CN117764632A

CN117764632A - Client management method and device, model training method and device

Info

Publication number: CN117764632A
Application number: CN202211115732.6A
Authority: CN
Inventors: 伍克远
Original assignee: 4Paradigm Beijing Technology Co Ltd
Current assignee: 4Paradigm Beijing Technology Co Ltd
Priority date: 2022-09-14
Filing date: 2022-09-14
Publication date: 2024-03-26

Abstract

The present disclosure relates to a client management method and apparatus, and a model training method and apparatus, where the client management method includes: determining a target customer from the plurality of customers based on the time of occurrence of the last at least one order of the plurality of customers for the predetermined category of merchandise; acquiring basic information of a target client and historical information of the target client related to a commodity of a preset category; predicting behavior information of the target client by using at least one machine learning model based on the acquired basic information and history information; and executing preset operation aiming at the target client according to the predicted behavior information.

Description

Client management method and device, model training method and device

Technical Field

The present disclosure relates to the field of artificial intelligence, and more particularly, to a client management method and apparatus, a model training method and apparatus, an electronic device, and a computer-readable storage medium.

Background

It is well known that in most industries, such as the fast-food industry (e.g., catering), the life cycle of customers can be divided into: the introduction phase, the growth phase, the maturation phase, the sleep phase and the loss phase. In these industries, customers are generally classified according to the length of time for their repurchase, and corresponding operations are performed for customers at different stages.

However, some industries have distinct customers and sales patterns. Taking the luxury industry as an example, since luxury goods are generally expensive, customers purchase very frequently (e.g., some customers purchased a new series of watches again only ten years), and thus, customer staging does not have a very clear boundary, if the corresponding operations for the customer are performed solely according to the length of the time for the re-purchase, it is likely that the corresponding operations performed for the customer are not appropriate and customer management cannot be performed accurately and properly. In view of this, there is a need for more efficient customer management methods for special industries such as luxury goods.

Disclosure of Invention

According to a first aspect of the present disclosure, there is provided a client management method including: determining a target customer from a plurality of customers based on the time of occurrence of the last at least one order of the plurality of customers for a predetermined category of merchandise; acquiring basic information of the target client and history information of the target client related to the commodity of the preset category; predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information; and executing preset operation aiming at the target client according to the predicted behavior information.

Optionally, the history information includes history order information about the target customer about the commodity of the predetermined category and history activity information about the target customer about the commodity of the predetermined category, wherein the predicting, based on the obtained basic information and the history information, behavior information of the target customer using at least one machine learning model includes: based on the acquired basic information, the historical order information and the historical activity information, predicting an interaction effect of the target client after being touched in each of a plurality of touch modes by using a first machine learning model as the predicted behavior information.

Optionally, the history information includes history order information about the target customer about the commodity of the predetermined category and history activity information about the target customer about the commodity of the predetermined category, wherein the predicting, based on the obtained basic information and the history information, behavior information of the target customer using at least one machine learning model includes: based on the acquired basic information, the historical order information and the historical activity information, predicting the probability that the target client participates in the activity of the commodity of the preset category after acquiring the commodity related activity information of the commodity of the preset category in a preset touch mode by using a second machine learning model as the predicted behavior information.

Optionally, the history information includes history order information about the target customer about the predetermined category of commodities and commodity information about the target customer historically purchased the predetermined category of commodities, wherein the predicting behavior information of the target customer using at least one machine learning model based on the obtained basic information and the history information includes: based on the acquired basic information, the history order information, and the commodity information, predicting a probability that the target customer purchases the commodity of the predetermined category again using a third machine learning model as the predicted behavior information.

Optionally, the performing a preset operation for the target client according to the predicted behavior information includes: determining a preferred touch mode of the target client from among the plurality of touch modes according to the predicted behavior information; providing the target customer with relevant information about the predetermined category of merchandise in a preferred reach of the target customer.

Optionally, the performing a preset operation for the target client according to the predicted behavior information includes: determining at least one client from the target clients based on the predicted behavior information; performing an operation related to the predetermined category of merchandise for the determined at least one customer.

Optionally, the performing, for the determined at least one customer, an operation related to the predetermined category of merchandise includes: and providing relevant information about the commodity of the preset category to the at least one client in a respective preference touch manner of the at least one client.

Optionally, the related information about the commodity of the predetermined category includes activity information about the commodity of the predetermined category and/or commodity information about the commodity of the predetermined category.

Optionally, the determining the target customer from the plurality of customers based on the occurrence time of the last at least one order of the plurality of customers for the predetermined category of commodities includes: sorting the plurality of customers according to the occurrence time of the last order of each customer of the plurality of customers for the predetermined category of merchandise; determining a first-ordered predetermined proportion of the plurality of customers as the target customer, wherein the predetermined proportion varies according to each brand of merchandise of the predetermined category.

Optionally, the preset operation for the target client is performed according to the predicted behavior information: and if the predicted behavior information meets the preset condition, outputting early warning prompt information.

Optionally, the client management method further includes: after providing the related information on the commodity of the predetermined category, feedback information of each customer provided with the related information is collected for the related information.

According to a second aspect of the present disclosure, there is provided a model training method comprising: determining a target customer from a plurality of customers based on the time of occurrence of the last at least one order of the plurality of customers for a predetermined category of merchandise; acquiring basic information of the target client, historical information of the target client related to the commodity of the preset category and real behavior information of the target client; predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information, and adjusting parameters of the at least one machine learning model according to the predicted behavior information and the real behavior information.

According to a third aspect of the present disclosure, there is provided a client management apparatus, which may include: a target customer determination unit configured to determine a target customer from a plurality of customers based on occurrence times of at least one recent order of the plurality of customers for a predetermined category of commodities; an information acquisition unit configured to acquire basic information of the target customer and history information of the target customer concerning the predetermined category of merchandise; a prediction unit configured to predict behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information; and the execution unit is configured to execute preset operation aiming at the target client according to the predicted behavior information.

Optionally, the history information includes history order information about the target customer about the commodity of the predetermined category and history activity information about the target customer about the commodity of the predetermined category, wherein the prediction unit is specifically configured to: based on the acquired basic information, the historical order information and the historical activity information, predicting an interaction effect of the target client after being touched in each of a plurality of touch modes by using a first machine learning model as the predicted behavior information.

Optionally, the history information includes history order information about the target customer about the commodity of the predetermined category and history activity information about the target customer about the commodity of the predetermined category, wherein the prediction unit is specifically configured to: based on the acquired basic information, the historical order information and the historical activity information, predicting the probability that the target client participates in the activity of the commodity of the preset category after acquiring the commodity related activity information of the commodity of the preset category in a preset touch mode by using a second machine learning model as the predicted behavior information.

Optionally, the history information includes history order information of the target customer about the commodity of the predetermined category and commodity information about the commodity of the predetermined category purchased by the target customer in history, wherein the prediction unit is specifically configured to: based on the acquired basic information, the history order information, and the commodity information, predicting a probability that the target customer purchases the commodity of the predetermined category again using a third machine learning model as the predicted behavior information.

