WO2016150354A1 - 一种对电子商务平台的用户进行分类的方法及系统 - Google Patents

一种对电子商务平台的用户进行分类的方法及系统 Download PDF

Info

Publication number
WO2016150354A1
WO2016150354A1 PCT/CN2016/076811 CN2016076811W WO2016150354A1 WO 2016150354 A1 WO2016150354 A1 WO 2016150354A1 CN 2016076811 W CN2016076811 W CN 2016076811W WO 2016150354 A1 WO2016150354 A1 WO 2016150354A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
grade
purchasing power
price
center
Prior art date
Application number
PCT/CN2016/076811
Other languages
English (en)
French (fr)
Inventor
邵佳帅
刘朋飞
牟川
Original Assignee
北京京东尚科信息技术有限公司
北京京东世纪贸易有限公司
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 北京京东尚科信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京京东尚科信息技术有限公司
Publication of WO2016150354A1 publication Critical patent/WO2016150354A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/06Buying, selling or leasing transactions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor

Definitions

  • the invention relates to the technical field related to electronic commerce, in particular to a method and a system for classifying users of an e-commerce platform according to purchasing power.
  • the user's purchasing power label is essential. For users with high purchasing power, when they choose the same type of goods, they often buy goods with higher quality and price. For example, a user with high purchasing power wants to buy a mobile phone, he will buy high-end mobile phones in Apple or Samsung brands; For users with low purchasing power, such as he wants to buy a headset, he will buy a low-end product that can be used for more than ten dollars or twenty dollars.
  • the definition of purchasing power of a user is the level of ability of the user to purchase the same type of goods.
  • the existing technologies are mostly based on the user's proportion of the number of purchases of high, medium and low-grade goods, and the users are divided into three levels of high, medium and low.
  • the specific method is: for each root commodity class, the highest 20% of the goods in the price segment is defined as high-end goods, and the lowest 20% in the price segment is low-end goods, 60% of which are in the middle. The product is a mid-end product. Then calculate the proportion of times each user buys high-end products, the proportion of times the mid-end products are purchased, and the number of times the low-end products are purchased. Finally, to see which level of purchases the user has in the largest proportion of purchases, the user is divided into the purchase level user group. In the end, the purchasing power is high, medium and low.
  • the final purchase power level is determined directly by the most proportion, which will reduce the accuracy.
  • the percentage of users A purchasing high, medium and low-end products is (0.8, 0.2, 0)
  • the percentage of users B purchasing high, medium and low-end products is (0.4, 0.3, 0.3). ).
  • users A and B are both users with high purchasing power, and in actual observation, we can easily find that user B is not like a high-end user. This is because the user's purchase ratio also plays a certain role in the user's final purchasing power, and the simple maximum proportion rule is not scientific.
  • the existing purchasing power level is generally divided into three levels: high, medium and low, with fewer classifications and less flexible use.
  • a method for classifying users of an e-commerce platform including:
  • the product grade determining step includes: dividing the products in the same category into high-grade products and non-high-end products based on price and sales distribution, and dividing non-high-end products into high-low items according to price from high to low.
  • Grade where x is a preset natural number greater than or equal to 1;
  • the user purchases the proportion calculation step including: calculating each user to purchase each grade of goods
  • the percentage of the user is vectorized based on each user to obtain a purchasing power vector for each user, the purchasing power vector is an x+1 dimensional vector, and each dimension corresponds to one grade;
  • the user categorization step includes: performing clustering operation on the purchasing power vector to obtain x+1 point clusters about the purchasing power vector, each point cluster corresponding to one grade, and corresponding points of the point clusters corresponding to the purchasing power vector corresponding to the user As the purchasing power of the user.
  • a system for classifying users of an e-commerce platform including:
  • the product grade determination module is configured to: classify the products in the same category into high-grade products and non-high-end products based on price and sales distribution, and divide non-high-end products into high-low items according to price from high to low. a low grade, where x is a preset natural number greater than or equal to 1;
  • the user purchase ratio calculation module is configured to: calculate a proportion of each user to purchase each grade of goods, and perform vectorization on the basis of each user to obtain a purchasing power vector of each user;
  • the user classification module is configured to: perform clustering operation on the purchasing power vector to obtain x+1 point clusters about the purchasing power vector, each point cluster corresponding to one grade, and corresponding to the point cluster of the purchasing power vector corresponding to the user The grade is used as the purchasing power of the user.
  • the invention makes the classification of the product grade more reasonable through the intelligent delineation of the product grade, and determines the purchasing power of the user based on the intelligently divided product grade, and classifies based on the purchasing power of the user, so that the classification of the purchasing power of the user is more accurate.
  • FIG. 1 is a flowchart of a method for classifying users of an e-commerce platform according to the present invention
