運用購物籃分析建立產品推薦系統

Abstract

資訊的發展使顧客可以透過網站快速的比較廠商及產品資訊，同時，網路也促成顧客之間的溝通更為頻繁，要如何善用資訊以培植企業與顧客的長久關係成為企業不應忽視的問題。近年來，推薦系統的發展不僅成為許多研究學者鑽研的目標，也引起業界不同應用領域的興趣與關注，然而多數企業在建置推薦系統時仍缺乏相關之技術及資源。故本研究試圖以簡單但科學的方法來建立產品推薦系統，以作為企業可以參考之依據。 本研究主要以購物籃的概念為基礎，以顧客過去的購買紀錄作為研究的依據，將分析主體劃分為三個部分，包括對整體顧客之分析、對分群顧客之分析以及對個別顧客之分析。分析方法主要有二：條件機率矩陣及層級貝氏Probit模型。本研究欲透過這兩種分析的方法以篩選出各種購物籃作為企業可向各類顧客推薦之產品組合，以建立顧客對企業之信賴，進而提升顧客關係。 研究結果不僅可以顯示出消費者在購買決策所呈現之各類產品之相關性，亦可分析產品之回購率進而作為企業推薦之依據。另一方面，結果顯示以層級貝式Probit模型之擊中率高於在無任何資訊下以平均機率之概念為基礎之擊中率。

The development of information enable customers to compare rapidly and communicate more frequently through the website. Due to the innovation of information technology, the enterprises have to respect an issue that how to use the information to cultivate the relationship between enterprises and customers. Recently, the development of the recommendation system has not only become the research target of professors also attract the concern of practice. However, most of enterprises build the recommendation system without sufficient technology and resources. Therefore, this article will construct the recommendation system with simple and scientific methods and as the reference system to enterprises. This article will adopt the concept of market basket analysis base on the trading records of customers, and we will focus on the analysis of aggregated customers, group customers, and individuals with the conditional probability matrix and Hierarchical Bayesian Probit model. We will use the analysis to filter out various market baskets as the recommended products to the different customers. The result can show not only the correlation between different products, but also the re-purchase rate of products. On the other hand, we find that the hitting rate of Hierarchical Bayesian Probit model is higher than the hitting rate of the average probability without any information.