WO2018059015A1 - 一种基于交易数据的客户分类方法及其系统 - Google Patents

一种基于交易数据的客户分类方法及其系统 Download PDF

Info

Publication number
WO2018059015A1
WO2018059015A1 PCT/CN2017/088724 CN2017088724W WO2018059015A1 WO 2018059015 A1 WO2018059015 A1 WO 2018059015A1 CN 2017088724 W CN2017088724 W CN 2017088724W WO 2018059015 A1 WO2018059015 A1 WO 2018059015A1
Authority
WO
WIPO (PCT)
Prior art keywords
transaction data
customer
purchase
tree
node
Prior art date
Application number
PCT/CN2017/088724
Other languages
English (en)
French (fr)
Inventor
陈小军
明勇
张文斌
彭思
黄哲学
张超
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Publication of WO2018059015A1 publication Critical patent/WO2018059015A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/01—Customer relationship services
    • G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
    • G06Q30/016—After-sales

Definitions

  • the invention relates to the field of data mining, in particular to a customer classification method based on transaction data and a system thereof.
  • Customer classification based on transaction data plays an important role in the retail industry. It analyzes customer purchase behavior through the products purchased by customers and provides retail decision-making marketing strategies for retail enterprises. When a customer purchases a product, he or she usually purchases multiple products at a time. Different products may have different products, but there may be products that are purchased. Therefore, the purpose of customer classification based on transaction data is to find out customers with similar purchase behaviors, and analyze the combination of the products that these customers frequently purchase from the retail records, thereby facilitating the matching of products, providing promotions, logistics, and the like.
  • the current customer classification method has based on the customer's geographic location and demographic characteristics information and hierarchical clustering based on transaction data, wherein hierarchical clustering based on transaction data is the main method.
  • the traditional clustering method based on customer's geographical location and demographic information has the problem that the data of these indicators are not easy to obtain, and the results of clustering based on these static feature attributes do not necessarily have the same purchasing behavior.
  • Hierarchical clustering based on transaction data needs to be calculated between two pairs in the distance calculation. The time complexity of the square level cannot effectively process the massive data in the current big data era, and when facing the data with higher dimensions, Processing efficiency will also drop rapidly.
  • the object of the present invention is to provide a customer classification method based on transaction data and a system thereof, aiming at solving the problem that the conventional method cannot process the current big data because the computational complexity is too high.
  • the invention provides a customer classification method based on transaction data, the method comprising:
  • the collected transaction data is preprocessed, and the preprocessed transaction data is separately coded according to layers;
  • All customer purchase trees created are clustered until the cluster center is no longer changed or the change is less than a preset threshold to obtain the result of the customer classification.
  • the transaction data includes: a membership card number, a transaction date, a transaction time, a product name, a product category name, and a sales amount.
  • the pre-processing includes performing a process of ignoring the transaction data with the missing information and performing unique processing on the transaction data of the information conflict, wherein the step of encoding the pre-processed transaction data separately by layer includes :
  • the pre-processed transaction data is respectively encoded by the layer name of the product name and the product category name, and the encoding method is coded in order from the first to the smallest according to the number, and the layers are independent from each other.
  • the step of creating a customer purchase tree for each customer according to the encoded hierarchical information specifically includes:
  • the transaction data is grouped by membership card number, and the consumption frequency, consumption amount, and latest consumption time of each commodity are counted to form an RFM value of each commodity;
  • the commodity purchased by each customer is constructed according to the commodity category, wherein the node of the purchase tree includes a root node, a leaf node and an internal node, the leaf node is a commodity name node, and the internal node is a commodity category name. node;
  • the transaction data that is not in the purchase tree is added to the purchase tree.
  • the step of clustering all the customer purchase trees that are created until the cluster center is no longer changed or the change is less than a preset threshold, so as to obtain the result of the customer classification specifically includes:
  • Update the initial center add the values of the corresponding nodes of the purchase tree divided into the same cluster, and take the average value as the RFM value of the corresponding node of the new center;
  • the present invention also provides a customer classification system based on transaction data, the system comprising:
  • a data collection module for collecting transaction data from a retail enterprise transaction database
  • a data preprocessing module configured to preprocess the collected transaction data, and encode the preprocessed transaction data by layers
  • a purchase tree generation module configured to create a customer purchase tree for each customer according to the encoded hierarchical information
  • the customer classification module is configured to cluster all the created customer purchase trees until the cluster center does not change or the change is less than a preset threshold to obtain the result of the customer classification.
  • the transaction data includes: a membership card number, a transaction date, a transaction time, a product name, a product category name, and a sales amount.
  • the pre-processing includes performing the process of ignoring the missing transaction data and the transaction data of the information conflict.
  • the data pre-processing module is specifically configured to:
  • the pre-processed transaction data is respectively encoded by the layer name of the product name and the product category name, and the encoding method is coded in order from the first to the smallest according to the number, and the layers are independent from each other.
  • the purchase tree generating module is specifically configured to:
  • the transaction data is grouped by membership card number, and the consumption frequency, consumption amount, and latest consumption time of each commodity are counted to form an RFM value of each commodity;
  • the commodity purchased by each customer is constructed according to the commodity category, wherein the node of the purchase tree includes a root node, a leaf node and an internal node, the leaf node is a commodity name node, and the internal node is a commodity category name. node;
  • the transaction data that is not in the purchase tree is added to the purchase tree.
  • the customer classification module is specifically configured to:
  • Update the initial center add the values of the corresponding nodes of the purchase tree divided into the same cluster, and take the average value as the RFM value of the corresponding node of the new center;
  • the technical solution provided by the invention can improve the effect of clustering, fully utilize the RFM information of the commodity and the hierarchical information carried, and greatly improve the practical use value.
  • the k-means principle can be used to process large-scale data efficiently and quickly, and the traditional method cannot solve the current big data because the computational complexity is too high.
  • FIG. 1 is a flow chart of a customer classification method based on transaction data according to an embodiment of the present invention
  • FIG. 2 is a structural diagram of an ambiguous commodity according to an embodiment of the present invention.
  • FIG. 3 is a diagram showing an example of a purchase tree in an embodiment of the present invention.
  • FIG. 4 is a schematic diagram showing the internal structure of a customer classification system 10 based on transaction data according to an embodiment of the present invention.
  • FIG. 1 is a flowchart of a customer classification method based on transaction data according to an embodiment of the present invention.
  • step S1 transaction data is collected from the retail enterprise transaction database.
  • the transaction data includes: a membership card number, a transaction date, a transaction time, a product name, a product category name, and sales.
