支援情境變動下的潛在目標用戶探勘系統之構建──以可停電力潛在用戶探勘為例

Abstract

隨著科技的日益發展與企業競爭環境的急遽變動，商品的選擇愈來愈多樣化再加上近年來客製化服務的提倡，顧客的消費行為模式不再是一成不變。然而，對於企業來說，服務廣大市場裡面的每個人無異是緣木求魚。以飲料公司舉例說明，儘管是一家頗具規模的大公司，也沒辦法開發出各種符合每個人口味的飲料，因為龐大的人口數和顧客間消費行為的歧異性會產生出很多不同的需求。為了讓資源如時間和成本等更有效率的利用以產生額外的效益，企業必須找出其產品或服務的目標市場，以從中獲得最大的利益，而這樣的過程稱為目標行銷。 市場區隔為目標行銷的首要步驟，有效率的市場區隔分析必須能發掘可衡量的、可得的、有價值的且能採取行動的市場區隔。一般而言市場區隔分析是由既有的資料去作區隔，再找出每個區隔的特徵行為，應用到廣大的客戶群。因而，難免遇到資料量不足的問題，而利用資料探勘技術硬式分群(hard clustering)作市場區隔時，所產生的分群就會落到無法衡量的窘境。本研究提出fuzzy c-means membership function權重分配方法來改善資料量不足所產生的分析偏誤。然而，當資料龐大時利用fuzzy c-means membership function權重分配，反而會受到雜亂的資料影響而效果不彰。因此，分群分析方法的選擇，必須根據分析資料的特性來決定。 此外，目前的目標行銷資料來源主要分為兩大類，分別為內部資料(internal sorces)和外部資料(external sources)，內部資料包括既有顧客的基本資料和消費行為，外部資料包括欲開發客戶的喜好與需求等，通常透過市調獲得。然而，不管是哪一種資料，其所代表的只是歷史資料或是目前狀態。因而，利用這些資料進行探勘分析出來的結果，僅能支援短期的決策。有鑑於此，本研究將以台電可停電力方案為例，擬提出情境變動分析資料探勘架構，希望透過此一架構來延長決策的支援期間。藉著市場區隔資料探勘方法的改善再加上情境變動分析流程，來改善現階段目標行銷的效益，並將流程標準化構建支援情境變動下的潛在目標用戶探勘系統，以提供有相同需求的產業做為參考。

Since consumer behavior is no longer fixed and competition among companies becomes even more intense with the improvement of technology, it is impossible for a company to satisfy everyone in the world. For example, it is rather expensive for a beverage company to offer various kinds of products to customers in order to satisfy them. Owing to diverse consumer behavior, it is economical to identify the target market. A company can benefit more by spending most of its resources on the target market. The process of identifying the target market is known as target marketing. Market segmentation is the first step of target marketing. In general, segmentation analysis uses hard clustering methods for the purpose of segmentation and identifies the characteristics of each segment in order to apply to other customers. Nonetheless, scant original data always make hard clustering methods worthless. In order to solve this problem, we use fuzzy c-means to improve the bias resulting from scarce data. Nevertheless, in case of an astronomical amount of data, the result obtained with fuzzy c-means is not as good as expected. Depending on different clustering methods, data characteristics always have a different effect on the analysis result. Furthermore, general target marketing merely supports short-term strategies, since its data is obtained from historical transactions or consumers’ current status. In order to support long-term strategies, we propose the process of target marketing analysis with scenario analysis and use data from the interruptible load measure of the electricity industry to verify the analysis process. Finally, according to the process, we implement the system to support target marketing analysis with scenario analysis.