Intl. Trans. in Op. Res. 17 (2010) 317331 DOI: 10.1111/j.1475-3995.2009.00712.

INTERNATIONAL TRANSACTIONS IN OPERATIONAL RESEARCH

Product selection in Internet business: a fuzzy approach

B. K. Mohantya and B. Aounib
b

Indian Institute of Management, O Sitapur Road, Lucknow 226 013, India, Decision Aid Research Group, School of Commerce and Administration, Faculty of Management, Laurentian University, Sudbury, ON, Canada P3E 2C6 Received 5 June 2007; received in revised form 11 August 2008; accepted 9 March 2009

Abstract In this paper, we propose a methodology which helps customers buy products through the Internet. This procedure takes into account the customers level of desire in the product attributes, which are normally fuzzy, or in linguistically dened terms. The concept of fuzzy number will be used to measure the degree of similarities of the available products to that of the customers requirements. The degrees of similarities so obtained over all the attributes give rise to the fuzzy probabilities and hence the fuzzy expected values of availing a product on the Internet as per the customers requirement. Attribute-wise the fuzzy expected values are compared with those of the available products on the Internet and the product that is closest to the customers preference is selected as the best product. The multi-attribute weighted average method is used here to evaluate and hence to select the best product.
Keywords: Internet business; multiple product attribute; customers preference; fuzzy probability; fuzzy numbers

1. Introduction Over the last decade many companies have focussed on using the Internet to commercialise their products and services. Some examples are available in Alter (2004). The development of advanced information technology, online business strategies and Internet tools have established more convenient ways to commercialise their business transactions eciently. Many organisations consider this a major paradigm shift from the traditional concept of across-the-table marketing to the world of easy and eortless business processing. Internet business is considered to be a technological driving force in the 21st century. For example, in online car purchasing, a customer in India can view cars which are available in the United States. If a car is to his/her liking, he/she can order the car over the Internet while sitting in India. In a similar way, a company can broadcast design or model changes to its customers across the globe by advertising on the Internet. Almost all multinational companies have incorporated IT-enabled systems including Internet business strategies in order to establish direct channels for organisational business
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies Published by Blackwell Publishing, 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main St, Malden, MA 02148, USA.

318

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

transactions, remote information retrieval, processing and management for agile marketing, and electronic trading. Online business has given a boost to companies in terms of responding rapidly and accommodating themselves to market changes. In any business environment (online or traditional), a company presents its products and features to their customers. A customer uses his product knowledge and the product information provided to compare dierent products and product features, and select the best product. This decision is made based upon information from marketing sources, sales personnel, brand information, product features, and the customers own nancial budget. Normally a customer gathers all this information from a variety of sources and combines it implicitly before making their decision. However, obtaining pertinent, consistent and up-to-date information is a complex and time-consuming process. With the help of the Internet, companies can provide various aspects of their business information to their customers. Internet tools aim to improve the way that this information is gathered, managed, distributed and presented to customers. However, for the successful implementation of the business process, we need to address two main diculties. Firstly, the customers product requirements and preferences are imprecise or fuzzy in nature, and secondly, product assessments by customers are based on multiple attributes. In practice, a customers product requirements are usually represented with multiple attributes that can be conicting, non-commensurable and fuzzy. In the traditional method of shopping, a sales assistant collects the customers fuzzy information about the desired product attributes and suggests an appropriate product. However, this is next to impossible to achieve in the Internet market as there is no sales assistant available for the job. Moreover, attribute-wise fuzzy inputs to e-business systems will show an input error. Although some of the e-commerce systems are based on the concepts of fuzzy logic, the majority of present day online business systems require precise data, non-ambiguous knowledge and exact information. Additionally, many of them do not follow any objectively dened procedure of obtaining the similarities between the customers desired attribute levels and the available products realisation level (Cao and Li, 2007; Lee and Widmeyer, 1986; Mohanty and Bhasker, 2005; Ryu, 1999). Non-fuzzy data assumptions in Internet business will be dicult to implement because imprecision is inherent in the customers mind and also in almost all business problem domains. Ignoring or approximating the customers fuzzy requirements and using existing precise models will not only show an unpredictable result but also lead to a business failure because of the misrepresentation of the customers real opinion in terms of their product requirements. For example in buying a shirt, a customer does not normally consider the precise percentage of cotton, the style and softness of the fabric and the durability of the shirt. However, a customer fuzzily expresses his requirements on these attributes. What is needed here is an e-business system which incorporates fuzzy representations of both the customers requirements and the products available and leads to a matching product, as in traditional showrooms. Our paper aims to develop such a model to address the issue of association of the customers fuzzy requirements and fuzzily dened product realisations on the Internet market. Another problem normally encountered in an online business process is due to the customers interest for non-digital product features. For example, a cars digital attributes include price, maintenance cost, and mileage, and can be easily expressed over the Internet. However, nondigital attributes are more dicult. In the same car problem, the non-digital attributes include seat comfort, surface polishing, and ease of driving. These attributes cannot be assessed over the
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

