產品屬性考慮集合對推薦效度之影響研究

Abstract

隨著全球化程度的增加，在市場上流通的品牌與產品越來越多，使得廠商想要窺探消費者的偏好益顯困難。但消費者受限於時間、金錢，乃至於資訊處理能力等因素的限制，必須先將可能的偏好屬性挑選出來，形成產品屬性的考慮集合，藉以方便相互比較，進行購買決策。而唯有進入消費者的考慮集合，此商品才有可能被購買，因此廠商如何了解消費者的偏好，進而推薦商品或是當作新產品研發的依據，便是一個重要的議題。考慮集合在行銷策略上有著重要的策略意涵，從考慮集合中，我們可以觀察到消費者的選擇策略、品牌忠誠度以及市場佔有率的變化。因此在本研究中，將利用兩種不同的方式來產生考慮集合，代表不同的策略意義。 在以往賣場人員推薦商品時，由於不知道消費者的偏好如何，只能隨機推薦，造成推薦效果不彰。但在資料庫行銷當中，資料庫記載著消費者的個人基本資料、購買紀錄，以及產品本身的資料，皆可提供行銷人員作分析。從資料中可以將消費者做適當的分群，了解其偏好結構，反映消費者之間的異質性與動態性。 有鑑於此，本論文將利用邏吉斯模式(Logit Model)及SUR模式（Seeming Unrelated Regressions）來建構出產品推薦系統，並利用保留樣本進行推薦效度的預測，比較兩種考慮集合在不同模式下的預測成功率（Hit Rate）。最後，結論與建議彚整各章之研究發現，發展策略意涵，並說明研究限制與未來研究方向。

To form a consideration set, consumers must pick out the preferable attributes that would help them to have easy-to-compare alternatives since they are limited by time, money and ability when dealing with information. A product could only be purchased when the product was taken into the consideration set. Therefore, it becomes crucial to any firms that take consumers’ preferences as the base of its product-buying recommendation and new products development plans. There are strong marketing strategic implications of a consideration set since the consideration set can be used at analyzing consumer behavior, brand loyalty and changes in market share. This paper attempts to construct two consideration sets in different modes, which stands for distinct strategic meanings. In the past, when a sales clerk was giving product-purchasing recommendations for customers, he could only make a random one to them due to his unawareness of his customers. However, with database marketing, purchasing history could be recorded and traced, and all information is updated and analyzed every time of a purchase. All these would benefit a firm by at least separating consumers into target groups, establishing marketing preference structures and reflecting heterogeneity and dynamics among consumers. In this regard, this paper is trying to utilize Logit Model and SUR Model to construct a new product recommendation system by giving the reserved sample to test the validity of recommendation and comparing the Hit Rate of these two consideration sets in both modes