Optionally, the execution unit is specifically configured to: determining a preferred touch mode of the target client from among the plurality of touch modes according to the predicted behavior information; providing the target customer with relevant information about the predetermined category of merchandise in a preferred reach of the target customer.

Optionally, the execution unit is specifically configured to: determining at least one client from the target clients based on the predicted behavior information; performing an operation related to the predetermined category of merchandise for the determined at least one customer.

Optionally, the execution unit is specifically configured to: and providing relevant information about the commodity of the preset category to the at least one client in a respective preference touch manner of the at least one client.

Optionally, the target client determining unit is specifically configured to: sorting the plurality of customers according to the occurrence time of the last order of each customer of the plurality of customers for the predetermined category of merchandise; determining a first-ordered predetermined proportion of the plurality of customers as the target customer, wherein the predetermined proportion varies according to each brand of merchandise of the predetermined category.

Optionally, the execution unit is specifically configured to: and if the predicted behavior information meets the preset condition, outputting early warning prompt information.

Optionally, the client management device further includes: and a feedback information collecting unit configured to collect feedback information for the relevant information for each customer provided with the relevant information after the relevant information on the commodity of the predetermined category is provided.

According to a fourth aspect of the present disclosure, there is provided a model training apparatus comprising: a determining unit configured to determine a target customer from a plurality of customers based on occurrence times of at least one recent order of the plurality of customers for a predetermined category of commodities; an acquisition unit configured to acquire basic information of the target customer, history information of the target customer regarding the predetermined category of merchandise, and real behavior information of the target customer; a training unit configured to: predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information; and adjusting parameters of the at least one machine learning model according to the predicted behavior information and the real behavior information.

Optionally, the history information includes history order information about the target customer about the commodity of the predetermined category and history activity information about the target customer about the commodity of the predetermined category, wherein the training unit is specifically configured to: based on the acquired basic information, the historical order information and the historical activity information, predicting an interaction effect of the target client after being touched in each of a plurality of touch modes by using a first machine learning model as the predicted behavior information.

Optionally, the history information includes history order information about the target customer about the commodity of the predetermined category and history activity information about the target customer about the commodity of the predetermined category, wherein the training unit is specifically configured to: based on the acquired basic information, the historical order information and the historical activity information, predicting the probability that the target client participates in the activity of the commodity of the preset category after acquiring the commodity related activity information of the commodity of the preset category in a preset touch mode by using a second machine learning model as the predicted behavior information.

Optionally, the history information includes history order information of the target customer about the predetermined category of commodities and commodity information about the target customer historically purchased the predetermined category of commodities, wherein the training unit is specifically configured to: based on the acquired basic information, the history order information, and the commodity information, predicting a probability that the target customer purchases the commodity of the predetermined category again using a third machine learning model as the predicted behavior information.

Optionally, the determining unit is specifically configured to: sorting the plurality of customers according to the occurrence time of the last order of each customer of the plurality of customers for the predetermined category of merchandise; determining a first-ordered predetermined proportion of the plurality of customers as the target customer, wherein the predetermined proportion varies according to each brand of merchandise of the predetermined category.

According to a fifth aspect of the present disclosure, there is provided an electronic device comprising: at least one processor; at least one memory storing computer-executable instructions, wherein the computer-executable instructions, when executed by the at least one processor, cause the at least one processor to perform a model training method and/or a customer management method as described above.

According to a sixth aspect of the present disclosure, there is provided a computer readable storage medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform a model training method and/or a customer management method as described above.

According to the model training method and device disclosed by the invention, at least one machine learning model capable of effectively predicting the behavior information of the target client can be trained.

According to the client management method and apparatus of the present disclosure, since after a target client is determined from a plurality of clients based on occurrence time of at least one last order of the clients for a predetermined category of commodity, behavior information of the target client is predicted using at least one machine learning model based on basic information of the target client and history information of the target client related to the predetermined category of commodity, and then a preset operation for the target client is performed according to the predicted behavior information, the preset operation for the target client can be performed more accurately or more pertinently, so that more efficient client management can be achieved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments consistent with the disclosure and, together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.

FIG. 1 is a flow chart of a model training method according to an embodiment of the present disclosure;

FIG. 2 is a flow chart of a customer management method according to an embodiment of the present disclosure;

FIG. 3 is a schematic diagram implementing customer management according to an embodiment of the present disclosure;

FIG. 4 is a block diagram of a model training apparatus according to an embodiment of the present disclosure;

FIG. 5 is a block diagram of a customer management device according to an embodiment of the present disclosure;

fig. 6 is a block diagram of an electronic device according to an embodiment of the present disclosure.

Detailed Description

In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The embodiments described in the examples below are not representative of all embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.

It should be noted that, in this disclosure, "at least one of the items" refers to a case where three types of juxtaposition including "any one of the items", "a combination of any of the items", "an entirety of the items" are included. For example, "including at least one of a and B" includes three cases side by side as follows: (1) comprises A; (2) comprising B; (3) includes A and B. For example, "at least one of the first and second steps is executed", that is, three cases are juxtaposed as follows: (1) performing step one; (2) executing the second step; (3) executing the first step and the second step.

As described in the background of the present disclosure, for special industries such as luxury goods, if a corresponding operation for a customer is simply performed according to the length of a repurchase time, there is a high possibility that a problem that the corresponding operation performed for the customer is not appropriate, and customer management cannot be accurately and properly performed.

To this end, the present disclosure proposes to first determine a target customer from a plurality of customers based on occurrence time of last at least one order of a predetermined category of goods by the plurality of customers, predict behavior information of the target customer using at least one machine learning model based on basic information of the target customer and history information of the target customer related to the predetermined category of goods after preliminary determination of the target customer, and then perform a preset operation for the target customer according to the predicted behavior information, thereby making it possible to perform the preset operation for the target customer more accurately or more pertinently, enabling more efficient customer management. In the present disclosure, the predetermined category of commodities is a commodity classified according to a certain attribute, for example, a commodity classified according to a brand, such as a certain brand of commodity, etc., or a commodity classified according to a consumer demand and/or feature, such as food, household appliances, daily necessities, children necessities, women necessities, etc., and a commodity classified according to a management angle, such as a main commodity, a general commodity, a mass-market commodity, a top-grade commodity, etc., which is not limited by the present disclosure, may be applied to any category of commodities in which it is difficult to divide a life cycle of a customer according to a length of a customer's repurchase time. In particular, in one embodiment of the present disclosure, the predetermined category of merchandise may be luxury or merchandise having luxury features.

Since the above-described client management method involves predicting behavior information of a target client using at least one machine learning model, a model training method according to an embodiment of the present disclosure will be described below with reference to fig. 1 first.

