Article Preview
Top1. Introduction
Target marketing requires marketers to (1) identify and profile distinct groups of buyers who differ in their needs and preferences (market segmentation), (2) select one or more segments to enter, and (3) establish and communicate an offer’s distinctive benefits for each target segment (Kotler & Keller, 2003). To extend customer life cycles, businesses apply various marketing plans in different customer clusters (called market segments) or fuzzy-based clusters (Terence et al., 2002). Businesses can yield higher growth rates from high-profit or expensive products or encourage customers to become long-term customers by implementing discovered rules (Berry & Linoff, 2004).
A market segment consists of a group of customers who share a similar set of needs and wants. Therefore, the behaviors of customers belonging to the same market segment are similar. To gain an insight into the behaviors of customers belonging to a specific market segment, sales patterns in various market segments must be determined first. Recency–frequency–monetary (RFM) scoring is a method of determining the scores of current customers according to their recency, frequency, and monetary values, and this method has proven highly effective when applied to marketing databases (Blattberg et al., 2008). Moreover, the RFM model can effectively process clustering according to customer value.
Integration of RFM scoring and frequent itemset mining techniques provides useful information for current customers. However, existing relevant studies have often focused on determining frequent itemsets in which sales patterns occur in a whole dataset, rather than focusing on several subdatasets clustered according to customer behaviors. In addition, specific sales features in various market segments are difficult to identify using existing approaches. How to identify sales features in different market segments is an appealing research topic.
This study first applies RFM scoring to segment customers into several groups (also called market segments) and further divide customer transactions into several sub-RFM datasets according to the customers’ RFM scores. Subsequently, we identify specific sales patterns (unique, common, and particular) from frequent RFM itemsets discovered from various sub-RFM datasets. First, we identify the unique sales patterns that only exist within a market segment. Second, we identify the common sales patterns that exist within all market segments. Finally, we apply a new criterion (Contrast Support, CS) to identify particular sales patterns from the discovered RFM itemsets in various market segments (also called equivalence-class RFM datasets).
The remainder of this paper is organized as follows. Section 2 reviews related studies. Problem definitions are provided in Section 3. The proposed algorithm and an example are illustrated in Section 4. Section 5 involves the use of survey data for a case study to demonstrate the usefulness of the proposed algorithm. Conclusions and directions for future research are provided in Section 6.