engagement history record; finally, a personalized product recommendation is made. This model is used for the analysis of mobile customer samples. The results of customer demand discrimination reflect its outcome with the correct percentile exceeding 80%. Compared to existing recommendation systems, the system can identify potential customers’ demands of unsought services/products. Furthermore, it is accurate and high in intelligence level. Keywords: Demand discrimination, knowledge capture, customer value hierarchy, recommendation systems

can also increase loyalty, for consumers will trust the mobile provider that make efficient recommendation.

1. INTRODUCTION With the development of the mobile telecommunication industry, overwhelming products/services are provided to the mobile customers, whose demand is diverse, multiplex and variable. During the interaction with customers, the mobile provider must normally make service suggestion in a short time to their customers and provide consumers with information to help them decide which products/services to engage. So the recommender systems available to mobile service providers are particularly useful.

Current common approaches for personalized recommendation systems are the content-based approach and collaborative filtering approach (Balabanovic and Shoham, 1997; Sarwar et al, 2000; Lawrence et al, 2001; Wu et al., 2001). Content-based systems provide recommendations by matching customer interests with product attributes. Collaborative systems utilize the overlap of preference ratings among customers for product recommendation.

As a fundamental step in most recommender system, the real-time and dynamic customer demand analysis technology is required by the mobile operators to efficiently and automatically respond to the customer demand. An effective recommender system can help boost the mobile service’s marketing in three ways (Schafer et al, 1999): (i) Finding potential customers. During the interaction with customers, the mobile service provider can convert browsers into buyers. The suggestion of provider will facilitate consumers to find products/services they wish to engage. (ii) Increasing Cross-sell. Recommender systems improve cross-sell by suggesting customers to engage additional services that are not originally engaged. (iii) Building Loyalty. Creating successful relationships between consumers

However, there are limitations of existing recommender systems. The content-based systems are confined to the contingency of customer interests, which results in homogeneous recommendation. Although collaborative systems have better capabilities than content-based systems in recommending heterogeneous products (e.g. beer and diaper example), they have a limitation for unsought services/products, which are new services or cold services that the consumer does not know about it at all or does not normally think of. Furthermore, both of these methods are based on data-driven analysis and on the assumption that similar customers make similar choices. However, recommendations based on individual customer’s purchase demand are seldom included in these approaches.

I.J. of SIMULATION Vol. 7 No 7

77

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

paper proposes the framework and process of a mobile service recommender system that identifies potential customers’ demands of unsought services/products by using customer demand analysis. We build a mobile customer demand analysis model and proposes ways to simulate customer value hierarchy and to capture customer demand knowledge. Firstly, a contour model of customer value hierarchy is obtained by investigation and specific interview; secondly, the significant attributes of customer value layers are screened out; then a customer demand discrimination model is built where the customer demand objective layer is the output of the model and the customer demand attribute layer is its input. A well-formed model can dynamically identify the customer demand objectives from their engagement history record; finally, personalized product recommendation based on customer value hierarchy is made. This model is used to analyze a sample of 122 mobile telecommunication customers. The results of customer demand discrimination reflect the outcome of the model with the correct percentile exceeding 80%. Compared to existing recommendation systems, the system can identify potential customers’ demands of unsought services/products. Furthermore, it is accurate and high in intelligence level.

Customer demand discrimination is a well-established methodology for the analysis of customer relationship management systems. The customer demand knowledge is descriptive information about customer preferences and consume behavior. However, in the actual marketing, not only the preference cannot be exactly defined by the customers, but also the preference can be erratic. Especially in the telecommunication industry, the customers’ demands are more variable and ambiguous because of various services and decreasing switching cost, etc. Furthermore, there are some factors that potentially impact the customer perceptive value, which are customer education background, market circumstance, customer emotion, etc (Boulton et al., 2000; Sharp, 1997). Existing service recommender methods are designed to identify the top engaging items based on the customer consume behaviors or to classify customers by using clustering analysis. They have a limitation for unsought services, which are new services or services that the consumer does not know about or does not normally think of. The complexity of customer demand discrimination is indicated in two aspects: firstly, a customer may belong to multiple sorts that are simply classified by demand attributes. Secondly, there are uncertain relationships between the customer demand attributes and consume decision-making. So the customer demand discrimination is a subject of customer classification under uncertain condition.

