CN112765467A

CN112765467A - Service recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN112765467A
Application number: CN202110069949.7A
Authority: CN
Inventors: 熊俊; 丁芬
Original assignee: Beijing Didi Infinity Technology and Development Co Ltd
Current assignee: Beijing Didi Infinity Technology and Development Co Ltd
Priority date: 2021-01-19
Filing date: 2021-01-19
Publication date: 2021-05-07

Abstract

The application provides a service recommendation method, a service recommendation device, an electronic device and a storage medium, wherein the method comprises the following steps: acquiring classification information of a plurality of users; aiming at each classification mode, according to the classification mode and the classification information of a plurality of users, dividing the plurality of users into a plurality of user groups; for each classification mode, determining the classification difference degree of the classification mode according to the difference degree of the utilization rates of the target travel service among the plurality of user groups obtained by classification according to the classification mode; determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes; and determining a target user group in the plurality of user groups corresponding to the target classification mode, and recommending a target trip service to the users in the target user group. According to the method and the device, feedback differences of different user groups for the target travel service can be determined through the classification mode, so that accurate service recommendation can be performed according to different scenes.

Description

Service recommendation method and device, electronic equipment and storage medium

Technical Field

The present application relates to the field of computer technologies, and in particular, to a service recommendation method and apparatus, an electronic device, and a storage medium.

Background

The network car booking platform is a platform for communicating a service request end (passenger) and a service providing end (driver), the service request end sends a car taking request on the network car booking platform, and then the network car booking platform generates an order according to the car taking request to send the order to the service providing end, so that the driver receives the order sent by the passenger.

To better serve passengers and drivers, net appointment platforms typically provide a variety of travel services for passengers and drivers to intervene in the taxi taking process to facilitate order completion. Currently, the method for analyzing the effect of an intervention (treatment) on an outcome (outcontrol) is a/B Test, but the above method considers the average experimental effect of all samples in an experiment, and the accuracy of the mean analysis is low.

Besides the above manner, the influence of intervention on the result can be evaluated by using an ANOVA regression model, and by adding an interactive item in the method, different experimental effects of a small-quantum population, namely, a gate (Conditional Average cause and Effect) can be evaluated.

Disclosure of Invention

In view of this, an object of the present application is to provide a service recommendation method, an apparatus, an electronic device, and a storage medium, which can determine feedback differences of different user groups for a target travel service, so as to perform accurate service recommendation for a differentiated scene.

According to an aspect of the present application, there is provided a service recommendation method including: the method comprises the steps of obtaining classification information of a plurality of users, wherein the classification information comprises attribute information of the users and order information of target travel service used by the users; for each classification mode, according to the classification mode and the classification information of the users, dividing the users into a plurality of user groups; for each classification mode, determining the classification difference degree of the classification mode according to the difference degree of the utilization rate of the target travel service among a plurality of user groups obtained by classification according to the classification mode; determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes; recommending the target travel service to users in a target user group corresponding to the target classification mode, wherein the target user group is determined from a plurality of user groups corresponding to the target classification mode.

According to another aspect of the present application, there is provided a service recommendation apparatus including: the information acquisition module is used for acquiring classification information of a plurality of users, wherein the classification information comprises attribute information of the users and order information of target travel service used by the users; the classification module is used for dividing the plurality of users into a plurality of user groups according to the classification information of the plurality of users according to each classification mode; a difference determining module, configured to determine, for each classification manner, a classification difference of the classification manner according to a difference of usage rates of the target travel service among the plurality of user groups classified according to the classification manner; the target classification determining module is used for determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes; and the service recommending module is used for recommending target travel service to users in a target user group corresponding to the target classification mode, wherein the target user group is determined from a plurality of user groups corresponding to the target classification mode.

An embodiment of the present application further provides an electronic device, including: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating via the bus when the electronic device is running, the machine-readable instructions when executed by the processor performing the steps of the service recommendation method as described above.

Embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to perform the steps of the service recommendation method as described above.

Embodiments of the present application further provide a computer program product, which includes a computer program/instruction, and the computer program/instruction, when executed by a processor, implement the steps of the service recommendation method as described above.

The service recommendation method provided by the embodiment of the application comprises the following steps: the method comprises the steps of obtaining classification information of a plurality of users, wherein the classification information comprises attribute information of the users and order information of target travel service used by the users; for each classification mode, according to the classification mode and the classification information of the users, dividing the users into a plurality of user groups; for each classification mode, determining the classification difference degree of the classification mode according to the difference degree of the utilization rate of the target travel service among a plurality of user groups obtained by classification according to the classification mode; determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes; recommending the target travel service to users in a target user group corresponding to the target classification mode, wherein the target user group is determined from a plurality of user groups corresponding to the target classification mode.

Compared with the prior art, the method and the device for classifying the target travel service can determine the feedback difference of different user groups for the target travel service, help to find the crowd with obvious feedback for the target travel service, and recommend the service according to the differentiated scene, so that the accuracy of service recommendation is improved.

Furthermore, the stability of classification is ensured by introducing a numerical distribution function, and the overfitting of classification is constrained by introducing a regularization function so as to ensure the stability and accuracy of a classification result.

In order to make the aforementioned objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.

Fig. 1 is a schematic structural diagram illustrating a service recommendation system provided in an embodiment of the present application;

FIG. 2 is a flow chart illustrating a method for service recommendation provided by an embodiment of the present application;

FIG. 3 is a flow chart illustrating another service recommendation method provided by an embodiment of the present application;

FIG. 4 illustrates a schematic diagram of a decision tree model provided by an embodiment of the present application;

fig. 5 is a schematic structural diagram illustrating a service recommendation apparatus provided in an embodiment of the present application;

fig. 6 shows a schematic structural diagram of an electronic device provided in an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. Every other embodiment that can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present application falls within the protection scope of the present application.

To enable those skilled in the art to utilize the present disclosure, the following embodiments are presented in conjunction with a specific application scenario "net appointment (e.g., express, carpool, or tailwind)". It will be apparent to those skilled in the art that the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the application. Although the present application is described primarily in the context of a net appointment, it should be understood that this is only one exemplary embodiment.

It should be noted that in the embodiments of the present application, the term "comprising" is used to indicate the presence of the features stated hereinafter, but does not exclude the addition of further features.

The terms "passenger," "requestor," "service requestor," and "customer" are used interchangeably in this application to refer to an individual, entity, or tool that can request or order a service. The terms "driver," "provider," "service provider," and "provider" are used interchangeably in this application to refer to an individual, entity, or tool that can provide a service. The term "user" in this application may refer to an individual, entity or tool that requests a service, subscribes to a service, provides a service, or facilitates the provision of a service. For example, the user may be a passenger, a driver, an operator, etc., or any combination thereof. In the present application, "passenger" and "passenger terminal" may be used interchangeably, and "driver" and "driver terminal" may be used interchangeably.

The terms "service request" and "order" are used interchangeably herein to refer to a request initiated by a passenger, a service requester, a driver, a service provider, or a supplier, the like, or any combination thereof. Accepting the "service request" or "order" may be a passenger, a service requester, a driver, a service provider, a supplier, or the like, or any combination thereof. The service request may be charged or free.

The term "targeted travel service" in this application may refer to a service provided by a driver, a service provider, or a supplier, etc., or any combination thereof. What accepts or uses the "targeted travel service" may be a passenger, a service requester, a driver, a service provider, a supplier, or the like, or any combination thereof. The target travel service may be charged or free.

