CN114969543B

CN114969543B - Popularization method, popularization system, electronic equipment and storage medium

Info

Publication number: CN114969543B
Application number: CN202210676425.9A
Authority: CN
Inventors: 杜运辉; 王轶凡; 陈冠霖; 李世雷; 封树超; 陈冠丞; 刘少江; 孔清清; 张钋; 王雪
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-06-15
Filing date: 2022-06-15
Publication date: 2023-08-25
Anticipated expiration: 2042-06-15
Also published as: CN114969543A

Abstract

The application provides a popularization method, a popularization system, electronic equipment and a storage medium, relates to the technical field of computers, and particularly relates to the field of artificial intelligence such as big data, deep learning and the like. The specific implementation scheme is as follows: acquiring user use data corresponding to a target strategy; user use data are respectively input into a plurality of gain models to obtain a first prediction result output by each gain model; selecting a target gain model from among the gain models according to first prediction results respectively corresponding to the gain models; and determining a target crowd corresponding to the target strategy according to the target gain model. Therefore, under the condition of comparing the same input data, the first prediction results of the gain models are compared, and the target gain model is selected from the gain models according to the comparison result, so that the accuracy of selecting the target gain model is improved, the target crowd for popularizing the target strategy is determined based on the selected target gain model, and the accuracy of strategy popularization is improved.

Description

Popularization method, popularization system, electronic equipment and storage medium

Technical Field

The application relates to the technical field of computers, in particular to the field of artificial intelligence such as big data, deep learning and the like, and in particular relates to a popularization method, a popularization system, electronic equipment and a storage medium.

Background

In practical application, besides the influence of some strategies of the concerned product on the target index, different influences brought by different groups of people receiving the strategies can influence the target which finally needs to be increased.

In the related technology, the gain evaluation method can effectively locate the part of users which are obviously affected by the strategy, but the gain method has a plurality of applied machine learning models, and gains caused by adopting different gain model popularization strategies can be different.

Therefore, how to improve the accuracy of policy popularization is a problem to be solved.

Disclosure of Invention

The application provides a popularization method, a popularization system, electronic equipment and a storage medium. The specific scheme is as follows:

according to an aspect of the present application, there is provided a promotion method including:

acquiring user use data corresponding to a target strategy;

the user use data are respectively input into a plurality of gain models to obtain a first prediction result output by each gain model;

selecting a target gain model from among the gain models according to first prediction results respectively corresponding to the gain models;

and determining a target crowd corresponding to the target strategy according to the target gain model.

According to another aspect of the present application, there is provided a promotion system including:

the model module is used for acquiring user use data corresponding to the target strategy sent by the platform module, inputting the user use data into a plurality of gain models respectively to acquire a first prediction result output by each gain model, and sending the first prediction result output by each gain model to the platform module;

the platform module is used for acquiring user use data corresponding to a target strategy, sending the user use data to the model module, acquiring a first prediction result output by each gain model and sent by the model module, selecting a target gain model from the gain models according to the first prediction results respectively corresponding to the gain models, and determining a target crowd corresponding to the target strategy according to the target gain model.

According to another aspect of the present application, there is provided an electronic apparatus including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein, the liquid crystal display device comprises a liquid crystal display device,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of the above embodiments.

According to another aspect of the present application, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method according to the above-described embodiments.

According to another aspect of the application, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the steps of the method described in the above embodiments.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the application or to delineate the scope of the application. Other features of the present application will become apparent from the description that follows.

Drawings

The drawings are included to provide a better understanding of the present application and are not to be construed as limiting the application. Wherein:

FIG. 1 is a schematic flow chart of a promotion method according to an embodiment of the present application;

fig. 2 is a schematic flow chart of a promotion method according to another embodiment of the present application;

FIG. 3 is a schematic flow chart of a promotion method according to another embodiment of the present application;

fig. 4 is a schematic flow chart of a promotion method according to another embodiment of the present application;

fig. 5 is a schematic structural diagram of a promotion system according to an embodiment of the present application;

Fig. 6 is a schematic structural diagram of a promotion system according to another embodiment of the present application;

FIG. 7 is a schematic diagram of a promotion system according to an embodiment of the present application;

FIG. 8 is a schematic overall flow chart according to an embodiment of the present application;

fig. 9 is a block diagram of an electronic device for implementing a promotional method of an embodiment of the present application.