319

Internet without experiencing them physically. Digitisation of non-digital attributes is beyond the scope of this paper. The proposed procedure only works on the product attribute information at hand and does not distinguish between digital and non-digital attributes. An additional diculty in handling Internet business is due to the fact that the customers product assessment is of multi-dimensional nature. That is, very often a customer would like to examine and analyse multiple aspects of the product before making their decision. For example, in the car-purchasing problem, the multiple attributes may be cost, design, colour, fuel consumption, resale value, engine type, etc. In this case, the customer has to select a car that oers the best match to his/her requirements. In the traditional market, a customer can discuss with a sales person and nally select a product. In this situation, a customer generally makes trade-os or compromises when convinced by the sales personnel that this is the only possible way of doing business in the given market prole. However, in the Internet market, there is no discussion and also it is dicult for a customer to express his views about the product attributes. This becomes more so, when product attributes are fuzzy, conicting and non-commensurable in nature. By considering the concepts of multiple-attribute decision-making, we have highlighted this case in the Internet market. The multi-attribute approach is one of the many models that have been developed to deal with the decision-making situations where several factors such as objectives, constraints, or attributes have to be considered simultaneously (Aouni and Kettani, 2001; Carlsson and Fuller, 1996; Martel and Aouni, 1998; Mohanty, 1998; Mohanty and Vijayaraghvan, 1995; Triantaphyllou, 2000). The proposed procedure could be qualied as a decision aid tool since it enables the customer to get more involved in the decision-making process. Personal shopping agents like Jango and DealTime (by http://www.dealtime.com) are early eorts to provide agent based shopping support to customers. These agents collect the price and desirable product attributes from the Internet for a specied product. Once the customer selects a product, these agents assist in identifying the best available deal. However, the buyers problem is in identifying a desirable product in the vast Internet market. Another system like decision guide by http://www.ActivityBuyersGuide.com assists the consumer by giving the best available product based on the product attribute requirements provided by the customer. This agent provides a list of matching products after obtaining the attribute-wise product requirements from the customer. This has its own limitations as a customer cannot dene his/her desired level of satisfaction in each product attribute. Further, in decision guide a customer is unable to state the importance of the product attributes or specify a range for the product attribute levels from minimum to maximum. These inputs enable the system to list appropriate products and present them to the customers. However, there might be a set of products having slight deviations from the above range but still oering an overall desirability to the customer, yet they are not included in the list. Thus, the satisfaction level that the customer is likely to derive from the deviated products becomes hidden and thereby makes it dicult to observe the products in totality and to assess the superiority of the listed products to that of the deviated ones. This comparison is necessary because at times a customer may like to give concession and compromise in a certain attribute and choose a product from the deviated list. The decision guide does not account for the customers compromising attitude. In Internet business, we have not come across any paper which represents product attributes in fuzzy sets and at the same time compares them to the customers fuzzily dened opinions to nd a
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