The Positioning technology used in the present application may be based on a Global Positioning System (GPS), a Global Navigation Satellite System (GLONASS), a COMPASS Navigation System (COMPASS), a galileo Positioning System, a Quasi-Zenith Satellite System (QZSS), a Wireless Fidelity (WiFi) Positioning technology, or the like, or any combination thereof. One or more of the above-described positioning systems may be used interchangeably in this application.

One aspect of the present application relates to a service recommendation system. The system can realize interaction among the server, the service request end and the service providing end, and can classify a plurality of users based on the classification information of the plurality of users in the target group to determine the feedback difference of different user groups obtained by classification for the target trip service, thereby carrying out personalized service recommendation for each user group with different feedback differences.

The "target travel service" in the present application may refer to a service for guiding a user to use a corresponding target travel mode, and may include, but is not limited to, any one or more of the following, for example: the system comprises a service for guiding a user to use express cars, a service for guiding the user to use carpools and a service for guiding the user to use windward driving.

In the embodiment of the application, the service provider can push the target travel service to a plurality of users in the target group within a predetermined time period, which is equivalent to applying intervention to the travel process of the users, and the feedback difference of a plurality of user groups obtained by classification based on the target classification mode for the target travel service can be determined by the service recommendation method of the application, that is, the feedback effect of different user groups for the applied intervention is determined. The application aims to find a classification mode which enables interference feedback differences of all user groups to be most obvious so as to carry out targeted service recommendation on different user groups.

It is noted that before the application of the present application, the effect of the intervention on the result is generally evaluated by using the a/B Test method in the prior art, but the a/B Test method considers the average experimental effect of all samples in the experiment, and the accuracy of the mean analysis is low.

Besides, in the prior art, the Effect of intervention on the outcome can be evaluated by Individual causal Effect (ITE), Average causal Effect (ATE), and Conditional Average causal Effect (pot).

At present, the ANOVA regression model can be used for estimating the CATE, and different experimental effects of few-quantum crowds, namely the CATE, can be estimated by adding an interactive item. However, the selection of the experimental indexes needs to be screened by an engineer with sophisticated experience, and the method is not suitable for the crossing of too many experimental indexes, and the too many crossing can make the variable selection and result interpretation of the regression equation become extremely tedious.

However, the service recommendation system provided by the present application may find the target classification method that can make the degree of difference between the classification results most significant according to the degree of difference in the usage rates of the target travel service among the plurality of user groups obtained by classification in different classification methods, so as to perform personalized service recommendation for each user group obtained by classification based on the target classification method.

Fig. 1 is a schematic structural diagram of a service recommendation system 100 according to an embodiment of the present application. For example, the service recommendation system 100 may be an online transportation service platform for transportation services such as taxi cab, designated drive service, express, carpool, bus service, driver rental, or regular service, or any combination thereof. The service recommendation system 100 may include one or more of a server 110, a network 120, a service requestor 130, a service provider 140, and a database 150.

In some embodiments, the server 110 may include a processor. The processor may process information and/or data related to the service request to perform one or more of the functions described herein. For example, the processor may determine the target vehicle based on a service request obtained from the service requester 130. In some embodiments, a processor may include one or more processing cores (e.g., a single-core processor (S) or a multi-core processor (S)). Merely by way of example, a Processor may include a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), an Application Specific Instruction Set Processor (ASIP), a Graphics Processing Unit (GPU), a Physical Processing Unit (PPU), a Digital Signal Processor (DSP), a Field Programmable Gate Array (FPGA), a Programmable Logic Device (PLD), a controller, a microcontroller Unit, a Reduced Instruction Set computer (Reduced Instruction Set computer), a microprocessor, or the like, or any combination thereof.

In some embodiments, the device types corresponding to the service request end 130 and the service providing end 140 may be mobile devices, such as smart home devices, wearable devices, smart mobile devices, virtual reality devices, or augmented reality devices, and the like, and may also be tablet computers, laptop computers, or built-in devices in motor vehicles, and the like.

In some embodiments, a database 150 may be connected to the network 120 to communicate with one or more components (e.g., the server 110, the service requester 130, the service provider 140, etc.) in the service recommendation system 100. One or more components in the service recommendation system 100 may access data or instructions stored in the database 150 via the network 120. In some embodiments, the database 150 may be directly connected to one or more components in the service recommendation system 100, or the database 150 may be part of the server 110.

The following describes in detail a service recommendation method provided in an embodiment of the present application with reference to the content described in the service recommendation system 100 shown in fig. 1.

Referring to fig. 2, a flowchart of a service recommendation method provided in the embodiment of the present application is shown, where the method may be executed by the server 110 in the service recommendation system 100, and the specific execution process is as follows:

s101, obtaining classification information of a plurality of users. Here, the classification information includes, but is not limited to, attribute information of the user and order information of the user using the target travel service.

For example, the service provider may push the target travel service to a plurality of users within a predetermined time period, and acquire attribute information of each user and order information of each user using the target travel service within the predetermined time period.

By way of example, the attribute information of the user may include, but is not limited to, at least one of: gender, age, driving age, school calendar, city, nationality, occupation. For example, the location of the user may be determined based on a positioning technique, and the city of the user may be determined based on the location of the user.

As an example, the order information of the user using the target travel service may include, but is not limited to, at least one of the following: the method comprises the following steps of response rate, response waiting time, driving receiving distance, online time of a user, total order resources (such as total amount of orders), historical order completion fluctuation index, order completion quantity, sensitivity index for target trip service, service stage of the user and group identification of the user. The group identity of the user may be used to indicate whether the user belongs to an experimental group or a control group.

Here, the historical single fluctuation index of the user may be used to represent the change of the user in the unit time of the number of finished products, and the sensitivity index for the target trip service may be used to represent the frequency of use of the user for the target trip service, and the sensitivity index may be in positive correlation with the frequency of use of the user for the target trip service.

In addition, in the present application, a plurality of service phases may also be divided in advance, and as an example, the plurality of service phases may include, but are not limited to: neonatal period, active period, fluid-loss period, active period.

In this case, it is determined which of the plurality of service stages the user is currently in based on the order information of the user. For example, taking the order information as the number of completed orders as an example, the service stage where the user is currently located may be determined based on the matching result of the number of completed orders of the user and the matching condition corresponding to different service stages.

S102, aiming at each classification mode, according to the classification mode and the classification information of a plurality of users, the users are divided into a plurality of user groups.

In this step, a plurality of user features may be determined according to the acquired classification information of the plurality of users, and the plurality of users may be divided by traversing each user feature.

In the present application, the classification dimensions (i.e., feature dimensions) of each classification method are the same, and the classification thresholds of the same classification dimensions corresponding to different classification methods are different.

Here, the multiple classification dimensions may refer to multiple user features, each of which corresponds to multiple classification thresholds, and based on this, all the classification manners may include multiple classification manners formed by combining multiple classification thresholds corresponding to different user features and different traversal orders for the multiple user features.

In an alternative embodiment, classification information for a plurality of users may be input into the decision tree model to obtain a plurality of user groups. At this time, the decision tree model is a multi-layer tree structure model, the split node of each layer corresponds to one user feature, and the user features are classified layer by layer to obtain a plurality of user groups.

As an example, the plurality of user characteristics may include at least one of: gender, age, driving age, school calendar, city, nationality, occupation, response rate, response waiting time, driving distance, online time of the user, total order resources, historical order completion fluctuation index, order completion quantity, sensitivity index for target travel service and service stage of the user.

S103, determining the classification difference degree of each classification mode according to the difference degree of the utilization rates of the target travel service among the plurality of user groups obtained by classification according to the classification mode.

