CN108399550B

CN108399550B - User grouping method

Info

Publication number: CN108399550B
Application number: CN201710067597.5A
Authority: CN
Inventors: 马添; 石野; 宋丕宇; 党白璐; 郑超; 刘俊
Original assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Jingdong Shangke Information Technology Co Ltd
Current assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Jingdong Shangke Information Technology Co Ltd
Priority date: 2017-02-07
Filing date: 2017-02-07
Publication date: 2021-05-25
Anticipated expiration: 2037-02-07
Also published as: CN108399550A

Abstract

The invention discloses a user grouping method, which comprises the following steps: determining one or more exclusive attributes of the item according to the browsing behavior of the user who has purchased the goods under the item within a first preset browsing time range; traversing and combining each commodity attribute value of the exclusive attribute with each commodity attribute value of other exclusive attributes to obtain a user group corresponding to each combination; and allocating users who have the intention of purchasing the goods under the category within the second preset browsing time range to the corresponding user groups. The exclusive attribute is a commodity attribute and is provided with a plurality of commodity attribute values, and the plurality of commodity attribute values completely cover all commodities in the class; the commodity attribute values of the same commodity attribute selected by the same user group in the process of browsing the commodities under the commodity category are most concentrated, and the commodity attribute values have lower contact degree with other commodity attribute values of the commodity attribute. By adopting the invention, the user groups can be distinguished according to the real purchasing demands of the users.

Description

User grouping method

Technical Field

The invention relates to the technical field of electronic commerce, in particular to a user grouping method.

Background

In recent years, with the rapid development of electronic commerce, each e-commerce company accurately recommends its intended commodity for the user based on user behavior data accumulated by its own platform, thereby promoting the user to make an order, becoming an important means for each company to improve the user conversion rate, and being a shortcut for the e-commerce company to optimize the user experience and enhance the user stickiness. In view of this, a set of BI function modules like "guess you like", "similar recommendations", etc. are applied by companies one after another according to user browsing, search data production.

Most of the existing BI functions are used for user classification and commodity recommendation through user images of regions, genders, ages and the like of users. The data base typically comes from the user's filled out personal information and IP, browser cache, etc. Meanwhile, user classification and commodity recommendation are performed through user behaviors, such as determining user preferences according to user browsing and purchased commodities. Marking goods or directly classifying users according to goods categories is generally adopted. And respectively applying browsing and purchasing data of the user to different recommendation scenes.

While current BI logic can fulfill the thousands of needs of people, it suffers from the following drawbacks due to its lack of efficient analysis and utilization of user behavior and commodity data:

(1) display resource limitation

The display logic of the recommendation function on the current line is as follows: when the goods which are manually selected for the user and exposed cannot meet the purchase demand or the goods are not in good, the goods which are possibly purchased by the user are displayed according to the user behavior data. Thus, such recommendation functions are only available at the bottom of the page or next to the merchandise. The method leads to the situation that a large number of high-quality resource positions on a page are not put in a personalized mode, and the putting of commodities on one side of thousands of people can interfere with potential purchasing users, so that the users lose.

(2) Inaccurate personalized recommendation

Most of the commodity recommendation dimensions on the current line are in the class level or directly and repeatedly show commodities browsed by a user. Since the cost of browsing the commodities is low when the users buy on the internet, the repeated display of the browsed commodities obviously does not have the conversion rate improving effect. And the recommendation accuracy of the commodity of the class level is difficult to guarantee because the attribute subdivision of the commodity is not carried out. Therefore, only personalized resource positions on the page face the current situations that the recommended commodities are not accurate and the conversion rate improvement effect is not ideal.

(3) Lack of generalization to target Consumer historical behavioral data

At present, personalized recommendation is carried out on a user, and only the browsing behavior of the user in a short term is directly used, and no analysis is carried out on data. Therefore, the e-commerce platform is not clear which commodity the user most wants to purchase, and accurate recommendation and resource allocation release for the user are not available. The behavior data of the user can help us to draw a conclusion, and generally, according to the frequency of browsing a certain category of commodities by the user, the user can judge that the user intentionally purchases the category of commodities and continuously recommends other commodities under the category, but the target commodity is not analyzed in a more detailed dimension through the browsing behavior of the target customer group.

