請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91461
標題: | 結合個人品牌選擇與群體購物籃分析改善在超市品牌產品現場推薦策略 Combining Personal Brand Choice and Market Basket Analysis in Group to Improve Brand Product On-Site Recommendation Strategy in Supermarket |
作者: | 王騰逵 Teng-Kuei Wang |
指導教授: | 張時中 Shi-Chung Chang |
關鍵字: | 品牌選擇,品牌選擇機率,分群購物籃分析,推薦系統,交易紀錄, Brand Choice,Brand Choice Probability,Market Basket Analysis,Recommender System,Transaction Records, |
出版年 : | 2024 |
學位: | 碩士 |
摘要: | 利用交易紀錄進行個人化推薦來協助顧客選擇是精準行銷裡相當普遍的手段。然而,目前超市透過貨架陳列、DM宣傳、現場銷售人員和POP廣告等等的管道,在現場向顧客推薦產品來增加顧客購買率的方式不足以達到個人化推薦。此外購物籃分析是常運用的推薦策略之一,利用交易紀錄分析顧客之間選購產品的關聯,藉此掌握顧客個人可能感興趣但尚未或久未購買過的產品。超市在現場也運用購物籃分析進行產品推薦的策略,例如:綑綁銷售,但如何善加利用來做現場個人化推薦,透過更精準的行銷來滿足的顧客個人需求。
除了基於購物籃分析顧客間產品選購的推薦外,根據William B. Dodds的研究,顧客的個人品牌選擇也是顧客在購買產品時要考量的因素之一。根據上述超市現場個人化推薦和個人品牌選擇的議題,提出在顧客進入或離開超市現場可利用超市APP和自助結帳系統,並加入考量顧客的個人品牌選擇來推薦產品,應有助於產生更準確的推薦策略。故本論文將探討利用超市交易紀錄做購物籃分析後,如何加入個人品牌選擇來實現一類別品牌產品的現場推薦。 本論文研究的目標就是幫助超市做現場推薦顧客未購買過的一類別品牌產品,主要研究問題(P)、相應挑戰(C)和新提出並設計解決方案(M)為以下幾點: P1 超市交易紀錄萃取一類別產品的顧客個人品牌選擇問題: 文獻中常用一類別產品下個人品牌選擇機率來評估顧客的品牌選擇,普遍採用交易紀錄中購買品牌的相對次數。但是顧客其他的消費行為可能也會影響個人品牌選擇機率,例如:顧客長期購買的品牌,那下次再購買的機率也會較高等等。若增加其他的消費行為勢必計算顧客的個人品牌選擇機率方式也會不同。我們具體細分為兩個子問題。 P1.1 從交易紀錄萃取消費行為問題: 需要從一類別產品的交易紀錄中找出顧客有哪些重點消費行為可運用來計算顧客的個人品牌選擇機率? C1.1 文獻中關於品牌選擇,反映在多元消費行為因素,各因素間相對的重要性有待量化,都造成萃取顧客交易紀錄中品牌選擇機率的挑戰。 M1.1 提出包括四個代表性消費行為的LRFM-PB模型。文獻中的LRFM模型為基礎分別為顧客的四個消費行為關係長度(L)、最近購買日(R)、購買頻率(F)和購買金額(M),但在不同情境LRFM的定義會有不同。我們是針對一類別產品下設計各個品牌的LRFM,故變化現行文獻中購買頻率(F)和購買金額(M)的定義為:F為消費的總次數和購買金額(M)為總花費金額。由萃取一類別產品下各品牌的LRFM顧客消費行為因素指標,作為估算品牌選擇機率的基礎。 P1.2 設計個人品牌價值指標並計算品牌選擇機率問題: 個人品牌價值指標可幫助計算品牌選擇機率,但文獻中鮮少以交易紀錄的消費行為所設計顧客個人的品牌價值指標與估算辦法。因此,採用的消費行為因素是否能作為一類別產品的品牌價值指標? C1.2 不易釐清量化後顧客多元且重要程度不同的消費行為因素對個人品牌價值的影響。要基於個人交易紀錄的消費行為模型中消費行為因素指標來設計顧客個人的品牌價值指標與估算品牌選擇機率,具挑戰性。 M1.2 基於顧客終身價值(CLV)是消費行為下的產物且能分析目標客群,提出顧客終身價值-產品品牌(CLV-PB)為品牌價值指標,定義為一類別產品中顧客對品牌的終身價值。原因有三個:(1)延續CLV能找出品牌或企業重視顧客、(2)假設CLV和購買意圖是正相關和(3)CLV可用LRFM來表示。CLV-PB計算為採一類別產品交易紀錄中,顧客對品牌的LRFM搭配問卷調查的層級分析法以問詢領域專家賦予LRFM的權重。 P2 群體購物行為有無作個人品牌選擇推薦一類別品牌產品資訊依據問題: 我們的目標是利用群體購物籃分析和個人品牌選擇機率來幫助超市做現場推薦顧客未購買過的一類別品牌產品,但是兩者之間是否存在關連性且應如何結合來做現場推薦策略? C2 須了解、掌握群體購物行為與個人品牌選擇的關聯性以便運用。 M2 以分群後的購物籃分析並運用信賴度乘以個人品牌選擇機率估計顧客在購買特定產品後,會購買某品牌產品的機率為超市推薦品牌產品策略依據,來提出具品牌面向的購物籃分析(MBAwB)。 本論文提出CLV-PB所計算個人品牌選擇機率和具品牌面向的購物籃分析相結合,簡稱MBAwB(CLV-PB)。為了驗證MBAwB(CLV-PB)在推薦品牌產品的有效性,把顧客購買過的品牌產品切割為50%訓練集和50%測試集,並與另外兩個推薦產品的品牌方法比較:以熱門的品牌(MBAwB(Random))和品牌購買次數(MBAwB(F))為推薦依據。利用松青超市在家庭清潔用品類中59位顧客的實際數據進行實驗。推薦得分率(10個品牌產品)在MBAwB(CLV-PB)為16.9%,且在推薦1個品牌產品就能發掘是顧客有興趣的。在個人化推薦下MBAwB(CLV-PB)的F1-score比另兩個好且高達47.5%,故MBAwB(CLV-PB)對超市現場個人化推薦策略是具應用潛力的。 本論文的貢獻因MBAwB(CLV-PB),新產生價值如下: 貢獻1. 歸類超市一類別產品交易紀錄中,顧客對品牌的消費行為LRFM。對超市的價值在於能掌握和利用顧客在一類別產品下對品牌的四個消費行為。 貢獻2. 新設計以加權LRFM線性總和的CLV-PB為顧客的品牌價值來計算品牌選擇機率。加入時間(L、R)與花費(M)、修正購買品牌次數(F)的定義和考量行為因素的重要性來調整單用購買品牌次數對一類別產品下的品牌選擇機率計算。 貢獻3. 新設計具品牌面向的購物籃分析(MBAwB)做為現場一類別品牌產品推薦策略。對顧客的價值在於現場能注意到其他同好群中自身未購買的產品且符合品牌喜好;對超市的價值在於發掘顧客喜好但未購買過的品牌產品,並透過文獻中提到的提高顧客個人化推薦準確度來增加的現場銷售量與滿意度。 The use of transaction records for personalized recommendations to assist customers in making choices is a common practice in precision marketing. However, the current methods used by supermarkets fall short in providing on-site personalized recommendation. Additionally, market basket analysis (MBA) is a common recommendation strategy that examines the associations between products purchased by customers, identifying products of potential interest that customers haven’t yet purchased or haven’t bought for a while. Thus, the challenge lies in effectively utilizing this method for on-site personalizing recommendations catering to individual customer needs to achieve precise marketing. In addition to recommending products based on MBA of customer product selections, a customer’s individual brand choice is alse one of the factors considered when making a purchase in William B. Dodds’ research. Building on the