Detailed Description

Exemplary embodiments of the present application will now be described with reference to the accompanying drawings, in which various details of the embodiments of the present application are included to facilitate understanding, and are to be considered merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

Artificial intelligence is the discipline of studying certain mental processes and intelligent behaviors (e.g., learning, reasoning, thinking, planning, etc.) of a person using a computer, both in the technical field of hardware and in the technical field of software. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing, and the like; the artificial intelligence software technology comprises a computer vision technology, a voice recognition technology, a natural language processing technology, a deep learning technology, a big data processing technology, a knowledge graph technology and the like.

Deep learning is a new research direction in the field of machine learning. Deep learning is the inherent regularity and presentation hierarchy of learning sample data, and the information obtained during such learning is helpful in interpreting data such as text, images and sounds. Its final goal is to have the machine have analytical learning capabilities like a person, and to recognize text, image, and sound data.

The following describes a popularization method, a system, an electronic device and a storage medium of the embodiment of the present application with reference to the drawings.

Fig. 1 is a schematic flow chart of a promotion method according to an embodiment of the present application.

The popularization method of the embodiment of the application can be executed by the popularization system of the embodiment of the application, the device can be configured in the electronic equipment to select the target gain model according to the comparison result of the gain models, and the target crowd for popularizing the target strategy is determined based on the target gain model, so that the accuracy of strategy popularization is improved.

The electronic device may be any device with computing capability, for example, may be a personal computer, a mobile terminal, a server, etc., and the mobile terminal may be, for example, a vehicle-mounted device, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, etc., which have various operating systems, touch screens, and/or display screens.

As shown in fig. 1, the popularization method includes:

step 101, user use data corresponding to a target policy is obtained.

In the application, after one strategy of the product is online, the use data of the user on the product can be obtained, so that the user use data corresponding to the strategy, namely the user use data corresponding to the target strategy, is obtained.

The user usage data may include, among other things, gain targets, user identifications, product identifications, target policy identifications, user characteristics, and the like. The user characteristics may include, among other things, user gender, age, preference, etc.

For example, where the product is an application and the policy is to expose a function of the application, the user usage data may include: target Y (gain target, e.g., user use time of the product), user id, product id_policy sid, whether the policy is valid (e.g., whether a product function is exposed to the user), user characteristics, etc.

In the technical scheme of the application, the acquisition, storage, application and the like of the related user personal information all conform to the regulations of related laws and regulations and do not violate the popular regulations.

In the application, the user use data comprises two types of user use data, one is a user who implements the strategy, and the other is a user who does not implement the strategy, namely, the user use data comprises experimental group data and control group data.

Step 102, user usage data is respectively input into a plurality of gain models to obtain a first prediction result output by each gain model.

In the application, the popularization system is integrated with a plurality of gain models, the acquired user use data can be respectively input into the gain models, and the first prediction result output by each gain model can be acquired.

The first prediction result may include a prediction result corresponding to each user. For example, the gain target is retention of an application program, and the first prediction result may include retention promotion probabilities corresponding to each user in the experimental group and the control group.

In the application, a plurality of gain models are obtained by adopting different modeling modes, thereby enriching the variety of the gain models.

Step 103, selecting a target gain model from among the gain models according to the first prediction results corresponding to the gain models.

According to the method and the device, the gain corresponding to each gain model can be determined according to the first prediction result corresponding to each gain model, and the model with the highest gain is selected from a plurality of gain models to be used as the target gain model.

Step 104, determining a target crowd corresponding to the target strategy according to the target gain model.

According to the method and the device, the use data of the users for the products can be obtained, the use data of the users are input into the target gain model, and the users with the predicted values larger than the preset threshold value can be used as target crowds.

After the target crowd is determined, the target strategy can be promoted to the target crowd, so that the accurate promotion of the target strategy is realized, and the gain is improved. For example, the target population is women over 25 years of age, and then the target policy may be enforced to female users over 25 years of age.