320

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

matching product. A shopping program given in Lee and Widmeyer (1986) aims at the selection of a desirable product through the Internet. In the situation of non-availability of the product the program suggests a product that is closest to the requested product in the taxonomy hierarchy as an alternative. The real shortcoming in this approach is that the search is conducted in a single generic product hierarchy. However, in real problems, the product selections are based on multiple attributes. For example, if a customer wants to select a car based on its price only, the cars can be arranged in an ascending order based on their prices. Similarly we can have another ordering of the cars based on its mileage attribute. However, if a customer prefers to order cars based on its price as well as mileage (with dierent degrees of importance) the ordering will be dierent. The improvement from a single hierarchy to a multiple one and the preference hierarchy therein is given in Ryu (1999). Electronic shopping support given in Ryu (1999) classies products based on product attributes, which are in multiple numbers. Depending on the attribute specications by the customer this procedure searches the Internet for the desired product. If a product is available with the prescribed attribute specications the customer will be satised. In case of nonavailability the procedure chooses the next available product which is closest to the target product. Closeness of the product is measured and compared through product attribute values. The author has introduced attribute exibility in order to classify dierent products with dierent attribute values into the same preference class. The main drawback in the methodology in Ryu (1999) is that exibility values are chosen subjectively. This idea was extended in Mohanty and Bhasker (2005) by deriving the customers exibility values objectively. However, this work also concentrates on the hierarchical satisfaction of the customers requirements and lacks simultaneous satisfaction of multiple attributes of the product. Although the procedure in Mohanty and Bhasker (2005) hierarchically classies the products according to the customers preferences, no similarity between the customers choice and the products availability is measured while making the classications. By representing the product attributes in fuzzy numbers, our paper also classies the products in a preference hierarchy based on multiple product attributes. Some other papers are available in the literature. In Cao and Li (2007), the consumer species the product preferences in each attribute and the system provides the products according to his/ her personal choices. However, the procedure of recommendation does not give the similarities between the buyers desired attribute levels and the products attainability in the market forum. In Miao et al. (2007), personalised recommendations are made in an interactive way. Although the process of interaction somewhat represents an assessment of similarity between the customers choices and the products availability, it does not indicate a concrete procedure for obtaining similarity measures. In Cheng et al. (2006), automated intelligent agents of the trading partners negotiate on several issues with the aim to arrive at a consensus business pact. Although the work (Cheng et al., 2006) maintains exibility in the agents strategies during the process of negotiation, it does not account for similarities among the agents views before arriving at a business deal. This paper is organised as follows: Section 2 is a summary of the procedure introduced in the paper. In Section 3, we give the fuzzy representation of product attributes as perceived by customers. Also in this section, we give the companys fuzzy viewpoint about product attributes. Section 4 gives the methodology of obtaining similarities between the attributes that customers require and those of the available products. Section 5 is devoted to problem formulation and the
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

321

solution procedure. In Section 6, we illustrate a numerical example to highlight the procedure. Finally, in Section 7 we present our conclusions.

2. Brief procedure Our paper has examined the above shortcomings and presents the following to overcome them. Product attribute values are incorporated in the model by taking the fuzzy requirements of the customer as inputs. (a) Consideration and implementation of multiple product attributes is addressed in the paper. (b) Each attribute of the available products is compared with the customers choices to obtain the degree of similarity between the available products and customers requirements. Let a customer require k fuzzily dened attributes of a product. Denote the customers desired preference levels of these attributes Ej ( j 5 1, 2, . . ., k) as LR fuzzy numbers E E1 ; E2 ; . . . ; Ek : Assume that there are m products available on the Internet. Let their attribute levels be specied as Bi Bi1 ; Bi2 . . . Bik for i 1; 2; . . . ; m: Bij represents the jth attribute level of the ith product. Bijs (8 i and j) are assumed here to be fuzzy numbers. Details about fuzzy numbers can be found in Dubois and Prade (1978). Initially take an attribute level Ej (say the jth attribute) as prescribed by the customer and match this attribute to the corresponding jth attribute Bij of all the products. The degree of similarity TEj|Bij between Ej and Bij (8i) can be obtained following the procedure given in Yager (1984). As per this procedure TEj|Bij is also a fuzzy number. The information on these degrees of similarities, along with the fuzzy number operations given in Yager (1984), helps the customer predict the probability and hence the expectation of availing a particular attribute level of Ej in the given market prole. Let the fuzzy number representation of the above expected values be ExpEj ExpEj l ; ExpEj n ; ExpEj r for j 1; 2; . . . ; k: 1 If Ii ( j) represents the matching or similarity degree between Bij and (Exp (Ej )), we can have the satisfaction level for the ith product as X Pi wj Ii j; for i 1; 2; . . . ; m; 2 where wj are the weights attached to the attributes Ej. The weights used here can be estimated by using the procedure given in Mohanty (1998). After calculating all P(i) values, we have the best product corresponding to the maximum P(i) value.