  • some information in the transaction data information provided by the retail enterprise is what we need, and some information is not needed. We only need to extract useful transaction data information, including membership card number, transaction date, transaction time, product name, product category name and sales.
  • the membership card number the retail enterprise handles the identification number of the membership card for the customer, and the product purchased by the same membership card number is considered to be the product purchased by the same customer, and the program distinguishes different customers according to the membership card number;
  • Transaction date The specific date when the product was purchased, including the date of the year, month and day;
  • Trading time The specific time when the goods are purchased, including time, minute and second information;
  • Product Name The name of the purchased product
  • Product category name category information such as subcategory, medium class, and large category to which the product belongs;
  • Sales The total amount of goods purchased by the customer.
  • step S2 the collected transaction data is preprocessed, and the preprocessed transaction data is separately encoded in layers.
  • the pre-processing includes performing ignoring processing on the missing transaction data and uniquely processing the transaction data in which the information conflicts, wherein the pre-processed transaction data is separately coded by layer.
  • the steps specifically include:
  • the pre-processed transaction data is respectively encoded by the layer name of the product name and the product category name, and the encoding method is coded in order from the first to the smallest according to the number, and the layers are independent from each other.
  • the transaction data in which the information is missing includes, for example, transaction data in which the membership card number is missing, transaction data in which the transaction date is missing, transaction data in which the transaction time is missing, transaction data in which the product name is missing, transaction data in which the product category name is missing, and transaction data in which the product category name is missing.
  • transaction data in which the membership card number is missing includes, for example, transaction data in which the transaction date is missing, transaction data in which the transaction time is missing, transaction data in which the product name is missing, transaction data in which the product category name is missing, and transaction data in which the product category name is missing.
  • Transaction data of information conflicts, for example, including ambiguous and ambiguous transaction data.
  • some commodities in the transaction data have ambiguity problems.
  • the commodity plate belongs to both the small kitchen and the porcelain.
  • the unique treatment is carried out, and the porcelain is removed, and only belongs to the kitchen.
  • all product names and their associated hierarchical category names are hierarchically and independently globally encoded.
  • step S3 a customer purchase tree is created for each customer based on the encoded hierarchical information.
  • the step S3 of creating a customer purchase tree for each customer according to the encoded hierarchical information specifically includes:
  • the transaction data is grouped by membership card number, and the consumption frequency, consumption amount, and latest consumption time of each commodity are counted to form an RFM value of each commodity;
  • the commodity purchased by each customer is constructed according to the commodity category, wherein the node of the purchase tree includes a root node, a leaf node and an internal node, the leaf node is a commodity name node, and the internal node is a commodity category name. node;
  • the transaction data that is not in the purchase tree is added to the purchase tree.
  • the RFM information of each customer is extracted using the widely used RFM model, while the hierarchical information of the data is retained, and the purchase information of each customer is represented by a tree structure.
  • the RFM model is an analytical model for customer relationship management (CRM), which is widely used. It is an important tool and means to measure customer value and customer profitability.
  • CRM customer relationship management
  • the transaction data of the retail enterprise is a complete set of transaction records, and the unremoved goods are extracted from the transaction record, and a commodity tree is established according to the hierarchical information carried therein, and all transaction data is pressed.
  • the membership card number is divided into groups, and the sliding window is set for the time slice.
  • the FRM value is calculated as the RFM value of the leaf node of the purchase tree, and then the RFM value of the leaf node is normalized according to the maximum and minimum values, and the RFM of the internal node is calculated.
  • the linearity of its child nodes is divided by the number of child nodes of the corresponding nodes in the commodity tree, thus establishing a purchase tree for each user from the bottom up.
  • the transaction data is grouped by membership card number, and the consumption frequency, consumption amount, and latest consumption time of each commodity are counted, that is, the RFM value of each commodity.
  • Each customer purchases a commodity to construct a purchase tree according to the product category.
  • the root node of the purchase tree is a root node
  • the leaf node is a commodity name node
  • the internal node is a category node
  • each node contains an RFM value.
  • the specific construction method of the purchase tree is to scan the transaction data in turn, and match the obtained product category information, the product name, and the purchase tree.
  • the leaf nodes of the purchase tree all contain RFM value information, and the RFM value of the internal node is linearly solved according to the RFM of the child node.
  • step S4 all the created customer purchase trees are clustered until the cluster center is no longer changed or the change is less than a preset threshold to obtain the result of the customer classification.
  • the step S4 of performing clustering on all the customer purchase trees that are created until the cluster center is no longer changed or the change is less than a preset threshold to obtain the result of the customer classification includes:
  • Update the initial center add the values of the corresponding nodes of the purchase tree divided into the same cluster, and take the average value as the RFM value of the corresponding node of the new center;
  • all the N purchase trees are divided into k classes, and k users of the class can be obtained, and the purchase tree and the initial center distance are sequentially calculated, and the value of the leaf node in the purchase tree is calculated, and the purchase is performed.
  • the obtained purchase tree has a distance value other than the root node, and the update center is updated to the purchase tree of the same cluster, and the value of the corresponding node is Add the average value as the RFM value of the corresponding node of the new center. If the corresponding node does not exist, assign a minimum value to it.
  • the k-means principle in the classification process of the customer purchase tree, the k-means principle can be used to efficiently and quickly process large-scale data, and the conventional method cannot solve the current big data because the computational complexity is too high.
  • the distance between the purchase tree and the center is calculated. By continuously iteratively calculating the distance between the purchase tree and the center until the center no longer changes or the change is lower than a certain threshold, the result of the user classification is finally obtained.
  • the invention provides a customer classification method based on transaction data.
  • the RFM model is used to extract the transaction information of the customer, and at the same time, the customer purchase tree is created according to the hierarchical information of the data, and then created by using the k-means pair.
  • the customer buys the tree for clustering.
  • the technical solution provided by the invention can improve the effect of clustering, fully utilize the RFM information of the commodity and the hierarchical information carried, and greatly improve the practical use value, and at the same time, in the process of classifying the customer purchase tree, using k
  • the -means principle can process large-scale data efficiently and quickly, and solves the problem that the traditional method cannot handle the current big data because the computational complexity is too high.
  • the customer classification system 10 based on transaction data mainly includes a data collection module 11, a data preprocessing module 12, a purchase tree generation module 13, and a customer classification module 14.
  • the data collection module 11 is configured to collect transaction data from a retail enterprise transaction database.