FIG. 1 is a flow chart of a model training method according to an embodiment of the present disclosure. Referring to fig. 1, in step S110, a target customer is determined from a plurality of customers based on the occurrence time of the last at least one order of the plurality of customers for a predetermined category of merchandise. According to an embodiment of the present disclosure, step S110 may first order the plurality of customers according to the occurrence time of the last order of each customer of the plurality of customers for the predetermined category of goods, and then determine the customer of the first ordered predetermined proportion of the plurality of customers as the target customer. For example, by analyzing the number of days on the date of the last order distance calculation for all customers under a certain luxury brand, after all customers are ranked in reverse order of days, the seventy percent or eighty percent of customers before the ranking may be determined as target customers. The identified target customer may be defined as a sleeping customer of the luxury brand. Alternatively, according to embodiments of the present disclosure, the predetermined ratio may vary depending on the category of merchandise, e.g., the predetermined ratio may vary depending on each brand of merchandise, and in particular, the predetermined ratio may be different for different luxury brands. Furthermore, the corresponding predetermined proportions may be determined according to the characteristics of different luxury brands. By the difference of the predetermined proportions, it is possible to more flexibly determine the sleeping customers of each brand.

After the target customer is determined, basic information of the target customer, history information of the target customer about a predetermined category of goods, and real behavior information of the target customer may be acquired at step S120.

According to the embodiments of the present disclosure, the basic information may include various personal information of the target client, for example, identity information of the client (name, sex, age, occupation, birthday, native place, address, marital situation, family member, etc.), preference information of the client (character feature, interest point, hobbies, etc.), and the like.

According to embodiments of the present disclosure, the historical information may include historical activity information of the target customer regarding the predetermined category of merchandise. Alternatively, the historical information may include historical order information for the target customer regarding the predetermined category of merchandise and historical activity information for the target customer regarding the predetermined category of merchandise. Alternatively, the history information may include history order information of the target customer regarding the predetermined category of goods and goods information regarding the target customer's history of purchased predetermined category of goods. Alternatively, the history information may include history order information of the target customer regarding the predetermined category of goods, history activity information of the target customer regarding the predetermined category of goods, and goods information regarding the target customer's history of purchased predetermined category of goods.

Wherein the historical order information may be any information related to the historical order of the target customer, including, for example, location and time information of occurrence of the historical order, respective statistics about the historical order, and the like. The historical activity information may be any information regarding the situation where the target customer historically participated in the relevant activities of the predetermined category of merchandise, for example, the historical activity information may include a means of touching the historical activity, merchandise for historical browsing or consultation, the number of times each product was browsed or consulted, historical activity participation time, place, frequency, etc. The commodity information may be information about the target customer history purchased commodity itself, such as commodity name and commodity attribute information (e.g., color, serial number, etc.), and the like.

The actual behavior information of the target customer is associated with historical information of the target customer regarding the predetermined category of merchandise, for example, for marking interactions or purchasing behavior of the target customer. Here, the real behavior information of the target client is associated with the history information of the target client regarding the predetermined category of merchandise, indicating that the real behavior information is the real behavior information of the target client in the event corresponding to the history information occurs. Here, the history information is generated based on various events, for example, the history activity information may be generated based on an event in which the target customer participates in the history activity, and the history order information may be generated based on an event in which the target customer purchases goods. For example, in the case where an event corresponding to the historical order information and the historical activity information occurs, when pushing a certain activity to a target client in a certain touch manner, the actual behavior information of the target client may be whether the target client participates in the activity.

It should be noted that the above basic information and the history information may include different contents according to the predicted targets of the machine learning model to be trained, and the actual behavior information may also be different according to the predicted targets.

For example, customer data (including, but not limited to, customer information, orders, browsing, clicking, communication with customer service, participation in offline activities, etc.) under all channels may be pre-integrated and stored, and basic information, historical information, and real behavior information may be obtained by processing such customer data. For example, in the luxury industry, there is now a common situation in which there are different unique customer identification information in different channels, such as on the online store cat and WeChat applet side, the same customer has two completely different identifications, which is disadvantageous for integrating all customer data, and this time, the unique customer identification needs to be unified and then used as one of the basic customer information, i.e., the same customer identification is used for the same customer in all channels, so that the customer data can be collected conveniently and comprehensively. In addition, for the private data, in order to protect the privacy of the client, if the client is to be used later, it is necessary to encrypt the private data, so that the private data cannot be processed in a plaintext form.

Training samples for training a machine learning model may include features and labels. The obtained basic information and history information can be used to construct features of the training sample, while the actual behavior information can be used as a marker of the training sample. As described above, the real behavior information may be different according to the prediction targets of the machine learning model. As an example, the prediction target may be an interactive effect of the target client after being touched in each of a plurality of touch manners, or may be whether the target client will participate in the activity of the predetermined category of commodity after acquiring the commodity-related activity information of the predetermined category of commodity in a predetermined touch manner, or may be whether the target client will purchase the predetermined category of commodity again, but is not limited thereto. For example, if the predicted target is whether the target customer will purchase the predetermined category of merchandise again, the real behavior information may be that the target customer purchased the predetermined category of merchandise again or did not purchase the predetermined category of merchandise again.

In step S130, behavior information of the target client is predicted using at least one machine learning model based on the acquired basic information and history information, and parameters of the at least one machine learning model are adjusted according to the predicted behavior information and the actual behavior information.

According to an embodiment, the at least one machine learning model may include at least one of a first machine learning model, a second machine learning model, and a third machine learning model, but is not limited thereto.

For example, the first machine learning model may be used to predict an interaction effect of the target client after being touched in each of a plurality of touch modes, and may also be referred to as a "touch mode interaction effect prediction model". The second machine learning model may be used to predict whether the target client will participate in the activity of the predetermined category of merchandise after obtaining merchandise-related activity information of the predetermined category of merchandise in a predetermined touch manner, and may also be referred to as an "activity behavior prediction model". The third machine learning model may be used to predict whether the target customer will repurchase a predetermined category of merchandise, and may also be referred to as a "repurchase probability prediction model".

The training of the first machine learning model, the second machine learning model, and the third machine learning model will be briefly described below, respectively.

The historical information may include historical order information for the target customer regarding the predetermined category of merchandise and historical activity information for the target customer regarding the predetermined category of merchandise if the first machine learning model is to be trained. In this case, predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and history information in step S130 may include: based on the acquired basic information, the historical order information and the historical activity information, predicting the interaction effect of the target client after being touched in each of a plurality of touch modes by using a first machine learning model as predicted behavior information.

For example, the basic information may include at least one of the following: age, sex, occupation, date of birth, date of wedding, registered city, registered store, current city, whether mobile phone number is reserved at registration, mailbox address is reserved at registration, communication address is reserved at registration, telephone contact, mail contact, etc., but is not limited thereto.

For example, the historical order information may include at least one of the following: the city of the nearest and earliest and full order, the store of the nearest and earliest and full order, the age of the nearest and earliest and full order, the amount of merchandise in the nearest and earliest and full order, the highest unit price of merchandise in the nearest and earliest and full order, the fraction of sales orders in the full order, the fraction of return orders in the full order, the fraction of maintenance orders in the full order, the total amount of sales orders in the full order, the total amount of maintenance orders in the full order, the total amount of return orders in the full order, the maximum number of purchase interval days in the full order, the minimum number of purchase interval days in the full order, the average number of purchase interval days in the full order, the median number of purchase interval days in the full order, whether the full order occurs before or after the middle western section holidays, etc., but is not limited thereto.

