整合顧客個人化與賣方獲利性之商品推薦系統

Abstract

目前既有的推薦系統(Recommender Systems)進行商品推薦時，主要考慮商品的 購買機率(product purchase probability)，而忽略了企業應用推薦系統的同時亦希望追 求獲利增加。因此，本三年計畫探討以企業（賣方）與顧客（買方）為不同出發點考 量之數種推薦系統。此外，亦有學者應用資料探勘(Data Mining)技術之關聯法則 (Association Rules)由購物籃(Market Basket)中探勘強物項(Frequent Itemsets)，以進行 商品推薦。然而，此方法亦僅透過強物項考慮顧客個人化(Customers』 Personalization) 與產品被顧客購買之頻率，而未考慮賣方之獲利性(Sellers』 Profitability)。 由賣方的角度而言，係依據整體產品購買機率(overall product purchase probability) 與產品獲利性(product profitability)進行推薦；由買方的角度而言，係依據個別顧客的 偏好(individual customer 』s preferences)推薦。本計畫結合買方觀點及賣方觀點建構商 品推薦系統。在同時考慮買方觀點及賣方觀點情況下，建構以協同過濾(Collaborative Filtering)為基礎與關聯法則為基礎之商品推薦系統，必須考慮多準則以推薦商品。所 以，本計畫亦應用資料包絡分析(Data Envelopment Analysis; DEA)發展多準則資料探 勘技術，並應用於商品推薦系統。本計畫同時以模擬資料與企業真實資料驗證所發展 之推薦系統並與傳統方法進行比較分析。推薦準確率(recommendation accuracy)與由 交叉銷售所得利潤(profit from cross-selling)為評估指標比較不同觀點的推薦系統。 依據上述，本三年計畫期望分析比較六種推薦系統，並且發展整合之商品推薦系 統以應用於實務上。此考量個人化推薦及/或商品獲利性之六種推薦系統包含： 1. Convenience perspective recommender system (CPRS); 2. Convenience plus profitability perspective recommender system (CPPRS); (本計畫發展 之系統) 3. Collaborative filtering perspective recommender system (CFRS); 4. Hybrid perspective recommender system (HPRS); (本計畫發展之系統) 5. Association rule perspective recommender system (APRS); 6. Multi-criteria association rule perspective recommender system (MCARS). (本計畫發展 之系統)

Companies need to shift from the old world of mass production where 「standardized products, homogeneous markets, and long product life and development cycles were the rule」 to the new world where 「variety and customization supplant standardized products」. This three-year project attempts to develop and analyze several recommender systems based on the perspectives of enterprises (sellers) and customers (buyers). From the sellers』 perspective, recommendations are made based on the overall product purchase probability and the product profitability; from the buyers』 perspective, recommendations are made based on ind ividual customers preferences and personalization. From the perspective of customers, recommender systems merely try to suggest suitable products to satisfy the needs of customers. However, enterprises employ recommender systems not only to satisfy customers』 needs but also to profit more. Therefore, the value (profit margin) of products should also be taken into consideration in developing recommender systems. The application of collaborative filtering recommender systems in the physical context of retailing will be analyzed and discussed in this project. In existing some recommender systems, association rule mining is employed to recommend products. However, due to the simultaneous consideration of the overall product purchase probability, the product profitability, customer preference and personalization, recommendations are made with multiple criteria. The recommendation approaches need be developed to resolve the issue of multi-criteria. In the previous studies regarding the discovery of subjectively interesting association rules, most approaches require manual input or interaction by asking users to explicitly indicate what kinds of rules are interesting and uninteresting. This project aims at using a non-parametric approach, Data Envelopment Analysis (DEA), to estimate the efficiency (interestingness or usefulness) of association rules with multiple criteria. Six recommender systems are compared in terms of recommendation accuracy and/or profit from cross-selling. They are as follows: 1. Convenience perspective recommender system (CPRS); 2. Convenience plus profitability perspective recommender system (CPPRS); 3. Collaborative filtering perspective recommender system (CFRS); 4. Hybrid perspective recommender system (HPRS); 5. Association rule perspective recommender system (APRS) 6. Multi-criteria association rule perspective recommender system (MCARS). CPPRS, HPRS and MCARS are novel perspectives, which additionally take sellers』 profitability measures into consideration. In the MCARS, DEA will be adopted to estimate the efficiency (interestingness or usefulness) of association rules with multiple criteria. In terms of recommendation accuracy and/or profit from cross-selling, comparisons can be made between CPRS and CPPRS, between CPRS and CFRS, between CFRS and HPRS, among CPRS, CFRS and HPRS, among CPPRS, CFRS and HPRS, and among APRS and MCARS. Furthermore, an integrated recommender system will be developed for real-world applications with supportive company in this project.