以交易紀錄與商品屬性的BPR個人化推薦應用-以一電商為例

Abstract

在電子商務發展快速的現代，消費者在網路上留下的購物資訊越來越多，其中包含了消費者各種的顯性回饋(explicit feedback)以及隱性回饋(implicit feedback)資料。 但是當電商處於發展階段，而尚未收集到完整的消費者購物行為資料的時候，如何利用BPR(Bayesian Personalized Ranking貝氏個人化排序)透過僅有的消費者何時買、買什麼的交易紀錄、以及商品屬性標籤來進行個人化的推薦，以達到經營會員的目的。並且在以個人購物週期作為計算的會員分群方式下，討論在(1)排除熱銷商品、(2)減少商品屬性標籤，以及(3)不同長短的交易時間區間的情況下，試圖找出優化推薦的方法。

With the fast growing of E-commerce industry, customers have been leaving a lot of buying information on the internet. It includes various kinds of explicit feedbacks or implicit feedbacks. While an E-commerce company is under the developing stage, it’s incapable of collecting the complete information about consumers’ behavior. However, the E-commerce company who only has the trading journal of what and when product’s been bought and the data of product attributes can make recommendations by the method of Bayesian Personalized Ranking to manage the customer relationships. Under the circumstance of different group members by the individual buying cycle time, we discussed the scenarios like excluding the hot sale products, reducing the product attributes, and the different lengths of time of the trading data to try to find out the ways to improve the recommendation.