In this step, for each classification mode, a plurality of user groups are obtained by classification according to the classification mode, and the degree of difference in the usage rate of the target travel service among the plurality of user groups is determined, so that the degree of difference in classification of the classification mode is determined. Here, the classification difference degree may be used to represent a difference degree of the usage rate of the target travel service according to the classification result obtained in the classification manner.

For example, the classification difference degree of each classification mode can be calculated according to the classification difference degree of the classification threshold value of each classification dimension under the classification mode. As an example, for each classification dimension, a classification difference of each classification threshold under the classification dimension may be determined, a target classification threshold is selected from all classification thresholds under the classification dimension, and the classification difference of the selected target classification threshold is determined as the classification difference of the classification dimension. Here, the classification threshold corresponding to the largest classification difference degree among the classification difference degrees of all the classification threshold values may be determined as the target classification threshold.

And S104, determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes.

In this step, the classification method having the largest degree of classification difference may be determined as the target classification method.

For example, the target classification may be determined from all classifications based on a ranking of the degree of classification dissimilarity of the different classifications. For different classification modes, ranking may be performed according to the sequence of the classification difference degrees from high to low, at this time, the classification mode corresponding to the first-ranked classification difference degree may be determined as the target classification mode, or ranking may be performed according to the sequence of the classification difference degrees from low to high, at this time, the classification mode corresponding to the last-ranked classification difference degree may be determined as the target classification mode.

And S105, recommending target travel service to the users in the target user group corresponding to the target classification mode.

In an example of the present application, the multiple users in step S101 may be partial users in the target group, and at this time, the target classification manner may be determined based on classification information of the partial users, and then all users in the target group are classified based on the determined target classification manner to obtain multiple user groups classified according to the target classification manner, and the target user group is determined from the multiple user groups. However, the present application is not limited to this, and the plurality of users in step S101 may be all users in the target group, and in this case, the target classification scheme may be determined using the classification information of all users in the target group.

In this step, a target user group may be determined from a plurality of user groups corresponding to the target classification manner, so as to recommend a target travel service to users in the target user group.

In an example, the target user group may be determined by: and selecting a target user group from the plurality of user groups according to the service recommendation priorities of different user groups corresponding to the target classification mode.

Here, the service recommendation priority of the target user group is higher than the service recommendation priorities of the other user groups. It should be understood that the service recommendation priority of the target user group is higher than the service recommendation priorities of other user groups, which may refer to that the service recommendation priority of the target user group is greater than a set value, or that the service recommendation priority of the target user group is greater than a middle value of the service recommendation priorities of all user groups, and the like.

That is, the target user group may be one of different user groups corresponding to the target classification manner, or may be a part of the user groups (in the case of multiple user groups) in the different user groups.

In an embodiment, the service recommendation priority of each user group may be determined according to the number of the users in the user group using the target travel service, and the service recommendation priority is positively correlated with the number of the users.

For example, the more the users in the user group use the target travel service, the higher the service recommendation priority of the user group, and the less the users in the user group use the target travel service, the lower the service recommendation priority of the user group. That is to say, the higher the determined service recommendation priority is, the higher the possibility that the user in the user group corresponding to the high service recommendation priority uses the target travel service is indicated, and the target travel service is subsequently recommended for the user in the user group with the high service recommendation priority, which is helpful for improving the utilization rate of the user in such user group for the target travel service.

Referring to fig. 3, a flowchart of another service recommendation method provided in the embodiment of the present application is shown, where the method may be executed by a server in the service recommendation system 100, and the specific execution process is as follows:

s201, obtaining classification information of a plurality of users. Here, the classification information includes, but is not limited to, attribute information of the user and order information of the user using the target travel service.

The description of S201 may refer to the description of S101 shown in fig. 2, and the same technical effect can be achieved, which is not described again.

S202, preprocessing the acquired classification information.

Here, the classification information may include numerical type classification information and non-numerical type classification information. Here, the classification information of the numerical type may mean that the classification information can be represented by a specific numerical value. For example, numerical type classification information may include age, driving age, response rate, user's online duration, historical end note volatility index, end note number, and the like.

A non-numeric type of classification information may mean that the classification information cannot be represented using a specific numeric value. For example, non-numerical types of classification information may include academic calendar, gender, profession, service phase in which the user is located, and so on.

In this step, the classification information for the numerical type may be preprocessed. By way of example, the pre-processing may include, but is not limited to, at least one of: supplement of missing values, elimination of abnormal values and correction of abnormal values. For example, the distribution of the five-point numbers of the classification information of the numerical value type may be checked, the missing values may be complemented by the mode, and the abnormal value data may be deleted or corrected.

And S203, performing characteristic engineering processing on the acquired classification information.

In this step, feature engineering may be performed for non-numerical types of classification information. By way of example, the feature engineering process may include: and performing 'binarization' operation on the classification information by adopting One Hot coding to form character string codes aiming at the classification information.

And S204, determining a plurality of user characteristics.

In this step, a plurality of user characteristics may be determined based on the acquired classification information of the plurality of users. For example, a plurality of user features are determined using the preprocessed numerical-type classification information and the feature-engineered non-numerical-type classification information.

S205, determining a plurality of classification threshold values of each user characteristic.

In this embodiment, the classification dimensions (i.e., feature dimensions) of each classification method are the same, and the classification thresholds of the same classification dimensions corresponding to different classification methods are different.

Here, the plurality of classification dimensions may refer to a plurality of user features, each user feature (i.e., each classification dimension) corresponding to a plurality of classification thresholds. That is, each user feature corresponds to a plurality of classification thresholds, and a target classification threshold of the user feature is selected from the plurality of classification thresholds.

The user features determined according to the classification information of the numerical type belong to the user features of the numerical type, the user features are taken as historical single fluctuation indexes as an example, the historical single fluctuation indexes of all users in a target group can be traversed according to the user features of the numerical type, sorting is carried out according to the sequence from small to large or from large to small, and the middle value of the adjacent historical single fluctuation indexes is selected as a classification threshold, so that a plurality of classification thresholds of the user features are formed.

For the user features of non-numerical type, all classification thresholds of the user features may be selected in an exhaustive manner to form a plurality of classification thresholds of the user features.

And S206, determining the classification difference degree of each classification threshold value of each user characteristic.

In the embodiment of the application, for each classification mode, according to the utilization rate of the target trip service by the users in the plurality of user groups obtained by classification according to the classification mode, the numerical distribution index of the classification mode is determined by using a numerical distribution function, for each classification mode, according to the utilization rate of the target trip service by the users in the plurality of user groups obtained by classification according to the classification mode, the regularization index of the classification mode is determined by using a regularization function, and for each classification mode, the classification difference degree of the classification mode is determined according to the determined numerical distribution index and the regularization index. As an example, the numerical distribution function is used to characterize the degree of difference of the usage rates corresponding to different classification modes, and the regularization function is used to control the classification overfitting.

For example, the degree of classification difference for each classification may be determined by: generating a plurality of candidate user populations based on the plurality of users; for each classification mode, determining a distribution difference index corresponding to each candidate user group in the classification mode according to the utilization rate of the target travel service used by each user group obtained by classifying each candidate user group according to the classification mode; for each classification mode, determining a numerical distribution index of the classification mode by using a numerical distribution function according to a distribution difference index corresponding to each candidate user group in the classification mode; aiming at each classification mode, determining the regularization index of the classification mode by utilizing a regularization function according to the corresponding utilization rate of each candidate user group in the classification mode; and determining the classification difference degree of each classification mode according to the numerical value distribution index and the regularization index of the classification mode.

Each of the candidate user groups includes some users of the plurality of users, that is, the users included between each of the candidate user groups may be repeated.

In the service recommendation method of the application, whether the classification results of the classification modes have significant differences or not can be judged, and then the classification difference degree of the classification modes with significant differences can be determined.

