基於超市交易紀錄推論消費者之新產品個人偏好

鄧惠平

doi:10.6342/NTU201702441

山西大学，您好！

感谢采购，人社期刊库｜人社论文库，期间：2024/01/01~2024/12/31

感谢试用，科学期刊库｜科学论文库，期间：2024/10/20~2025/01/20

浏览高级检索

关闭
检索检索高级检索出版物检索
浏览期刊主题浏览核心期刊浏览期刊拼音浏览

关闭

查询符号(半角)	查询符号意义说明
空格	表示为「AND」两个查询词之交集
双引号 ( " " )	片语以双引号标示开始及结束，而且只寻找出现顺序相同的字词，例： " image process "
?	表示一个字母切截，输入两个?表两个字母，依此类推，例：输入「Appl?」，查得结果应为appl e , appl y … ( 常用于英文字查询 )
*	表示不限字母切截，由0~n. 例：输入「appl*」，查得结果应为appl e , appl es , appl y , appl ied , appl ication … ( 常用于英文字查询 )
AND、OR、NOT	布尔逻辑组合关键字，用来扩大或缩小查询范围的技巧。 (1) AND ：缩小查询范围 (2) OR ：扩大查询范围(3) NOT：排除不相关的范围

DOI 是数位物件识别码（ D igital O bject I dentifier ）的简称，
为物件在网路上的唯一识别码，可用于永久连结并引用目标物件。

使用DOI作为永久连结

每个DOI号前面加上「 http://dx.doi.org/ 」便成为永久网址。
如以DOI号为 10.5297/ser.1201.002 的文献为例，此文献的永久连结便是： http://dx.doi.org/ 10.5297/ser.1201.002 。
日后不论出版单位如何更动此文献位置，永久连结所指向的位置皆会即时更新，不再错失重要的研究。

引用含有DOI的文献

有DOI的文献在引用时皆应同时引用DOI。若使用APA、Chicago以外未规范DOI的引用格式，可引用DOI永久连结。

DOI可强化引用精确性、增强学术圈连结，并给予使用者跨平台的良好使用经验，目前在全世界已有超过五千万个对象申请DOI。如想对DOI的使用与概念有进一步了解，请参考（）。

书目管理工具

相关连结

问题回报

全文下载

基於超市交易纪录推论消费者之新产品个人偏好

Infer Individual Customer Preference for a New Product Based on Supermarket Transaction History

邓惠平 , 硕士　　导师：张时中　　共同指导教授：黄亭凯

英文 DOI： 10.6342/NTU201702441

购买频率；新产品偏好推理；本体论模型；产品个人偏好；预测潜在消费者；萃取产品特徵；语义模型；交易纪录； Frequency of Choice (FOC) ； New Product Preference Inference ； Ontology Model ； Personal Preference for Products (PPFP) ； Potential Customer Identification ； Product Feature Extraction ； Semantic Model ； Transaction Data

0 个人觉得这篇文章赞

摘要 │ 参考文献 (95) │ 国际计量

摘要〈TOP〉

市场上的产品不断推陈出新。因着消费者有越来越多的产品选择性，且随着数据分析的进步，今预测消费者对新产品的个人偏好在精准营销中越显其重要。本论文研究以超市的行销为范畴，藉由分析交易数据萃取消费者的偏好。

现有的交易数据分析方法，主要是基於他人的纪录来预测消费者对产品的偏好。基於交易纪录的个人化偏好推理鲜少考虑。应如何从消费者的个人历史交易纪录，预测其对新产品的个人偏好？下列为回答此问题相对应的挑战：
（i）交易纪录常以表格中的数字呈现，应如何从中萃取个人偏好资讯？
（ii）面对新产品与现有产品之间的差异，且未有新产品的购买纪录下，我们应如何利用消费者交易纪录中的产品偏好，推断其对於新产品的偏好及购买决策？

本论文设计了一套以语义为基础的方法─基於交易纪录的个人偏好推理引擎（TPIE）─以应对问题和挑战：
（1）从交易纪录中萃取个人偏好资讯
Samuelson的显示性偏好理论（Revealed preference theory, 1972）认为消费者的偏好可由过去的购买纪录、选择观察而得。当某消费者满意於某一项产品，他极有可能重复购买。基於此考量，我们认为消费者对较常购买的产品，有较高的偏好。因此从交易纪录中，我们计算消费者选择某产品的频率，视其为消费者对该产品的个人偏好（Personal Preference for Products, PPFP）。然而，受限於不同产品类别间购买频率的差异，本论文一次仅分析一种类别的产品。

