藉由文字探勘從線上評論找出最佳的產品屬性組合

Finding the Optimal Product Attribute Configuration through the Application of Text Mining to the on-line Reviews

新產品開發的過程是在尋找可以賣得最好的產品屬性組合，又稱之理想構型。傳統的方法包括感性工學、結構設計、先驅使用者及同理心設計等，除了耗時外，最終得到的屬性構型也不一定是理想構型。迴歸方程式偏重在各屬性對效用的線性影響;聯合分析則進一步找出各屬性對效用的不同影響型態，文獻上對於如何找出產品屬性的理想構型仍存在缺口。本研究利用文字探勘技術，將線上留言轉成數字資料，結合類神經網路捕捉非線性關係的優點，找出屬性構型與銷售量的對應關係，進而建構該產品屬性的理想構型。在學術上結合質與量的分析，提供決定產品屬性理想構型的新方法，在實務上可加速新產品開發，並對於新產品開發方向與改善現有產品，提供具體改善方向。

The process of developing a new product is a search for the ideal configuration of product attributes that will sell best. Though traditional methods such as Kansei engineering, configurator design, lead user and empathic design are time consuming, the derived final products are not necessarily the ideal configuration. Hedonic regression focuses on the linear effect of each attribute on utility; conjoint analysis goes further to find out how each attribute affects utility in different ways, and there is still a gap in the literature on how to find the ideal configuration of a product. This study converts online reviews into numerical data by text mining and combines the advantages of neural network in capturing non-linear relationships by finding out the correspondence between attributes and sales, and then construct the ideal configuration of the product. In academia, the combination of qualitative and quantitative analysis provides a new method to determine the ideal product configuration. In practice, the method proposed in this study not only accelerates the speed of new product development, but also provides specific directions for the development process of new products and the direction of improvement for existing products.