In the embodiment of the application, the obtained user use data corresponding to the target strategy is respectively input into the gain models to obtain the first prediction result output by each gain model, and the target gain model is selected from the gain models according to the comparison result by comparing the first prediction results of the gain models under the condition of the same input data, so that the accuracy of selecting the target gain model is improved, the target crowd for popularizing the target strategy is determined based on the selected target gain model, and the accuracy of strategy popularization and the gain are improved.

Fig. 2 is a schematic flow chart of a promotion method according to another embodiment of the present application.

As shown in fig. 2, the popularization method includes:

step 201, user usage data corresponding to a target policy is obtained.

Step 202, user usage data is respectively input into a plurality of gain models to obtain a first prediction result output by each gain model.

In the present application, steps 201 to 202 are similar to those described in the above embodiments, and thus are not repeated here.

Step 203, determining a first evaluation result corresponding to each gain model according to the first prediction result corresponding to each gain model.

In the application, the same evaluation index can be adopted to evaluate the first prediction results corresponding to the gain models respectively, so as to obtain the first evaluation result corresponding to each gain model.

Wherein the evaluation index may comprise a monotonicity score, an average causality effect, an individual causality effect, a Qini curve, etc.

Step 204, selecting a target gain model from among the gain models according to the first evaluation result corresponding to each gain model.

In the application, the first evaluation results of the gain models can be compared to select the gain model with the largest gain from the gain models, namely, select the target gain model from the gain models.

Taking the evaluation index as a monotonicity score as an example, if the first prediction result includes a retention lifting probability after each user policy, then a mean value of retention lifting probabilities corresponding to each user corresponding to each gain model may be taken as an overall lifting probability, a ratio between the overall lifting probability and a true lifting probability may be taken as a monotonicity score, and a gain model with the highest monotonicity score may be taken as a target gain model. The actual boost probability may be the difference between the actual retention of the experimental group and the actual retention of the control group.

Step 205, determining a target crowd corresponding to the target strategy according to the target gain model.

In the present application, step 205 is similar to that described in the above embodiments, and thus will not be described herein.

In the embodiment of the application, when the target gain model is selected from the multiple gain models, the first evaluation result of each gain model can be determined according to the first prediction result of each gain model, and the gain model with better gain effect can be selected from the multiple gain models by comparing the first evaluation results of the multiple gain models, so that the accuracy of gain model selection is improved.

Fig. 3 is a schematic flow chart of a promotion method according to another embodiment of the present application.

As shown in fig. 3, the popularization method includes:

step 301, user usage data corresponding to the target policy is obtained.

Step 302, user usage data is respectively input into a plurality of gain models to obtain a first prediction result output by each gain model.

Step 303, selecting a target gain model from among the gain models according to the first prediction results corresponding to the gain models.

In the present application, steps 301 to 303 are similar to those described in the above embodiments, and thus are not described herein.

And step 304, generating an explanatory tree diagram corresponding to the target gain model according to the first prediction result corresponding to the target gain model.

According to the method and the device, an explanatory tree diagram corresponding to the target gain model can be generated by calling the model according to the first prediction result corresponding to the target gain. For example, a policy learning tree or decision tree may be invoked to generate an explanatory tree graph.

In the application, each tree node in the explanatory tree diagram corresponding to the target model can comprise the gain size corresponding to the user characteristics, whether a strategy is adopted or not, and the like.

Step 305, determining a target user feature corresponding to the user gain greater than the threshold according to the user gain corresponding to each user feature in the explanatory tree diagram.

According to the user gain method and the user gain device, the user gain which is larger than the threshold value can be determined according to the user gain corresponding to each user feature in the explanatory tree diagram, and the user feature with the user gain larger than the threshold value is used as the target user feature.

For example, the target user is characterized by an age of 25 years, 26 years, 28 years, 30 years.

Step 306, determining a target crowd corresponding to the target strategy according to the target user characteristics.

According to the application, the user conforming to the target user characteristics can be determined according to the target user characteristics, and the user can be used as a target crowd corresponding to the target strategy.

For example, a target user is characterized by an age of 25 years, 26 years, 28 years, 30 years, etc., and users older than 25 years may be considered as target groups.