  • the transaction data includes: a membership card number, a transaction date, a transaction time, a product name, a product category name, and sales.
  • some information in the transaction data information provided by the retail enterprise is what we need, and some information is not needed. We only need to extract useful transaction data information, including membership card number, transaction date, transaction time, product name, product category name and sales.
  • the membership card number the retail enterprise handles the identification number of the membership card for the customer, and the product purchased by the same membership card number is considered to be the product purchased by the same customer, and the program distinguishes different customers according to the membership card number;
  • Transaction date The specific date when the product was purchased, including the date of the year, month and day;
  • Trading time The specific time when the goods are purchased, including time, minute and second information;
  • Product Name The name of the purchased product
  • Product category name category information such as subcategory, medium class, and large category to which the product belongs;
  • Sales The total amount of goods purchased by the customer.
  • the data pre-processing module 12 is configured to pre-process the collected transaction data, and encode the pre-processed transaction data into layers.
  • the pre-processing includes performing the process of ignoring the missing transaction data and the transaction data of the information conflict.
  • the data pre-processing module 12 is specifically configured to:
  • the pre-processed transaction data is respectively encoded by the layer name of the product name and the product category name, and the encoding method is coded in order from the first to the smallest according to the number, and the layers are independent from each other.
  • the transaction data in which the information is missing includes, for example, transaction data in which the membership card number is missing, transaction data in which the transaction date is missing, transaction data in which the transaction time is missing, transaction data in which the product name is missing, transaction data in which the product category name is missing, and transaction data in which the product category name is missing.
  • transaction data in which the membership card number is missing includes, for example, transaction data in which the transaction date is missing, transaction data in which the transaction time is missing, transaction data in which the product name is missing, transaction data in which the product category name is missing, and transaction data in which the product category name is missing.
  • Transaction data of information conflicts, for example, including ambiguous and ambiguous transaction data.
  • some commodities in the transaction data have ambiguity problems.
  • the commodity plate belongs to both the small kitchen and the porcelain.
  • the unique treatment is carried out, and the porcelain is removed, and only belongs to the kitchen.
  • all product names and their associated hierarchical category names are hierarchically and independently globally encoded.
  • the purchase tree generation module 13 is configured to create a customer purchase tree for each customer according to the encoded hierarchical information.
  • the purchase tree generating module 13 is specifically configured to:
  • the transaction data is grouped by membership card number, and the consumption frequency, consumption amount, and latest consumption time of each commodity are counted to form an RFM value of each commodity;
  • the commodity purchased by each customer is constructed according to the commodity category, wherein the node of the purchase tree includes a root node, a leaf node and an internal node, the leaf node is a commodity name node, and the internal node is a commodity category name. node;
  • the transaction data that is not in the purchase tree is added to the purchase tree.
  • the RFM information of each customer is extracted using the widely used RFM model, while the hierarchical information of the data is retained, and the purchase information of each customer is represented by a tree structure.
  • the RFM model is an analytical model for customer relationship management (CRM), which is widely used. It is an important tool and means to measure customer value and customer profitability.
  • CRM customer relationship management
  • the transaction data of the retail enterprise is a complete set of transaction records, and the unremoved goods are extracted from the transaction record, and a commodity tree is established according to the hierarchical information carried therein, and all transaction data is pressed.
  • the membership card number is divided into groups, and the sliding window is set for the time slice.
  • the FRM value is calculated as the RFM value of the leaf node of the purchase tree, and then the RFM value of the leaf node is normalized according to the maximum and minimum values, and the RFM of the internal node is calculated.
  • the linearity of its child nodes is divided by the number of child nodes of the corresponding nodes in the commodity tree, thus establishing a purchase tree for each user from the bottom up.
  • the transaction data is grouped by membership card number, and the consumption frequency, consumption amount, and latest consumption time of each commodity are counted, that is, the RFM value of each commodity.
  • Each customer purchases a commodity to construct a purchase tree according to the product category.
  • the root node of the purchase tree is a root node
  • the leaf node is a commodity name node
  • the internal node is a category node
  • each node contains an RFM value.
  • the specific construction method of the purchase tree is to scan the transaction data in turn, and match the obtained product category information, the product name, and the purchase tree.
  • the leaf nodes of the purchase tree all contain RFM value information, and the RFM value of the internal node is linearly solved according to the RFM of the child node.
  • the customer classification module 14 is configured to cluster all the created customer purchase trees until the cluster center does not change or the change is less than a preset threshold to obtain the result of the customer classification.
  • the customer classification module 14 is specifically configured to:
  • Update the initial center add the values of the corresponding nodes of the purchase tree divided into the same cluster, and take the average value as the RFM value of the corresponding node of the new center;
  • all the N purchase trees are divided into k classes, and k users of the class can be obtained, and the purchase tree and the initial center distance are sequentially calculated, and the value of the leaf node in the purchase tree is calculated, and the purchase is performed.
  • the obtained purchase tree has a distance value other than the root node, and the update center is updated to the purchase tree of the same cluster, and the value of the corresponding node is Add the average value as the RFM value of the corresponding node of the new center. If the corresponding node does not exist, assign a minimum value to it.
  • the k-means principle in the classification process of the customer purchase tree, the k-means principle can be used to efficiently and quickly process large-scale data, and the conventional method cannot solve the current big data because the computational complexity is too high.
  • the distance between the purchase tree and the center is calculated. By continuously iteratively calculating the distance between the purchase tree and the center until the center no longer changes or the change is lower than a certain threshold, the result of the user classification is finally obtained.
  • the customer classification system 10 based on the transaction data provided by the invention uses the RFM model to extract the transaction information of the customer according to the transaction data of the retail enterprise, and simultaneously creates a customer purchase tree according to the hierarchical information of the data, and then uses the k-means pair. Create a customer purchase tree for clustering.
  • the technical solution provided by the invention can improve the effect of clustering, fully utilize the RFM information of the commodity and the hierarchical information carried, and greatly improve the practical use value, and at the same time, in the process of classifying the customer purchase tree, using k
  • the -means principle can process large-scale data efficiently and quickly, and solves the problem that the traditional method cannot handle the current big data because the computational complexity is too high.
  • the experimental data set of the present invention is the actual transaction record data of a large supermarket chain, with a total of 1997,636 records and 79,832 purchase trees.
  • the running time of the technical solution of the present invention is reduced by more than 40% compared with the traditional hierarchical clustering algorithm.