For example, the historical activity information may include at least one of: cities when clients participate in the historical activities, ages when participating in the historical activities, seasons when participating in the historical activities, total times when participating in the historical activities, whether orders were generated in the historical activities, whether contact addresses were updated in the historical activities, and the like.

The customer may be reached through various means of access such as short messages, customer service telephones, mail, third party platforms, etc., for example, providing the customer with merchandise activity advance notice information, merchandise information, etc. For example, in the short message touch mode or the email touch mode, a link may be designed, where the link may be used for the purposes of opening an activity, starting a micro-message applet, browsing an official webpage, etc., and meanwhile, the link may be attached with unique identification information of each client, and all feedback data after the client is touched may be collected as real behavior information of the client. By the method, participation accurate to the granularity of the client can be obtained. These collected feedback data may be saved as markers for training samples, including but not limited to in the form of offline files or databases.

According to an embodiment, the first machine learning model may be a two-class prediction model for predicting an interaction effect of the target client after being touched in each of a plurality of touch modes. For the first machine learning model, the real behavior information may be whether the target client has interaction after being touched in each of a plurality of touch modes in the event of the event corresponding to the historical order information and the historical activity information. The customer having interaction (or feedback) after being touched by some touch means may be, for example, participating in a current activity including, but not limited to: on-line and off-line reservations, check-ins, updates in activities or stay in contact, etc. The customer has no interaction (or no feedback) after being touched by some touch means, including but not limited to: no online-offline reservation, check-in, update in the campaign, or leave contact details, etc. occur. For example, where client a was reached by a sms on day 10 and day 1 of the last year, the training sample for training the first machine learning model may be structured as < client a, day 10 of the year, label= {0,1} >. label indicates that the actual behavior information of the client (i.e., with or without interaction) is a sign of the first training sample used for training the first machine learning model, label=0 indicates that the client has no interaction, and the sample corresponding to label=1 indicates that the client has interaction, and the sample corresponding to label=1 is a positive sample.

When predicting an interaction effect of a target client after being touched in each of a plurality of touch modes by using a first machine learning model based on acquired basic information, historical order information and historical activity information, first, feature processing is required to be performed on the basic information, the historical order information and the historical activity information to construct a first training sample. Specifically, for example, the features of the first training sample may be constructed by discretizing continuous features in the base information, the historical order information, and the historical activity information and encoding the discrete features in the base information, the historical order information, and the historical activity information. For example, all continuous features may be binned according to a 10 quantile number, and all discrete features may be unithermally encoded to construct features of the first training sample. After the feature is constructed and processed, the format of the first training sample may be, for example, in the form of < customer a, feature 1, …, feature N, label= {0,1} >. Then, the first training sample can be input into a first machine learning model to predict the interaction effect of the target client after being touched in each touch mode in a plurality of touch modes as predicted behavior information. After predicting the behavior information corresponding to each touch mode, parameters of the first machine learning model can be adjusted by comparing the predicted behavior information corresponding to each touch mode with the corresponding real behavior information until the model prediction loss is minimum, and the trained first machine learning model is obtained.

After the training of the first machine learning model is completed, the effect of the first machine learning model can also be verified by using a verification set, for example, by randomly selecting 20% of samples in the verification set as verification data, and calculating the ROC or AUC of the verification data by using the model, wherein a value of the ROC or AUC is closer to 1, which indicates that the classification performance of the model is better.

The training of the first machine learning model was described above, and the training of the second machine learning model is described below.

The historical information may include historical order information for the target customer regarding the predetermined category of merchandise and historical activity information for the target customer regarding the predetermined category of merchandise if the second machine learning model is to be trained. In this case, predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and history information in step S130 may include: based on the acquired basic information, history order information, and history activity information, a probability that a target client participates in the activity of a commodity of a predetermined category after acquiring commodity-related activity information of the commodity of the predetermined category in a predetermined touch manner is predicted using a second machine learning model as predicted behavior information.

According to an embodiment, the second machine learning model may be a classification prediction model for predicting a probability that the target client participates in the activity of the commodity of the predetermined category after obtaining the commodity-related activity information of the commodity of the predetermined category in a predetermined touch manner. For the second machine learning model, the actual behavior information may be whether or not the target customer participates in the activity of the commodity of the predetermined category after obtaining the commodity-related activity information of the commodity of the predetermined category in a predetermined touch manner in the event that the event corresponding to the historical order information and the historical activity information occurs. Here, the predetermined touch manner may be a touch manner of a short message, a customer service phone, a mail, or the like. For example, a customer may be provided with advance notice information of a large luxury item in a touch-up manner such as a customer service telephone. According to an embodiment, activities involving a predetermined category of merchandise include, but are not limited to: on-line and off-line reservations, check-ins, updates in activities or stay in contact, etc. Activities that do not participate in the predetermined category of merchandise include, but are not limited to: no online-offline reservation, check-in, update in the campaign, or leave contact details, etc. occur. For example, where customer a was reached by customer phone on day 10 and 1 of the last year, then a second training sample for training a second machine learning model may be constructed shaped as < customer a, day 10, label= {0,1} >. label indicates the actual behavioral information of the customer (i.e., whether engaged in an activity or not), which is a marker of the second training sample. label=0 indicates that the client is not engaged in an activity, and the sample corresponding to the label=0 is a negative sample, whereas label=1 indicates that the client is engaged in an activity, and the sample corresponding to the label=1 is a positive sample.

It should be noted that the basic information, the historical order information, and the historical activity information used in training the second machine learning model may be the same as or different from the basic information, the historical order information, and the historical activity information used in training the first machine learning model, but the actual behavior information for the first machine learning model and the second machine learning model may be different.

When predicting, based on the acquired basic information, history order information, and history activity information, a probability that a target customer participates in an activity of a commodity of a predetermined category after acquiring commodity-related activity information of the commodity of the predetermined category in a predetermined touch manner using a second machine learning model, first, it is necessary to perform feature processing on the basic information, history order information, and history activity information to construct a second training sample. Specifically, for example, the features of the second training sample may be constructed by discretizing continuous features in the base information, the historical order information, and the historical activity information and encoding the discrete features in the base information, the historical order information, and the historical activity information. For example, all continuous features may be binned according to a 10 quantile number, and all discrete features may be unithermally encoded to construct features of the second training sample. After the feature is constructed and processed, the format of the second training sample may be, for example, in the form of < customer a, feature 1, …, feature N, label= {0,1} >. Then, a second training sample may be input into the second machine learning model to predict, as predicted behavior information, a probability that the target client participates in the activity of the commodity of the predetermined category after obtaining the commodity-related activity information of the commodity of the predetermined category in the predetermined touch manner. After the behavior information corresponding to the preset touch mode is predicted, parameters of the second machine learning model can be adjusted by comparing the predicted behavior information with the real behavior information until the model prediction loss is minimum, and the trained second machine learning model is obtained.