2. FRAMEWORK FOR PERSONALIZED SERVICE RECOMMENDER SYSTEM The supporting framework for the process of personalized service recommendation is presented in figure 1. The subsequent sections of this paper will explore it in detail.

Previous studies about customer demand discrimination mainly focus on these subjects: firstly, predicting customer preferences and repeat-purchase patterns by consume data analysis (Simpson et al., 2001; Shih et al., 2005); secondly, analyzing the antecedents and consequences of consume behavior and customer loyalty (Srinivasan et al., 2002; Inoue et al., 2003); thirdly, classifying customers by using clustering analysis (Wan et al., 2005). The shortcoming of this method is that it is subjective, not intellectually challenging, and involves a large amount of manual work.

data from EIS Investigations, specific interviews

Objective layer

Attribute layer

Customer demand model

Products/serv ices data

Reduction Fuzzy clustering Key attributes of customer demand

Consequence layer

Customer related data

Woodruff, Burns and Goodstein proposed the CVD (Customer Value Determination) and built the correlation among the customer demand attribute layer, the consequence layer, and the objective layer (Woodruff, 1997; Burns, 1990). This method proposes a way to identify customer demands based on their purchase attributes. However, the authors did not present technical tools to implement the CVD knowledge capture. According to the research work of Woodruff, the implementation requires very large numbers of questionnaires and customer interviews, as well as high expenditures in CVD.

Knowledge capture and update Recommendation

Modeling of customer demand discrimination Customer demand

Modeling Matching Engine

Unsought Products /services feature data

Figure 1: Framework for personal service recommendation Definition of customer value layer—According to Woodruff’s CVD theory, which suggested that customer demand hierarchy contains the objective layer, the

Based on the background of the mobile industry, this

I.J. of SIMULATION Vol. 7 No 7

Definition of customer value layer

78

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

consequence layer and the attribute layer, the mobile customer value hierarchy is defined.

Attribute layer—The Attribute layer specifies the usage of mobile services.

Reduction--This step is to find the significant attributes of customer value layers, which is a group of attribute layer variables that influence the customer demand objectives. That is solved by the fuzzy cluster analysis.

3.2 Determination of Mobile Customer Value Hierarchy Model In order to uncover the attribute-consequence-objective chain, an in-depth interview technique called “laddering” was developed [9]. Subsequently, Walker and Olson developed a paper-and-pencil version [12]. Laddering refers to a 2-stage process. (1) Elicite the salient criteria for products/services discrimination, which is feasibly achieved by direct specific interview. In this step, concrete attributes (e.g. price) and abstract attributes (e.g. efficiency) are identified. (2) Elicite the salient attributes (concrete or abstract) form the top to the bottom of the customer value layer, which will reveal the entire means-end structure (this step is called the laddering probes). This stage is achieved by continuous enquiring such questions as “How important is the service to you? And why?” The response of each customer value layer is used as the basis for further questioning. This iterative questioning is a means to “abstract the subject up to the top of customer value layer” until the objective layer is determined.

Modeling—A customer demand discrimination model is built. A well-formed model can dynamically identify the customer demand objectives from their engagement history record. This is achieved by adopting a methodology based on neural networks. Matching and recommendation--Based on the performances of product attributes, the degree of a product match to a customer’s goals can be estimated. Then the system sorts the products/services according to their matching degree and derives the top-N recommendations. 3. CUSTOMER VALUE HIERARCHY MODEL 3.1 Contour Model of Customer Value Hierarchy Woodruff enhanced the CVD indicating how customers consider products in a hierarchical structure. The customer value hierarchy is presented in figure 2.