According to the method and the device for determining the target crowd, the explanatory tree diagram corresponding to the target gain model can be generated according to the first prediction result corresponding to the target gain model, the target user characteristics corresponding to the user gains larger than the threshold value are determined according to the user gains corresponding to the user characteristics in the explanatory tree diagram, and the target crowd corresponding to the target strategy is determined based on the target user characteristics, so that the target crowd with better gain effect can be determined according to the explanatory tree diagram corresponding to the target gain model, and the method and the device are simple and convenient.

Fig. 4 is a schematic flow chart of a promotion method according to another embodiment of the present application.

As shown in fig. 4, the popularization method further includes:

step 401, obtaining training data corresponding to a plurality of candidate gain models and target strategies.

In the application, before the target gain model is determined, a plurality of candidate gain models and training data corresponding to the target strategy can be obtained.

The training data corresponding to the target strategy is similar to the user use data corresponding to the target strategy, the training data also comprises control group data and experimental group data, and the training data can be used for modeling.

In order to facilitate training, when training data is acquired, if the acquired data does not accord with the standard specification, the data can be processed so as to ensure that the training data obtained after processing accords with the standard specification. For example, the standard specification is that training data needs to include information such as target Y, user id, target policy id, user characteristics, etc.

In practical application, the gain models are relatively large, and in order to facilitate calculation, a plurality of candidate gain models can be screened out from a large number of gain models in advance. When the method is implemented, a plurality of initial gain models can be obtained, the plurality of initial gain models can be respectively trained by utilizing a standard data set, a plurality of trained gain models are obtained, and a plurality of models are selected from the plurality of trained gain models to serve as candidate gain models based on evaluation results of the plurality of trained gain models. Therefore, a plurality of candidate gain models can be screened out through training a plurality of initial gain models, and the subsequent calculated amount is reduced.

And step 402, training the plurality of candidate gain models by using training data to obtain a plurality of gain models.

According to the application, each candidate gain model can be trained by using training data, so as to obtain a corresponding gain model. The modeling modes of the gain models of each candidate model are different, and the gain model can be obtained by adopting the corresponding modeling modes.

For example, a differential model based on a double model is to model experimental group data and control group data in training data respectively, and the control group data and the experimental group data can be modeled together.

In the application, when the candidate gain model is trained, the training can be performed in a deep learning mode, and compared with other machine learning methods, the deep learning has better performance on a large data set.

Step 403, test data corresponding to the target policy is obtained.

The test data corresponding to the target strategy is similar to the training data corresponding to the target strategy, and is the usage data of some users for the product obtained after the target strategy of the product is online, and the test data also comprises experimental group data and control group data.

In the application, after training the training data corresponding to the target strategy to obtain a plurality of gain models, test data corresponding to the target strategy can be obtained, or after the target strategy of the product is on line, use data of some users for the product can be obtained, and one part of the obtained use data of the users is used as training data, and the other part is used as test data.

Step 404, inputting the test data to the gain models respectively to obtain a second prediction result output by each gain model.

After the plurality of gain models are acquired, the effect evaluation can be performed on the plurality of gain models. In the application, test data corresponding to a target strategy are acquired, and the test data are respectively input into a plurality of gain models to acquire a second prediction result output by each gain model.

And step 405, evaluating the second prediction result corresponding to each gain model by using the same evaluation index to determine a second evaluation result corresponding to each gain model.

Wherein the evaluation index may comprise an average causal effect, an individual causal effect, a group causal effect, a Qini curve, a boost curve, a gain curve, etc.

In the application, the values of one or more evaluation indexes can be calculated according to the second prediction result of each gain model, so as to obtain the second evaluation result corresponding to each gain model. Thus, the gain effect of each gain model can be determined based on the second evaluation result.

In order to facilitate visual observation of the gain effect of the model or visual comparison of the gain effects of the gain models, the second evaluation results of one or more gain models can be displayed. For example, an evaluation index graph of a gain model with the best gain effect can be displayed, wherein the evaluation index graph comprises a grouping causal effect graph, a Qini graph, a lifting graph, a gain graph and the like.

According to the embodiment of the application, the plurality of candidate gain models can be trained by utilizing the training data corresponding to the target strategies to obtain the plurality of gain models, so that the plurality of gain models corresponding to different strategies can be obtained by training based on the training data of different target strategies, the test data corresponding to the target strategies can be respectively input into the plurality of gain models to obtain the second prediction results of the plurality of gain models, and the plurality of gain models are evaluated by adopting the unified evaluation index, so that the gain effects of the gain models under the same input can be conveniently compared.