Disclosure of Invention

The invention aims to provide a user grouping method which can distinguish user groups according to real purchasing demands of users.

In order to achieve the above object, the present invention provides a user grouping method, which is applied to users who have a desire to purchase goods of the same category, and comprises:

determining one or more exclusive attributes of the item according to the browsing behavior of the user who has purchased the goods under the item within a first preset browsing time range;

traversing and combining each commodity attribute value of the exclusive attribute with each commodity attribute value of other exclusive attributes to obtain a user group corresponding to each combination;

allocating users who have the intention of purchasing the commodities under the category within a second preset browsing time range to corresponding user groups;

the exclusive attribute is a commodity attribute and is provided with a plurality of commodity attribute values, and the plurality of commodity attribute values completely cover all commodities in the class; the commodity attribute values of the same commodity attribute selected by the same user group in the process of browsing the commodities under the commodity category are most concentrated, and the commodity attribute values have lower contact degree with other commodity attribute values of the commodity attribute.

In summary, the user clustering method provided by the present invention performs data statistics according to the browsing behavior of the user who has purchased the commodity under the category within the first predetermined browsing time range, and uses the attribute of the commodity selected in the browsing behavior set as the exclusive attribute; according to the traversal combination of the exclusive attribute commodity attribute values, obtaining a user group corresponding to each combination; and finally, distributing the users who have the intention of purchasing the commodities under the category within the second preset browsing time range to the corresponding user groups. Thereby realizing the accurate subdivision of the user group. Compared with the prior art, the invention summarizes historical behavior data of the target customer group and subdivides the commodity attributes, thereby effectively realizing the subdivision of the group according to the purchase demand of the user.

Drawings

Fig. 1 is a flow chart of a user grouping method according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Embodiment one user grouping

Fig. 1 is a schematic flow chart of a user grouping method according to the present invention, which is applied to users who have a desire to purchase goods of the same category, and the method includes:

step 11, determining one or more exclusive attributes of the item according to the browsing behavior of the user who has purchased the goods under the item within a first preset browsing time range;

the method for determining one or more exclusive attributes of the item according to the browsing behavior of the user who has purchased the goods under the item within the first preset browsing time range comprises the following steps:

s111, counting the average browsing stock quantity unit (SKU) number and the average browsing SKU number of each commodity attribute value of the same commodity attribute of a user who has purchased the commodity in the category within a first preset browsing time range;

a SKU is a physically indivisible unit of stock in a minimum, typically one SKU for each item. For example, SKU 1: long iris 55 inches, SKU 2: TCL55 inches, SKU 3: 55 inches for music, assuming that a user browses SKU1 five times, SKU2 fifteen times and SKU3 five times, it can be determined that the number of times SKU is browsed is 20 and the number of times SKU is browsed is 3.

And S112, if the ratio of the average browsing SKU number of the user to the maximum value of the average browsing SKU number in all the commodity attribute values of the commodity attribute in the user group corresponding to each commodity attribute value of the commodity attribute is larger than a preset threshold value, determining that the commodity attribute is an exclusive attribute.

Step 12, traversing and combining each commodity attribute value of the exclusive attribute with each commodity attribute value of other exclusive attributes to obtain a user group corresponding to each combination;

step 13, distributing users who have the intention of purchasing the commodities under the category within a second preset browsing time range to corresponding user groups;

the method for distributing the users who have the willingness to purchase the commodities under the category in the second preset browsing time range to the corresponding user groups comprises the following steps:

s131, counting the average browsing SKU number and the average browsing SKU number of each commodity attribute value of users who have purchased commodities under the category in a first preset browsing time range for the same combination;

s132, for each exclusive attribute in the combination, obtaining a commodity attribute value with the maximum value ratio in all commodity attribute values of the exclusive attribute according to the average browsing SKU number and the average browsing SKU number of each commodity attribute value; setting a screening threshold corresponding to the commodity attribute value with the maximum value ratio according to the maximum value ratio;

s133, distributing the users with the average browsing SKU number and the average browsing SKU number of the users exceeding each screening threshold value in the combination in the second preset browsing time range to the user group corresponding to the combination.