Based on the complete chain of customer value layers, the next step in the procedure is to shift the layers from the individual perspective to the aggregate perspective of a group of customers. We can accomplish this step by using association methods to find the association rules among attributes of different layers or by cumulating the “connection” times of two adjacent layers’ attributes. Based on the mobile customer interview made in Zhang Jiajie, this paper constructs the mobile customer value hierarchy presented in figure 3. The factors of objective layer and attribute layer are defined as the variable a （i=1, 2…29）which is presented in table 1.

Objective layer Customer‘s goal and purpose

Consequence layer

Desired consequences in use situation

i

Attribute layer Desired products/services attributes and performances

Table 1: Mobile customer value hierarchy Figure 2: Customer value hierarchy

Objectives

Consequences

Attributes

Communicative Convenient short massage service Object (a26) communication, call waiting call diversion little secretary voice mail box U-net Business High quality, Object(a27) knight service, Routine service high standing, Ticket booking Uni-colour E convenient E- bank business Stock exchange Mobile purchase Recreational Identification, Color ring back tone Object(a28) Fashion, mobile ring pleasure, mobile picture Have fun E-game chat mobile movie

From the bottom of the customer value hierarchy, customers firstly consider the attributes and availability of products. At the second layer, customers begin to make expectations according to these attributes. At the top layer, customers form expectations about the realization of their aim. In this paper, the mobile customer value hierarchy consists of the customer demand objective layer, the consequence layer and the attribute layer. The objective layer-- The objective layer includes the ultimate motivations of customers engaging in mobile telecommunication services. Customers may have multiple motivations in the objective layer.

a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13 a14 a15 a16 a17 a18

Consequence layer—The Consequence layer represents the customer experience of mobile services.

I.J. of SIMULATION Vol. 7 No 7

79

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

Informational Object(a29)

knowledge, in time, information

Objectives

News service Weather info Travel info Finance info Physical news Entertainment info U-map

Consequences

recommendation approach is to uncover the relationship between products/services that they engaged (attributes layer) and the customers’ actual goals (objective layer), which enable the mobile providers to identify the customer’s goal from his/her consume history. So we reduce the customer value hierarchy to an attribute-objective map. Each attribute is connected directly to the objective if there is a path between them, while the consequences are ignored. A brief example is illustrated in figure 4.

a19 a20 a21 a22 a23 a24 a25

Attributes Short massage

O1

Call waiting Communicative Object

Convenient communication

O2

Call diversion

C1

Little secretary

C2

T3

Voice mail box U-net

A1

A2

Business Object

High quality

Ticket booking

knight service

Uni-colour E

high standing

E-bank

Convenient business

Stock exchange

C1

A1

Identification

Mobile ring

Fashion

Mobile picture

Pleasure

E-game

Have fun

Chat

A6

A2

A3

C2

A4

A5

A6

4.2 Significant Attributes Analysis of Customer Value Hierarchy The significant attributes of customer value hierarchy are the key attribute variables of the attribute layer which distinctly correlate to the objective layer. Because of the large numbers of mobile telecommunication products/services and the relatively small percentage of the mobile services/products engagement, the original data of customer value hierarchy is high dimensional sparse feature data. Therefore, this step mainly consists of reducing the data dimension for customer demand analysis. This paper adopts the fuzzy cluster analysis method to find the significant attribute and accomplish reduction.

News service Weather info Knowledge

Travel info

In time

Finance info

Information

A5

Figure 4: Attribute-objective map

Mobile movie

Informational Object

A4

Mobile purchase Color ring back tone

Recreational Object

A3

Flating

Routine service

Physical news U-map Entertainment info

4.2.1. The Principles of Significant Attributes Analysis Figure 3: Mobile customer value hierarchy According to the rough set theory, data of the customer value objective layer and attribute layer can be defined as S= (U, A, V, f). Here: U= {u1, u2,…, un}: the set of customers where n is the total number of customers. A= {a1, a2,…, am}: the set of variables of the objective layer and of the attribute layer. A = C ∪ D , where C is the characteristics set of the attribute layer, and D is the characteristics set of the objective layer. V is the set of the customer attribute parameters. The value of f (uj, ai) indicates the value of uj about ai.