Landscapes

  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Finance (AREA)
  • Marketing (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于交易数据的客户分类方法及其系统，属于数据挖掘技术领域，方法包括：从零售企业交易数据库中采集交易数据；将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码；根据编码后的层次信息为每个客户创建客户购买树；对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。本方案能够高效快速的处理大规模数据，解决了传统方法因为计算复杂度太高而不能处理当前大数据的问题。

Description

一种基于交易数据的客户分类方法及其系统 技术领域
本发明涉及数据挖掘领域 ，尤其涉及一种基于交易数据的客户分类方法及其系统。
背景技术
基于交易数据的客户分类在零售行业有着重要的运用，它通过客户所购买的商品来分析客户的购买行为进而为零售企业提供商业决策营销战略。客户在购买商品的过程中通常会一次购买多个商品，不同的客户购买的商品中有可能有不同的商品，但也有可能有都购买的商品。所以，基于交易数据的客户分类目的是找出有相似购买行为的客户，从零售记录中分析出这些客户经常购买的商品的组合，从而有利于商品的摆架搭配、提供促销、物流等。
在现有技术中，当前客户的分类方法有基于客户的地理位置和人口学特征信息以及基于交易数据的层次聚类，其中基于交易数据的层次聚类是主要方法。但是，传统的基于客户的地理位置和人口学特征信息的聚类方法存在着这些指标数据具有私有性不容易获取的问题，同时基于这些静态特征属性聚类的结果不一定有相同的购买行为，基于交易数据的层次聚类在距离计算的时候需要两两之间进行计算，平方级的时间复杂度，不能有效处理当前大数据时代下的海量数据，同时在面对维度较高的数据时，处理效率也会迅速下降。
 技术问题
有鉴于此，本发明的目的在于提供一种基于交易数据的客户分类方法及其系统，旨在解决传统方法因为计算复杂度太高而不能处理当前大数据的问题。
 技术解决方案
本发明提出一种基于交易数据的客户分类方法，所述方法包括：
从零售企业交易数据库中采集交易数据；
将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码；
根据编码后的层次信息为每个客户创建客户购买树；
对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。
优选的，所述交易数据包括：会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
优选的，所述预处理包括对信息缺失的交易数据进行忽略处理和对信息冲突的交易数据进行唯一性处理，其中，所述将经过预处理后的交易数据分别按层进行编码的步骤具体包括：
对经过预处理后的交易数据分别对商品名称、商品类别名称按层进行编码，且编码方式为按数字从1开始从小到大依次进行编码，层与层之间相互独立。
优选的，所述根据编码后的层次信息为每个客户创建客户购买树的步骤具体包括：
对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，以形成每个商品的RFM值；
将每个客户购买的商品根据商品类别构建出购买树，其中，所述购买树的节点包括根节点、叶子节点和内部节点，所述叶子节点为商品名称节点，所述内部节点为商品类别名称节点；
依次扫描交易数据，将获取的商品名称、商品类别名称与购买树做匹配；
将不在购买树中的该笔交易数据加入购买树中。
优选的，所述对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果的步骤具体包括：
a、从购买树中随机选择k颗购买树作为初始中心；
b、初始化权重树；
c、依次计算购买树中节点的RFM值；
d、依次计算购买树和每个中心距离的加权和，并将该购买树划分到加权和最小中心所在的簇；
e、更新初始中心，将划分到同一个簇的购买树对应结点的值相加后取均值作为新中心对应结点的RFM值；
f、重复上述过程c、d和e，直到聚类中心不再发生变化或变化小于预设的阈值为止。
另一方面，本发明还提供一种基于交易数据的客户分类系统，所述系统包括：
数据采集模块，用于从零售企业交易数据库中采集交易数据；
数据预处理模块，用于将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码；
购买树生成模块，用于根据编码后的层次信息为每个客户创建客户购买树；
客户分类模块，用于对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。
优选的，所述交易数据包括：会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
优选的，所述预处理包括对信息缺失的交易数据进行忽略处理和对信息冲突的交易数据进行唯一性处理，其中，所述数据预处理模块具体用于：
对经过预处理后的交易数据分别对商品名称、商品类别名称按层进行编码，且编码方式为按数字从1开始从小到大依次进行编码，层与层之间相互独立。
优选的，所述购买树生成模块具体用于：
对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，以形成每个商品的RFM值；
将每个客户购买的商品根据商品类别构建出购买树，其中，所述购买树的节点包括根节点、叶子节点和内部节点，所述叶子节点为商品名称节点，所述内部节点为商品类别名称节点；
依次扫描交易数据，将获取的商品名称、商品类别名称与购买树做匹配；
将不在购买树中的该笔交易数据加入购买树中。
优选的，所述客户分类模块具体用于：
a、从购买树中随机选择k颗购买树作为初始中心；
b、初始化权重树；
c、依次计算购买树中节点的RFM值；
d、依次计算购买树和每个中心距离的加权和，并将该购买树划分到加权和最小中心所在的簇；
e、更新初始中心，将划分到同一个簇的购买树对应结点的值相加后取均值作为新中心对应结点的RFM值；
f、重复上述过程c、d和e，直到聚类中心不再发生变化或变化小于预设的阈值为止。
 有益效果