  • Figure 2 is a flow chart of the work of dividing high-end goods
  • FIG. 3 is a flowchart of a work performed by a clustering operation according to the present invention.
  • FIG. 4 is a structural block diagram of a system for classifying users of an e-commerce platform according to the present invention.
  • FIG. 1 is a flowchart of a method for classifying users of an e-commerce platform according to the present invention, including:
  • Step S101 includes: classifying products in the same category into high-grade products and non-high-end products based on price and sales distribution, and classifying non-high-end products into high-low grades according to price from high to low.
  • x is a preset natural number greater than or equal to 1;
  • Step S102 comprising: calculating a proportion of each user purchasing each grade of goods, and performing vectorization on the basis of each user to obtain a purchasing power vector of each user, wherein the purchasing power vector is an x+1 dimensional vector. And each dimension corresponds to one grade;
  • Step S103 includes: performing clustering operation on the purchasing power vector to obtain x+1 point clusters related to the purchasing power vector, each point cluster corresponding to one grade, and the corresponding grade of the point cluster of the corresponding purchasing power vector of the user as the user Purchasing power grade.
  • the present invention proposes a more reasonable method of dividing the purchasing power level of a user.
  • the invention does not simply distinguish the product grades according to the commodity price, but comprehensively evaluates the product grade according to the commodity price and the corresponding merchandise sales volume. Since the merchandise grade finally determines the purchasing power grade of the user, the comprehensive evaluation of the merchandise grade is A comprehensive assessment of the user's purchasing power level.
  • the level evaluation of the corresponding purchasing power of the user is also automatically adjusted, so that the classification of the purchasing power of the user is more accurate, thereby greatly improving the user's experience in the website.
  • the step S101 includes: determining a grade of a% of the price in the same category as the high grade, wherein a is obtained by the following method:
  • X+1 is 5.
  • the user's purchasing ability is divided into high, high, medium, and There are five grades in the low and low grades. Therefore, in the process of labeling the merchandise, we also mark all the merchandise as X grades, preferably five grades of high grade, high grade, medium grade, medium grade and low grade.
  • the "one size fits all" approach mentioned in the background art makes very high grades of many categories very sparse. Therefore, for the classification of high-end goods, this embodiment automatically adjusts the division percentage according to the price and sales distribution of each category of goods.
  • a category such as a third-class category
  • select a product that is a% before the price segment as a high-end product where a is preferably three, and you can choose 5, 10, or 20, that is, a% can be 5%, 10, or 20%.
  • the specific value of a needs to calculate two indicators m and n first, where m is the price statistic value of the category y% of the category price, and n is the price statistic of the commodity price of the category in the previous y%. Value (do not repeat, if a product is bought multiple times, it is also counted).
  • y is preferably 10%
  • the price statistic of the previous y% is preferably the highest value among the remaining commodities after the y% of the goods are removed.
  • Step S201 calculating a price statistic value m of the category 10 price of 10%
  • Step S202 calculating a price statistical value of the top 10% of the commodity price of the product in the recent period
  • the grades of the remaining commodities are non-high-end, which is divided into X grades (X is a natural number greater than 1). Preferably, it can be divided into four grades: high, medium, low, and low. Take the last 20% of the price segment, the upper, middle and middle lows are the first 1/3 after removing the high and low, the third and the third.
  • the step S102 includes:
  • the proportion of each grade purchased by each user is taken as an x+1-dimensional vector, and the purchasing power vector of each user is obtained.
  • a user buys two products A and B in high-end goods, and the number of purchases is k1 and k2 respectively, and the prices are respectively p1 and p2, and the user purchases k1 ⁇ ln(p1)+k2 ⁇ ln in high-grade ( P2), where ln represents the logarithm of the value in parentheses.
  • P2 high-grade
  • the step S103 includes:
  • Step S301 comprising: randomly selecting the purchasing power vector of x+1 users from the purchasing power vectors of all users as the first center, and executing step S302;
  • Step S302 comprising: respectively calculating the remaining Euclidean distances of all the purchasing power vectors to x+1 centers, each purchasing power vector is respectively assigned to the point cluster with the smallest Euclidean distance from the center, step S303 is performed;
  • Step S303 including: calculating the arithmetic mean of each dimension for all the purchasing power vectors in the x+1 point clusters as the center of the point cluster, step S304;
  • Step S304 comprising: recalculating the purchasing power vector of all users according to the center obtained in step S304, and calculating the Euclidean cluster with each center, and classifying each purchasing power vector into a point cluster having the smallest Euclidean distance from the center. Go to step S305;
  • Step S305 including: if the center of each point cluster no longer changes, step S306 is performed, otherwise step S303 is performed;
  • Step S306 calculating a grade corresponding to the corresponding point cluster according to the center of the x+1 point clusters.
