以圖形分析及圖神經網路改善電商網站使用者經驗

Abstract

VoucherCloud的統計[22]顯示，有92.6%的消費者表示，好的網站頁面設計是影響購買決策的主要因素，使用者導向設計(User-Centered Design)已是目前電商網站設計的主流趨勢。為了不斷改善使用者體驗，電商網站透過蒐集分析工具追蹤使用者行為，分析使用者與商品的關係，網站維護者在網頁上架後根據分析結果調整介面布局、網站流程及程式效能等。然而由於使用者行為紀錄沒有標準化的格式，資料蒐集過程往往受限於單一平台，蒐集到的行為資料難以深化運用。因此，本研究以標準資料格式xAPI (eXperience Application Programming Interface)為基礎開發一個使用者行為的圖形分析系統。相對於現有以統計分析方法為主，且僅聚焦於使用者與產品之間關係的電商網站分析工具，我們透過轉換使用者行為資料為圖形資料，並運用圖形方法Pattern Matching、Graph Similarity及圖神經網路(Graph Neural Network)進行資料分析，可有效分析使用者與網站內容的互動關係，根據使用者瀏覽歷程進行個人化商品推薦、尋找產品購買關聯內容區塊，並預測內容之間的關聯關係。此分析結果可提供網站維護者維運管理時，網頁重構的建議，並調整線下營運安排，有效節省成本。我們以實際線上運作的網站log進行實驗，實驗結果顯示本系統可以有效縮短使用者在整個購買流程路徑以改善使用者經驗。

According to VoucherCloud's survey [22], nearly 93% of customers regard the good UI/UX design as the key factor of online shopping. User-Centered Design has become a mainstream trend of e-commerce website design. To improve the user experience, e-commerce websites track users’ behaviors. They analyze the relationship between users and products. Then the maintainer tunes the website’s layout, workflow, performance according to these analysis results. However, the behavior tracking and its data collection are often platform-dependent as the lack of the standard format for recording user’s behaviors. Moreover, conventional e-commerce web analysis tools usually focus on statistical analysis with the relationship between users and products. As user behavior data are graphical data, we develop a graphical analysis system for user behavior based on xAPI (eXperience Application Programming Interface) standard data format. In this paper, we convert xAPI user behavior data into graphical data for analyzing by graphical methods and graph neural network. We analyze the relationship between users and web contents for personal recommendation via user browsing history. The proposed method can also find key contents of products and predict link between contents. These analysis results can provide suggestions for website reconfiguration in DevOps, and adjust the offline operation arrangement to save costs. We analyze real e-commerce web log data in the experiment, and the result shows our system can shorten the user’s shopping process effectively to improve user experience.