In order to achieve the above embodiment, the embodiment of the present application further provides a popularization system. Fig. 5 is a schematic structural diagram of a promotion system according to an embodiment of the present application.

As shown in fig. 5, the promotion system 500 includes: a platform module 510 and a model module 520, wherein,

the model module 520 is configured to obtain user usage data corresponding to the target policy sent by the platform module 510, input the user usage data into the multiple gain models respectively, obtain a first prediction result output by each gain model, and send the first prediction result output by each gain model to the platform module 510.

The platform module 510 is configured to obtain user usage data corresponding to a target policy, send the user usage data to the model module 520, obtain a first prediction result output by each gain model sent by the model module 520, select a target gain model from among a plurality of gain models according to the first prediction results respectively corresponding to the gain models, and determine a target crowd corresponding to the target policy according to the target gain model.

In the present application, after a policy of a product is online, the platform module 510 may obtain usage data of the product by a user, so as to obtain usage data of the user corresponding to the policy, that is, usage data of the user corresponding to a target policy.

In the present application, the popularization system 500 has a plurality of gain models, and the model module 520 may input the user usage data corresponding to the target policy into the gain models, respectively, and may obtain the first prediction result output by each gain model.

The model module 520 sends the first prediction result output by each gain model to the platform module 510, the platform module 510 can determine the gain size corresponding to each model according to the first prediction result corresponding to each gain model, select the model with the highest gain from a plurality of gain models as a target gain model, obtain the usage data of some users for the product, input the usage data of the users into the target gain model, and can use the users with the prediction value larger than the preset threshold as a target crowd.

In one embodiment of the present application, based on the promotion system shown in fig. 5, the platform module 510 is configured to determine a first evaluation result corresponding to each gain model according to a first prediction result corresponding to each gain model, and select a target gain model from among multiple gain models according to the first evaluation result corresponding to each gain model.

In the present application, the platform module 510 may adopt the same evaluation index to evaluate the first prediction results corresponding to the multiple gain models respectively, obtain the first evaluation result corresponding to each gain model, and compare the first evaluation results of the multiple gain models, so as to select the gain model with the largest gain from the multiple gain models, that is, select the target gain model from the multiple gain models.

In the embodiment of the application, the platform module can determine the first evaluation result corresponding to each gain model according to the first prediction result corresponding to each gain model, select the target gain model from a plurality of gain models according to the first evaluation result corresponding to each gain model, and select the gain model with better gain effect from the plurality of gain models by comparing the evaluation results of the plurality of gain models, thereby improving the accuracy of selection.

In one embodiment of the present application, based on the promotion system shown in fig. 5, the platform module 510 is configured to generate an explanatory tree diagram corresponding to the target gain model according to the first prediction result corresponding to the target gain model, determine a target user feature corresponding to a user gain greater than a threshold according to the user gain corresponding to each user feature in the explanatory tree diagram, and determine a target crowd according to the target user feature.

In the present application, the platform module 510 may generate an explanatory tree diagram corresponding to the target gain model by calling the tree model according to the first prediction result corresponding to the target gain. The tree model may be a policy learning tree, a decision tree, or the like.

In the application, each tree node in the explanatory tree diagram corresponding to the target model comprises the gain size corresponding to the user characteristic, whether a strategy is adopted or not, and the like.

According to the method and the system, the platform module can determine the user gain larger than the threshold according to the user gain corresponding to each user feature in the explanatory tree diagram, and the user gain larger than the threshold is used as the target user feature according to the user features of the user gain, and the user conforming to the target user feature is determined according to the target user feature and is used as the target crowd corresponding to the target strategy.

In the embodiment of the application, the platform module 510 can generate the explanatory tree diagram corresponding to the target gain model according to the first prediction result corresponding to the target gain model, determine the target user characteristics corresponding to the user gains greater than the threshold according to the user gains corresponding to the user characteristics in the explanatory tree diagram, and determine the target crowd according to the target user characteristics, thereby determining the target crowd with better gain effect according to the explanatory tree diagram corresponding to the target gain model, and being simple and convenient.