（2）基於本体论（Ontology）的产品特徵萃取方法及衡量产品相似度
为了预测消费者对新产品的购买决定，我们参考他们对「与新产品相似的既有产品」的偏好。为了定义两产品之间的相似性，我们采用本体论模型，从「产品资料档」中萃取产品特徵（feature）。并依照资料档中所提供的资讯，赋予每项特徵其产品属性（attribute）及属性值（attribute value），以藉由计算属性之间的差异来衡量两个产品之间的相似度。然而，产品的每项特徵重要性不完全相同；因此，属性值之间的级距（interval）应有不同的权重。为决定各属性之属性值的级距，我们从历史消费纪录观察每位消费者最喜好的三样产品，其中未改变的属性所属的产品特徵，即是消费者最看重的产品特徵。最後，加总所有消费者看重每项产品特徵的次数作为产品特徵重要性的比例，并将其视为属性的值（attribute value）级距的权重。最後，我们以产品间，特徵属性值之差异加权总和，计算产品相似度。

（3）预测新产品的个人偏好
我们假设两样产品越相似，消费者对其偏好也越接近。为了推断消费者对於新产品的偏好，我们从消费者购买过的产品中，挑出与新产品最相似的产品。计算「消费者对该产品的PPFP」与「产品相似度」的乘积，视其为新产品预期的PPFP。求得新产品的预期PPFP後，我们需要制定一个阈值来决定是否视消费者为新产品的潜在消费者；例如：设定阈值为「消费者过去购买过的产品中，最喜爱的产品之PPFP的一半」。

为了评估TPIE预测潜在消费者的有效性，我们利用松青超市的两组产品类别的实际数据进行实验。第一组数据为牛奶类，共有65个牛奶产品。我们挑选其中一样产品，视其为新产品，并利用其他64个产品的偏好，判断消费者的对於新产品的偏好。每样牛奶产品轮流作为新产品，最後取65样产品的结果平均值，我们正确地预测约75.01%的消费者是否为潜在顾客。我们的实验结果优於利用群体消费者偏好预测方法所得的结果。第二组数据─冰棒类，预测的准确率约为57.14%。由於数据稀疏问题，预测结果较利用群体偏好预测结果差。

本论文根据产品本体论方法所设计的TPIE个人偏好推论引擎，具体贡献在於以消费者购买纪录，预测其对新产品的个人偏好购买决策。具体贡献包括：
（1）利用消费者的购买频率，从交易数据中萃取个人偏好；
（2）基於产品的本体论模型，找出产品的特徵和属性，以衡量产品之间的相似度；
（3）假设若消费者对於一新产品的偏好程度达过去最喜爱之产品的一半时，即可视其为新产品的潜在购买者。
（4）正确预测75.01%的消费者是否为牛奶类新产品的潜在消费者。

并列摘要〈TOP〉

New products are introduced into markets at a fast pace. With customers being increasingly selective in product choices and with the advancement of data analytics revolution, personal preference prediction has grown in importance in precision marketing of new products. In this thesis research, we consider the problem domain of supermarket marketing and sales, with the transaction data analysis as our approach to extract customers’ preferences.

Current methods of transaction data analysis predict a customer’s preference for products based mainly on other customers’ records. Preference inference based on personal transaction data is barely considered. How are we going to predict a customer’s preference for a new product only based on his/her own transaction history? The challenges to address this question are as follows: 
(i) Transaction data is usually numbers in a tabular form. How do we extract personal preference information from the transaction data?
(ii) A new product is more or less different from the existing products, and there is no purchase record of the new product yet. How could we infer a customer’s preference and purchase decision for a new product by exploiting his or her own preferences for products in the transaction database?

This thesis designs an innovative semantic-based methodology – Transaction data-based Personal preference Inference Engine (TPIE) in response to the problems and challenges: 
(1) Personal Preference Information Extraction from Transaction Data
Exploiting the revealed preference theory by Paul Samuelson, 1972, the preferences of a customer can be revealed by the purchasing choices. It is based on  the fact that when a customer is satisfied with a product, he/she would repeat the purchase; thus, the more frequent a product is purchased, the better one likes it. From each customer’s transactions, we calculate the frequency every product is purchased, and consider it to be the customer’s personal preference for products (PPFP). However, due to the purchase frequency differences among categories, we analyze category by category.