  • the clustering analysis is performed on the data of all the users to divide the purchasing power level of the user.
  • each cluster of points represents a class of purchasing power.
  • the users of the purchasing power vector included in each point cluster are separately classified into a database, thereby completing the classification of the purchasing power of the user.
  • the step S306 includes:
  • the high-level selection sub-step includes: selecting the current to-be-determined grade to be high-grade;
  • the step determining sub-step includes: determining, in the center of the point cluster of the undetermined level, the current determination of the point cluster corresponding to the center having the largest dimension corresponding to the current to-be-determined level as the current to-be-determined level;
  • the other file selection sub-steps include: if there is a point cluster of undetermined grades, selecting the next sequential grade of the current to-be-determined grade as the current to-be-determined grade, performing the grade determination sub-step, otherwise ending.
  • the method further includes:
  • the classification recommendation step includes: when receiving the user's access information, obtaining the user's purchasing power level, and recommending the corresponding grade of the product to the user according to the purchasing power level.
  • the purchasing power of the user can be known through the database, and the product of the corresponding price is recommended according to the level of the purchasing power of the user. For example, a user with a high purchasing power is recommended to give him a high-end, smart phone when browsing the mobile phone; on the contrary, a user with a low purchasing power rating recommends him a low-priced, practical mobile phone.
  • FIG. 4 is a structural block diagram of a system for classifying users of an e-commerce platform according to the present invention, including:
  • the product grade determination module 401 is configured to divide the products in the same category into high-grade products and non-high-end products based on price and sales distribution, and divide non-high-end products into high-order items according to price from high to low. To a lower grade, where x is a preset natural number greater than or equal to 1;
  • the user purchase ratio calculation module 402 is configured to: calculate a percentage of each user's purchase of each grade item, and vectorize the percentage based on each user to obtain a purchasing power vector of each user, the purchasing power vector An x+1-dimensional vector, and each dimension corresponds to a grade;
  • the user classification module 403 is configured to perform a clustering operation on the purchasing power vector to obtain x+1 point clusters about the purchasing power vector, and each point cluster corresponds to one file Secondly, the corresponding grade of the point cluster in which the user's corresponding purchasing power vector is located is taken as the purchasing power level of the user.
  • the commodity grade determination module is configured to determine a grade of a% of items in the same category before the price segment as a high grade, wherein a is obtained by the following method:
  • the user purchase percentage calculation module is configured to:
  • the proportion of each grade purchased by each user is taken as an x+1-dimensional vector, and the purchasing power vector of each user is obtained.
  • the user classification module includes:
  • the central initialization sub-module is configured to: randomly select x+1 users' purchasing power vectors from the purchasing power vectors of all users as the first center, and execute an initial categorization sub-module;
  • the initial categorization sub-module is configured to: respectively calculate the Euclidean distances of all the remaining purchasing power vectors to x+1 centers, and assign each purchasing power vector to a point cluster with the smallest Euclidean distance from the center, and execute Center update submodule;
  • the central update submodule is configured to: measure all purchasing power vectors in x+1 point clusters Calculating the arithmetic mean of each dimension as the center of the point cluster, performing a classification update sub-module;
  • the categorization update sub-module is configured to: recalculate the purchasing power vector of all users according to the center obtained by the categorization update sub-step with the Euclidean cluster of each center, and assign each purchasing power vector to the center with the European In the cluster of points with the smallest distance, the convergence judgment sub-module is executed;
  • a convergence determination sub-module configured to: if the center of each point cluster no longer changes, execute a profile determination sub-module, otherwise execute a center update sub-module;
  • the grade determining submodule is configured to: calculate a grade corresponding to the corresponding point cluster according to the center of the x+1 point clusters.
  • the grade determination sub-module is configured to:
  • the high-grade selection sub-module is configured to: select a current grade to be determined as high-grade;
  • a level determining sub-module configured to: determine, in a center of a point cluster of an undetermined level, a current point cluster corresponding to a center having a largest dimension corresponding to a current to-be-determined level as a current to-be-determined level;
  • the other file selection sub-module is configured to: if there is a point cluster of undetermined grades, select the next sequential grade of the current to-be-determined grade as the current to-be-determined grade, execute the grade determination sub-module, and otherwise end.
  • system further comprises:
  • the classification recommendation module is configured to: when receiving the user's access information, obtain the user's purchasing power level, and recommend the corresponding grade product to the user according to the purchasing power level.