For example, determining the degree of classification difference for each classification method may include: for each classification mode, determining the number of candidate user groups meeting requirements in the plurality of candidate user groups in the classification mode, wherein the distribution difference index of the candidate user groups meeting the requirements falls in a preset confidence interval; for each classification mode, determining whether the classification result obtained by classifying according to the classification mode has significant difference according to the number of candidate user groups meeting the requirements; and determining the classification difference degree of the classification modes with the significant difference. Here, classification methods that do not have significant differences can be discarded without calculating the degree of classification difference.

In a preferred embodiment of the present application, the classification difference of each classification method is calculated according to the classification difference of the classification threshold of each classification dimension in the classification method, and a process of determining the classification difference of the classification threshold of each classification dimension is described below.

For example, a splitting objective function may be constructed in advance, for each user feature (i.e., each classification dimension), a splitting objective function value of each classification threshold of the user feature is calculated, and the calculated splitting objective function value is determined as a classification difference degree of the corresponding classification threshold.

In an example, the splitting objective function may include, but is not limited to, a numerical distribution function and a regularization function. For each classification threshold of each user characteristic, the numerical distribution function can be used for representing the difference degree of the utilization rate of the target travel service of the user group obtained by classification according to the classification threshold, and the regularization function can be used for controlling classification overfitting.

For example, for each user feature, a numerical distribution index and a regularization index are determined by using a numerical distribution function and a regularization function respectively according to the usage rate of the target travel service by the users in a plurality of user groups obtained by classifying according to each classification threshold of the user feature, and the classification diversity of each classification threshold of the user feature is determined according to the determined numerical distribution index and regularization index.

Here, the value distribution function in the split objective function represents the usage rate distribution of the target travel service by different user groups in the classification result, and based on this, in the present application, for each classification threshold, multiple times of extraction may be performed from multiple users corresponding to the classification threshold based on the principle of random sampling, so as to obtain multiple candidate user groups, for example, a boottrap random sampling method, that is, a repeated sampling method with back-put may be employed to obtain multiple candidate user groups.

And aiming at each classification threshold, respectively determining a numerical distribution index and a regularization index of the classification threshold according to the utilization rate of the target travel service by a first candidate user group and a second candidate user group which are obtained by classifying each candidate user group according to the classification threshold, and determining the classification difference of the classification threshold according to the determined numerical distribution index and the regularization index. That is, sampling is performed multiple times for each classification threshold, and the usage distribution at the classification threshold can be obtained.

In an optional embodiment, before determining the classification difference of each classification threshold for each user characteristic, the service recommendation method of the present application may further include:

for each classification threshold of the specified user characteristics, determining the number of qualified candidate user groups selected from the plurality of candidate user groups under the classification threshold, wherein the distribution difference index of the qualified candidate user groups falls within a preset confidence interval. The distribution difference index is used for representing the difference degree of the utilization rate of the target travel service of the first candidate user group and the second candidate user group obtained by classification according to the classification threshold; and for each classification threshold, determining whether the classification result obtained under the classification threshold has significant difference according to the number of the candidate user groups meeting the requirements.

And if the number of the candidate user groups meeting the requirements is not less than (greater than or equal to) the set threshold, determining that the classification results obtained under the classification threshold have significant differences, and continuously determining the classification difference degree of the classification threshold with the significant differences. If the number of the candidate user groups meeting the requirements is smaller than the set threshold, the classification result obtained under the classification threshold is determined to have no significant difference, and at this time, the classification threshold can be discarded without continuously determining the classification difference degree of the classification threshold. As an example, the predetermined confidence interval may be a confidence interval capable of determining a significant difference between the classification results, for example, the predetermined confidence interval may include 5% and 95% confidence intervals, but the application is not limited thereto, and a value range of the predetermined confidence interval may be adjusted by a person skilled in the art as needed.

By the method, the segmentation mode with insignificant statistics can be removed in advance before the classification difference is determined, so that the processing efficiency is improved.

A specific process of determining the degree of classification dissimilarity for each classification threshold for each user feature is described below.

In the embodiment of the present application, the classification result of each classification threshold has two branches, for example, taking the user characteristic as the age as an example, it is assumed that the multiple classification thresholds corresponding to the ages may include 30 years, 40 years and 50 years, that is, 30 years is used as the classification threshold for classification, 40 years is used as the classification threshold for classification, 50 years is used as the classification threshold for classification, and the like. For classification with 30 years as a classification threshold, two branches of the classification result are one branch less than or equal to 30 years and one branch more than 30 years, and so on, and the application does not give an example.

For example, for each classification threshold of a specified classification dimension, each candidate user group is divided into a first candidate user group and a second candidate user group according to the classification threshold, in which case, the first candidate user group may refer to one of the classification results classified according to the classification threshold, and the second candidate user group may refer to the other of the classification results classified according to the classification threshold.

And aiming at each classification threshold of the specified classification dimension, respectively determining a distribution difference index corresponding to each candidate user group under the classification threshold according to a first split index of a first candidate user group and a second split index of a second candidate user group which are divided by the classification threshold.

Here, the first split index may be used to characterize usage of the target travel service by users in the first candidate user group, and the second split index may be used to characterize usage of the target travel service by users in the second candidate user group.

Specifically, for each classification threshold of the specified classification dimension, the numerical distribution index of the classification threshold is determined according to the average value of the distribution difference indexes and the variance of the distribution difference indexes corresponding to all candidate user groups under the classification threshold.

And aiming at each classification threshold of the specified classification dimension, determining a numerical distribution index of the classification threshold by using a numerical distribution function according to the distribution difference index corresponding to each candidate user group under the classification threshold.

Specifically, for each classification threshold of a specified classification dimension, calculating the variance of the first split index of the classification threshold based on the first split indexes of the first candidate user groups corresponding to all candidate user groups under the classification threshold; for each classification threshold of the specified classification dimension, calculating the variance of the second split index of the classification threshold based on the second split indexes of the second candidate user groups corresponding to all the candidate user groups under the classification threshold; for each classification threshold of a specified classification dimension, a regularization index of the classification threshold is determined based on a constraint parameter, a variance of the first split index, and a variance of the second split index.

As an example, the magnitude of the constraint parameter is positively correlated with the degree to which the regularization function affects the classification overfitting. That is, the larger the constraint parameter, the greater the degree of constraint of the regularization function on the classification overfitting, and the smaller the constraint parameter, the smaller the degree of constraint of the regularization function on the classification overfitting.

Sometimes, the abnormal and unstable scenes are identified by the model due to the fact that the random features of the individual are identified by the model in a wrong mode, and the generalization capability is weakened, aiming at the problem, a regularization function is introduced into a splitting target function, overfitting of the model is controlled by the regularization function, namely the standard deviation of two branches after classification cannot be too large, and the strength of the regularization function to overfitting control is controlled by introducing a constraint parameter lambda.

For each classification threshold of a given classification dimension, determining a classification difference of the classification threshold based on a numerical distribution index and a regularization index of the classification threshold. For example, for each classification threshold, the difference between the numerical distribution index and the regularization index may be determined as the classification variance for that selection.

Here, specifying a classification dimension may refer to any one of a plurality of classification dimensions, that is, determining a classification difference degree of each classification threshold value of each classification dimension by repeating the above-described steps for each classification dimension.

And S207, determining a target classification mode based on the classification difference degree.

For example, after the classification difference degree of each classification threshold value of the designated classification dimension is determined in the above manner, a target classification threshold value may be selected from all classification threshold values, and the classification difference degree of the selected target classification threshold value is determined as the classification difference degree of the designated classification dimension.