3. Fuzzy representation of the product attributes Most of the decision making in the real world takes place in an environment in which goals, constraints, and possible consequent actions are not known precisely. Fuzzy logic deals
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

322

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

quantitatively with the qualitatively dened items in the decision-making process. In order to make business decisions more customer focussed, a company needs to analyse a customers fuzzily dened requirements to nd matching products for the customer. To do this, the company needs to maintain exibility in its products to satisfy diverse groups of customers. For example, in a car purchasing problem customers requirements for the attribute cost will dier from one customer to another, and might be in fuzzy terms such as around US$20,000, close to US$25,000, approximately US$22,000, or more or less US$23,000, etc. In the scenario of diverse customer price requirements, a salesperson may like to promote the companys products in various ways to dierent groups of customers. That is, if the price of a car as per the companys judgement is US$21,000 (say), the company may claim that it is the most suitable car for a particular group of customers whose desired price level is US$25,000, of course with a satisfaction level or a matching degree. Similarly the company can derive dierent matching degrees for dierent groups of customers with respect to the cost attribute. That is, as per the companys appraisal, each product can be claimed as a suitable product for each customer group with a matching degree or a satisfaction level. In other words, the company, on its own, may look into the desires of the various customer groups in terms of the product attributes and then dene the ranges for attribute deviations. These deviations from the actual attribute values help companies make their products more exible and customer focussed in the Internet market. Note that the customers desires and the companys assessments in terms of dening attribute exibilities can be represented through fuzzy numbers. Example 3.1. In any product purchase, a customer normally assesses a product through multiple numbers of attributes and species their requirements in fuzzy terms. In the car purchasing problem, a customer may opt for attributes such as cost, re-sale value, mileage, comfort, maintenance cost, etc. By choosing the attribute cost, the customers fuzzily dened views might be dened in the following ways.     (A, mA): Cost should be around US$20,000, (B, mB): Cost should be approximately US$22,000, (C, mC): Cost should be closer to US$23,000, (D, mD): Cost should be more or less US$25,000.

In the above the italic words are fuzzy terms. They are denoted by the fuzzy sets A, B, C, and D. Without loss of generality these fuzzy terms can be represented as fuzzy numbers as shown below. A; mA 15; 20; 25; B; mB 18; 22; 24; C; mC 17; 23; 25 and D; mD 22; 25; 27; where mA, mB, mC and mD represent the membership functions of the fuzzy numbers A, B, C and D, respectively. These are graphically shown in Fig. 1. In Fig. 1, the horizontal axis represents car prices and the vertical axis represents the customers satisfaction level (membership function of the fuzzy numbers) corresponding to a price. Taking the fuzzy number around US$20,000, we can say that a buyer is fully satised (with membership value one) if the car price is US$20,000 and his/her satisfaction level gradually decreases when the
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

1.2 1 0.8 0.6 0.4 0.2 0 10 15 20 Price 25 30 Satisfaction Level

323

A B C D

Fig. 1. Buyers requirements of prices in fuzzy terms.