After the second machine learning model is trained, the effect of the second machine learning model can also be verified by using a verification set, for example, by randomly selecting 20% of samples in the verification set as verification data, and calculating the ROC or AUC of the verification data by using the model, wherein a value of ROC or AUC is closer to 1 indicates that the classification performance of the model is better.

After the second machine learning model and the training of the second machine learning model are introduced, the training of the third machine learning model is described below.

If the third machine learning model is to be trained, the history information may include historical order information for the target customer regarding the predetermined category of merchandise and merchandise information for the target customer regarding the predetermined category of merchandise purchased historically. In this case, predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and history information in step S130 may include: based on the acquired basic information, history order information, and commodity information, the probability that the target customer purchases the commodity of the predetermined category again is predicted by using the third machine learning model as predicted behavior information.

For example, the historical order information may include at least one of the following: the city of the nearest and earliest and full order, the store of the nearest and earliest and full order, the age of the nearest and earliest and full order, the amount of goods in the nearest and earliest and full order, the highest unit price of goods in the nearest and earliest and full orders, the fraction of sales orders in the full order, the fraction of return orders in the full order, the fraction of maintenance orders in the full order, the total amount of sales orders in the full order, the total amount of maintenance orders in the full order, the total amount of return orders in the full order, the maximum number of purchase interval days in the full order, the minimum number of purchase interval days in the full order, the average number of purchase interval days in the full order, the median number of purchase interval days in the full order, whether the full order occurs before and after the middle and western section false days, the full purchased goods, etc., but is not limited thereto.

For example, the merchandise information may include at least one of the following: whether the commodity purchased by the target customer is 5 or 10 before the sale, the serial number of the commodity, the line of the commodity, the metal proportion in the commodity, the precious stone proportion in the commodity, the color of the commodity, the date of the first marketing of the commodity and the like.

According to an embodiment, the third machine learning model may be a classification prediction model for predicting a probability of re-purchasing the predetermined classification of the good. For the third machine learning model, the actual behavior information may be whether the target customer purchases the predetermined category of the commodity again in the event corresponding to the history order information and the commodity information. For example, the actual behavior information may be that the target customer generated at least one sales order within 12 months after the current order (e.g., customer A generated a sales order on day 2020-10-01 and again generated a sales order during time range 2020-10-02-2021-10-01), or may not generate any sales order within 12 months after the current order.

When predicting the probability that the target customer purchases the commodity of the predetermined category again using the third machine learning model based on the acquired basic information, history order information, and commodity information, first, it is necessary to perform feature processing on the basic information, history order information, and commodity information to construct a third training sample for training the third machine learning model. Specifically, for example, the features of the third training sample may be constructed by discretizing continuous features in the basic information, the historical order information, and the merchandise information, and encoding discrete features in the basic information, the historical order information, and the historical activity information. For example, all continuous features may be binned according to a 10 quantile number, and all discrete features may be unithermally encoded to construct features of the third training sample. After the feature is constructed and processed, the format of the third training sample may be, for example, in the form of < customer a, feature 1, …, feature N, label= {0,1} >. Subsequently, a third training sample may be input into a third machine learning model to predict a probability that the target customer will again purchase the predetermined category of merchandise as predicted behavior information. Finally, parameters of the third machine learning model can be adjusted by comparing the predicted behavior information with the real behavior information until the model prediction loss is minimum, and the trained third machine learning model is obtained.

After the training of the third machine learning model is completed, the effect of the third machine learning model may also be verified by using a verification set, for example, by randomly selecting 20% of samples in the verification set as verification data, and calculating ROC or AUC of the verification data by using the model, wherein a value of ROC or AUC is closer to 1 indicates that the classification performance of the model is better.

In the above, the training of at least one machine learning model involved in the client management method according to the present disclosure has been described, and according to the model training method of the present disclosure, at least one machine learning model capable of effectively predicting the behavior information of the target client can be trained.

Hereinafter, a customer management method using at least one machine learning model will be described. Fig. 2 is a flow chart of a customer management method according to an embodiment of the present disclosure.

Referring to fig. 2, in step S210, a target customer is determined from a plurality of customers based on the occurrence time of the last at least one order of the plurality of customers for a predetermined category of merchandise. For example, the plurality of customers may first be ranked according to the time of occurrence of the last order for the predetermined category of merchandise for each of the plurality of customers, and then the first ranked predetermined percentage of the plurality of customers may be determined to be the target customer. Here, the predetermined ratio may be changed according to each brand of the commodity of the predetermined category, and by the difference of the predetermined ratio, the sleeping customer of each brand may be more flexibly determined.

Step S210 is the same as the operation of step S110 described above, except that the time point referred to by the last at least one order in step S210 is the time point when the current prediction is performed, and the time point referred to by the last at least one order in step S110 is a certain time point in history, so details thereof will not be repeated here.

Next, in step S220, basic information of the target customer and history information of the target customer concerning a predetermined category of merchandise are acquired. In step S230, behavior information of the target client is predicted using at least one machine learning model based on the acquired basic information and history information.

The description of the basic information and the history information has been described above with respect to fig. 1, except that the history information of step S220 is the history information with respect to the point of time at the time of the current prediction, and the history information of step S120 is the history information corresponding to a certain point of time in the history, so that the basic information and the history information will not be described here again, and the relevant details can be found in the description of the corresponding contents above.

According to an embodiment, the at least one machine learning model may comprise at least one of a first machine learning model, a second machine learning model and a third machine learning model trained using the model training method above. The first machine learning model may be used to predict an interaction effect of the target client after being reached in each of a plurality of reach modes. The second machine learning model may be used to predict a probability that the target client will participate in the activity of the commodity of the predetermined category after obtaining commodity-related activity information of the commodity of the predetermined category in a predetermined touch manner. A third machine learning model may be used to predict the probability that the target customer will again purchase a predetermined category of merchandise.

According to an embodiment, the history information may include historical order information of the target customer regarding the predetermined category of goods and historical activity information of the target customer regarding the predetermined category of goods. In this case, step S230 may include: based on the acquired basic information, the historical order information and the historical activity information, predicting the interaction effect of the target client after being touched in each of a plurality of touch modes by using a first machine learning model as predicted behavior information.

For example, for an upcoming event, after the target customer is determined in the manner of step S210, the characteristics of the first prediction sample may be constructed based on the obtained base information, the historical order information, and the historical event information. The manner of constructing the features of the first prediction sample is the same as the manner of constructing the features of the first training sample described above, and will not be described here again. For example, the format of the first prediction sample may be in the format < client a, feature 1, …, feature N >. Then, each first prediction sample is predicted by using the trained first machine learning model, and a probability value of an interval [0,1] is output for each first prediction sample. The higher the probability value, the more likely the customer participates in the interaction after being touched in a certain touch manner, and the more likely the customer participates in the interaction in terms of actual business expression, which indicates that the customer has higher visibility to the brand and is willing to learn more information of the brand. Thus, the behavior information predicted by the first machine learning model can be used not only for determining the preference touch manner of the target client, but also for determining the importance of the client to the commodity of the preset category.

