快速流行產業之個人化互動平台推薦系統

Abstract

資訊科技的快速發展，促使許多服飾企業紛紛導入智慧型商店之概念，引進許多科技應用至實體店面。加上服飾業者採用快速流行經營策略，使得產品多樣化，生命週期縮短。目前導入服飾業者實體店面之互動平台與然而目前尚無針對此設施之商品推薦系統。構建推薦系統時，若採用傳統分析方式進行商品推薦，將會發生冷啟動(Cold Start)與過度特定化(over-specialization)的問題，冷啟動問題將會降低推薦系統之效率，而過度特定化則會使得推薦系統過度集中商品推薦種類，無法有效提供多樣化之商品推薦。為解決上述之問題，本研究構建之個人化互動平台推薦系統，藉由分析顧客對商品本質的喜好，推薦符合顧客偏好的商品，解決快速流行產業產品層面之冷啟動問題。並透過關聯法則分析找出偏好相似之顧客群，藉此推薦顧客未曾購買過，但購買機率高之商品，以解決過度特定化之問題。此推薦系統結合資料探勘技術、協同過濾(Collaborative Filtering)與以內容為基礎式過濾(Content-based Filtering)方法，應用於個人化互動平台的服飾商品推薦。 本研究之推薦系統架構中，依資料與分析目的之不同，將此推薦系統分成四個子模組。分別為分店交易資料、顧客基本資料、顧客歷史交易資料與互動行為資料四個模組。可依情境的不同將模組相互搭配，以分析出顧客歷史偏好與即時偏好，推薦符合顧客需求之商品，完成個人化之商品推薦。 企業亦可透過本推薦系統的推論結果制定行銷及銷售策略。例如，在營運成本的允許下，將顧客喜好的關聯商品一起促銷，藉此增加消費者同時購買的意願，與提高企業之銷售績效。此推薦系統能夠加速結合實體店面之科技應用，達到整合之效果，並有效解決傳統商品推薦之問題，提供未來智慧商店科技整合應用之建議。

Because of the rapid development of information technology, many clothing enterprises try to implement the concept of smart store. These enterprises introduce a number of technology applications to their physical stores. In addition, these enterprises take “Fast Fashion” as their business strategy so that their clothing products would be diversity and have a quite short life cycle. The Kiosk is a new facility in physical clothing store and there is no recommendation system for Kiosk. The traditional recommendation system is not an efficient way for Kiosk because of the Cold Start and over-specialization problem. The Cold Start problem will decrease the performance of recommendation system. The over-specialization problem can only focus on some products while making recommendation for customers. In order to solve the problems mentioned above, the recommendation system proposed in this study analyzes the nature of clothing products. The proposed system tries to learn about customers’ preferences for the nature of products and make recommendations. We can reduce the impact of Cold Start problem with this approach. This study solves the over-specialization problem by making recommendation lists based on association rule method. This proposed recommendation system that combine data mining, collaborative filtering and content-based filtering would apply to Kiosk in fast fashion industry. In this study, the architecture of the recommendation system is more flexible. According to the data and purposes, this recommendation system is divided into four sub-modules such as the branch transactions, customer data, transaction data and customer interaction data. Users can constitute the structure from distinct modules at will to meet the requirement in different situations. To provide personalized recommendations, system will analyze historical transactions and interaction data among users to learn their preferences and behaviors and then predicts what kind of the products users need. Not only provide a personalized and meaningfully ordered list, enterprises can also deduce and develop marketing and sales strategy from the recommendation system in this study. For example, For example, in case of the operating costs is not a limitation, enterprises can hold some bundle sales by taking customer preferences and association rules as reference. This strategy may strengthen customer’s purchase intention and improve business operation performance.