(2) Ontology-based Feature Extraction and Similarity Measurement
To predict customers’ purchase decisions for a new product, we infer on their preferences for existing products which are similar to the new product. In order to define the similarity between two products, we adopt an ontology model-based approach to derive product features from the product profile table. The attributes and the associated attribute values of each feature are then assigned based on information in the product profile table. Similarity between two products is calculated by measuring the attribute differences. Since the importance of individual product features varies, the intervals of attribute values of individual features should be weighted differently. To decide the intervals of attribute values, we seek out each customer’s most cared feature by sorting through the unchanged attribute of their three most favored products. And count the feature each customer cares the most as the attribute interval weight ratios of importance of each feature. We then calculate the product similarity by the weighted sum of feature attribute value differences between the new product and the existing products.

(3) Personal Preference Prediction of New Products
We reason that the more similar two products are, the closer customer’s preferences for them will be. To infer a customer’s preference for the new product, we identify the similarity between the new product and the purchased products. The customer’s expected PPFP for the new product is a sum of PPFP multiplied by the similarity of each purchased product. Finally, we select a threshold of PPFP for deciding whether one customer may be considered a potential buyer of the new product, for example, half of the PPFP of his/her favorite product from the purchase history.

To evaluate the effectiveness of predicting potential customers of TPIE, two real datasets from Matsusei Supermarket are chosen for experiment. First dataset is tested on milk category, consisting of 65 products. We divide the database into the training dataset of 64 products and a test product as the selected new product, i.e., each milk product takes turn to be the new product. In the validation experiment, the outcome outperforms the group preference approach. We correctly predict 75.01% of the customers that whether he/she is a potential customer to the new milk product. The second dataset is tested on popsicle category, with 57.14% accuracy rate. However, the result is worse than group preference approach because of data sparsity.
	The contribution of this thesis is an innovative design of a semantic-based ontological methodology, TPIE, for predicting personal preferential purchase decision to a new product based on personal transaction data. Specifically, contributions include:
(1)	Personal preference extraction from the transaction data, based on customer’s frequency of choice;
(2)	Representing each product based on ontology model product features extraction and attributes assignment for similarity measurements;
(3)	Assume when a customer’s expected new product PPFP reaches half of the PPFP of the most favored purchased product, he/she is considered to be a potential customer of the new product.
(4)	Achievement of identifying whether a customer is potential customer of a new milk product with 75.01% correction rate.

参考文献 ( 95 ) 〈TOP〉

[AmF99] Bernd Amann and Irini Fundulaki. “Integrating Ontologies and Thesauri to Build RDF Schemas.” The Third European Conference on Research and Advanced Technology for Digital Libraries. 234–253. 1999.
连结：
[ASS15] Robert Arp, Barry Smith, and Andrew D. Spear. “Building Ontologies with Basic Formal Ontology.” MIT Press. 2015.
连结：
[BaS97] Marko Balabanović and Yoav Shoham. “Content-Based, Collaborative Recommendation.” March 1997/Vol. 40, No. 3 Communications of the ACM. 1997.
连结：
[Bhu11] Nadia Bhuiyan. “A framework for successful new product development.” Concordia University. 2011.
连结：
[BML89] Stanley J. Baran, Jin Ja Mok, Mitchell Land and Tae Young Kang. “You are What You Buy: Mass-Mediated Judgement of People’s Worth.” Journal of Communication. 1989.
连结：

国际计量

〈TOP〉

E-mail ：

文章公开取用时，将寄通知信至您填写的信箱地址

E-mail ：

使用DOI作为永久连结

引用含有DOI的文献

DOI可强化引用精确性、增强学术圈连结，并给予使用者跨平台的良好使用经验，目前在全世界已有超过五千万个对象申请DOI。 如想对DOI的使用与概念有进一步了解，请参考 （ ） 。

DOI可强化引用精确性、增强学术圈连结，并给予使用者跨平台的良好使用经验，目前在全世界已有超过五千万个对象申请DOI。如想对DOI的使用与概念有进一步了解，请参考（）。