Alternatively, in the case where the history information includes the history order information of the target customer about the predetermined category of goods and the history activity information of the target customer about the predetermined category of goods, step S230 may include: based on the acquired basic information, history order information, and history activity information, a probability that a target client participates in the activity of a commodity of a predetermined category after acquiring commodity-related activity information of the commodity of the predetermined category in a predetermined touch manner is predicted using a second machine learning model as predicted behavior information.

For upcoming campaigns, after the target customer is determined in the manner of step S210, features of the second prediction sample may be constructed based on the obtained base information, the historical order information, and the historical campaign information. The manner of constructing the features of the second prediction sample is the same as the manner of constructing the features of the second training sample above, and will not be described here again. For example, the format of the second prediction sample may be in the format < client a, feature 1, …, feature N >. Then, each second prediction sample is predicted by using the trained second machine learning model, and a probability value of an interval [0,1] is output for each second prediction sample. The greater the probability value, the greater the likelihood that the target client will participate in the activity, and conversely the lesser the likelihood that the target client will participate in the activity. For example, a luxury large campaign cannot invite every customer and can only be opened to high potential customers, so that with the behavioral information predicted by the second machine learning model, high potential customers can be mined ahead of time, avoiding misplacing of excellent resources.

Optionally, the history information includes history order information of the target customer regarding the predetermined category of merchandise and merchandise information of the target customer for the predetermined category of merchandise purchased. In this case, step S230 may include: based on the acquired basic information, history order information, and commodity information, the probability that the target customer purchases the commodity of the predetermined category again is predicted by using the third machine learning model as predicted behavior information.

For example, after the target customer is determined in the manner of step S210, the characteristics of the third prediction sample may be constructed based on the acquired basic information, the historical order information, and the commodity information. The manner of constructing the features of the third prediction sample is the same as the manner of constructing the features of the third training sample described above, and will not be described here again. For example, the format of the third prediction sample may be in the format < client a, feature 1, …, feature N >. Then, each third prediction sample is predicted using the trained third machine learning model, and a probability value of an interval [0,1] is output for each third prediction sample. The greater the probability value, the greater the likelihood that the target customer will purchase the merchandise again, and vice versa. Therefore, the behavior information predicted by the third machine learning model can be utilized to predict the possibility of the repurchase of the target client in advance and to pertinently formulate the marketing strategy according to the possibility of the repurchase.

After predicting the behavior information of the target client using at least one machine learning model, a preset operation for the target client may be performed according to the predicted behavior information at step S240.

For example, in the case that the first machine learning model is used to predict the interaction effect of the target client after being touched in each of the multiple touch modes in step S230 as the predicted behavior information, step S240 may include: determining a preferential touch mode of the target client among multiple touch modes according to the predicted behavior information; the target customer is provided with relevant information about the predetermined category of merchandise in a target customer's preferred reach. For example, if the first machine learning model predicts the interaction effect of the target client after being touched in the short message touch manner, the email touch manner, the telephone touch manner and the third party platform touch manner, respectively, and if it is determined that the target client prefers the email touch manner according to the comparison of the interaction effects of the four touch manners, relevant information about the predetermined category of merchandise may be provided to the target client in the triggering manner of the email in the future. As an example, the related information about the commodity of the predetermined category may include activity information about the commodity of the predetermined category and/or commodity information of the commodity of the predetermined category. Because the relevant information is provided for the client in a touch manner preferred by the client, the possibility that the client participates in interaction after obtaining the relevant information is relatively high, the accuracy and the appropriateness of client management are improved, and the client management method is beneficial to creating greater commercial value.

For another example, in the case where the second machine learning model predicts, as predicted behavior information, a probability that the target client participates in the activity of the commodity of the predetermined category after obtaining the commodity-related activity information of the commodity of the predetermined category in the predetermined touch manner in step S230, or in the case where the third machine learning model predicts, as predicted behavior information, a probability that the target client purchases the commodity of the predetermined category again in step S230, step S240 may include: determining at least one client from the target clients based on the predicted behavior information; an operation related to a predetermined category of merchandise is performed for the determined at least one customer. For example, the at least one customer may be a customer with a high likelihood of participating in an activity or a high likelihood of purchasing goods again. Further, performing an operation related to the predetermined category of merchandise for the determined at least one customer may include: and providing relevant information about the commodity of the preset category to at least one client in a respective preference touch mode of the at least one client. For example, the respective preference touch patterns of the at least one client may be determined based on behavior information of the at least one client predicted using the first machine learning model. According to an embodiment, the related information about the commodity of the predetermined category may include activity information about the commodity of the predetermined category and/or commodity information of the commodity of the predetermined category.

Since at least one customer having a high possibility of participating in an activity or purchasing goods again can be further determined from among the target customers according to the predicted behavior information, and then only relevant information on the goods of a predetermined category can be provided to the determined customer in a targeted manner, it is possible to mine out high potential customers, avoiding misplacing excellent goods or activity resources. Further, since relevant information about the predetermined category of goods can be transmitted thereto in a manner of being touched by the respective preferences of the at least one customer, the at least one customer is more likely to accept and give feedback for the relevant information.

According to an embodiment, although not shown, the customer management method shown in fig. 2 may further include: after providing the related information on the commodity of the predetermined category, feedback information of each customer provided with the related information is collected for the related information. For example, the feedback information may be whether the customer participates in the activity of the predetermined category of goods after obtaining the goods-related activity information of the predetermined category of goods in a predetermined touch manner, or whether the customer purchases the goods again after being provided with the related information on the predetermined category of goods. Such feedback information may be used to further update the corresponding machine learning model.

Optionally, performing a preset operation for the target client according to the predicted behavior information may include: and if the predicted behavior information meets the preset condition, outputting early warning prompt information. For example, if the first machine learning model predicts that the interaction effect of the target client for multiple touch modes is poor, an early warning prompt is output, so that the loss of the client can be determined in time, and a targeted response strategy is adopted.

In the present disclosure, the behavior information of the target client may be predicted by using one of the first machine learning model, the second machine learning model, and the third machine learning model alone, or by any combination of the first machine learning model, the second machine learning model, and the third machine learning model.

For ease of understanding, the overall architecture of performing model training and customer management with at least one machine learning model including the first machine learning model, the second machine learning model, and the third machine learning model is briefly described below with reference to fig. 3.

Referring to fig. 3, in order to perform more efficient customer management for customers of a predetermined category of commodity, first, data preparation is required. For example, the prepared data may include basic information and history information (e.g., including history activity information, history order information, and merchandise information) of a plurality of customers, real behavior information (e.g., whether customers have interactions after being touched in a touch manner, participate in an activity, purchase merchandise of a predetermined category again), and the like. The information may be data from a local source or data from a third party. The data can be acquired and processed with assurance that the data privacy is not compromised. In the data preparation stage, because the data of the same customer may come from different data platforms, and the unique identifications of the customers on the different data platforms may not be consistent, the unique identifications of the customers may be redefined and unified in the data preparation stage.