本发明提供的技术方案可以提高聚类的效果，充分利用商品的RFM信息和携带的层次信息，大大提高实际使用价值。同时在客户购买树的分类过程中，利用k-means原理，能够高效快速的处理大规模数据，解决了传统方法因为计算复杂度太高而不能处理当前大数据的情况。
 附图说明
图1为本发明一实施方式中基于交易数据的客户分类方法流程图；
图2为本发明一实施方式中有二义性商品结构图；
图3为本发明一实施方式中购买树示例图；
图4为本发明一实施方式中基于交易数据的客户分类系统10的内部结构示意图。
 本发明的实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
以下将对本发明所提供的一种基于交易数据的客户分类方法进行详细说明。
请参阅图1，为本发明一实施方式中基于交易数据的客户分类方法流程图。
在步骤S1中，从零售企业交易数据库中采集交易数据。
在本实施方式中，所述交易数据包括：会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
在本实施方式中，零售企业提供的交易数据信息中有些信息是我们需要的，有些信息是我们不需要的。我们只需要从中提取有用的交易数据信息，包括会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
其中，会员卡号：零售企业为客户办理会员卡的标识号，同一个会员卡号购买的商品即认为是同一个客户购买的商品，本方案根据会员卡号区分不同的客户；
交易日期：购买商品时的具体日期，包含年月日信息；
交易时间：购买商品时的具体时间，包含时分秒信息；
商品名称：购买商品的名称；
商品类别名称：商品所属的小类、中类和大类等类别信息；
销售额：客户所购买商品的总的金额。
在步骤S2中，将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码。
在本实施方式中，所述预处理包括对信息缺失的交易数据进行忽略处理和对信息冲突的交易数据进行唯一性处理，其中，所述将经过预处理后的交易数据分别按层进行编码的步骤具体包括：
对经过预处理后的交易数据分别对商品名称、商品类别名称按层进行编码，且编码方式为按数字从1开始从小到大依次进行编码，层与层之间相互独立。
在本实施方式中，信息缺失的交易数据，例如包括会员卡号缺失的交易数据、交易日期缺失的交易数据、交易时间缺失的交易数据、商品名称缺失的交易数据、商品类别名称缺失的交易数据、销售额缺失的交易数据等等多种情况。
信息冲突的交易数据，例如包括有二义性以及多义性的交易数据等多种情况。其中，对交易数据中有些商品有二义性的问题，如图2所示，商品盘子既属于商品小类厨房又属于瓷器，这里进行唯一性处理，去掉属于瓷器，仅保留属于厨房。为了节省存储空间和提高计算效率，对所有的商品名称和所属的层次类别名称进行分层独立全局编码。
在步骤S3中，根据编码后的层次信息为每个客户创建客户购买树。
在本实施方式中，所述根据编码后的层次信息为每个客户创建客户购买树的步骤S3具体包括：
对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，以形成每个商品的RFM值；
将每个客户购买的商品根据商品类别构建出购买树，其中，所述购买树的节点包括根节点、叶子节点和内部节点，所述叶子节点为商品名称节点，所述内部节点为商品类别名称节点；
依次扫描交易数据，将获取的商品名称、商品类别名称与购买树做匹配；
将不在购买树中的该笔交易数据加入购买树中。
在本实施方式中，基于交易数据，利用广泛使用的RFM模型，提取出每个客户的RFM信息，同时保留数据的层次信息，根据每个客户的购买信息用树形结构表示。其中RFM模型是客户关系管理（CRM）方面的分析模型，被广泛应用，它是衡量客户价值和客户创利能力的重要工具和手段，该模型通过一个客户的近期购买行为、购买的总体频率以及花了多少钱三项指标来描述该客户的价值状况。
在本实施方式中，零售企业的交易数据是交易记录的全集，从该交易记录中抽取出无重复无遗漏的商品，依据其带有的层次信息建立一颗商品树，对所有的交易数据按会员卡号划分成组，以月为时间片设置滑动窗口，统计出FRM值即为购买树叶子节点的RFM值，然后对叶子节点的RFM值根据最大最小值进行归一化处理，内部节点的RFM为其孩子节点的线性和分别除以商品树中对应节点的孩子节点的数量，这样自底向上建立每个用户的购买树。其中，对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，即每个商品的RFM值。每个客户购买的商品可以根据商品类别构造出购买树，如图3所示，购买树的根节点为root节点，叶子节点为商品名称节点，内部节点为类别节点，每个节点含有RFM值。购买树的具体构建方法为，依次扫描交易数据，将获取的商品类别信息、商品名称和购买树做匹配，如果不在购买树中，将其加入购买树，如果在则不加入，这样扫描完后，购买树的叶子节点都含有RFM值信息，内部节点的RFM值根据孩子节点的RFM进行线性求解得到。
在步骤S4中，对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。
在本实施方式中，所述对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果的步骤S4具体包括：
a、从购买树中随机选择k颗购买树作为初始中心；
b、初始化权重树；
c、依次计算购买树中节点的RFM值；
d、依次计算购买树和每个中心距离的加权和，并将该购买树划分到加权和最小中心所在的簇；
e、更新初始中心，将划分到同一个簇的购买树对应结点的值相加后取均值作为新中心对应结点的RFM值；
f、重复上述过程c、d和e，直到聚类中心不再发生变化或变化小于预设的阈值为止。
在本实施方式中，对全部的N颗购买树划分成k个类，可以得到k个类的用户，依次计算购买树和所述初始中心距离，计算购买树中叶子结点的值，对购买树中的每个结点计算完成后，得到的购买树除了root结点以外的其他结点都有了距离值，更新初始中心，将划分到同一个簇的购买树，对应结点的值相加后取均值作为新中心对应结点的RFM值，如果相应的节点不存在，则赋值一个最小值给它。
在本实施方式中，在客户购买树的分类过程中，利用k-means原理，能够高效快速的处理大规模数据，解决了传统方法因为计算复杂度太高而不能处理当前大数据的情况。根据Bregman信息熵计算购买树与中心的距离，通过不断迭代计算购买树与中心的距离直到中心不再发生变化或变化低于一定的阀值结束，最后得到用户分类的结果。