In an embodiment of the present application, the layer-by-layer classification may be performed based on a plurality of user characteristics to obtain a plurality of user groups. The split node of each layer corresponds to a user characteristic, aiming at the split node of each layer, after the target classification threshold of each user characteristic is determined through the process, the classification difference degrees of the target classification threshold of each user characteristic are compared, the user characteristic corresponding to the maximum classification difference degree is determined as the split node of the layer, and the classification mode of the split node of the layer is determined based on the target classification threshold of the user characteristic corresponding to the maximum classification difference degree.

And forming a target classification mode aiming at a plurality of users by using the user characteristics corresponding to the split nodes of each layer and the classification mode corresponding to the split nodes of each layer determined in the above mode.

And S208, recommending the target travel service to the users in the target user group corresponding to the target classification mode.

In this step, a target user group may be determined from a plurality of user groups corresponding to the target classification manner, so as to recommend a target travel service to users in the target user group. Here, the target travel service is a service for guiding users in the target user group to use the target travel pattern to promote the order completion, for example, in the form of issuing a subsidy (such as a distribution coupon) for the lab group user.

The description of S208 may refer to the description of S105 shown in fig. 2, and the same technical effect can be achieved, which is not described again.

In addition, the service recommendation method of the present application may further include: and adjusting the target travel service, and recommending the adjusted target travel service to other user groups.

Here, the adjustment of the target travel service may refer to various means capable of contributing to the improvement of the utilization rate of the target travel service, and the manner of adjusting the target push service may include, but is not limited to: and adjusting the experimental objects, the experimental variables and the like in multiple aspects to improve the utilization rate of the target trip service recommended to the users in other user groups.

In a preferred embodiment of the present application, the above classification process for multiple users in the target group can be implemented by using a decision tree model. The classification process for multiple users in a target group based on a decision tree model is described below with reference to FIG. 4.

Fig. 4 shows a schematic diagram of a decision tree model provided by an embodiment of the present application.

The decision tree model is a multilayer tree structure formed by a plurality of split nodes and a plurality of leaf nodes, each split node in the split nodes corresponds to a user characteristic, each leaf node in the leaf nodes corresponds to a user group and a service recommendation index of the user group, and the service recommendation index is used for representing the number of the user in the user group using the target travel service.

That is, the decision tree model is a tree structure constructed based on the association between the user features and the user groups, which in this example is a binary tree, i.e., each split node has two branches.

The process of constructing the decision tree model is described below.

In this example, a random forest thought boottrap may be adopted to extract N training sets from all users of the target group, train the decision tree model based on the N training sets, extract M test sets from all users of the target group, and verify the classification accuracy of the decision tree model based on the M test sets.

And if the classification precision of the decision tree model meets the preset classification precision requirement, inputting the classification information of all users in the target group into the decision tree model so as to classify all users in the target group. And if the classification precision of the decision tree model does not meet the preset classification precision requirement, continuing to train the decision tree model.

That is, the classification information of a part of users in the target group may be utilized to train the decision tree model, the classification information of another part of users in the target group is utilized to verify the classification accuracy of the decision tree model, and the decision tree model meeting the preset classification accuracy requirement is utilized to classify all the users in the target group.

The specific training process is as follows: obtaining classification information of a plurality of users in a target group, and determining a plurality of user characteristics based on the obtained classification information, which can be written as: x is the number of₁,…,x_nWhere n represents the number of user features and determines the grouping status of each user (experimental or control).

For a first layer of the decision tree, the selectable user features corresponding to the first layer are all user features, and each user feature corresponds to a plurality of classification thresholds.

In this example, it is assumed that there are three user features (n ═ 3), such as user feature a, user feature B, and user feature C, where user feature a corresponds to 3 classification thresholds (corresponding to 3 candidate classification manners), user feature B corresponds to 2 classification thresholds, and user feature C corresponds to 4 classification thresholds.

In this case, a plurality of users may be classified according to 3 classification thresholds corresponding to the user feature a, respectively, for the user feature a, and the classification difference degrees of the 3 classification thresholds corresponding to the user feature a may be determined, respectively.

For example, for a first classification threshold of the user feature a, multiple candidate user groups are obtained by performing multiple extractions from multiple candidate users (all users in the target group corresponding to the first layer) corresponding to the classification threshold in a manner of putting back random sampling, and for each candidate user group, a first split index of a first candidate user group (i.e., a first branch) and a second split index of a second candidate user group (i.e., a second branch) obtained by classifying the candidate user group according to the first classification threshold are determined.

In this example, the usage rate of the target travel service by the users in each candidate user group is characterized by a Return On Investment (ROI) as a split index. In this case, the first split index may refer to a return on investment resulting from a target outbound service being recommended to each user in the first candidate user group, and the second split index may refer to a return on investment resulting from a target outbound service being recommended to each user in the second candidate user group.

At present, the direct modeling of the ROI has a very big challenge, firstly, the index of the ROI cannot sum up or calculate an average value, so that the Gini coefficient and the splitting standard such as Impurity and the like adopted by the traditional CART tree model cannot be directly used. And the ROI index has the characteristic of instability, which brings certain difficulty for modeling.

In order to solve the above problems, in the present application, a decision tree model is constructed based on user characteristics, and an ROI is used as an evaluation index for evaluating the degree of difference in classification modes of each split node in the decision tree model. In this example, the splitting objective function of the decision tree model is constructed based on the ROI, and the classification mode that maximizes the splitting objective function value is determined as the final classification mode of the splitting node. That is, each split node in the decision tree model is split with the goal of maximizing the split objective function (or maximizing the return on investment), so as to mine the ROI scene with significant difference.

In this example, assuming that attention is paid to the change of the ending resource (e.g., ending transaction amount, which may also be referred to as ending GMV) brought by recommending the target travel service for each user in the target group within a predetermined time period, taking the target travel service as an example of subsidizing the target travel mode (e.g., distributing a coupon), at this time, the splitting index of each branch may be calculated by using the following formula:

in equation (1), ROI represents the division index of a branch obtained by classification according to a classification threshold, GMV_treatmentRepresenting the total amount of completed single transaction, GMV, of the users belonging to the experimental group in the candidate user group corresponding to the branch_controlRepresents the sum of the completed orders, Subsidy, of the users belonging to the control group in the candidate user group corresponding to the branch_treatmentRepresents Subsidy cost, susidy, of users belonging to the experimental group in the candidate user group corresponding to the branch_controlAnd showing subsidy cost of the users belonging to the control group in the candidate user group corresponding to the branch.

Through the formula (1), the calculated ROI can be used for measuring the single transaction total increment caused by subsidy, namely subsidy cost is considered, the maximum increment GMV under the condition of total budget constraint can be obtained on the basis, and the problem that the existing algorithm only considers GMV promotion but does not consider actual cost of subsidy is solved, namely the existing algorithm has the condition that high subsidy brings high profit.

It should be understood that the foregoing example is described by taking a single resource as an example, but the present application is not limited thereto, and may also be applied to scene mining on different points of interest, and only the y value in modeling needs to be modified, and the y value may be a continuous variable or a discrete variable. The application scenario is the sum increment of finished single traffic brought by subsidies in a preset time period, and if y is modified into the vehicle output rate of the driver in the preset time period, a scene of vehicle output rate differentiation brought by subsidies can be excavated.

Continuing with the above example, for the first classification threshold of the user feature a, after the first split index of the first candidate user group and the second split index of the second candidate user group under the classification threshold are calculated by the above formula (1), the distribution difference index corresponding to each candidate user group under the classification threshold is respectively calculated.

