自我組織特徵映射圖是一個優秀的資料探勘工具,它可以將高維度的輸入圖樣投影至低維度的拓撲方格上,而且能夠供給人們目視與探查資料群集的性質,此外,若資料量增加時,研究者也可透過拓撲節點的增加精確地進行數量分析。在此篇論文中,本研究針對電子商務及行銷學上相當著名的RFM變數進行顧客分群的動作,過往的公司企業通常直接採用一些通用的非監督式分群法來區隔顧客,所得到的結果雖尚能稱善,但本研究秉持建立更優良的分群方式以完成更精確的顧客分群過程之精神,嚐試採用廣義K-平均法與模糊C-平均法對拓撲網路圖進行分群,這樣的過程稱之為兩階層式的分群法,換言之,本研究先使用自我組織特徵映射圖的技術產生拓撲網路圖,接著再以不同的分群方式對該拓撲網路圖進行第二階段的分群,實驗結果顯示,不論是以效率或速度的層面觀之,此種分群方法比傳統式直接分群的結果更好。
Self-organizing feature map (SOFM) is a distinguished data mining tool for academic or practice. It projects the input data on a two or three-dimensional grid called prototypes that help to visualize effectively and explore characteristics of them. While the number of data is large, the researchers can increase the number of node to facilitate quantitative analysis of them. In this paper, we focus on the RFM variables and use different approaches to segment the SOFM prototypes. In particular, general K-means and fuzzy C-means are applied. We execute the technique of SOFM to generate the prototypes in the first place. Then we adopt segment methods in the second phase. The experiment result demonstrates that it performs well when compared with direct segment of the RFM variable.