1.2 Satisfaction Level 1 0.8 0.6 0.4 0.2 0 0 5 10 15 Price 20 25 30

Fig. 2. Companys price assessment in fuzzy terms.

car price deviates from US$20,000 and becomes zero when the price is either below US$15,000 or above US$25,000. If a company has a car with price US$23,000, the companys analyst may present the cost of the car for all the customer groups in a single fuzzy term (fuzzy number) as ( Price, mPrice) 5 (15, 23, 25) with membership function mPRICE. Graphically this is shown in Fig. 2. As before the horizontal axis represents the companys own assessed exible price for dierent customer groups and the vertical axis represents the expected satisfaction level of the company (membership values). Note that in Fig. 2 the company has fuzzi-ed the attribute cost after scanning the market demand and the customers requirements on the cost attribute. The companys fuzzily dened attribute is meant for all the customer groups with variant price aspirations. In fact this is assessed subjectively by the companys experts without following an analytical or logical procedure.

4. Similarity measure In this section, we measure the degrees of similarity between the attribute levels of the available products on the Internet and the desired attribute levels of the customer. As the customers interest is based on multiple product attributes, which are generally conicting, noncommensurable and fuzzy in nature, it is very dicult to nd similarities amongst the attributes
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

324

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

when choosing a product. However, in our paper we address this problem and obtain similarities by representing product attributes in fuzzy numbers and following the technique given in Yager (1984). The procedure is given below. Let B1, B2, . . ., Bm be m products available on the Internet. Let these products be assessed in terms of the satisfaction levels of k attributes. We denote the jth attribute of the ith product as Bij so the ith product in terms of its attributes is Bi Bi1 ; Bi2 . . . Bik : Similarly we can have the customers desired levels ( preference levels) of the product attributes as E E1 ; E2 ; . . . ; Ek : Ej ( j 5 1, 2, . . ., k) represents the customers desired level in the jth attribute. We have assumed here that the product attributes are in the form of LR fuzzy numbers (Dubois and Prade, 1978; Mohanty and Bhasker, 2005). Thus, we have Ej Ejl ; Ejn ; Ejr for j 1; 2; . . . ; k and Bij Bijl ; Bijn ; Bijr

for i 1; 2; . . . ; m and j 1; 2; . . . ; k: In order to determine the satisfaction levels of the customers regarding the available products, we need to calculate the similarities between the attribute levels Ej and the product attributes Bij (for i 5 1, 2, . . ., m and j 5 1, 2, . . ., k). Mathematically the attribute-wise comparisons for the ith product to that of the customers requirements are given as: Bi1 $ E1 ; Bi2 $ E2 ; Bik $ Ek : The symbol $ is used here to represent the similarity. By generalising equation (3) we have Bij $ Ej for i 1; 2; . . . ; m and j 1; 2; . . . ; k: 4 Following the method given in Yager (1984), we derive the degree of similarity between Ej and Bij: TEjjBij i Max: Bij x; Ej x i: Here, TEj|Bij represents the extent to which the customers satisfaction level in the jth attribute matches Bij. Note that TEj|Bij is also a fuzzy number (Yager, 1984). By varying over all products, for the attribute j and the customers preference on the jth attribute, we can have the following set of similarity measures, which are in the form of fuzzy numbers. TEjjB1j ; TEjjB2j ; . . . ; TEjjBmj : Equation (6) represents the similarity degrees between the jth attributes of Bij and Ej.
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

325

Given this scenario, the customer may like to know the likelihood that he/she will get a product with the desired attribute levels. This can be predicted by calculating the probability of availing the attribute level Ej, Pf (Ej ). Here, the probability needs to be calculated under fuzzy observations Bij (8 i, j) and with the fuzzily dened customers requirements Ej. Following Yager (1984), we have calculated the probabilities as given below X TEjjBij : 7 Pf Ej 1=m The term Pf (Ej ) reprents the probability of obtaining the customers preference level in the jth attribute Ej. Since each TEj|Bij are LR fuzzy numbers their addition and hence the derived probabilities are again LR fuzzy numbers. Thus we have Pf Ej Pf Ej l ; Pf Ej n ; Pf Ej r : The expected value of the jth attribute under the given market prole can be calculated as Expf Ej Pf Ej l Ejl ; Pf Ej n Ejn ; Pf Ej r Ejr ; 8j: 9 8

From equation (9) it is clear that the expected value is a fuzzy set of fuzzy numbers. The membership function of the fuzzy set represents the exibility behaviour or compromising attitude of the customer while purchasing a product.