For example, for each candidate user group, a difference between a first split index of a first candidate user group and a second split index of a second candidate user group at the classification threshold may be determined as a distribution difference indicator corresponding to the candidate user group.

As an example, the distribution difference index corresponding to each candidate user group may be calculated using the following formula:

ROI_gap＝ROI_left-ROI_right (2)

in formula (2), ROI_gapDistribution variance indicator, ROI, representing a population of candidate users_leftA first splitting index, ROI, representing a first group of candidate users into which the group of candidate users is divided, i.e. the splitting index of the left node in the binary tree classification structure_rightA second split index representing a second group of candidate users into which the group of candidate users is partitioned, i.e., a split index of a right node in the binary tree classification structure.

In this case, the split objective function can be represented using the following formula:

in equation (3), MAX represents the splitting objective function value, i.e., the degree of classification variance of the classification threshold, and the right side of the equation is dividedThe subtrahend is a numerical distribution function, the subtrahend is a regularization function,

mean value, σ, representing distribution difference indicators corresponding to a plurality of candidate user groups_gapVariance, σ, representing distribution variance indicators corresponding to a plurality of candidate user groups_ROIleftRepresenting the variance, σ, of the first split indices corresponding to a plurality of candidate user groups_ROIrightAnd expressing the variance of the second split indexes corresponding to the plurality of candidate user groups, and expressing a constraint parameter by lambda.

Based on the method, the sub-scene classification modes with different experiment effects can be found, and meanwhile, the regularization term is added to ensure the stability of sub-scene segmentation and improve the generalization capability of the sub-scene segmentation.

Similarly, for the user feature B, classifying the users according to the 2 classification thresholds corresponding to the user feature B, and obtaining the classification difference degrees of the 2 classification thresholds for the user feature B through the above steps. And classifying the users according to the 4 classification threshold values corresponding to the user characteristic C aiming at the user characteristic C, and obtaining the classification difference degrees of the 4 classification threshold values aiming at the user characteristic C through the steps.

And comparing the classification difference degrees of all classification threshold values of all the user characteristics, determining the user characteristic corresponding to the maximum classification difference degree as the split node of the layer, and determining the target classification threshold value of the user characteristic corresponding to the maximum classification difference degree as the classification mode of the split node of the layer. For example, assuming that the maximum classification difference corresponds to the second classification threshold of the user feature a, the split node of the first layer of the decision tree model may be determined as the user feature a, and the second classification threshold of the user feature a may be determined as the classification manner of the split node of the first layer.

In the method, the distribution difference indexes of the two branches are calculated for each classification threshold of all user characteristics, the distribution difference index distribution is described by adopting the thought of random forests, meanwhile, the fitting items are added to ensure the splitting stability, and finally, the classification mode with the most obvious ROI difference is effectively excavated.

And aiming at a second layer of the decision tree model, the second layer comprises two splitting nodes, the selectable user features corresponding to each splitting node are N-1 user features, namely, other user features except the user features corresponding to the determined splitting nodes of the first layer in all the user features, and each user feature corresponds to a plurality of classification thresholds.

Taking the above example as an example, the split node of the first layer of the decision tree model is the user feature a, one split node of the second layer of the decision tree model corresponds to the first candidate user group obtained by classifying according to the second classification threshold of the user feature a, and the other split node of the second layer of the decision tree model corresponds to the second candidate user group obtained by classifying according to the second classification threshold of the user feature a.

For a split node in the second layer of the decision tree model, the classification difference degrees of different classification thresholds corresponding to the user feature B and the user feature C can be respectively determined by referring to the above process, so as to determine the user feature corresponding to the split node and the classification manner corresponding to the classification node based on the classification difference degrees. Similarly, the user characteristics corresponding to another split node of the second layer of the decision tree model and the classification mode corresponding to the another classification node are determined.

In this example, for one split node in the second layer of the decision tree model, it is assumed that the one split node is determined as the user feature B according to the first candidate user group obtained by classifying according to the second classification threshold of the user feature a, and the first classification threshold of the user feature B is determined as the classification mode of the one split node in the second layer. For another split node of the second layer of the decision tree model, it is assumed that another split node is determined as a user feature C according to a second candidate user group obtained by classification according to a second classification threshold of the user feature a, and the second classification threshold of the user feature C is determined as a classification mode of another split node of the second layer. Then, at the third level of the decision tree model, the first candidate user group and the second candidate user group obtained by classifying according to the first classification threshold of the user feature B, and the first candidate user group and the second candidate user group obtained by classifying according to the second classification threshold of the user feature C may be continuously classified, and so on, after all the user features are traversed, the classification of multiple users in the target group is completed.

Here, in the decision tree model, the user features corresponding to any split node in the path from the root node to the leaf node are not repeated, that is, the selectable user features corresponding to the first layer are all the user features, the selectable user features corresponding to the second layer are other user features except the determined user features corresponding to the first layer, and so on. Thus, after traversing all the user features is finished, the final decision tree model is obtained.

The classification mode corresponding to the finally obtained decision tree model can be determined as a target classification mode, the user groups corresponding to each leaf node of the decision tree model are a plurality of user groups obtained by classification according to the target classification mode, the service recommendation index of each user group is calculated, and the target user group is determined based on the service recommendation index. In this example, the return on investment may be used as a service recommendation index to characterize the amount of orders used by the users in each user group for the target travel service.

For example, the user group corresponding to the service recommendation index not less than the preset threshold may be determined as a target user group, the user group corresponding to the service recommendation index less than the preset threshold may be determined as another user group, and the target travel service may be recommended to the user in the target user group corresponding to the target classification manner.

That is to say, the finally determined target user group is a crowd with obvious feedback for the target travel service, and if the target travel service pointer distributes the coupon to the express travel service, the target user group can indicate the crowd with high return on investment brought by the distributed coupon.

According to the service recommendation method, the return on investment is used as a classification standard for determining the classification mode of each user characteristic, a plurality of user groups with ROI differences can be effectively identified, different branches obtained through classification need to have statistically significant differences, and a preset confidence interval is given. In addition, a regularization function is constructed based on the variances of different branches to eliminate abnormal and unstable scenes and prevent overfitting of the model.

Based on the same inventive concept, a service recommendation device corresponding to the service recommendation method is also provided in the embodiments of the present application, and as the principle of solving the problem of the device in the embodiments of the present application is similar to the service recommendation method described above in the embodiments of the present application, the implementation of the device may refer to the implementation of the method, and repeated details are not repeated.

Referring to fig. 5, which is a schematic structural diagram of a service recommendation device according to an embodiment of the present application, the service recommendation device 300 includes: an information acquisition module 301, a classification module 302, a difference determination module 303, a target classification determination module 304 and a service recommendation module 305; wherein the content of the first and second substances,

an information obtaining module 301, configured to obtain classification information of multiple users. Here, the classification information includes attribute information of the user and order information of the user using the target travel service.

A classifying module 302, configured to, for each classification manner, divide the multiple users into multiple user groups according to the classification information of the multiple users according to the classification manner.

The classification module 302 may determine a plurality of user features according to the obtained classification information of the plurality of users, and divide the plurality of users by traversing each user feature.

In an alternative embodiment, the classification module 302 may input classification information for a plurality of users into the decision tree model to obtain a plurality of user groups. At this time, the decision tree model is a multi-layer tree structure model, the split node of each layer corresponds to one user feature, and the user features are classified layer by layer to obtain a plurality of user groups.

A difference determining module 303, configured to determine, for each classification manner, a classification difference of the classification manner according to a difference of usage rates of the target travel service used among the plurality of user groups classified according to the classification manner.