After the data preparation is completed, training of the machine learning model may be performed using the data. For example, first, a target customer may be determined from a plurality of customers based on the time of occurrence of the last at least one order of the plurality of customers for a predetermined category of merchandise. Next, the first machine learning model, the second machine learning model, and the third machine learning model may be trained based on the basic information of the target customer and the history information of the target customer regarding the predetermined category of merchandise, respectively. Specifically, for example, a first machine learning model may be trained based on basic information, historical order information, historical activity information, and corresponding real behavior information (whether a target customer has interacted with after being touched in some manner of touch). The first machine learning model is trained to predict an interactive effect of the target client after being reached in each of a plurality of reach modes. The second machine learning model may also be trained with the base information, the historical order information, the historical activity information, and the corresponding actual behavior information (whether the target customer is engaged in an activity), and may be trained to predict a probability that the target customer will engage in an activity of a predetermined category of merchandise after obtaining merchandise-related activity information for the predetermined category of merchandise in a predetermined touch manner. Further, a third machine learning model may be trained based on the base information, the historical order information, the merchandise information, and the corresponding real behavior information (whether the target customer purchases the merchandise of the predetermined category again), and the third machine learning model may be trained to predict a probability that the target customer purchases the merchandise of the predetermined category again.

Next, in the model application phase, application experiments can be purposefully performed in batches. Specifically, for example, the first machine learning model may be used to predict an interaction effect of the target client after being touched in each of multiple touch modes, and determine a preferred touch mode of the target client according to the prediction result. Then, for example, the second machine learning model may be utilized to predict the probability that the target client will participate in the activity of the predetermined category of merchandise after obtaining merchandise-related activity information of the predetermined category of merchandise in such a preferential touchdown manner. Next, a portion of the customers with a high probability of participating in the activity may be determined from the target customers based on the prediction results of the second machine learning model, and for these customers, relevant activity information about the merchandise may be provided to them in their respective preferred touch manner. In addition, the third machine learning model can be further utilized to predict the probability of the customers purchasing goods again, and if the probability of one or some of the customers purchasing goods again is determined to be higher according to the prediction result of the third machine learning model, the goods information to be promoted can be provided for the customers in a targeted manner, so that the customers can conveniently learn the latest goods information in time and then purchase goods.

In addition, after corresponding operations are performed on the clients according to the prediction results of the first machine learning model, the second machine learning model and the third machine learning model, real feedback information of the clients can be further collected, the collected feedback information can be further enriched with data for performing iterative optimization (i.e. performing model update training) on the trained machine learning model, so that training of the machine learning model with more accurate prediction effect is facilitated. According to the needs, the real feedback of the customer can be collected by carrying out multiple experiments by changing different variables in the experiments, for example, the communication technology, the picture type, the strength of preferential strength, the promoted commodity series, the online and offline activities of the promotion and the like which are involved in the touch mode can be changed, the feedback can be purposefully collected, and a more accurate customer label can be obtained according to the feedback.

As described above, the customer management method according to the embodiment of the present disclosure has been described with reference to fig. 2 and 3, according to the above-described customer management method, since after a target customer is determined from a plurality of customers based on the occurrence time of at least one recent order of a predetermined category of goods by the plurality of customers, behavior information of the target customer is predicted using at least one machine learning model based on basic information of the target customer and history information of the target customer related to the predetermined category of goods, and then a preset operation for the target customer is performed according to the predicted behavior information, a corresponding operation can be performed more accurately or more specifically, so that more efficient customer management can be achieved.

Having described the model training method and the client management method of the embodiment of the present disclosure above, the model training apparatus and the client management apparatus of the embodiment of the present disclosure are briefly described below with reference to fig. 4 and 5.

Fig. 4 is a block diagram of a model training apparatus according to an embodiment of the present disclosure.

Referring to fig. 4, the model training apparatus 400 may include a determining unit 410, an acquiring unit 420, and a training 430. Specifically, the determining unit 410 may be configured to determine the target customer from the plurality of customers based on occurrence times of the last at least one order of the plurality of customers for the predetermined category of merchandise. The acquisition unit 420 may be configured to acquire basic information of the target customer, history information of the target customer regarding a predetermined category of merchandise, and real behavior information of the target customer. The training unit 430 may be configured to: predicting behavior information of the target client by using at least one machine learning model based on the acquired basic information and history information; and adjusting parameters of at least one machine learning model according to the predicted behavior information and the real behavior information. Since the model training method shown in fig. 1 may be performed by the model training apparatus 400 shown in fig. 4, and the determining unit 410, the acquiring unit 420, and the training 430 may perform operations corresponding to step S110, step S120, and step S130 in fig. 1, respectively, any relevant details concerning the operations performed by the units in fig. 4 may be referred to the corresponding description concerning fig. 1, and will not be repeated here.

Further, it should be noted that, although the model training apparatus 400 is described above as being divided into units for performing the respective processes, it is clear to those skilled in the art that the processes performed by the respective units described above may be performed without any specific division of units or without explicit demarcation between the units by the model training apparatus 400. In addition, the model training apparatus 400 may further include other units, for example, a data preprocessing unit, and the like.

Fig. 5 is a block diagram of a customer management device according to an embodiment of the present disclosure.

Referring to fig. 5, the client management apparatus 500 may include a target client determining unit 510, an information acquiring unit 520, a predicting unit 530, and an executing unit 540. Specifically, the target customer determination unit 510 may be configured to determine the target customer from the plurality of customers based on the occurrence time of the last at least one order of the plurality of customers for the predetermined category of merchandise. The information acquisition unit 520 may be configured to acquire basic information of the target customer and history information of the target customer regarding a predetermined category of merchandise. The prediction unit 530 may be configured to predict behavior information of the target client using at least one machine learning model based on the acquired basic information and history information. The execution unit 540 may be configured to perform a preset operation for the target client according to the predicted behavior information. Alternatively, although not shown in fig. 5, the customer management apparatus 500 may further include a feedback information collecting unit, which may be configured to collect feedback information for the relevant information for each customer provided with the relevant information after the relevant information about the predetermined category of goods is provided.

Since the client management method shown in fig. 2 may be performed by the client management apparatus 500 shown in fig. 5, and the target client determination unit 510, the information acquisition unit 520, the prediction unit 530, and the execution unit 540 may perform the steps S210, S220, S230, and S240 in fig. 2, respectively, any relevant details concerning the operations performed by the units in fig. 5 may be referred to the corresponding description concerning fig. 2, and will not be repeated here.

Further, it should be noted that, although the client management apparatus 500 is described above as being divided into units for performing the respective processes, it is clear to those skilled in the art that the processes performed by the respective units described above may be performed without any specific division of units or without explicit demarcation between the units by the client management apparatus 500. In addition, the client management device 500 may further include other units, for example, a storage unit, a data preprocessing unit, and the like.