Abstract

一种对电子商务平台的用户进行分类的方法及系统,包括:将同一品类中的商品基于价格和销量分布确定出档次为高档的商品,同一品类中的其他商品的档次确定为非高档,将档次为非高档的商品按照价格从高到低依次分为x个从高到低的档次(S101),其中,x为预设的大于或等于1的自然数;计算每个用户购买每个档次商品的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量,所述购买力向量为x+1维向量,且每一维与一个档次对应(S102);对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次(S103)。该方法对用户购买力的分类更加准确。

Description

一种对电子商务平台的用户进行分类的方法及系统 技术领域
本发明涉及电子商务相关技术领域,特别是对电子商务平台的用户依据购买力来进行分类的方法及系统。
背景技术
随着电商行业的飞速发展,满足用户的个性化购物需求也变得迫在眉睫。在用户浏览购物的过程中给用户推荐合理的商品将会大大提升用户体验。然而一个个性化的购物推荐系统的背后需要大量的用户标签来支撑。其中,用户的购买力标签是必不可少。对于购买力高的用户,在挑选同一类商品时,往往买的是品质和价格都较高的商品,比如一个购买力高的用户想要买一部手机,他会买苹果或三星品牌中的高端手机;而对于购买力低的用户,比如他想要买一个耳机,他会买十几块钱或者二十块钱能满足一般使用即可的低端产品。由此,对于一个用户的购买力的定义是,用户在购买同一类商品时,支付能力的高低。
对于电商领域用户的购买力区分,现有的技术大多基于用户对高、中、低档商品购买次数占比的方法,把用户划分为高中低三个档次(level)。具体做法是:对每一个底级品类(root commodity class),价格段处在最高的20%的商品定义为高端商品,价格段处在最低的20%的商品为低端商品,中间60%的商品为中端商品。然后计算每个用户的购买高端商品的次数占比,购买中端商品的次数占比,购买低端商品的次数占比。最后看用户在哪个等级商品的购买次数占比最大,则把该用户划分到该购买等级用户群当中。最终得到购买力高、中、低三个等级。
现有技术的缺点主要有四个方面:
1)在区分价格段高、中、低时,使用的商品的价格段,但是实际情况中,很多品类的高价格段的购买情况非常稀疏,甚至没有销量,所以这样的“一刀切”规则很容易导致结果集分布的不均衡。
2)计算用户在各个档次购买商品占比的时,计算购买频次占比,没有加入商品本身的价格因素,导致准确率降低。比如,一个用户虽然买了很多某品类的高档的商品,但是这个品类的价格本身就很低(尿布、家居用品等),那么把他和其他买高档手机、电脑等价格昂贵的用户群中,自然会有失公平。
3)在得到用户高、中、低三种商品购买占比的时候,直接用占比最多的来确定最终的购买力等级,会降低准确率。比如,一个用户A在购买高、中、低端商品的占比分别是(0.8,0.2,0),而用户B在购买高、中、低端商品的占比分别是(0.4,0.3,0.3)。根据现有的判断方法,用户A和B都是购买力高的用户,而实际观察,我们会很容易发现,用户B并不像是高端用户。这是因为用户的购买占比在分布上也对用户最终的购买力也应该起着一定的作用,而简单的最大占比规则判断,并不科学。
4)现有的购买力的等级一般分为高中低三个级别,分类较少,使用起来不够灵活。
发明内容
基于此,有必要针对现有技术中对用户购买力的分类不准确的技术问题,提供一种对电子商务平台的用户依据其购买力进行更为准确的分类的方法及系统。
一种对电子商务平台的用户进行分类的方法,包括:
商品档次确定步骤,包括:将同一品类中的商品基于价格和销量分布分为高档的商品和非高档的商品,将非高档的商品按照价格从高到低依次分为x个从高到低的档次,其中,x为预设的大于或等于1的自然数;
用户购买占比计算步骤,包括:计算每个用户购买每个档次商品 的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量,所述购买力向量为x+1维向量,且每一维与一个档次对应;
用户归类步骤,包括:对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次。
一种对电子商务平台的用户进行分类的系统,包括:
商品档次确定模块,被配置为:将同一品类中的商品基于价格和销量分布分为高档的商品和非高档的商品,将非高档的商品按照价格从高到低依次分为x个从高到低的档次,其中,x为预设的大于或等于1的自然数;
用户购买占比计算模块,被配置为:计算每个用户购买每个档次商品的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量;
用户归类模块,被配置为:对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次。
本发明通过对商品档次的智能划定,使得商品档次的划定更为合理,并基于智能划分的商品档次确定用户购买力,基于用户购买力进行分类,使得对用户购买力的分类更加准确。
附图说明
图1为本发明一种对电子商务平台的用户进行分类的方法的工作流程图;
图2为进行高档商品划分的工作流程图;
图3为本发明进行聚类运算的工作流程图;
图4为本发明一种对电子商务平台的用户进行分类的系统的结构模块图。
具体实施方式
下面结合附图和具体实施例对本发明做进一步详细的说明。
如图1所示为本发明一种对电子商务平台的用户进行分类的方法的工作流程图,包括:
步骤S101,包括:将同一品类中的商品基于价格和销量分布分为高档的商品和非高档的商品,将非高档的商品按照价格从高到低依次分为x个从高到低的档次,其中,x为预设的大于或等于1的自然数;
步骤S102,包括:计算每个用户购买每个档次商品的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量,所述购买力向量为x+1维向量,且每一维与一个档次对应;
步骤S103,包括:对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次。