Thus, the user grouping method of the present invention is completed.

The exclusive attribute is an attribute which is very important for grouping the users, is a commodity attribute and has a plurality of commodity attribute values, and the commodity attribute values completely cover all commodities in the class; the commodity attribute values of the same commodity attribute selected by the same user group in the process of browsing the commodities under the commodity category are most concentrated, and the commodity attribute values have lower contact degree with other commodity attribute values of the commodity attribute. In summary, the exclusive attribute is characterized by being centralized and exclusive, that is, centralized means that browsing behaviors of the same user group are centralized on a certain item attribute value of the item attribute; the exclusive property means that when a certain commodity attribute value is selected in a browsing behavior set of the same user group, other commodity attribute values of the commodity attribute are rarely selected. The commodity attributes include basic attributes such as brand, price, color, etc. for representing natural information carried by the commodity itself; extended attributes for indicating commodity characteristic information, such as usage, level, and the like; and image attributes for representing abstract information of the commodity, such as small assets, businesses, etc. Each product attribute and product attribute value differ according to the type of product. Therefore, the exclusive attribute may be brand, use, or business. One category may have a plurality of exclusive attributes, and the exclusive attributes of the category need to be determined according to the historical browsing behavior of the user. The more exclusive attributes a category has, the more exclusive attributes a subdivided user group must have, the more accurate the user group is, and the more easily the purchasing preference of the target user is analyzed.

For clarity of the present invention, the following description will be made by taking specific scenarios.

The application of the exclusive property relates to a plurality of categories such as televisions, air conditioners, water heaters, computers, mobile phones and the like, and one of the categories is selected for illustration.

1. Determining exclusive attributes and subdividing user populations

Taking a television as an example, when the category is television, firstly, the commodity attributes of the television are briefly classified as follows:

basic properties: price, brand, color, size;

and (3) expanding the attribute: display technology, use, manufacturer type, networking function;

image attribute: definition takes precedence, sound effects take precedence.

Each of the above-mentioned commodity attributes may be analyzed as an exclusive attribute.

If the product attribute is a use, the product attribute value of the product attribute may include large-room use, small-room use, and bedroom use. If the commodity attribute is a manufacturer type, the commodity attribute value of the commodity attribute can comprise a domestic brand, a joint venture brand and an internet brand.

Most users purchase the television, the time from browsing to ordering is not more than 15 days, so that the users who order to purchase the television within the last 3 days are intercepted, and the browsing behaviors of the users 15 days before are analyzed.

The exclusive attribute of the television can gradually count the browsing behaviors of the user on each commodity attribute of the television through a machine learning method, namely counting the average browsing Stock Keeping Unit (SKU) number and the average browsing SKU number of each commodity attribute value of the same commodity attribute in the previous 15 days of the user who places an order to purchase the television in the last 3 days. This embodiment is described by way of example only with respect to the number of views of the SKU for clarity of description.

Table 1 shows the average number of SKU views per purchased user in the usage dimension.

	For browsing a large living room	For browsing small living rooms	For browsing bedroom	Ratio of maximum value
					Purchased large living room	59.1	22.8	4.6	68％
Purchased for living room	3.6	75.2	10.2	84％
					Purchased for bedroom use	1.1	11.6	46.9	79％

TABLE 1

Table 2 shows the average SKU views for each purchased user in the vendor type dimension.

TABLE 2

As can be seen from tables 1 and 2, a user who has a preference for a certain item does not browse other products of the same dimension. For example, a user who has purchased a large living room has an average number of times of viewing a television for the large living room of 59.1, a maximum percentage of all product attribute values for use of up to 68%, and a small number of times of viewing a television for the small living room and a television for a bedroom. A user who has purchased a home brand views the home brand 53.8 on average, accounts for up to 60% of all product attribute values of the manufacturer type, and views a joint-good brand and an internet brand only rarely. Other commercial attributes for television were also analyzed as in table 1, table 2. Finally, the two commodity attributes of the application and the manufacturer type are obtained through analysis and are suitable as exclusive attributes of the television, and a plurality of commodity attribute values of the application completely cover all commodities of the television; the commodity attribute values of the same commodity attribute selected by the same user group in the process of browsing the television commodities are most concentrated, and the commodity attribute values have lower contact degree with other commodity attribute values of the commodity attribute. A plurality of commodity attribute values of the manufacturer type completely cover all commodities of the television; the commodity attribute values of the same commodity attribute selected by the same user group in the process of browsing the television commodities are most concentrated, and the commodity attribute values have lower contact degree with other commodity attribute values of the commodity attribute.