Fig. 6 is a schematic structural diagram of a promotion system according to another embodiment of the present application.

As shown in fig. 6, the promotion system 600 includes: platform module 610, model module 620, data module 630, and evaluation module 640.

The model module 620 is configured to obtain user usage data corresponding to the target policy sent by the platform module 610, input the user usage data into the multiple gain models respectively, obtain a first prediction result output by each gain model, and send the first prediction result output by each gain model to the platform module 610.

The platform module 610 is configured to obtain user usage data corresponding to a target policy, send the user usage data to the model module 620, obtain a first prediction result output by each gain model sent by the model module 620, select a target gain model from among a plurality of gain models according to the first prediction results respectively corresponding to the gain models, and determine a target crowd corresponding to the target policy according to the target gain model.

In the present application, the platform module 610 and the model module 620 function similarly to the platform module 510 and the model module 520 in the above embodiments, and thus, this step is repeated.

The data module 630 is configured to obtain training data corresponding to the target policy, and send the training data to the model module 620.

The model module 620 is configured to obtain training data sent by the data module 630, obtain a plurality of candidate gain models from the evaluation module 640, and respectively train the plurality of candidate gain models by using the training data to obtain a plurality of gain models.

In the present application, the data module 630 may obtain training data corresponding to the target policy, and send the training data to the model module 620, where the model module 620 obtains the training data corresponding to the target policy sent by the data module 630, and obtains a plurality of candidate gain models from the evaluation module 640, and trains each candidate gain model by using the training data, so as to obtain a corresponding gain model.

For the convenience of training, in the present application, standard specifications of training data may be specified, and the data module 630 ensures that the training data meets the standard specifications when acquiring the training data. For example, the standard specification is that training data needs to include information such as target Y, user id, target policy id, user characteristics, etc.

In the application, the modeling modes of each candidate model gain model are different, and the gain model can be obtained by adopting the corresponding modeling modes. For example, a differential model based on a double model is to model experimental group data and control group data in training data respectively, and the control group data and the experimental group data can be modeled together.

In the present application, the data module 630 may support feature data with csv format as input data, support local data and cluster HDFS (Hadoop Distributed File System ) data reading, and may directly call the data interface with instructions, for example: -s CSV input FILE, or-HDFS hdfs_file is used to obtain data at the cluster after the cluster client is set.

In addition, the data module 630 may also access the generated analog data, where the analog data is generated various types of data, such as nonlinear intervention collocation linear feature data or linear intervention collocation nonlinear feature data.

In the embodiment of the application, the model module can train a plurality of candidate gain models by utilizing the training data corresponding to the target strategies to obtain a plurality of gain models, so that training can be performed based on the training data of different target strategies to obtain a plurality of gain models corresponding to different strategies.

In one embodiment of the present application, based on the popularization system shown in fig. 6, the model module 620 may be further configured to obtain test data corresponding to the target policy, input the test data to a plurality of gain models respectively, obtain a second prediction result output by each gain model, and send the second prediction result output by each gain model to the evaluation module 640. And the evaluation module 640 is configured to obtain the second prediction result output by each gain model sent by the model module, and evaluate the second prediction result corresponding to each gain model by using the same evaluation index, so as to determine the second evaluation result corresponding to each gain model.

Therefore, the test data are respectively input into the gain models to obtain the second prediction results of the gain models, and the gain models are evaluated by adopting the unified evaluation index, so that the gain effects of the gain models under the same input can be conveniently compared.

After the multiple gain models are obtained, the multiple gain models may be evaluated for effectiveness using the evaluation module 640. In the present application, the model module 620 obtains the test data corresponding to the target policy, and inputs the test data into the multiple gain models respectively, so as to obtain the second prediction result output by each gain model, and sends the second prediction result output by each gain model to the evaluation module 640.

The test data is similar to the training data, and is the usage data of the product by the user obtained after the target strategy of the product is online, and the test data also comprises experimental group data and control group data.

The evaluation module 640 may calculate a value of one or more evaluation indexes according to the second prediction result of each gain model, so as to obtain a second evaluation result corresponding to each gain model. Thus, the gain effect of each gain model can be determined based on the second evaluation result.