本发明提出了一种更合理的划分用户购买力等级的方法。本发明并不是简单地仅仅依据商品价格来区分商品档次,而是根据商品价格和相应的商品销量来综合评估商品档次,由于商品档次最终确定用户的购买力档次,因此对商品档次的综合评估就是对用户购买力等级的综合评估。通过对高档商品基于价格和销量分布自动调节的划分方法,使得相应地用户购买力的等级评估也实现了自动调节,使得用户购买力的分类更为准确,从而大大地提升了用户在网站中的体验。
在其中一个实施例中,所述步骤S101,具体包括:将同一品类中价格段前a%的商品的档次确定为高档,其中a是采用如下方法获得的:
选择三个或三个以上取值范围为在0~100之间的待选择值,最大的待选择值作为最大选择值,最小待选择值为最小选择值,其他的待选择值为中间选择值,令m为同一品类中价格的前y%的价格统计值,令n为同一品类中最近预设时间段内有销量的商品中价格的前y% 的价格统计值,如果m大于n超过预设第一阈值,则选择a为最大选择值,如果n大于m超过预设第二阈值,则选择a为最小选择值中的最小值,其他情况,选择a为中间选择值中的一个,其中,y小于最大选择值且大于最小选择值。
首先,要对所有商品进行标注,因为购买力模型最终要将用户的购买能力分为X+1个等级,优选地,X+1为5,则用户的购买能力分为高、偏高、中、中低、低五个等级,所以在对商品标注的过程中,我们也要将所有商品也标为X个档次,优选地为高档、偏高挡、中档、中低档、低档五个档次。由于在背景技术中提到的“一刀切”方法会使很多品类的高档商品非常稀疏。所以对高档商品划分,本实施例根据各个品类商品的价格与销量分布情况自动调节划分百分比。对于一个品类,例如三级品类,选取处于价格段前a%的商品为高档商品,其中,a优选为三个,可以选择5、10或者20,即a%可以为5%、10或20%。a的具体取值需要先计算两个指标m、n,其中m为该品类价格位于前y%的价格统计值,n为该品类近一段时间内有销量的商品价格位于前y%的价格统计值(不去重复,如果一个商品被买多次,也计算在内)。y优选为10%,前y%的价格统计值优选为将前y%的商品去除后剩余商品中的最高值。这样,如果m>>n,说明该品类价格前10%商品销量不好,高端商品的阈值应该提高,即a=20%;如果n>>m,说明该品类的价格前10%商品销量很好,应该少取一些,降低阈值,即a=5%;其他情况a=10%。由此就做到了高端商品的判断阈值可以根据销售的实际情况进行自调节。m的值应介于a取值的中间,但是不一定是中间值。这里面的取值都是可以改变的。具体的流程图如图2所示:
步骤S201,计算该品类价格位于10前10%的价格统计值m;
步骤S202,计算该品类近一段时间内有销量的商品价格位于前10%的价格统计值;
步骤S203,如果m>>n,即m大于n超过第一阈值,第一阈值可以取一个较大的范围,则a%=20%,否则执行步骤S204;
步骤S204,如果n>>m,及n大于m超过第二阈值,第二阈值可以取一个较大的范围,则a%=5%,否则a%=10。
对于高端商品,还可以加入一些规则性的补充,比如奢侈品类、高端的非必须产品(如智能设备等)或者商品单价非常高的商品。
剩下的商品的档次为非高端,将其划分成X个档次(X为大于1的自然数),优选地,可以划分为偏高、中、中低、低四个档次,其中,低档商品可以取价格段的后20%的商品,偏高档、中档、中低档分别是去掉高档和低档后的前1/3,中间的1/3和后面的1/3。
在其中一个实施例中,所述步骤S102,具体包括:
对于每个用户,计算该用户在每个档次的下单量与对应价格取对数后的乘积作为该用户在该档次的购买量,计算该用户所有档次的购买量总和,计算每个用户在每个档次的购买量占该用户的购买量总和的比例作为用户购买每个档次商品的占比;
对于每个用户,将每个用户购买每个档次商品的占比作为一个x+1维的向量,得到每个用户的购买力向量。
计算每个用户在各个档次商品购买的占比,会得到一个X+1维的向量,优选地为五维向量(x1,x2,x3,x4,x5),称为购买力向量。其中xi代表该用户在档次为第i档的商品购买的占比。在计算每个档次占比xi的时候,不是计算在每个档次购买商品频次j(即下单量)的占比,而是计算用户在每个档次购买商品的频次×ln(价格)(即下单量与价格取对数后的乘积)的占比。比如某个用户在高档商品中共买了两个商品A、B,分别买的次数为k1、k2,价格分别为p1、p2,则用户在高档购买为k1×ln(p1)+k2×ln(p2),其中,ln表示对括号内数值的对数。按照这个方法计算出该用户在每个档次的购买量后,再求出每个档次的购买量占总的购买量的占比即可。这里加入了商品的价格取log作为权重,从而解决了背景技术中提到的现有技术的第二个缺点。加入了 商品本身的价格因素,即使用户在低价位品类买了很多相对高档的商品,也会因为价格权重低,得到调节。
如图3所示,在其中一个实施例中,所述步骤S103,具体包括:
步骤S301,包括:从所有用户的购买力向量中随机选取x+1个用户的购买力向量作为最开始的中心,执行步骤S302;
步骤S302,包括:分别计算其余的所有购买力向量到x+1个中心的欧氏距离,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行步骤S303;
步骤S303,包括:对x+1个点簇中所有购买力向量计算关于每个维度的算术平均数作为该点簇的中心,执行步骤S304;
步骤S304,包括:将所有用户的购买力向量按照步骤S304得到的中心重新计算与每个中心的欧氏聚类,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行步骤S305;
步骤S305,包括:如果每个点簇的中心不再变化,则执行步骤S306,否则执行步骤S303;
步骤S306,根据x+1个点簇的中心计算对应点簇所对应的档次。
本实施例不是简单地根据用户购买某个档次商品的数量占比大小来决定用户购买力,而是对所有用户的数据进行聚类分析来划分用户购买力等级。
通过聚类方法对用户的购买力向量进行聚类后,得到了X+1个点簇,优选为五个点簇。经观测,明显可以看出每个点簇代表着一种档次的购买力人群。将每个点簇所包括的购买力向量的用户分别归类保存到数据库中,从而完成用户购买力的分类。