In order to subdivide the user groups, the user groups corresponding to each combination are obtained by the users under the combined action of the two exclusive attributes.

Table 3 shows the average number of SKU views per purchased user given the combination of the two exclusive attributes.

TABLE 3

As can be seen from table 3, by traversing and combining each commodity attribute value (for large living room, small living room, and bedroom) of the use exclusion attribute with each commodity attribute value (for domestic brand, joint venture brand, and internet brand) of the manufacturer type exclusion attribute, the user can be subdivided into 9 more accurate groups, which are respectively large living room + domestic product, large living room + joint venture, large living room + internet, small living room + domestic product, small living room + joint venture, small living room + internet, bedroom + joint venture, and bedroom + internet.

Through the data analysis, the two exclusive attributes do not interfere with each other, and under the combined action of the two exclusive attributes of the application and the manufacturer type, the purchase target distinction among the user groups is more obvious. Taking the user group who has bought the large living room and the domestic television as an example, the average number of browsing the large living room is 14.9, the average number of browsing the small living room is 5, the average number of browsing the bedroom is 1.2, and the maximum value of all the commodity attribute values used by the large living room is up to 71%. The average number of browsing home brands was 16.7, the average number of browsing joint-venture brands was 3, the average number of browsing internet brands was 1.4, and the percentage of home brands to the maximum of all the product attribute values of the manufacturer type was as high as 79%.

2. Screening target user groups

And grouping the target users who browse but have not purchased by using the two exclusive attributes of the obtained purposes and the types of the manufacturers.

First, a screening range is determined, and users who wish to purchase a television are screened before grouping, so that users who have both the following two points need to be screened. According to statistics, when the number of television SKUs browsed by a general user is 15, the order is placed only when the browsing time is 15 days, so that the following steps are set according to the decision time of the user for purchasing the television:

(1) user browsing TV but not purchasing within 15 days

(2) Browsing users with television SKU numbers less than 15

And (4) regarding the users meeting the two conditions as the users who really have purchasing intention at present. It is assumed that 490 ten thousand users who have a desire to purchase in this embodiment.

According to table 3, a filtering threshold is set for each subdivided user population. For example, users may be screened according to 60% of the average value in the target dimension of the group, that is, users who browse a large living room more than 14.9 × 60% to 9 times and browse a domestic brand more than 16.7 × 60% to 10 times, and are divided into the large living room + domestic user group. By analogy, 490 users with willingness to purchase are assigned to 9 segment groups. At this time, about 16 thousands of high-potential users who purchase products with corresponding combined product attribute values are screened out from 490 thousands of users who browse unpurchased products. The specific distribution is shown in table 4:

TABLE 4

And subsequently, when the high-potential users in the table 4 are recommended to be commodities and recommended to be commodity display, the commodity recommendation can be carried out according to the user group to which each user belongs.

Example two Commodity recommendation

In order to utilize the screened users, the invention respectively gives different promotion benefits to the screened users of different types so as to promote the conversion.

And counting the SKU which is browsed by the user most frequently in a second preset browsing time range in the user group to which the user belongs, and selecting the SKU if the SKU is in the promotion SKU list.

If the SKU is not in the promotion SKU list, judging the promotion SKU list according to the user group to which the user belongs; if the user group to which the user belongs corresponds to one SKU in the promotion SKU list, selecting the SKU; and if the user group to which the user belongs corresponds to a plurality of SKUs in the promotion SKU list, selecting the SKUs, or judging the promotion SKU list according to the commodity attribute value.