本发明提供的一种基于交易数据的客户分类方法，根据零售企业的交易数据，利用RFM模型，提取客户的交易信息，同时根据数据的层次信息，创建客户购买树，然后使用k-means对创建的客户购买树进行聚类。与传统的方法相比，本发明提供的技术方案可以提高聚类的效果，充分利用商品的RFM信息和携带的层次信息，大大提高实际使用价值，同时在客户购买树的分类过程中，利用k-means原理，能够高效快速的处理大规模数据，解决了传统方法因为计算复杂度太高而不能处理当前大数据的情况。
请参阅图2，所示为本发明一实施方式中基于交易数据的客户分类系统10的结构示意图。在本实施方式中，基于交易数据的客户分类系统10主要包括数据采集模块11、数据预处理模块12、购买树生成模块13以及客户分类模块14。
数据采集模块11，用于从零售企业交易数据库中采集交易数据。
在本实施方式中，所述交易数据包括：会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
在本实施方式中，零售企业提供的交易数据信息中有些信息是我们需要的，有些信息是我们不需要的。我们只需要从中提取有用的交易数据信息，包括会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
其中，会员卡号：零售企业为客户办理会员卡的标识号，同一个会员卡号购买的商品即认为是同一个客户购买的商品，本方案根据会员卡号区分不同的客户；
交易日期：购买商品时的具体日期，包含年月日信息；
交易时间：购买商品时的具体时间，包含时分秒信息；
商品名称：购买商品的名称；
商品类别名称：商品所属的小类、中类和大类等类别信息；
销售额：客户所购买商品的总的金额。
数据预处理模块12，用于将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码。
在本实施方式中，所述预处理包括对信息缺失的交易数据进行忽略处理和对信息冲突的交易数据进行唯一性处理，其中，所述数据预处理模块12具体用于：
对经过预处理后的交易数据分别对商品名称、商品类别名称按层进行编码，且编码方式为按数字从1开始从小到大依次进行编码，层与层之间相互独立。
在本实施方式中，信息缺失的交易数据，例如包括会员卡号缺失的交易数据、交易日期缺失的交易数据、交易时间缺失的交易数据、商品名称缺失的交易数据、商品类别名称缺失的交易数据、销售额缺失的交易数据等等多种情况。
信息冲突的交易数据，例如包括有二义性以及多义性的交易数据等多种情况。其中，对交易数据中有些商品有二义性的问题，如图2所示，商品盘子既属于商品小类厨房又属于瓷器，这里进行唯一性处理，去掉属于瓷器，仅保留属于厨房。为了节省存储空间和提高计算效率，对所有的商品名称和所属的层次类别名称进行分层独立全局编码。
购买树生成模块13，用于根据编码后的层次信息为每个客户创建客户购买树。
在本实施方式中，所述购买树生成模块13具体用于：
对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，以形成每个商品的RFM值；
将每个客户购买的商品根据商品类别构建出购买树，其中，所述购买树的节点包括根节点、叶子节点和内部节点，所述叶子节点为商品名称节点，所述内部节点为商品类别名称节点；
依次扫描交易数据，将获取的商品名称、商品类别名称与购买树做匹配；
将不在购买树中的该笔交易数据加入购买树中。
在本实施方式中，基于交易数据，利用广泛使用的RFM模型，提取出每个客户的RFM信息，同时保留数据的层次信息，根据每个客户的购买信息用树形结构表示。其中RFM模型是客户关系管理（CRM）方面的分析模型，被广泛应用，它是衡量客户价值和客户创利能力的重要工具和手段，该模型通过一个客户的近期购买行为、购买的总体频率以及花了多少钱三项指标来描述该客户的价值状况。
在本实施方式中，零售企业的交易数据是交易记录的全集，从该交易记录中抽取出无重复无遗漏的商品，依据其带有的层次信息建立一颗商品树，对所有的交易数据按会员卡号划分成组，以月为时间片设置滑动窗口，统计出FRM值即为购买树叶子节点的RFM值，然后对叶子节点的RFM值根据最大最小值进行归一化处理，内部节点的RFM为其孩子节点的线性和分别除以商品树中对应节点的孩子节点的数量，这样自底向上建立每个用户的购买树。其中，对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，即每个商品的RFM值。每个客户购买的商品可以根据商品类别构造出购买树，如图3所示，购买树的根节点为root节点，叶子节点为商品名称节点，内部节点为类别节点，每个节点含有RFM值。购买树的具体构建方法为，依次扫描交易数据，将获取的商品类别信息、商品名称和购买树做匹配，如果不在购买树中，将其加入购买树，如果在则不加入，这样扫描完后，购买树的叶子节点都含有RFM值信息，内部节点的RFM值根据孩子节点的RFM进行线性求解得到。
客户分类模块14，用于对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。
在本实施方式中，所述客户分类模块14具体用于：
a、从购买树中随机选择k颗购买树作为初始中心；
b、初始化权重树；
c、依次计算购买树中节点的RFM值；
d、依次计算购买树和每个中心距离的加权和，并将该购买树划分到加权和最小中心所在的簇；
e、更新初始中心，将划分到同一个簇的购买树对应结点的值相加后取均值作为新中心对应结点的RFM值；
f、重复上述过程c、d和e，直到聚类中心不再发生变化或变化小于预设的阈值为止。
在本实施方式中，对全部的N颗购买树划分成k个类，可以得到k个类的用户，依次计算购买树和所述初始中心距离，计算购买树中叶子结点的值，对购买树中的每个结点计算完成后，得到的购买树除了root结点以外的其他结点都有了距离值，更新初始中心，将划分到同一个簇的购买树，对应结点的值相加后取均值作为新中心对应结点的RFM值，如果相应的节点不存在，则赋值一个最小值给它。
在本实施方式中，在客户购买树的分类过程中，利用k-means原理，能够高效快速的处理大规模数据，解决了传统方法因为计算复杂度太高而不能处理当前大数据的情况。根据Bregman信息熵计算购买树与中心的距离，通过不断迭代计算购买树与中心的距离直到中心不再发生变化或变化低于一定的阀值结束，最后得到用户分类的结果。