5. Solution procedure This section is devoted to the procedure of articulating the customers preference level of the product attributes and selecting the best available product in the Internet market. For each product attribute the customer expects at least the level prescribed by the expected value given in equation (9). This needs a comparison of the available products attribute levels to the desired expectation. For a minimisation (maximisation) type attribute (of say Bij ), the comparison can be expressed as: To what extent is the attribute Bij ( jth attribute of the ith product) less (greater) than the expected value given in (9)? Mathematically this can be written as Max: Min: Bij x; ExpEj fyg Ii j; x)y: 10

Similarly for an attribute which is of maximisation type we can have the mathematical equation as: Max: Min: Bij x; ExpEj fyg Ii j; x*y: 11

In equations (10) and (11) we have made fuzzy number comparisons (Dubois and Prade, 1978). The quantity Ii ( j) represents the degree of satisfaction of the jth attribute of the ith product. Continuing this procedure over all the attributes and over all the products we have m sets of
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

326

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

satisfaction levels. They are as given below Ii 1; Ii 2; . . . ; Ii k for i 1; 2; . . . ; m: 12 If all the Ii( j)s (for every j) are unity, all the attributes of the ith product are very much up to expectation and the ith product can be selected as the best product. Unfortunately, in normal cases this does not occur as the attributes conict with each other. Generally, for a particular product one attribute may achieve to the fullest extent, another to some extent, and another may be to a negligible extent. In this situation, a customer needs to compromise, and the degree of compromise depends on his/her attribute preferences. If Wj represents the weights attached to the attribute j, following the weighted average method given in Mohanty, (1998), we can have the satisfaction level for the ith product as X 13 Pi Wj Ii j i 1; 2; . . . ; m: Similarly we can obtain the satisfaction level for other products. The product corresponding to the highest satisfaction level is selected as the best product. This can be obtained through the following equation: P Max: P1; P2; . . . ; Pm: 14 Here, P represents the best product as per the customers preference. The next best product and so on can be obtained by maximising again over P(i) values for the remaining products. 6. Numerical example In this section, we have illustrated a numerical example to highlight the procedure. Here, we have taken three products B1, B2, and B3, which are available on the Internet. A customer assesses these products based on three attributes. We can represent these product attributes as Bi Bi1 ; Bi2 ; Bi3 for i 1; 2; 3: Bij represents the jth attribute of the ith product. Let Bi1, Bi3 be of minimisation type and Bi2 of maximisation type attributes. The attribute values are given as fuzzy numbers For the product B1: B11 5 (1, 2.5, 4), B12 5 (0.6, 0.7, 0.8), B13 5 (4, 5, 6). For the product B2: B21 5 (2, 4, 6), B22 5 (0.6, 0.7, 0.9), B23 5 (3, 5, 6). For the product B3: B31 5 (2, 3, 6), B32 5 (0.5, 0.6, 0.9), B33 5 (2, 3, 5).
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