The method for judging the promotion SKU list according to the commodity attribute value comprises the following steps:

judging the promotion SKU list according to the commodity attribute value with the highest browsing frequency in the user group to which the user belongs within the second preset browsing time range; selecting a SKU if the item attribute value corresponds to the SKU in the list of promotional SKUs; and if the item attribute value corresponds to a plurality of SKUs in the promotion SKU list, selecting the plurality of SKUs, or judging the promotion SKU list according to other item attribute values which are browsed by the user in the user group to which the user belongs within a second preset browsing time range until the SKU in the promotion SKU list is matched, otherwise, abandoning the user.

Using the television example, the list of promotional SKUs is shown in Table 5:

TABLE 5

After screening, the users who have three users who meet the user screening rule of the television categories are divided into bedroom and domestic user groups as high-potential users, in the user groups, the SKU with the highest browsing frequency is the WeChat 40 inch A, but the sales promotion SKU list has no television, and the sales promotion SKU list is judged according to the bedroom and the domestic of the user group to which the users belong: if the user group to which the user belongs corresponds to one SKU in the promotion SKU list, selecting the SKU; and if the user group to which the user belongs corresponds to a plurality of SKUs in the promotion SKU list, selecting the SKUs, or judging the promotion SKU list according to the commodity attribute value. In this embodiment, bedroom + domestic corresponds to three SKUs in the promotion SKU list.

In this embodiment, SKUs that are exactly matched to the list of promotional SKUs may be matched to two SKUs in the list of promotional SKUs, assuming that the SKU brand that was most frequently viewed by the user in the user group to which the user belongs within the second predetermined viewing time period is a hai letter. Further matching the SKU size, and assuming that the SKU size with the highest number of viewings by the user in the user group to which the user belongs within the second predetermined viewing time range is 42 inches, corresponding to one SKU in the list of promoted SKUs. Therefore, the SKU of the promotion SKU list, namely, the Credit B, 42 inches, the home and the bedroom is selected and recommended to the user for three. The user Zhang III finally enjoys the sales promotion preferential special price of the WeChat B money. And by analogy, each screened high-potential user is matched with the corresponding promotion SKU.

If the size of the SKU is matched with a plurality of SKUs in the promotion SKU list, selecting a SKU with a relatively large browsing frequency of the user; if the sizes are not matched to be consistent, the following steps can be further carried out: and matching absolute values of the size difference, and selecting the SKU with the smallest difference, wherein if a plurality of SKUs with the same absolute value exist in the promotion SKU list, the number of browsing times of the user is relatively large. If no suitable SKU match is successful, the user is abandoned.

That is, the matching rules are flexible and can be set according to specific categories. The specific matched commodity attribute value can be flexibly set according to the specific category.

Example three recommended merchandise display

In order to enable the matched user to receive the promotion information in time, the selected SKU and the corresponding preferential price can be sent to the user in a short message, mail and application software APP pushing mode. The method mainly prompts the user that the user is selected to enjoy the exclusive preferential price of a certain SKU, and informs the user of the effective expiration date of the price to remind the user to place an order as soon as possible.

In addition, the user screened in the first embodiment may be applied to other scenarios. For example, on the activity map and focus map of the website home page and APP home page, and in various activity pages on the e-commerce platform, the selected SKU and the preferential price corresponding to the SKU are displayed according to the user group to which the user belongs.

For example, in a large activity page of home appliances, for a bedroom + home user, the first screen shows a television with the highest browsing frequency, and the last screens show some SKUs of the bedroom + home brand. Therefore, the method is closer to the purchase demand of the user, so that the browsing and searching paths of the user are saved, and meanwhile, the interference of non-target SKUs on the user can be reduced, and the conversion rate is influenced.

In summary, the present invention discloses a user group division method based on exclusive attributes, which is existed but not well analyzed and utilized, and generates subdivided user groups by using multidimensional exclusive attributes, so that the divided user groups do not overlap or affect each other, and all commodities can be completely covered. The target users are found out based on the exclusive attribute grouping, the method of accurate commodity recommendation is applied, recommended commodities are released in the high-quality resource positions of the user pages, and the user conversion rate is greatly improved.