本发明提供的一种基于交易数据的客户分类系统10，根据零售企业的交易数据，利用RFM模型，提取客户的交易信息，同时根据数据的层次信息，创建客户购买树，然后使用k-means对创建的客户购买树进行聚类。与传统的方法相比，本发明提供的技术方案可以提高聚类的效果，充分利用商品的RFM信息和携带的层次信息，大大提高实际使用价值，同时在客户购买树的分类过程中，利用k-means原理，能够高效快速的处理大规模数据，解决了传统方法因为计算复杂度太高而不能处理当前大数据的情况。
本发明的实验数据集为某大型连锁超市的实际交易记录数据，共有1997636条记录，79852颗购买树。通过对比传统的层次聚类算法，本发明技术方案的运行时间比传统的层次聚类算法减少40%以上。
另外，本领域普通技术人员可以理解实现上述各实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，相应的程序可以存储于一计算机可读取存储介质中，所述的存储介质，如ROM/RAM、磁盘或光盘等。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种基于交易数据的客户分类方法，其特征在于，所述方法包括：
    从零售企业交易数据库中采集交易数据；
    将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码；
    根据编码后的层次信息为每个客户创建客户购买树；
    对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。
  2. 如权利要求1所述的基于交易数据的客户分类方法，其特征在于，所述交易数据包括：会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
  3. 如权利要求2所述的基于交易数据的客户分类方法，其特征在于，所述预处理包括对信息缺失的交易数据进行忽略处理和对信息冲突的交易数据进行唯一性处理，其中，所述将经过预处理后的交易数据分别按层进行编码的步骤具体包括：
    对经过预处理后的交易数据分别对商品名称、商品类别名称按层进行编码，且编码方式为按数字从1开始从小到大依次进行编码，层与层之间相互独立。
  4. 如权利要求3所述的基于交易数据的客户分类方法，其特征在于，所述根据编码后的层次信息为每个客户创建客户购买树的步骤具体包括：
    对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，以形成每个商品的RFM值；
    将每个客户购买的商品根据商品类别构建出购买树，其中，所述购买树的节点包括根节点、叶子节点和内部节点，所述叶子节点为商品名称节点，所述内部节点为商品类别名称节点；
    依次扫描交易数据，将获取的商品名称、商品类别名称与购买树做匹配；
    将不在购买树中的该笔交易数据加入购买树中。
  5. 如权利要求4所述的基于交易数据的客户分类方法，其特征在于，所述对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果的步骤具体包括：
    a、从购买树中随机选择k颗购买树作为初始中心；
    b、初始化权重树；
    c、依次计算购买树中节点的RFM值；
    d、依次计算购买树和每个中心距离的加权和，并将该购买树划分到加权和最小中心所在的簇；
    e、更新初始中心，将划分到同一个簇的购买树对应结点的值相加后取均值作为新中心对应结点的RFM值；
    f、重复上述过程c、d和e，直到聚类中心不再发生变化或变化小于预设的阈值为止。
  6. 一种基于交易数据的客户分类系统，其特征在于，所述系统包括：
    数据采集模块，用于从零售企业交易数据库中采集交易数据；
    数据预处理模块，用于将采集到的交易数据进行预处理，并将经过预处理后的交易数据分别按层进行编码；
    购买树生成模块，用于根据编码后的层次信息为每个客户创建客户购买树；
    客户分类模块，用于对创建的所有客户购买树进行聚类，直到聚类中心不再改变或者变化小于预设的阈值为止，以获取客户分类的结果。
  7. 如权利要求6所述的基于交易数据的客户分类系统，其特征在于，所述交易数据包括：会员卡号、交易日期、交易时间、商品名称、商品类别名称以及销售额。
  8. 如权利要求7所述的基于交易数据的客户分类系统，其特征在于，所述预处理包括对信息缺失的交易数据进行忽略处理和对信息冲突的交易数据进行唯一性处理，其中，所述数据预处理模块具体用于：
    对经过预处理后的交易数据分别对商品名称、商品类别名称按层进行编码，且编码方式为按数字从1开始从小到大依次进行编码，层与层之间相互独立。
  9. 如权利要求8所述的基于交易数据的客户分类系统，其特征在于，所述购买树生成模块具体用于：
    对交易数据按会员卡号聚合分组，统计每个商品的消费频率、消费金额、最近一次消费时间，以形成每个商品的RFM值；
    将每个客户购买的商品根据商品类别构建出购买树，其中，所述购买树的节点包括根节点、叶子节点和内部节点，所述叶子节点为商品名称节点，所述内部节点为商品类别名称节点；
    依次扫描交易数据，将获取的商品名称、商品类别名称与购买树做匹配；
    将不在购买树中的该笔交易数据加入购买树中。
  10. 如权利要求9所述的基于交易数据的客户分类系统，其特征在于，所述客户分类模块具体用于：
    a、从购买树中随机选择k颗购买树作为初始中心；
    b、初始化权重树；
    c、依次计算购买树中节点的RFM值；
    d、依次计算购买树和每个中心距离的加权和，并将该购买树划分到加权和最小中心所在的簇；
    e、更新初始中心，将划分到同一个簇的购买树对应结点的值相加后取均值作为新中心对应结点的RFM值；
    f、重复上述过程c、d和e，直到聚类中心不再发生变化或变化小于预设的阈值为止。
PCT/CN2017/088724 2016-09-29 2017-06-16 一种基于交易数据的客户分类方法及其系统 WO2018059015A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610865276.5 2016-09-29
CN201610865276.5A CN106529968B (zh) 2016-09-29 2016-09-29 一种基于交易数据的客户分类方法及其系统