For example, the classification difference degree of each classification mode can be calculated according to the classification difference degree of the classification threshold value of each classification dimension under the classification mode. As an example, the difference determining module 303 may determine, for each classification dimension, a classification difference of each classification threshold in the classification dimension, select a target classification threshold from all classification thresholds in the classification dimension, and determine the classification difference of the selected target classification threshold as the classification difference of the classification dimension. Here, the classification threshold corresponding to the largest classification difference degree among the classification difference degrees of all the classification threshold values may be determined as the target classification threshold.

And the target classification determining module 304 is configured to determine a target classification manner from all classification manners based on the classification difference degrees of different classification manners.

For example, the target classification determination module 304 may determine the classification manner with the largest degree of classification difference as the target classification manner.

In an example, the target classification determination module 304 may determine the target classification from the total classification based on a ranking of the classification dissimilarity of the different classifications. For different classification modes, ranking may be performed according to the sequence of the classification difference degrees from high to low, at this time, the classification mode corresponding to the first-ranked classification difference degree may be determined as the target classification mode, or ranking may be performed according to the sequence of the classification difference degrees from low to high, at this time, the classification mode corresponding to the last-ranked classification difference degree may be determined as the target classification mode.

The service recommending module 305 recommends the target travel service to the users in the target user group corresponding to the target classification manner.

For example, the service recommending module 305 may determine a target user group from a plurality of user groups corresponding to the target classification manner, so as to recommend the target travel service to the users in the target user group.

In an example, the service recommendation module 305 may determine the target user group by: and selecting a target user group from the plurality of user groups according to the service recommendation priorities of different user groups corresponding to the target classification mode.

Here, the service recommendation priority of the target user group is higher than the service recommendation priorities of the other user groups. The target user group may be one of different user groups corresponding to the target classification manner, or may be a part of the different user groups (in the case of multiple user groups).

In a possible implementation, the information obtaining module 301 may be further configured to: and preprocessing the acquired classification information and performing characteristic engineering processing. Here, the classification information may include numerical type classification information and non-numerical type classification information. The information obtaining module 301 may perform preprocessing on the numerical type classification information, and may perform feature engineering processing on the non-numerical type classification information.

In a possible implementation, the classification module 302 is specifically configured to: a plurality of user characteristics may be determined based on the obtained classification information of the plurality of users, and a plurality of classification thresholds for determining each user characteristic may be determined.

The user features determined according to the classification information of the numerical type belong to the user features of the numerical type, taking the user features as the historical single fluctuation indexes as an example, for the user features of the numerical type, the classification module 302 may traverse the historical single fluctuation indexes of all users in the target group, sort the historical single fluctuation indexes in the order from small to large or from large to small, select the middle value of the adjacent historical single fluctuation indexes as a classification threshold, and thereby form a plurality of classification thresholds of the user features.

For non-numerical types of user features, classification module 302 may select all classification thresholds for the user feature in an exhaustive manner to form a plurality of classification thresholds for the user feature.

In a possible implementation, the difference determining module 303 may be further configured to: a classification variance for each classification threshold for each user feature is determined.

In a preferred embodiment of the present application, the difference determining module 303 may pre-construct a splitting objective function, calculate a splitting objective function value of each classification threshold of each user feature (i.e., each classification dimension) for each user feature, and determine the calculated splitting objective function value as the classification difference of the corresponding classification threshold.

For example, for each user feature, the difference degree determining module 303 determines a numerical distribution index and a regularization index by using a numerical distribution function and a regularization function respectively according to the usage rate of the target travel service by the users in the user groups classified according to each classification threshold of the user feature, and determines the classification difference degree of each classification threshold of the user feature according to the determined numerical distribution index and regularization index.

The description of the processing flow of each module in the device and the interaction flow between the modules may refer to the related description in the above method embodiments, and will not be described in detail here.

The embodiment of the application discloses a TS1 and a service recommendation method, which comprises the following steps:

the method comprises the steps of obtaining classification information of a plurality of users, wherein the classification information comprises attribute information of the users and order information of target travel service used by the users;

for each classification mode, according to the classification mode and the classification information of the users, dividing the users into a plurality of user groups;

for each classification mode, determining the classification difference degree of the classification mode according to the difference degree of the utilization rate of the target travel service among a plurality of user groups obtained by classification according to the classification mode;

determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes;

recommending the target travel service to users in a target user group corresponding to the target classification mode, wherein the target user group is determined from a plurality of user groups corresponding to the target classification mode.

The TS2, the method according to TS1, wherein recommending the target travel service to the user in the target user group corresponding to the target classification manner includes:

selecting the target user group from a plurality of user groups corresponding to the target classification mode according to the service recommendation priority of each user group corresponding to the target classification mode; the service recommendation priority of the target user group is determined according to the amount of the user in the user group using the target trip service, and the service recommendation priority is positively correlated with the amount of the user in the user group using the target trip service;

and recommending the target travel service to the users in the target user group corresponding to the target classification mode.

TS3, the method according to TS1, wherein the determining, for each classification method, a classification difference degree of the classification method according to a difference degree of usage rates of the target travel service among the plurality of user groups classified according to the classification method includes:

for each classification mode, determining a numerical distribution index of the classification mode by using a numerical distribution function according to the utilization rate of users in a plurality of user groups to the target travel service, wherein the users are classified according to the classification mode;

for each classification mode, determining a regularization index of the classification mode by using a regularization function according to the utilization rate of users in a plurality of user groups obtained by classification according to the classification mode to the target travel service;

for each classification mode, determining the classification difference degree of the classification mode according to the determined numerical distribution index and the regularization index; the numerical distribution function is used for representing the difference degree of the utilization rates corresponding to different classification modes, and the regularization function is used for controlling classification overfitting.

TS4, the method of TS3, wherein the degree of classification difference for each classification is determined by:

generating a plurality of candidate user populations based on the plurality of users; each of the candidate user groups comprises a part of users in the plurality of users;

for each classification mode, determining a distribution difference index corresponding to each candidate user group in the classification mode according to the utilization rate of the target travel service used by each user group obtained by classifying each candidate user group according to the classification mode;

for each classification mode, determining a numerical distribution index of the classification mode by using a numerical distribution function according to a distribution difference index corresponding to each candidate user group in the classification mode;

aiming at each classification mode, determining the regularization index of the classification mode by utilizing a regularization function according to the corresponding utilization rate of each candidate user group in the classification mode;

and determining the classification difference degree of each classification mode according to the numerical value distribution index and the regularization index of the classification mode.

TS5, the method of TS4, wherein determining a degree of classification variance for each classification method comprises:

for each classification mode, determining the number of candidate user groups meeting requirements in the plurality of candidate user groups in the classification mode, wherein the distribution difference index of the candidate user groups meeting the requirements falls in a preset confidence interval;

for each classification mode, determining whether the classification result obtained by classifying according to the classification mode has significant difference according to the number of candidate user groups meeting the requirements;

and determining the classification difference degree of the classification modes with the significant difference.

TS6, the method of TS4, wherein the classification dimension of each classification is the same; the classification thresholds for the same classification dimension for different classification approaches are different,

the classification difference degree of each classification mode is calculated according to the classification difference degree of the classification threshold value of each classification dimension under the classification mode;

the classification difference degree of the classification threshold value of the specified classification dimension is calculated according to the following mode:

aiming at each classification threshold of the specified classification dimension, determining a numerical distribution index of the classification threshold by using a numerical distribution function, and determining a regularization index of the classification mode by using a regularization function;

for each classification threshold of a specified classification dimension, determining the classification difference degree of the classification threshold based on the numerical distribution index and the regularization index of the classification threshold;

and selecting a target classification threshold from all classification thresholds according to the classification difference of each classification threshold of the specified classification dimension, and determining the classification difference of the selected target classification threshold as the classification difference of the specified classification dimension.