4 MOBILE CUSTOMER DEMAND ANALYSE AND THE KNOWLEDGE CAPTURE 4.1 Constructing attribute-objective map Customer buying behavior is often depicted as purposeful and goal-oriented. Recommender systems intend to provide products satisfying each customer’s goal. Therefore, the core of our proposed

I.J. of SIMULATION Vol. 7 No 7

80

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

a i ( i = 6 , 7 ,8 ,9 ,10 ,11 ,12 ) is a ij ( j = 0,1) . The matrix of the

The significant attributes analysis is solved by fuzzy clustering. The process of the analysis includes the following steps:

numerical

rij =

∑ (a k =1

m

m

ik

a jk )

( ∑ a )( ∑ a i =1

2 ik

k =1

2 jk

)

expressed

as:

Secondly, calculate the fuzzy similarity matrix R. The result is expressed in equation (2): 1

Step2. Calculate the fuzzy similarity matrix R. As shown in equation (1) the research adopts the cosine distance measure as the method of similarity measurement of the study objects. 2

is T

Step1. Partition customer set A into D and C. Consider the numerical character of attribute ai in attributes set C, and represent attribute ai as a（j=1,2,…,k） . Here k is the ij number of incoordinate value of attribute ai.

m

characters

⎡42 11 50 44 48 49 49⎤ ⎢ 8 39 0 6 2 1 1 ⎥ ⎣ ⎦

0.447

0.447 0.982 0.999 0.989 0.986 0.986 1

0.271 0.399 0.311 0.291 0.291

0.982 0.271

[R] = 0.999

(1)

Step3. Calculate the fuzzy transitive closure t(R) of the fuzzy correlation matrix R. Use the cluster method to analyze t(R) with intercept λ and determine the significant attributes set.

1

0.991 0.999

0.399 0.991

1

1

1

0.996 0.993 0.993

0.989 0.311 0.999 0.996

1

1

1

0.986 0.291 0.986 0.291

1 1

1 1

1 1

1 1

0.993 0.993

(2)

Thirdly, calculate the fuzzy transitive closure t(R) of the fuzzy correlation matrix R with the square method (He and Li, 1999). If the fuzzy correlation matrix can be expressed as R = ( rij ) n× n , then R o R = (tij ) n×n ,

4.2.2 The Process of Data Analysis

n

t ij = max(min(rik , rkj )) . k =1

The investigation gave 150 questionnaires out to the individual mobile customers in Zhang Jiajie. 122 effective sheets of questionnaire were retrieved. The rate of response efficiency is 81.3%.

If [ R ]2 o [ R ]2 = [ R ]2 , then the fuzzy transitive closure [t ( R )] = [ R ] 2 . The result of this calculation and the significant attributes of four objective layers are presented in table 2. k

The questionnaire contains two parts: (1) questions about the importance of the customer objects in the objective layer. Five scores are adopted to fill out the questions: one indicates the least important and five indicates the most important. (2) Questions on whether the customers have engaged the services of the attribute layer. The questionnaire enumerates products/services of the attribute layers corresponding to a given object of the objective layer. Then we transform the answer into data: 1 means the customer has engaged the products/services and 0 means the customer hasn’t engaged the products/services.

objective layer

Recreational Object Informational Object

0.95

C

Cc ={a1,a2,a3,a4 ,a6,a7 ,a9,a13,a14,a15,a16,a19,a20}

4.3 Mobile Modeling

Customer

Demand

Discrimination

The customer demand hierarchy is applied to the analysis of customer demand. This is achieved by adopting a methodology based on BP-neural networks.

