用戶行為分析與購買預測

Abstract

顧客的行為分析是近年學術與業界關切的議題，顧客的消費旅程數位化，轉換為數據紀錄下來，讓學者與企業得以深入分析顧客行為，更加了解顧客的需求，並制定相對應的策略，進而與顧客維繫更長的關係；相對應的管理架構如海盜模型、機器學習分析法的應用也更加廣泛。本研究試建立模型，透過分析訪客在電子商務網站的瀏覽行為，預測其是否進行購買，並建議企業制定行銷策略。 本研究以訪客瀏覽網站的不同行為次數作為變數，並考量行為時間點的代表意義給予權重；以經過加權之瀏覽行為作為特徵，以下個月是否進行購買為標籤，利用機器學習進行訓練與測試。在測試的演算法中，決策樹法在精準行銷的條件下有較高的召回率；模型做決策的過程以用戶經加權之瀏覽個別商品頁面次數作為主要判斷依據，越高則購買的可能性越高。藉此，本研究建議品牌挑選模型預測成為購買客的訪客作為目標對象，進行折價券與購物籃分析以提升購買人數與客單價。

Customer behavior analysis has been an important topic for academics and the industry in recent years. Customer journey has digitalized and recorded as data, so that scholars and companies can analyze customer behavior, understand customer needs, develop strategies, and strengthen customer relationship. Business framework like Pirate Model, and machine learning are generally implied. This paper establishes the model to predict whether users purchase by analyzing their behavior of browsing e-commerce website, and suggests marketing strategies. This paper sets the counts of different behaviors of browsing as variables and gives weight considering behavior timing. The model sets the weighted variables as feature and sets whether the user purchase next month as label, using machine learning for training and testing. In testing algorithms, decision tree method has a higher recall rate under the condition of precision marketing. The decision tree method makes the decision mainly based on the weighted counts of user browsing product pages. The higher the weighted counts, the higher the possibility of purchasing. So, this paper suggests the firm target users predicted by model, and set coupons or do basket analysis to increase the number of customers and the transaction value.