與傳統商店相比，網路商店較容易取得顧客的基本資料及購物資料。藉由分析顧客資料，網路商店才能夠更進一步瞭解顧客購物行為模式。在本文中，我們定義了一個衡量顧客忠誠度的指標，SPLS (標準產品忠誠度)，使用的是顧客購物紀錄以衡量每一位顧客對某一產品的忠誠度。SPLS和忠誠顧客的個人背景資料合併成為集群分析的輸入資料，而集群分析產生的輸出即是將忠誠顧客分為數群。集群分析的結果使得在同一集群內的忠誠顧客背景資料及購物行為相似性最高，集群間的忠誠顧客相似性最低。接著就是藉由相似性分析衡量非忠誠顧客的個人背景資料與哪一個忠誠顧客集群的相似性最高，若找到所屬集群則可進行下一步的忠誠度預測。忠誠度預測方式是將一預估的SPLS值指定給一非忠誠顧客，以預測此顧客購買該產品的可能性。所有非忠誠顧客所得的SPLS預測值中若高於門檻值則視為潛在顧客，該產品就會被推薦給這些顧客。另外，在忠誠顧客分析方面，我們使用參照圖(reference map)來確認忠誠顧客對現有產品的偏好。參照圖除了可以用來做忠誠顧客對現有產品的偏好分析之外亦可以用於新產品推薦的狀況。總結此篇論文是提出一系統化的模組用以分析線上消費者的購物行為及提供專為忠誠顧客及潛在顧客的推薦策略。 Comparing with the traditional store, the online store can keep the track of customers’ purchasing records and personal information. By analyzing these customers’ records, online store can have a better understanding of their customers’ profile and purchasing behavior. In this paper, we define a standard product loyalty status, or SPLS, using customers’ purchasing records to evaluate each customer’s loyalty to a certain product. SPLS is incorporated with loyal customers’ personal backgrounds as the input of cluster analysis that divides loyal customers into different groups. Loyal customers in the same groups have similar purchasing behavior and personal backgrounds. Similarity analysis measures the similarity of backgrounds between a non-loyal customer and groups of loyal customers in order to find this customer’s belonged group. Then, an expected SPLS value is assigned to this non-loyal customer to estimate his/her probability of purchasing a certain product. Customers who have expected SPLS value larger than a threshold are regarded as potential customers. Marketing specialists should recommend the product to potential customers. Loyal customers, on the other hand, are analyzed with the reference map to identify their preference for current product line. The reference map is used to provide suggestions under the condition of new-product-launch. Overall, this paper proposes a systematic model to construct online customer profile and a recommendation strategy for loyal customers and potential customers.