Given the set of objective layer D={ a 27 }，the set of attribute layer is C={ a 6 , a7 , a8 , a9 , a10 , a11 , a12 }，and the

I.J. of SIMULATION Vol. 7 No 7

0.99

attributes cluster Significant attributes {a1}{a2,a3,a4} a1,a2,a3,a4 {a5,a7} {a6,a9},{a7}, a6,a7,a9 {a8,a10,a11,a12} {a13,a14}{a15} a13,a14,a15 {a16}{a17,a18} a16 a19,a20 {a19},{a20} {a21,a22,a23,a24 a25}

From a total 25 products/services, 13 products/services were found to have distinct correlation with the customer demand objectives. This conclusion can help the operators to implement powerful marketing strategies. Therefore, the set of significant C attributes can be expressed as:

~~ , V = { 1 , 0 ,5, 4,3, 2,1} U = {u1 , u 2 , …，u 50 } , ~~~~~~~ ~ ~ ~ ~ ~ ~ …… ~~~~~~ , u1 = (1,1,0,0,0,1,0,5),u2 = (0,1,0,0,0,0,1), u50 = ( 0,1, 0, 0, 0, 0,1) ~ ，…… ~, ~, f (u1 , a 27 ) = 5 f (u1 , a6 ) = 1 f (u1 , a7 ) = 1 f (u1 , a8 ) = 0

of

λ

Communicative 0.91 Object Business Object 0.99

Firstly, data of the customer business demand objective layer and attribute layer can be defined as: , S = (U, A, V, f ) A = {a6 , a7 , a8 , a9 , a10 , a11 , a12 ,...a27 }

representation

k

Table 2: Significant attributes of attribute layer

In order to be less costly and easily applied, 50 questionnaires are chosen to form the analysis sample. By taking the business object related attribute layer as an example, the process of finding the significant attributes of the attribute layer is illustrated.

numerical

k

k

attribute

81

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

4.3.1 The Principle of Modeling of mobile customer demand discrimination A customer demand discrimination model is built where the customer demand objective layer ( ) is the output of the model k ∈ D U = [u 1 k , u 2 k ,... u nk ] and significant attributes of the attribute layer U = [u 1 j , u 2 j ,... u nj ] ( j ∈ C C ) are the input.

Hidden layer

Oj

Wjk

…

…

…

…

….

Wij

unk

Ok

Figure 5: The model of BP-neutral network for customer demand discrimination 4.3.2 The Process of Data Analysis The process of data analysis is accomplished by using Clementine 8.0. This paper uses 61 customers as the training samples set and 61 customers as the contrastive samples set. The model is trained with α = 0.9 and η = 0.01 , and finally when the number of hidden layer is 5, the best training result is achieved. The accurate percentage of forecast is 80.517%.

W =

− 0 . 04

− 0 . 63

0 . 51

0 . 34

0 . 67

− 0 . 73

0.32

- 0.22

- 0.72

- 0.82 0 . 26

− 1 . 18 − 0 . 94

0 . 06 0 . 68

0 . 44 − 0 . 20

− 1 . 02 − 0 . 70

− 0 . 37

− 0 . 06

0 . 21

− 0 . 03

− 0 . 70

− 0 . 33

− 0 . 26

− 0 . 35

− 0 . 21

− 0 . 26

0 . 02

− 0 . 62

− 0 . 56

0 . 16

0 . 39

0 . 14

0 . 86

− 0 . 43

− 0 . 58

− 0 . 15

− 0 . 02

0 . 06

− 0 . 70

− 0 . 48

− 0 . 29

0 . 35 − 0 . 79

− 0 . 45 − 0 . 08

0 . 27 − 0 . 73

− 0 . 79 0 . 04

0 . 59 − 0 . 35

1 . 26

− 0 . 42

0 . 16

− 1 . 02

− 0 . 19

0 . 53

− 0 .1

0 . 15

− 1 . 48

0 . 75

0.724 − 0.195

− 0.424 − 1.577

0.509 − 0.247

0.992

− 0.537

− 0.395

1.527

5. MATCHING AND RECOMMENDATION In this step, the recommender system decides the characteristics of the product or service that could fit the customer’s purchase objectives as identified in the prior steps. Here, we assume that the performances of each products/services that satisfy the consumer’s objectives (such as business object, etc.) have been scored by product experts. Then the degree to which a product