327

Let the customers preferences on the product attributes 1, 2 and 3 be as follows: E1 5 (1, 3, 5), E2 5 (0.2, 0.8, 0.9), E3 5 (2, 4, 6). From equation (5), we have derived the following degrees of similarity of the product attribute 1, with the customers desire level E1: TE1jB11 f0=0; 0:53=0:4; 1=0:75; 0:94=0:8; 0:65=1g 0; 0:75; 1; TE1jB21 f0:5=0; 0:84=0:4; 1=0:5; 0:7=0:8; 0:5=1g 0:12; 0:5; 1:5; TE1jB31 f0:3=0; 0:6=4; 0:9=0:8; 1:0=1:0g 0:5; 1; 0: Degrees of similarity of the product attribute 2, with the customers desire level E2: TE2jB12 f0=0; 0=0:4; 1=0:75; 0:8=0:8; 1=0:85; 0:7=0:9; 0=1g 0:6; 0:85; 1; TE2jB22 f0=0; 0:2=0:4; 1=0:75; 0:3=0:6; 0:8=0:8; 0:5=1g 0:37; 0:83; 1:8; TE2jB32 f0=0; 0=0:4; 0:6=0:6; 1=0:67; 0:75=0:8; 0:35=1g 0:5; 0:67; 1:16: Degrees of similarity of the product attribute 3, with the customers desire level E3: TE3jB13 f0=0; 0:72=0:4; 1=0:5; 0:50:8; 0=1g 0:13; 0:5; 1; TE3jB23 f0:2=0:1; 0:6=0:3; 1=0:5; 0:68=0:8; 0:5=1g 0; 0:5; 1:5; TE3jB33 f0=0; 0:1=0:1; 1=0:5; 0:78=0:7; 0:5=1g 0:05; 0:5; 1:4: Using equation (7), we can have the probabilities of attaining the attribute levels E1, E2 and E3 as follows (after limiting the values between 0 and 1): Pf E1 0; 0:75; 1:0; Pf E2 0:49; 0:78; 1:0; Pf E3 0:08; 0:5; 1:0: Note that the above probabilities are in the form of fuzzy numbers. Now using equation (9) we can nd the customers expected values in dierent attributes under the given market prole as: Exp E1 0; 2:25; 5; Exp E2 0:1; 0:62; 0:9; Exp E3 0:16; 2; 6: In order to see whether the available products live up to the expectation, the customer needs to compare each product attribute to the fuzzily dened expected values given in equation (19). This comparison is done by considering that Bi1, Bi3 are of minimisation type and Bi2 is of maximisation type. Following equations (3)(5) we have:
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

15

16

17

18

19

328

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

For the product B1: Deg [B114Exp (E1)] 5 I1(1) 5 1.0, Deg [B12XExp (E2)] 5 I1(2) 5 1.0, Deg [B134Exp (E3)] 5 I1(3) 5 0.4. For the product B2: Deg [B214Exp (E1)] 5 I2(1) 5 0.66, Deg [B22XExp (E2)] 5 I2(2) 5 1.0, Deg [B234Exp (E3)] 5 I2(3) 5 0.6. For the product B3: Deg [B314Exp (E1)] 5 I3(1) 5 0.85, Deg [B32XExp (E2)] 5 I3(2) 5 0.97, Deg [B334Exp (E3)] 5 I3(3) 5 0.81. The above expressions are shown in Figs 3, 4 and 5. In general a customers preferences are better reected by attaching weights to the product attributes. By using equation (13) and taking the weights as W1 5 0.33, W2 5 0.34 and W3 5 0.33 we have the product satisfaction levels according to equation (6) as follows: For the product B1: 1(0.33)11(0.34)10.6 (0.33) 5 0.76. For the product B2: 0.66(0.33)11(0.34)10.81(0.33) 5 0.82. For the product B3: 0.85(0.33)10.97(0.34)10.81(0.33) 5 0.88.
1.2 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 7 Exp(E1) B11 B21 B31

Fig. 3. Comparison of product attribute 1 with the expectation Exp (E1).

r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

1.2 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1
Exp(E2) B12 B22 B32

329

Fig. 4. Comparison of product attribute 2 with the expectation Exp (E2).

1.2 1 0.8 0.6 0.4 0.2 0 0 1 2 3 4 5 6 7
Exp(E3) B13 B23 B33

Fig. 5. Comparison of product attribute 3 with the expectation Exp (E3).

Using equation (14) we will have the best product corresponding to the maximum product satisfaction level. Thus we have Max: 0:76; 0:82; 0:88 0:88: The value 0.88 corresponds to the product B3. Hence as per the customers preference the best product in the Internet market is B3. The second best is B2 corresponding to the level 0.82 and third is B1 with satisfaction level 0.76.