TS7, the method of TS6, wherein the numerical distribution index for each classification threshold specifying a classification dimension is calculated as follows:

for each classification threshold of a specified classification dimension, dividing each candidate user group into a first candidate user group and a second candidate user group according to the classification threshold;

aiming at each classification threshold of the specified classification dimension, respectively determining a distribution difference index corresponding to each candidate user group under the classification threshold according to a first split index of a first candidate user group and a second split index of a second candidate user group divided by the classification threshold; the first split index is used for representing the utilization rate of users in a first candidate user group to the target travel service, and the second split index is used for representing the utilization rate of users in a second candidate user group to the target travel service;

aiming at each classification threshold of the specified classification dimension, determining a numerical distribution index of the classification threshold by using a numerical distribution function according to a distribution difference index corresponding to each candidate user group under the classification threshold;

the regularization index for each classification threshold for a given classification dimension is computed as follows:

and aiming at each classification threshold of the specified classification dimension, determining the regularization index of the classification threshold by utilizing a regularization function according to the first split index of the first candidate user group and the second split index of the second candidate user group which are divided by the classification threshold.

TS8, the method of TS7, wherein the numerical distribution index for each classification threshold for a given classification dimension is determined by:

and aiming at each classification threshold of the specified classification dimension, determining the numerical distribution index of the classification threshold according to the average value of the distribution difference indexes and the variance of the distribution difference indexes corresponding to all candidate user groups under the classification threshold.

TS9, the method of TS7, wherein the regularization index for each classification threshold specifying a classification dimension is determined by:

aiming at each classification threshold value of the specified classification dimension, calculating the variance of the first split index of the classification threshold value based on the first split indexes of the first candidate user groups corresponding to all candidate user groups under the classification threshold value;

for each classification threshold of the specified classification dimension, calculating the variance of the second split index of the classification threshold based on the second split indexes of the second candidate user groups corresponding to all the candidate user groups under the classification threshold;

and aiming at each classification threshold of the specified classification dimension, determining the regularization index of the classification threshold based on a constraint parameter, the variance of the first split index and the variance of the second split index, wherein the size of the constraint parameter is in positive correlation with the influence degree on the classification overfitting.

TS10, the method according to TS1, wherein the attribute information includes at least one of: gender, age, driving age, school calendar, city, nationality, occupation;

the order information includes at least one of: the method comprises the following steps of response rate, response waiting time, driving receiving distance, online time of a user, total order resources, historical order completion fluctuation indexes, order completion quantity, sensitivity indexes aiming at target trip service, service stage of the user and group identification of the user.

The embodiment of the application discloses TS11, a service recommendation device, this service recommendation device includes:

the system comprises an information acquisition module, a data processing module and a data processing module, wherein the information acquisition module is used for acquiring classification information of a plurality of users, and the classification information comprises attribute information of the users and order information of target travel service used by the users;

the classification module is used for dividing the plurality of users into a plurality of user groups according to the classification information of the plurality of users according to each classification mode;

the system comprises a difference degree determining module, a difference degree determining module and a judging module, wherein the difference degree determining module is used for determining the classification difference degree of each classification mode according to the difference degree of the utilization rates of the target travel service among a plurality of user groups obtained by classification according to the classification mode;

the target classification determining module is used for determining a target classification mode from all classification modes based on the classification difference degrees of different classification modes;

and the service recommending module is used for recommending the target travel service to the users in the target user group corresponding to the target classification mode, wherein the target user group is determined from a plurality of user groups corresponding to the target classification mode.

TS12, the device according to TS11, wherein the service recommendation module selects the target user group from the plurality of user groups corresponding to the target classification manner according to the service recommendation priority of each user group corresponding to the target classification manner; the service recommendation priority of the target user group is determined according to the amount of the user in the user group using the target trip service, and the service recommendation priority is positively correlated with the amount of the user in the user group using the target trip service;

The TS13, the apparatus according to TS11, wherein the difference determining module determines, for each classification manner, a distribution difference indicator of the classification manner according to a usage rate of the target travel service used by users in the plurality of user groups classified according to the classification manner;

TS14, the apparatus of TS13, wherein the dissimilarity determination module determines the degree of classification dissimilarity for each classification by:

The TS15 is the device according to the TS14, wherein the difference degree determining module determines the number of qualified candidate user groups in the plurality of candidate user groups in each classification mode, the distribution difference indexes of the qualified candidate user groups fall into a preset confidence interval, and the probability of the true values of the distribution difference indexes of the preset confidence interval is larger than a set threshold value;

TS16, the apparatus of TS14, wherein the classification dimensions of each classification are the same; the classification thresholds for the same classification dimension for different classification approaches are different,

TS17, the apparatus of TS16, wherein the numerical distribution index for each classification threshold specifying a classification dimension is calculated as follows:

TS18, the apparatus of TS17, wherein the dissimilarity determination module determines a numerical distribution index for each classification threshold for a specified classification dimension by:

TS19, the apparatus of TS17, wherein the dissimilarity determination module determines a regularization index for each classification threshold specifying a classification dimension by:

TS20, the apparatus of TS11, wherein the attribute information includes at least one of: gender, age, driving age, school calendar, city, nationality, occupation;

Referring to fig. 6, fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. As shown in fig. 6, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

The memory 420 stores machine-readable instructions executable by the processor 410, when the electronic device 400 runs, the processor 410 communicates with the memory 420 through the bus 430, and when the machine-readable instructions are executed by the processor 410, the steps of the service recommendation method in the method embodiments shown in fig. 2 and fig. 3 may be performed.

An embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the service recommendation method in the method embodiments shown in fig. 2 and fig. 3 may be executed.

In particular, the storage medium can be a general-purpose storage medium, such as a removable disk, a hard disk, or the like, and the computer program on the storage medium can execute the service recommendation method when executed.

An embodiment of the present application further provides a computer program product, which includes a computer program/instruction, and when the computer program/instruction is executed by a processor, the steps of the service recommendation method in the method embodiments shown in fig. 2 and fig. 3 are implemented.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the exemplary embodiments of the present application, and are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A service recommendation method, comprising:

2. The method of claim 1, wherein determining the classification difference degree of each classification method according to the difference degree of the usage rate of the target travel service among the plurality of user groups classified according to the classification method comprises:

3. The method of claim 2, wherein the degree of classification variance for each classification is determined by:

4. The method of claim 3, wherein determining the degree of classification difference for each classification method comprises:

5. The method of claim 3, wherein the classification dimensions of each classification are the same; the classification thresholds for the same classification dimension for different classification approaches are different,

6. The method of claim 5, wherein the numerical distribution index for each classification threshold for a given classification dimension is calculated as follows:

7. A service recommendation device, comprising:

the information acquisition module is used for acquiring classification information of a plurality of users, wherein the classification information comprises attribute information of the users and order information of target travel service used by the users;

a difference determining module, configured to determine, for each classification manner, a classification difference of the classification manner according to a difference of usage rates of the target travel service among the plurality of user groups classified according to the classification manner;

and the service recommending module is used for recommending target travel service to users in a target user group corresponding to the target classification mode, wherein the target user group is determined from a plurality of user groups corresponding to the target classification mode.

8. An electronic device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when the electronic device is operating, the processor executing the machine-readable instructions to perform the steps of the method of any of claims 1 to 6.

9. A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, is adapted to carry out the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising computer program/instructions, characterized in that the computer program/instructions, when executed by a processor, implement the steps of the method according to any of claims 1 to 6.