The model training apparatus 400 and the client management apparatus 500 may be the same apparatus or may be different apparatuses. If the model training apparatus 400 and the client management apparatus 500 are the same apparatus, the determination unit 410 in fig. 4 and the target client determination unit 510 in fig. 5 may be the same unit, and the acquisition unit 420 in fig. 4 and the information acquisition unit 520 in fig. 5 may be the same unit.

Referring to fig. 6, an electronic device 600 may include at least one memory 601 and at least one processor 602 storing computer-executable instructions that, when executed by the at least one processor 602, cause the at least one processor to perform the model training method and/or the customer management method as described above.

By way of example, the electronic device may be a PC computer, tablet device, personal digital assistant, smart phone, or other device capable of executing the above-described set of instructions. Here, the electronic device is not necessarily a single electronic device, but may be any device or an aggregate of circuits capable of executing the above-described instructions (or instruction set) singly or in combination. The electronic device may also be part of an integrated control system or system manager, or may be configured as a portable electronic device that interfaces with either locally or remotely (e.g., via wireless transmission).

In an electronic device, a processor may include a Central Processing Unit (CPU), a Graphics Processor (GPU), a programmable logic device, a special purpose processor system, a microcontroller, or a microprocessor. By way of example, and not limitation, processors may also include analog processors, digital processors, microprocessors, multi-core processors, processor arrays, network processors, and the like.

The processor may execute instructions or code stored in the memory, wherein the memory may also store data. The instructions and data may also be transmitted and received over a network via a network interface device, which may employ any known transmission protocol.

The memory may be integrated with the processor, for example, RAM or flash memory disposed within an integrated circuit microprocessor or the like. In addition, the memory may include a stand-alone device, such as an external disk drive, a storage array, or any other storage device usable by a database system. The memory and the processor may be operatively coupled or may communicate with each other, for example, through an I/O port, a network connection, etc., such that the processor is able to read text stored in the memory.

In addition, the electronic device may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the electronic device may be connected to each other via a bus and/or a network.

According to an embodiment of the present disclosure, there may also be provided a computer-readable storage medium storing instructions, wherein the instructions, when executed by at least one processor, cause the at least one processor to perform a model training method and/or a customer management method according to an embodiment of the present disclosure. Examples of the computer readable storage medium herein include: read-only memory (ROM), random-access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random-access memory (RAM), dynamic random-access memory (DRAM), static random-access memory (SRAM), flash memory, nonvolatile memory, CD-ROM, CD-R, CD + R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD + R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, blu-ray or optical disk storage, hard Disk Drives (HDD), solid State Disks (SSD), card-type memories (such as multimedia cards, secure Digital (SD) cards or ultra-fast digital (XD) cards), magnetic tapes, floppy disks, magneto-optical data storage devices, hard disks, solid state disks, and any other devices configured to store computer programs (or software) and any associated data, data text and data structures in a non-transitory manner and to provide the computer programs and any associated data, data text and data structures to a processor or computer to enable the processor or computer to execute the programs. The instructions or computer programs in the computer-readable storage media described above can be run in an environment deployed in a computer device, such as a client, host, proxy device, server, etc., and further, in one example, the computer programs and any associated data, data text, and data structures are distributed across networked computer systems such that the computer programs and any associated data, data text, and data structures are stored, accessed, and executed in a distributed fashion by one or more processors or computers.

The foregoing description of various exemplary embodiments of the present application has been presented for purposes of illustration and description, and is not intended to be exhaustive or to limit the application to the precise embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the invention. The scope of the application should, therefore, be determined with reference to the appended claims.

Claims

1. A customer management method comprising:

determining a target customer from a plurality of customers based on the time of occurrence of the last at least one order of the plurality of customers for a predetermined category of merchandise;

acquiring basic information of the target client and history information of the target client related to the commodity of the preset category;

predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information;

and executing preset operation aiming at the target client according to the predicted behavior information.

2. The customer management method as claimed in claim 1, wherein the history information includes history order information of the target customer about the predetermined category of goods and history activity information of the target customer about the predetermined category of goods,

Wherein said predicting behavior information of said target client using at least one machine learning model based on said obtained base information and said history information comprises:

based on the acquired basic information, the historical order information and the historical activity information, predicting an interaction effect of the target client after being touched in each of a plurality of touch modes by using a first machine learning model as the predicted behavior information.

3. The customer management method as claimed in claim 1, wherein the history information includes history order information of the target customer about the predetermined category of goods and history activity information of the target customer about the predetermined category of goods,

based on the acquired basic information, the historical order information and the historical activity information, predicting the probability that the target client participates in the activity of the commodity of the preset category after acquiring the commodity related activity information of the commodity of the preset category in a preset touch mode by using a second machine learning model as the predicted behavior information.

4. The customer management method as claimed in claim 1, wherein the history information includes history order information of the target customer regarding the predetermined category of goods and commodity information regarding the target customer historically purchased the predetermined category of goods,

based on the acquired basic information, the history order information, and the commodity information, predicting a probability that the target customer purchases the commodity of the predetermined category again using a third machine learning model as the predicted behavior information.

5. The customer management method as claimed in claim 1, wherein the determining a target customer from the plurality of customers based on occurrence time of at least one recent order of the plurality of customers for the predetermined category of goods comprises:

sorting the plurality of customers according to the occurrence time of the last order of each customer of the plurality of customers for the predetermined category of merchandise;

determining a first-ordered predetermined proportion of the plurality of customers as the target customer, wherein the predetermined proportion varies according to each brand of merchandise of the predetermined category.

6. A model training method, comprising:

acquiring basic information of the target client, historical information of the target client related to the commodity of the preset category and real behavior information of the target client;

predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information, and adjusting parameters of the at least one machine learning model according to the predicted behavior information and the real behavior information.

7. A customer management apparatus comprising:

a target customer determination unit configured to determine a target customer from a plurality of customers based on occurrence times of at least one recent order of the plurality of customers for a predetermined category of commodities;

an information acquisition unit configured to acquire basic information of the target customer and history information of the target customer concerning the predetermined category of merchandise;

a prediction unit configured to predict behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information;

And the execution unit is configured to execute preset operation aiming at the target client according to the predicted behavior information.

8. A model training apparatus comprising:

a determining unit configured to determine a target customer from a plurality of customers based on occurrence times of at least one recent order of the plurality of customers for a predetermined category of commodities;

an acquisition unit configured to acquire basic information of the target customer, history information of the target customer regarding the predetermined category of merchandise, and real behavior information of the target customer;

a training unit configured to: predicting behavior information of the target client using at least one machine learning model based on the acquired basic information and the history information; and adjusting parameters of the at least one machine learning model according to the predicted behavior information and the real behavior information.

9. An electronic device, comprising:

at least one processor;

at least one memory storing computer-executable instructions,

wherein the computer executable instructions, when executed by the at least one processor, cause the at least one processor to perform the method of any of claims 1 to 6.

10. A computer-readable storage medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform the method of any of claims 1 to 6.