And the weight matrix from hidden layer to output layer is expressed as:

I.J. of SIMULATION Vol. 7 No 7

T = 0.852 0.976

It can be indicated that the model can discriminate the customer demand with a high percentage of accuracy. The sort order of customer demand objects that were discriminated by the model are in accordance with the customer’s actual objects’ sort order, even though the conclusion is not absolutely correct. That is because the conclusion is influenced by the dividing line of customer demand. Furthermore, the model can indicate a mensurable conclusion based on customer motivation that has the advantage of stability and reliability. The model can help the mobile operators carry out a “one to one” customer service strategy.

The weight matrix from input layer to hidden layer W is described as: − 0 .9

− 0.459

customer a26 a27 a28 a29 customer demand object 1 Actual 5.0 2.0 2.0 3.0 Communication object, demand Information object Analytical 4.9 3.1 2.6 3.8 Communication object, conclusion Business object Information object 2 Actual 5.0 3.0 2.0 4.0 Communication object, demand Business object, Information object Analytical 4.9 3. 2.9 3.8 Communication object, conclusion Business object, Information object 3 Actual 5.0 3.0 3.0 5.0 Communication object, demand Business object, recreation object, information object Analytical 4.9 2.6 3.1 3.9 Communication object, conclusion recreation object, information object

u2k

unj

0.973

0.363

Table 3: Comparison between analytical conclusions and actual demands

u1k

u2j

0.921

− 1.252

A well-formed model can dynamically identify the classification of customer demand objectives from their demand attributes. 3 customers are randomly selected from the sample and the comparison between the analytical conclusions and the actual demands is presented in table 3. We assume 3 as the dividing line.

Output layer

u1j

0.393

0.145

4.3 Customer Demand Knowledge Capture

The structure of the neural network is shown in figure 5. wij is the generic weight from the input layer to the hidden layer; wjk is the generic weight from the hidden layer to the output layer. The number nodes in the hidden layer is set between 3 and 20. Input layer

0.918

82

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

Multi-Attribute Learning Mechanisms”. In the International Conference Integration of Knowledge Intensive Multi-Agent Systems ( KIMAS '03). IEEE, 634-639 Lawrence, R. D., Almasi, G. S., Kotlyar, V., Viveros, M. S., Duri, S. S., 2001, “Personalization of supermarket product recommendations”, Data Mining and Knowledge Discovery, 5(1), pp.11–32. Sarwar, B., Karypis, G., Konstan, J., Riedl, J., 2000, “Analysis of recommendation algorithms for e-Commerce”, In Proceedings of the Second ACM Conference on Electronic Commerce, Minneapolis, Minnesota, USA ,Pp. 158–167. Schafer J.B., Konstan J.A. and Riedl. J., 1999, “Recommender systems in e-commerce”, In Proceedings of the First ACM Conference on Electronic Commerce, Denver, Pp.158-166. Sharp, B. and A. Sharp, 1997. “Loyalty programs and their impact on repeat-purchase loyalty patterns”. International Journal of Research in Marketing, No.14 , 473-486. Shih, Y.Y.; D.R. Liu; Hybrid. 2005. “Recommendation Approaches: Collaborative Filtering Via Valuable Content Information.” In Proceedings of the 38th Hawaii International Conference on System Sciences, 217 Simpson. P.M; J.A. Siguaw; T.L. Baker. 2001. “Value Creation Supplier Behaviors and Their Impact on Reseller-Perceived Value”. Industrial Marketing Management, No.30, 119-134 Srinivasan, S.S.; R. Andeson; and K. Ponnavolu. 2002, “Customer Loyalty in E-commerce: an Exploration of Its Antecedents and Consequences.” Journal of Retailing, No.78, 41-50 Wan, Y.; Y. Qian; J. Li; and L. Cui. 2005. “The Analysis of Customer Value Dimensions’ Difference of the Bank’s Interacting Channel.” In The Second IEEE Conference on Service Systems and Service Management，Institute of Electrical and Electronice Engineer Inc， 193-198 Woodruff, R.B. 1997. “Customer Value: The next source for competitive advantage.” Academy of Marketing science. Spring,Vol.25, Issue 2, 139 Wu, Y. H., Chen, Y. C., Chen, A., 2001, “Enabling personalized recommendation on the web based on user interests and behaviors”, In Proceedings of the 11th International Workshop on Research Issues in Data Engineering, Heidelberg, Germany , Pp. 17–24.