Publications (1)

Publication Number Publication Date
WO2018059015A1 true WO2018059015A1 (zh) 2018-04-05

Family

ID=58344578

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/088724 WO2018059015A1 (zh) 2016-09-29 2017-06-16 一种基于交易数据的客户分类方法及其系统

Country Status (2)

Country Link
CN (1) CN106529968B (zh)
WO (1) WO2018059015A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI678671B (zh) * 2018-05-15 2019-12-01 第一商業銀行股份有限公司 金流分群方法及其系統
CN111047336A (zh) * 2019-12-24 2020-04-21 太平金融科技服务（上海）有限公司 用户标签推送、用户标签展示方法、装置和计算机设备
CN111626614A (zh) * 2020-05-27 2020-09-04 国网上海市电力公司 一种基于电费回收的用户分类方法
CN111985577A (zh) * 2020-09-02 2020-11-24 深圳壹账通智能科技有限公司 基于人工智能的客户价值标注方法、装置、设备及介质
CN116797253A (zh) * 2022-12-13 2023-09-22 乖乖数字科技(苏州)有限公司 一种基于客户资源的分类管理方法
CN116975539A (zh) * 2023-08-16 2023-10-31 杭州火奴数据科技有限公司 基于聚类算法的营销数据存储管理系统

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106529968B (zh) * 2016-09-29 2021-05-14 深圳大学 一种基于交易数据的客户分类方法及其系统
CN107480708A (zh) * 2017-07-31 2017-12-15 微梦创科网络科技（中国）有限公司 一种复杂模型的聚类方法及系统
CN107608792B (zh) * 2017-09-12 2020-09-01 中国联合网络通信集团有限公司 资源调度方法与装置
CN108108747B (zh) * 2017-09-21 2020-07-28 西安交通大学 一种解决大数据聚类的基于视觉原理的聚类方法
CN108230029A (zh) * 2017-12-29 2018-06-29 西南大学 客户交易行为分析方法
CN110458298A (zh) * 2018-05-03 2019-11-15 上海博泰悦臻网络技术服务有限公司 车辆及配件的分析维护系统、及客户在线管理系统和方法
CN110503446A (zh) * 2018-05-16 2019-11-26 江苏天智互联科技股份有限公司 基于聚类算法的电商平台的客户分类方法与决策方法
CN108664653A (zh) * 2018-05-18 2018-10-16 拓普暨达（广州）基因精准医疗科技有限公司 一种基于K-means的医疗消费客户自动分类方法
CN109145018A (zh) * 2018-07-10 2019-01-04 武汉海云健康科技股份有限公司 一种基于大数据分析用户购药挖掘用户价值的方法
CN111328012B (zh) * 2018-12-14 2021-08-27 中国移动通信集团广东有限公司 一种基于楼宇WiFi立体分层定位方法及装置
CN111222686A (zh) * 2019-11-21 2020-06-02 施甸县保施高速公路投资开发有限公司 优化高速公路服务区业态的方法
CN112184303A (zh) * 2020-09-25 2021-01-05 中国建设银行股份有限公司 基于聚类算法的目标信息推送方法、装置及存储介质
CN113781108A (zh) * 2021-08-30 2021-12-10 武汉理工大学 一种电商平台客户细分方法、装置、电子设备及存储介质
CN114708003B (zh) * 2022-04-27 2023-11-10 西南交通大学 一种异常数据检测方法、装置、设备及可读存储介质
CN115392406B (zh) * 2022-10-28 2023-04-28 山东智豆数字科技有限公司 一种基于历史交易信息的用户分类管理方法
CN116342168B (zh) * 2023-05-23 2023-08-04 山东灵动电子商务有限公司 一种信息大数据智能采集管理系统
CN116681434A (zh) * 2023-06-07 2023-09-01 科睿特软件集团股份有限公司 基于年卡防盗刷算法的行为管理系统及其方法
CN116562769B (zh) * 2023-06-15 2024-06-21 深圳爱巧网络有限公司 一种基于货物属性分类的货物数据分析方法及系统
CN117522419B (zh) * 2024-01-08 2024-03-22 广东深玎科技有限公司 一种应用于客户关系管理系统的资源分配方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102136123A (zh) * 2011-03-15 2011-07-27 中国工商银行股份有限公司 一种银行目标客户识别系统
CN202093568U (zh) * 2011-03-15 2011-12-28 中国工商银行股份有限公司 一种银行目标客户识别系统
CN105791242A (zh) * 2014-12-24 2016-07-20 阿里巴巴集团控股有限公司 对象类型的识别方法、服务器、客户端和系统
CN106127493A (zh) * 2016-06-23 2016-11-16 深圳大学 一种分析用户交易行为的方法及装置
CN106529968A (zh) * 2016-09-29 2017-03-22 深圳大学 一种基于交易数据的客户分类方法及其系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103530341A (zh) * 2013-10-08 2014-01-22 广州品唯软件有限公司 物品信息生成推送方法和系统
CN103731738A (zh) * 2014-01-23 2014-04-16 哈尔滨理工大学 基于用户群组行为分析的视频推荐方法及装置
CN105631709A (zh) * 2015-12-26 2016-06-01 深圳大学 一种购物篮分析方法及系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102136123A (zh) * 2011-03-15 2011-07-27 中国工商银行股份有限公司 一种银行目标客户识别系统
CN202093568U (zh) * 2011-03-15 2011-12-28 中国工商银行股份有限公司 一种银行目标客户识别系统
CN105791242A (zh) * 2014-12-24 2016-07-20 阿里巴巴集团控股有限公司 对象类型的识别方法、服务器、客户端和系统
CN106127493A (zh) * 2016-06-23 2016-11-16 深圳大学 一种分析用户交易行为的方法及装置
CN106529968A (zh) * 2016-09-29 2017-03-22 深圳大学 一种基于交易数据的客户分类方法及其系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YE, XIAOMING ET AL.: "Retail Customer Segmentation Based on Data Mining", MODERN MANAGEMENT SCIENCE, 30 June 2006 (2006-06-30), pages 63, 64, 82, ISSN: 1007-368X *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI678671B (zh) * 2018-05-15 2019-12-01 第一商業銀行股份有限公司 金流分群方法及其系統
CN111047336A (zh) * 2019-12-24 2020-04-21 太平金融科技服务（上海）有限公司 用户标签推送、用户标签展示方法、装置和计算机设备
CN111626614A (zh) * 2020-05-27 2020-09-04 国网上海市电力公司 一种基于电费回收的用户分类方法
CN111985577A (zh) * 2020-09-02 2020-11-24 深圳壹账通智能科技有限公司 基于人工智能的客户价值标注方法、装置、设备及介质
CN116797253A (zh) * 2022-12-13 2023-09-22 乖乖数字科技(苏州)有限公司 一种基于客户资源的分类管理方法
CN116797253B (zh) * 2022-12-13 2024-03-01 乖乖数字科技(苏州)有限公司 一种基于客户资源的分类管理方法
CN116975539A (zh) * 2023-08-16 2023-10-31 杭州火奴数据科技有限公司 基于聚类算法的营销数据存储管理系统
CN116975539B (zh) * 2023-08-16 2024-03-19 杭州火奴数据科技有限公司 基于聚类算法的营销数据存储管理系统

Also Published As

Publication number Publication date
CN106529968B (zh) 2021-05-14
CN106529968A (zh) 2017-03-22

Similar Documents

Publication Publication Date Title
WO2018059015A1 (zh) 一种基于交易数据的客户分类方法及其系统
Aldino et al. Comparison of market basket analysis to determine consumer purchasing patterns using fp-growth and apriori algorithm
WO2021016978A1 (zh) 电信网络告警预测方法及系统
Elalfi et al. Extracting rules from trained neural network using GA for managing E-business
WO2013185518A1 (zh) 一种电视直播流中的广告段检测方法
CN112381598B (zh) 产品服务信息推送方法及装置
Chai et al. Analysis of spatial-temporal behavior pattern of the share bike usage during COVID-19 pandemic in Beijing
Chen et al. A fuzzy coherent rule mining algorithm
US20220180400A1 (en) Information processing apparatus, display system, and information processing method
WO2023136492A1 (ko) 이커머스 데이터 기반 신용 평가 스코어 카드 결정 방법 및 이러한 방법을 수행하는 장치
TW202141386A (zh) 消費數據處理方法與系統
Pandit et al. A review on clothes matching and recommendation systems based on user attributes
CN111833073A (zh) 基于K-Means++算法的航空公司客户细分方法
US20170132648A1 (en) Anonymous reporting of multiple venue location data
Weng Knowledge discovery of digital library subscription by RFC itemsets
Gupta et al. Data mining framework for IoT applications
CN114331569A (zh) 一种商业空间内不同场景的用户消费行为分析方法及系统
Yurtay et al. Dashboard application model in supplier evaluation by using artificial immune system and data mining methods
JP6249500B2 (ja) マッチングシステム、マッチング方法及びプログラム
Xiao et al. Value segmentation of airline customers based on K-means and neighborhood rough set
Srinivas et al. Customer stress prediction in telecom industries using machine learning
TW202026989A (zh) 推銷方法、推銷裝置、電腦裝置及存儲介質
Myung et al. A Supply Chain Management Process Modeling for an Agriculture Marketing Information System
Bing et al. Research and application of association rules methods in data mining for commercial sales analysis
Pavithra et al. Comparative study of effective performance of association rule Mining in Different Databases

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17854469

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 30/07/2019)

122 Ep: pct application non-entry in european phase

Ref document number: 17854469

Country of ref document: EP

Kind code of ref document: A1