7. Conclusion This paper has introduced a model which takes into account the customers opinion in linguistic terms while purchasing a product on the Internet. Very often customers express their assessments or opinions in fuzzy terms. Customers fuzzy behaviours and their expectations are modelled here by using the concepts of fuzzy logic and fuzzy probability respectively. Additionally, products attributes as perceived by customers are also incorporated into the model. The appropriation of these additional concepts makes the online business more practical and corresponds to real world business decision-making situations.
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

330

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

The methodology in our paper helps customers in the following ways. (1) In general customers assess products by their attributes, which are usually conicting, noncommensurable and fuzzy in nature. Our model considers all these aspects and suggests a viable product(s) to the customer. (2) Based on customers fuzzy preference of product attributes, a companys available products are also dened fuzzily. Using this method the companys products are compared with the customers requirements and products with high matching degrees are presented to the customer. Thus, the customer can select a product of his/her own choice with a maximum possible satisfaction level. (3) The earlier electronic shopping agents like Jango, DealTime (http://www.dealtime.com) and decision guide (http://www.activityBuyersGuide.com) assist customers in nding the best available product based on the product attributes required by the customers. However, these shopping agents do not have any feature to entertain the fuzziness of the customers views about the product attributes. Simultaneous consideration of multiple numbers of product attributes is an added diculty. Our proposed procedure handles these problems by using the concepts of fuzzy sets, fuzzy probability and multi-attribute decision analysis. This paper has the following limitations and these topics can be taken as a scope for future research: (1) Our methodology does not consider the non-digital attributes inherent in the products. The explanation of non-digital attributes on the Internet is a challenging problem in Internet business. (2) It is explained in Section 2 that in order to make business decisions more customer focussed, a company needs to analyse fuzzily dened customer requirements about product features and also assess its own available products in a exible way. In the companys point of view we have not used any methodology to derive product attribute exibility values. This is merely done by looking into the market scenario, customers attribute-wise requirements and the companys own judgement.

Acknowledgement The authors would like to thank the anonymous referees for their valuable comments which have improved the paper signicantly.

References
Alter, S., 2004. Information Systems: the Foundation of E-Business. Pearson Education (Singapore) Publishers, New Delhi. Aouni, B., Kettani, O., 2001. Goal programming model: A glorious history and promising future. European Journal of Operational Research 133, 225231. Cao, Y., Li, Y., 2007. An intelligent fuzzy-based recommendation system for consumer electronic products. Expert Systems with Applications 33, 230240.
r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies

B. K. Mohanty and B. Aouni / Intl. Trans. in Op. Res. 17 (2010) 317331

331

Carlsson, C., Fuller, R., 1996. Fuzzy multiple criteria decision making: Recent developments. Fuzzy Sets and Systems 78, 139153. Cheng, C.-B., Chan, C.-C.H., Lin, K.-C., 2006. Intelligent agents for e-market place: Negotiations with issue trade-os by fuzzy inference systems. Decision Support Systems 42, 626638. Dubois, D., Prade, H., 1978. Operations on fuzzy numbers. International Journal of Systems Science 9, 613626. Lee, R.M., Widmeyer, G.R., 1986. Shopping in the electronic market place. Journal of Management Information Systems 2, 2135. Martel, J.-M., Aouni, B., 1998. Diverse imprecise goal programming formulations. Journal of Global Optimization 12, 2434. Miao, C., Yang, Q., Fang, H., Goh, A., 2007. A cognitive approach for agent-based personalized recommendations. Knowledge Based Systems 20, 397405. Mohanty, B.K., 1998. Assigning weights to multiple criteria a fuzzy approach. Journal of Fuzzy Mathematics 6, 209 222. Mohanty, B.K., Bhasker, B., 2005. Product classications in Internet business. Decision Support Systems 38, 611619. Mohanty, B.K., Vijayaraghvan, T.A.S., 1995. A multiobjective programming problem and its equivalent goal programming problem with appropriate priorities and aspiration levels a fuzzy approach. Computers and Operations Research 22, 529541. Ryu, Y.U., 1999. A hierarchical constraint satisfaction approach to product selection for electronic shopping support. IEEE, Transactions on Systems, Man and CyberneticsA 29, 525532. Triantaphyllou, E., 2000. Multi-Criteria Decision Making Methods: A Comparative Study. Kluwer Academic Publishers, Dordrecht. Yager, R.R., 1984. Probabilities from fuzzy observations. Information Science 32, 131.

r 2009 The Authors. Journal compilation r 2009 International Federation of Operational Research Societies