优选地,所述步骤S306,具体包括:
获取x+1个点簇的中心,按照档次高低顺序依次执行如下子步骤:
高档选择子步骤,包括:选择当前待确定档次为高档;
档次确定子步骤,包括:从未确定档次的点簇的中心中,将与当前待确定档次对应的维最大的中心所对应的点簇的当前确定为当前待确定档次;
其他档选择子步骤,包括:如果还有未确定档次的点簇,则选择当前待确定档次的下一顺序档次作为当前待确定档次,执行档次确定子步骤,否则结束。
在其中一个实施例中,还包括:
分类推荐步骤,包括:当接收到用户的访问信息,获取用户的购买力档次,根据购买力档次向用户推荐相应档次的商品。
当完成了用户购买力的分类后,当一个老用户访问网站,就可以通过数据库知道该用户的购买力情况,根据用户购买力的档次推荐相应价位的商品。比如一个购买力档次高的用户在浏览手机,就推荐给他高端、智能手机;相反的,一个购买力等级很低的用户,则推荐给他一些价位低、实用的手机。
如图4所示为本发明一种对电子商务平台的用户进行分类的系统的结构模块图,包括:
商品档次确定模块401,被配置为:将同一品类中的商品基于价格和销量分布分为高档的商品和非高档的商品,将非高档的商品按照价格从高到低依次分为x个从高到低的档次,其中,x为预设的大于或等于1的自然数;
用户购买占比计算模块402,被配置为:计算每个用户购买每个档次商品的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量,所述购买力向量为x+1维向量,且每一维与一个档次对应;
用户归类模块403,被配置为:对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档 次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次。
在其中一个实施例中,所述商品档次确定模块被配置为:将同一品类中价格段前a%的商品的档次确定为高档,其中a是采用如下方法获得的:
选择三个或三个以上取值范围为在0~100之间的待选择值,最大的待选择值作为最大选择值,最小待选择值为最小选择值,其他的待选择值为中间选择值,令m为同一品类中价格的前y%的价格统计值,令n为同一品类中最近预设时间段内有销量的商品中价格的前y%的价格统计值,如果m大于n超过预设第一阈值,则选择a为最大选择值,如果n大于m超过预设第二阈值,则选择a为最小选择值中的最小值,其他情况,选择a为中间选择值中的一个,其中,y小于最大选择值且大于最小选择值
在其中一个实施例中,所述用户购买占比计算模块被配置为:
对于每个用户,计算该用户在每个档次的下单量与对应价格取对数后的乘积作为该用户在该档次的购买量,计算该用户所有档次的购买量总和,计算每个用户在每个档次的购买量占该用户的购买量总和的比例作为用户购买每个档次商品的占比;
对于每个用户,将每个用户购买每个档次商品的占比作为一个x+1维的向量,得到每个用户的购买力向量。
在其中一个实施例中,所述用户归类模块包括:
中心初始化子模块,被配置为:从所有用户的购买力向量中随机选取x+1个用户的购买力向量作为最开始的中心,执行初始归类子模块;
初始归类子模块,被配置为:分别计算其余的所有购买力向量到x+1个中心的欧氏距离,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行中心更新子模块;
中心更新子模块,被配置为:对x+1个点簇中所有购买力向量计 算关于每个维度的算术平均数作为该点簇的中心,执行归类更新子模块;
归类更新子模块,被配置为:将所有用户的购买力向量按照归类更新子步骤得到的中心重新计算与每个中心的欧氏聚类,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行收敛判断子模块;
收敛判断子模块,被配置为:如果每个点簇的中心不再变化,则执行档次确定子模块,否则执行中心更新子模块;
档次确定子模块,被配置为:根据x+1个点簇的中心计算对应点簇所对应的档次。
在其中一个实施例中,所述档次确定子模块被配置为:
获取x+1个点簇的中心,按照档次高低顺序依次执行如下子模块:
高档选择子模块,被配置为:选择当前待确定档次为高档;
档次确定子模块,被配置为:从未确定档次的点簇的中心中,将与当前待确定档次对应的维最大的中心所对应的点簇的当前确定为当前待确定档次;
其他档选择子模块,被配置为:如果还有未确定档次的点簇,则选择当前待确定档次的下一顺序档次作为当前待确定档次,执行档次确定子模块,否则结束。
在其中一个实施例中,该系统还包括:
分类推荐模块,被配置为:当接收到用户的访问信息,获取用户的购买力档次,根据购买力档次向用户推荐相应档次的商品。
以上所述实施例仅表达了本发明的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对本发明专利范围的限制。应当指出的是,对于本领域的普通技术人员来说,在不脱离本发明构思的前提下,还可以做出若干变形和改进,这些都属于本发明的保护范围。 因此,本发明专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种对电子商务平台的用户进行分类的计算机实现的方法,包括:
    商品档次确定步骤,包括:将同一品类中的商品基于价格和销量分布分为高档的商品和非高档的商品,将非高档的商品按照价格从高到低依次分为x个从高到低的档次,其中,x为预设的大于或等于1的自然数;
    用户购买占比计算步骤,包括:计算每个用户购买每个档次商品的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量,所述购买力向量为x+1维向量,且每一维与一个档次对应;
    用户归类步骤,包括:对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次。
  2. 根据权利要求1所述的方法,其中,所述商品档次确定步骤包括:将同一品类中价格段前a%的商品的档次确定为高档,其中a是采用如下方法获得的:
    选择三个或三个以上取值范围为在0~100之间的待选择值,最大的待选择值作为最大选择值,最小待选择值为最小选择值,其他的待选择值为中间选择值,令m为同一品类中价格的前y%的价格统计值,令n为同一品类中最近预设时间段内有销量的商品中价格的前y%的价格统计值,如果m大于n超过预设第一阈值,则选择a为最大选择值,如果n大于m超过预设第二阈值,则选择a为最小选择值中的最小值,其他情况,选择a为中间选择值中的一个,其中,y小于最大选择值且大于最小选择值。
  3. 根据权利要求1所述的方法,其中,所述用户购买占比计算步骤包括:
    对于每个用户,计算该用户在每个档次的下单量与对应价格取对 数后的乘积作为该用户在该档次的购买量,计算该用户所有档次的购买量总和,计算每个用户在每个档次的购买量占该用户的购买量总和的比例作为用户购买每个档次商品的占比;
    对于每个用户,将每个用户购买每个档次商品的占比作为一个x+1维的向量,得到每个用户的购买力向量。
  4. 根据权利要求1所述的方法,其中,所述用户归类步骤包括:
    中心初始化子步骤,包括:从所有用户的购买力向量中随机选取x+1个用户的购买力向量作为最开始的中心,执行初始归类子步骤;
    初始归类子步骤,包括:分别计算其余的所有购买力向量到x+1个中心的欧氏距离,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行中心更新子步骤;
    中心更新子步骤,包括:对x+1个点簇中所有购买力向量计算关于每个维度的算术平均数作为该点簇的中心,执行归类更新子步骤;
    归类更新子步骤,包括:将所有用户的购买力向量按照归类更新子步骤得到的中心重新计算与每个中心的欧氏聚类,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行收敛判断子步骤;
    收敛判断子步骤,包括:如果每个点簇的中心不再变化,则执行档次确定子步骤,否则执行中心更新子步骤;
    档次确定子步骤,包括:根据x+1个点簇的中心计算对应点簇所对应的档次。
  5. 根据权利要求1所述的方法,其中,还包括:
    分类推荐步骤,包括:当接收到用户的访问信息,获取用户的购买力档次,根据购买力档次向用户推荐相应档次的商品。
  6. 一种对电子商务平台的用户进行分类的系统,包括:
    商品档次确定模块,被配置为:将同一品类中的商品基于价格和销量分布分为高档的商品和非高档的商品,将非高档的商品按照价格 从高到低依次分为x个从高到低的档次,其中,x为预设的大于或等于1的自然数;
    用户购买占比计算模块,被配置为:计算每个用户购买每个档次商品的占比,将所述占比基于每个用户进行向量化,得到每个用户的购买力向量,所述购买力向量为x+1维向量,且每一维与一个档次对应;
    用户归类模块,被配置为:对所述购买力向量进行聚类运算,得到x+1个关于购买力向量的点簇,每个点簇分别对应一个档次,将用户对应的购买力向量所在点簇相应的档次作为用户的购买力档次。
  7. 根据权利要求6所述的系统,其中,所述商品档次确定模块,被配置为:将同一品类中价格段前a%的商品的档次确定为高档,其中a是采用如下方法获得的:
    选择三个或三个以上取值范围为在0~100之间的待选择值,最大的待选择值作为最大选择值,最小待选择值为最小选择值,其他的待选择值为中间选择值,令m为同一品类中价格的前y%的价格统计值,令n为同一品类中最近预设时间段内有销量的商品中价格的前y%的价格统计值,如果m大于n超过预设第一阈值,则选择a为最大选择值,如果n大于m超过预设第二阈值,则选择a为最小选择值中的最小值,其他情况,选择a为中间选择值中的一个,其中,y小于最大选择值且大于最小选择值
  8. 根据权利要求6所述的系统,其中,所述用户购买占比计算模块,被配置为:
    对于每个用户,计算该用户在每个档次的下单量与对应价格取对数后的乘积作为该用户在该档次的购买量,计算该用户所有档次的购买量总和,计算每个用户在每个档次的购买量占该用户的购买量总和的比例作为用户购买每个档次商品的占比;
    对于每个用户,将每个用户购买每个档次商品的占比作为一个x+1维的向量,得到每个用户的购买力向量。
  9. 根据权利要求6所述的系统,其中,所述用户归类模块包括:
    中心初始化子模块,被配置为:从所有用户的购买力向量中随机选取x+1个用户的购买力向量作为最开始的中心,执行初始归类子模块;
    初始归类子模块,被配置为:分别计算其余的所有购买力向量到x+1个中心的欧氏距离,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行中心更新子模块;
    中心更新子模块,被配置为:对x+1个点簇中所有购买力向量计算关于每个维度的算术平均数作为该点簇的中心,执行归类更新子模块;
    归类更新子模块,被配置为:将所有用户的购买力向量按照归类更新子步骤得到的中心重新计算与每个中心的欧氏聚类,将每个购买力向量分别划归到与中心的欧氏距离最小的点簇中,执行收敛判断子模块;
    收敛判断子模块,被配置为:如果每个点簇的中心不再变化,则执行档次确定子模块,否则执行中心更新子模块;
    档次确定子模块,被配置为:根据x+1个点簇的中心计算对应点簇所对应的档次。
  10. 根据权利要求6所述的系统,进一步包括:
    分类推荐模块,被配置为:当接收到用户的访问信息,获取用户的购买力档次,根据购买力档次向用户推荐相应档次的商品。
PCT/CN2016/076811 2015-03-20 2016-03-18 一种对电子商务平台的用户进行分类的方法及系统 WO2016150354A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510126137.6 2015-03-20
CN201510126137.6A CN104715409A (zh) 2015-03-20 2015-03-20 一种电子商务用户购买力分类方法及系统

