108 C.-S. Hwang, Y.-C. Su, and K.-C. tseng P=(,f,,f where n is the number of product features(n=12, in our experiment. ) Each feature is assigned a value of l if a product possesses that feature, and 0 otherwise. To fully understand which product features a customer is of interest, in the second step, we compute the customer product preference profile from the customer trans tion data and the product profile. The product preference profile of customer k is described as CPP=(CPP,CPP,…,CPP) where CPP represents the preference profile of feature i for customer k and is ob- tained by summing up the feature information from the products purchased by cus tomer k. CPp can be defined as CPP= where T represents the set of products purchased by customer k. Finally, the customer preference profile is generated by integrating three portions f information: the customer product preference profile, the customer information and customer purchasing behavior. The customer preference profile is defined as CP=(CPP, CIk, CT) =CPk,CP2,…,CP) where CI,=(CIi, CIF) contains customer ks information including member level and gender and CT: =(R,, Fr, M, contains the information about customer k's purchasing behavior measured by recent Table I shows an example of the customer preference profile Table 1. The Customer Preference Profile 是 [0O2|10.33103301007800019108 C.-S. Hwang, Y.-C. Su, and K.-C. Tseng ( , , , ) n j j j j P f f " f 1 2 = (1) where n is the number of product features (n=12, in our experiment.). Each feature is assigned a value of 1 if a product possesses that feature, and 0 otherwise. To fully understand which product features a customer is of interest, in the second step, we compute the customer product preference profile from the customer transac￾tion data and the product profile. The product preference profile of customer k is described as ( , , , ) n CPPk CPPk CPPk " CPPk 1 2 = (2) where i CPPk represents the preference profile of feature i for customer k and is ob￾tained by summing up the feature information from the products purchased by cus￾tomer k. i CPPk can be defined as k j T i j i k T f CPP k ∑∈ = (3) where Tk represents the set of products purchased by customer k. Finally, the customer preference profile is generated by integrating three portions of information: the customer product preference profile, the customer information and customer purchasing behavior. The customer preference profile is defined as ( , , , ) ( , , ) 1 2 17 k k k k k k k CP CP CP CP CPP CI CT = " = (4) where ( , ) 1 2 CI k = CI k CI k contains customer k’s information including member level and gender and ( , , ) CTk = Rk Fk Mk contains the information about customer k’s purchasing behavior measured by the RFM (recency, frequency, and monetary) information. Table 1 shows an example of the customer preference profile. Table 1. The Customer Preference Profile C_no 1 f 2 f … 12 f level sex Recency Frequency Monetary 002 1 0.33 … 0.33 0 1 0.078 0 0.019