satisfies a customer’s objectives can be estimated using the Pearson correlation, which is presented in equation (3). After all products have been evaluated, the system sorts the products according to their satisfaction rate and recommends the target customers for the top-N services/products. corrab =

∑ (r − r )(r − r ) ∑ (r − r ) ∑ (r − r ) i

ai

a

bi

i

ai

a

i

(3)

b

2

bi

2

b

6. CONCLUSIONS Based on the background of the mobile industry, this paper proposes the framework and process of a mobile service recommender system that identifies potential customers’ demands of unsought services/products by using customer demand analysis. This paper proposes a mobile customer demand analysis model and develops a mobile customer value hierarchy to capture customer demand knowledge. A well-formed model could identify the customer demand objectives dynamically from their engagement record; then a personalized product recommendation based on the customer value hierarchy is made. The method is used to analyze a sample of 122 mobile telecommunication customers. We make a contour model of mobile customer value layer, and screen out 13 key variables of attribute layer. The results of customer demand discrimination reflect the outcomes of the survey with the correct percentile exceeding 80%. Compared to customer clustering analysis, this methodology is accurate and high in intelligence level. ACKNOWLEDGEMENT The research presented in this paper is funded by the National Natural Science Foundation of China (Project No.70371056) and the Key Laboratory of Information Management & Information Economy of Ministry of Education of the People’s Republic of China. REFERENCES Balabanovic, M., Shoham, Y., 1997, “Fab: content-based, collaborative recommendation”, Communications of the ACM, 40(3), Pp.66–72. Boulton, R., B. Libert, and S.M. Samek. 2000. Cracking the Value code: How Successful Businesses Are Creating Wealth in the New Economy. New York: Harper Collins Publishers Inc., 288 Burs M., R. Jane, and B. Woodruff. 1990. “Value: An Integrative Perspective”. In Proceedings of the Society for Consumer Psychology. Washington ： American Psychological Association, 59-64 He Qing and Li Hongxing. 1999. “Fuzzy Equivalent Matrices in Fuzzy Clustering”. Systems Engineering-Theory & Practice, No.4, 9-12 Inoue, A., S. Takahashi; K. Nishimatsu; and H. Kawano. 2003. “Service Demand Analysis Using

I.J. of SIMULATION Vol. 7 No 7

AUTHOR BIOGRAPHIES SI YAJING became a research student of Beijing University of Posts and Telecommunications in 2003, where she started her PhD studies under the supervision of Professor Shu Huaying. Now she is a lecturer in the Economics and Management Department at the Beijing Material Institute. Her current research interests include mathematical modelling and simulation relating to

83

ISSN 1473-804x online, 1473-8031 print

S. YAJING et al.: CUSTOMER VALUE HIERARCHY

customer behavior and customer relationship management. Her e-mail address is: [email protected].

SHU HUAYING is a professor in School of Economics and Management, Beijing University of Posts and Telecommunications, where he is now leading a large research group in the Key Laboratory of Information Management & Information Economy of Ministry of Education of the People’s Republic of China. His main research fields are information economics, information system and decision support system of communications enterprises, communications operation theory etc.

QI JIAYIN is an associate professor in the School of Economics and Management, Beijing University of Posts and Telecommunications. Her main research fields are customer relationship management, communications operation theory and application etc.

I.J. of SIMULATION Vol. 7 No 7

84

ISSN 1473-804x online, 1473-8031 print