issues of on-site personalized recommendations in supermarket and individual brand choice, a proposal is made to utilize the supermarket app and self-checkout system when customers enter or leave the store. By incorporating considerations of a customer''s individual brand choice, this approach aims to enhance the accuracy of recommendation strategy. The main goal of our thesis is to assist supermarkes in making on-site recommendation for brand products that customers haven’t purchased within a category. The primary research problems (P), associated challenges (C), and newly proposed methods (M) are listed below: P1 The problem of brand choice extration. In literature, the probability of individual brand choice is commonly used to assess a customer’s brand choice within a category, and it often employs the relative frequency of purchased brands in transaction records. However, other consumer behaviors might also influence the probability of individual brand choice. We can categorize this into two specific sub-problems. P1.1 Which key consumer behaviors can be identified to calculate the probability of individual choice in transcation records? C1.1 The literature reflects that in terms of brand choice, the relative importance among various factors of consumer behavior needs quantification, posing a challenge in extracting the probability of brand choice from customer transaction records M1.1 We propose the LRFM-PB model incorporating four representative consumer behavior Length (L), Recency (R), Frequency (F), and Monetary (M) within a category. P1.2 Although the individual brand value aids in calculating brand choice probabilities, the literature has limited examples of customer-specific brand value and estimation methods based on transaction records of consumer behavior. Can consumer behavior factors that we selected serve as a brand value? C1.2 Quantifying the impact of diverse consumer behaviors on individual brand value, based on transaction records, presents a challenging task for estimating brand choice probabilities. M1.2 Customer Lifetime Value (CLV), the derivative of consumer behavior, allows for the analysis of a business''s target audience. The proposed Customer Lifetime Value - Product Brand (CLV-PB) is defined as the brand value within a category. P2 Is there a correlation between the grouped MBA and brand choice and how should they be combined to formulate an on-site recommendation strategy? C2 Understanding and mastering the correlation between grouped behaviors and individual brand choices is crucial for effective utilization. M2 We propose Market basket analysis with brand (MBAwB). The paper introduces MBAwB(CLV-PB), a method combining personal brand choice with CLV and MBA. To validate the effectiveness of MBAwB(CLV-PB), we campare with two other methods: MBAwB(Random) and MBAwB(F). The experiment utilizes actual data from 59 customers in the household cleaning products. Under personalized recommendations, MBAwB(CLV-PB) achieves an F1-score 47.5% higher than the other two methods, indicating the significant potential application of MBAwB(CLV-PB). Contribution1. Analyzing customer transactions for a specific product category in a supermarket using LRFM to understand brand consumption behavior. Contribution2. The newly designed CLV-PB calculates brand choice probabilities based on customer brand value using a weighted LRFM linear sum. Contribution 3. The new design employs MBAwB as an on-site strategy for recommending products within a category. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91461 |
DOI: | 10.6342/NTU202400140 |
全文授權: | 同意授權(限校園內公開) |
顯示於系所單位: | 工業工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-112-1.pdf 目前未授權公開取用 | 3.09 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。