By the technical scheme, the e-commerce platform can be helped to finish user locking more accurately, and the marketing effect is amplified, so that the user conversion rate is improved. And meanwhile, the system helps the user to more directly acquire the marketing information of the psychographic commodity, so that more benefits are obtained. Specific measures include, but are not limited to, the following:

(1) and optimizing the commodity ordering of the list page. And optimizing the sequence of the listing pages of the e-commerce platform according to the behavior characteristics of the user, and preferentially displaying the commodities which the user wishes to purchase in front of the listing pages of the e-commerce platform so as to reduce the interference of other irrelevant commodities to the user.

(2) And optimizing the commodity screening label on the list page. Various exclusive attributes are displayed in a list page screening box as commodity labels, so that users can screen heart instrument commodities more accurately, and the commodity selection efficiency of the users is improved.

(3) The split flows operate limited marketing locations. And carrying out flow distribution on the limited display positions in the promotion activity page. Different target commodities are displayed at the same position for users with different purchasing demands, accurate matching between the users and the commodities is optimized, and purchasing efficiency of the users is improved.

(4) Personalized delivery of marketing strategies. The browsing behavior of the user is monitored, the high-potential user is automatically identified, the purchasing time of the high-potential user is pre-judged, and marketing methods with different strategies are supplemented.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A user grouping method is applied to users who have the desire to purchase goods in the same category, and comprises the following steps:

the exclusive attribute is a commodity attribute and is provided with a plurality of commodity attribute values, and the plurality of commodity attribute values completely cover all commodities in the class; the commodity attribute values of the same commodity attribute selected by the same user group in the process of browsing the commodities under the commodity category are most concentrated, and the commodity attribute values have lower contact degree with other commodity attribute values of the commodity attribute;

counting the average browsing SKU number and the average browsing SKU number of each commodity attribute value of the same commodity attribute of a user who has purchased commodities under the category within a first preset browsing time range;

and if the maximum ratio of the average browsing SKU number of the user to the average browsing SKU number in all the commodity attribute values of the commodity attribute in the user group corresponding to each commodity attribute value of the commodity attribute is larger than a preset threshold value, determining that the commodity attribute is an exclusive attribute.

2. The method of claim 1, wherein the method of assigning users within the second predetermined browsing time range who have a desire to purchase items of the category to the corresponding group of users comprises:

counting the average browsing SKU number and the average browsing SKU number of each commodity attribute value of users who have purchased commodities under the category in a first preset browsing time range for the same combination;

for each exclusive attribute in the combination, obtaining a commodity attribute value with the maximum value ratio in all commodity attribute values of the exclusive attribute according to the average browsing SKU number and the average browsing SKU number of each commodity attribute value; setting a screening threshold corresponding to the commodity attribute value with the maximum value ratio according to the maximum value ratio;

and allocating the users with the average browsing SKU number and the average browsing SKU number of the users exceeding each screening threshold value in the combination in a second preset browsing time range to the user group corresponding to the combination.

3. The method of claim 1, further comprising: and counting a SKU which is browsed by the user most frequently in a second preset browsing time range in the user group to which the user belongs, and selecting the SKU if the SKU is in the promotion SKU list for promotion.

4. The method of claim 3, further comprising: if the SKU is not in the promotion SKU list, judging the promotion SKU list according to the user group to which the user belongs; if the user group to which the user belongs corresponds to one SKU in the promotion SKU list, selecting the SKU; and if the user group to which the user belongs corresponds to a plurality of SKUs in the promotion SKU list, selecting the SKUs, or judging the promotion SKU list according to the commodity attribute value.

5. The method of claim 4 wherein determining a list of promotional SKUs based on item attribute values comprises:

6. The method of claim 3, 4 or 5, wherein the selected SKU and its corresponding preferential price are sent to the user by short message, mail, application APP push.

7. A method as claimed in claim 3, 4 or 5 wherein the selected SKU and its corresponding offer price are presented in an activity page in dependence upon the group of users to which the user belongs.

8. The method according to claim 1, wherein the commodity attributes include a basic attribute for representing natural information carried by the commodity itself, an extended attribute for representing commodity characteristic information, and an image attribute for representing commodity abstract information; each product attribute and product attribute value differ according to the type of product.

9. An electronic device, comprising:

one or more processors;

storage means for storing one or more programs;

when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-8.

10. A computer-readable medium, on which a computer program is stored, wherein the program, when executed by a processor, implements the method of any one of claims 1-8.