Publications (1)

Publication Number Publication Date
WO2016150354A1 true WO2016150354A1 (zh) 2016-09-29

Family

ID=53414708

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/076811 WO2016150354A1 (zh) 2015-03-20 2016-03-18 一种对电子商务平台的用户进行分类的方法及系统

Country Status (3)

Country Link
CN (1) CN104715409A (zh)
HK (1) HK1208946A1 (zh)
WO (1) WO2016150354A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111709812A (zh) * 2020-06-19 2020-09-25 山东财经大学 基于用户动态分类的电商平台商品推荐方法及系统
CN112348300A (zh) * 2020-03-05 2021-02-09 北京沃东天骏信息技术有限公司 用于推送信息的方法和装置
CN115935068A (zh) * 2022-12-12 2023-04-07 杭州洋驼网络科技有限公司 互联网平台的商品推荐方法以及装置
CN115953166A (zh) * 2022-12-27 2023-04-11 鑫恒绅企业服务(无锡)有限公司 基于大数据智能匹配的客户信息管理方法及系统

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104715409A (zh) * 2015-03-20 2015-06-17 北京京东尚科信息技术有限公司 一种电子商务用户购买力分类方法及系统
CN106445941A (zh) * 2015-08-05 2017-02-22 北京奇虎科技有限公司 一种网站提供对象的推荐方法和装置
CN105516353B (zh) * 2016-01-06 2018-12-25 沈思远 一种网络信息推广方法和系统
CN105894019A (zh) * 2016-03-30 2016-08-24 北京京东尚科信息技术有限公司 一种数据库数据分类装置和方法
CN106204163B (zh) * 2016-07-29 2021-09-03 北京京东尚科信息技术有限公司 一种用户属性特征的确定方法和装置
CN107133843A (zh) * 2017-04-25 2017-09-05 丹露成都网络技术有限公司 一种基于协同过滤的商品推荐方法
CN107545470B (zh) * 2017-08-30 2021-04-30 北京京东尚科信息技术有限公司 数据处理方法及系统
CN110020135B (zh) * 2017-11-09 2023-01-31 阿里巴巴集团控股有限公司 一种需求确定方法、资源推荐方法和相关装置
CN108109042A (zh) * 2017-12-22 2018-06-01 中国联合网络通信集团有限公司 业务优惠信息推送方法及装置
CN109272373B (zh) * 2018-08-02 2021-09-14 创新先进技术有限公司 一种基于计算机的品牌推荐方法
CN113016169A (zh) * 2019-01-15 2021-06-22 深圳市欢太科技有限公司 信息推送方法及相关产品
CN110245977B (zh) * 2019-05-21 2023-10-20 中国平安人寿保险股份有限公司 一种业务升级方法、设备及计算机可读存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20010014868A1 (en) * 1997-12-05 2001-08-16 Frederick Herz System for the automatic determination of customized prices and promotions
CN102567900A (zh) * 2011-12-28 2012-07-11 尚明生 一种向客户推荐商品的方法
CN103729383A (zh) * 2012-10-16 2014-04-16 阿里巴巴集团控股有限公司 商品信息的推送方法和装置
CN104715409A (zh) * 2015-03-20 2015-06-17 北京京东尚科信息技术有限公司 一种电子商务用户购买力分类方法及系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5256362B1 (ja) * 2012-03-30 2013-08-07 楽天株式会社 情報処理装置、情報処理方法、情報処理プログラム及び記録媒体
CN103426102A (zh) * 2013-08-02 2013-12-04 苏州两江科技有限公司 基于本体分类的商品特征推荐方法
CN103412948B (zh) * 2013-08-27 2017-10-24 北京交通大学 基于聚类的协同过滤的商品推荐方法及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20010014868A1 (en) * 1997-12-05 2001-08-16 Frederick Herz System for the automatic determination of customized prices and promotions
CN102567900A (zh) * 2011-12-28 2012-07-11 尚明生 一种向客户推荐商品的方法
CN103729383A (zh) * 2012-10-16 2014-04-16 阿里巴巴集团控股有限公司 商品信息的推送方法和装置
CN104715409A (zh) * 2015-03-20 2015-06-17 北京京东尚科信息技术有限公司 一种电子商务用户购买力分类方法及系统

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112348300A (zh) * 2020-03-05 2021-02-09 北京沃东天骏信息技术有限公司 用于推送信息的方法和装置
CN111709812A (zh) * 2020-06-19 2020-09-25 山东财经大学 基于用户动态分类的电商平台商品推荐方法及系统
CN115935068A (zh) * 2022-12-12 2023-04-07 杭州洋驼网络科技有限公司 互联网平台的商品推荐方法以及装置
CN115935068B (zh) * 2022-12-12 2023-09-05 杭州洋驼网络科技有限公司 互联网平台的商品推荐方法以及装置
CN115953166A (zh) * 2022-12-27 2023-04-11 鑫恒绅企业服务(无锡)有限公司 基于大数据智能匹配的客户信息管理方法及系统
CN115953166B (zh) * 2022-12-27 2024-04-02 鑫恒绅企业服务(无锡)有限公司 基于大数据智能匹配的客户信息管理方法及系统

Also Published As

Publication number Publication date
HK1208946A1 (zh) 2016-03-18
CN104715409A (zh) 2015-06-17

Similar Documents

Publication Publication Date Title
WO2016150354A1 (zh) 一种对电子商务平台的用户进行分类的方法及系统
CN109087177B (zh) 向目标用户推荐商品的方法、装置及计算机可读存储介质
CN107169052B (zh) 推荐方法及装置
JP5965911B2 (ja) オンライン取引プラットフォームに基づくデータ処理
CN106779809B (zh) 一种大数据平台的价格信息优化组合方法与系统
TWI512653B (zh) Information providing method and apparatus, method and apparatus for determining the degree of comprehensive relevance
CN110580649B (zh) 一种商品潜力值的确定方法和装置
CN103136683A (zh) 计算产品参考价格的方法、装置及产品搜索方法、系统
US20190108571A1 (en) System and method for efficiently determining relevant products for display in an online marketplace
CN108205768A (zh) 数据库建立方法和数据推荐方法及装置、设备和存储介质
TW201501059A (zh) 訊息推薦方法及裝置
CN107403345A (zh) 畅销商品预测方法及系统、存储介质及电子终端
CN107122980A (zh) 识别商品所属类目的方法和装置
CN104615721B (zh) 用于基于退货关联信息推荐商品的方法和系统
CN111738805B (zh) 基于行为日志的搜索推荐模型生成方法、设备和存储介质
CN110930214A (zh) 一种用于满减活动的商品排序方法
CN105468628B (zh) 一种排序方法及装置
CN105630836A (zh) 搜索结果的排序方法和装置
CN110111179A (zh) 药品组合的推荐方法、装置及计算机可读存储介质
WO2017041226A1 (zh) 一种商品信息推送方法
US20240062267A1 (en) Systems and methods for determining price bands and user price affinity predictions using machine learning architectures and techniques
CN112488863A (zh) 一种在用户冷启动场景下的险种推荐方法及相关设备
CN111445133A (zh) 一种材料管理方法、装置、计算机设备和存储介质
CN113781106B (zh) 商品运营数据分析方法、装置、设备及计算机可读介质
US20210090105A1 (en) Technology opportunity mapping

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: 16767731

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 16/01/2018)

122 Ep: pct application non-entry in european phase

Ref document number: 16767731

Country of ref document: EP

Kind code of ref document: A1