For visual comparison purposes, the platform module 610 may also present a second evaluation of one or more gain models in the present application. For example, an evaluation index graph of a gain model with the best gain effect can be displayed, wherein the evaluation index graph comprises a grouping causal effect graph, a Qini graph, a lifting graph, a gain graph and the like.

In practical applications, there are a large number of gain models, based on which, in one embodiment of the present application, based on the popularization system shown in fig. 6, the evaluation module 640 may screen the model module 620 for a plurality of candidate gain models from a large number of gain models for the purpose of calculation.

In implementation, the evaluation module 640 may obtain a plurality of initial gain models, and may respectively train the plurality of initial gain models using the standard data set to obtain a plurality of trained gain models, and select a plurality of models from the plurality of trained gain models as a plurality of candidate gain models based on the evaluation results of the plurality of trained gain models. Therefore, a plurality of candidate gain models can be screened out through training a plurality of initial gain models, and the subsequent calculated amount is reduced.

In one embodiment of the application, the promotion system as shown in FIG. 6 may also include an validation module 650.

And an validation module 650, configured to promote the target policy to the target crowd. Therefore, the accurate popularization of the target strategy is realized, and the gain is improved.

For example, the target population is women over 25 years of age, and then the target policy may be enforced to female users over 25 years of age.

In order to facilitate understanding of the foregoing embodiments, the promotion system is further described with reference to fig. 7, and fig. 7 is a schematic frame diagram of the promotion system according to an embodiment of the present application.

As shown in fig. 7, the promotion system includes a platform module, a data module, a model module, an evaluation module, and an validation module.

The platform module is responsible for displaying input and output of the whole flow; the data module is responsible for effectively importing data aiming at a flow experiment into the model module; the model module is responsible for model training and evaluation; and the effective module is finally responsible for pushing the whole implementation or the target crowd strategy online. The above modules are described below.

And (3) a platform module: after the strategy is online, product use data are collected, an interface of a model module is called, a model comparison result is obtained, and profit confirmation can be carried out.

In addition, in the platform module shown in fig. 7, the original policy index and the new policy index may be, for example, retention rate, a usage period of a product by a user, etc., a true value, such as a true retention rate, required for comparing a gain effect may be determined based on the policy index, a target gain model with an optimal gain effect may be determined according to the gain effect comparison result, and then, a benefit confirmation is performed by using the target gain model. The gain confirmation method is used for comparing the user gain selected by the model with the original strategy. The platform module can also record, display and count the benefits.

Under the condition that the gain effect comparison accords with the expectation, the strategy can be directly adopted for the product user, and after the strategy is adopted, a new strategy index can be obtained for the gain effect comparison.

The data module can ensure that the received training data all meet the standard, support the input data to be the characteristic data in csv format, support the local data and the cluster HDFS data to read, and can also be accessed to the generated analog data.

In fig. 7, the data module may acquire service flow data based on policy configuration of the platform module, extract user behavior features from the service flow data through the feature extraction operator, and obtain external features from external data (such as portraits, etc.), and synthesize the user behavior features and the external features to obtain user granularity data for model training. External features herein may be understood as user features, and user granularity data, that is, training data corresponding to the target policy in the above embodiment.

In fig. 7, the evaluation module may perform model training on the algorithm in the candidate algorithm set (i.e., performing model training on the plurality of initial gain models) by using the historical verification data and the generated data, that is, the standard data set, and evaluate the model effect by using the evaluation indexes in the evaluation index set, and add the model meeting the expectation to the formal algorithm set according to the evaluation result, and re-perform the investigation, development and submission of the gain algorithm for the algorithm not meeting the expectation. The candidate algorithm set is obtained through investigation, development and submission of a gain algorithm, and the formal algorithm set comprises the candidate gain models.

The model module is responsible for model loading and training, the model module can train the models in the formal algorithm set by using the user granularity data, namely train the models screened by the evaluation module, evaluate the model effect by using the evaluation indexes in the evaluation index set of the evaluation model, obtain the model interpretation tree diagram, and the platform module can display the generalized characteristics or the business benefit evaluation of the model interpretation tree diagram.

In addition, the computing architecture interface in the model implementation interface realizes single machine training and high concurrency training, and aiming at the high concurrency training, the popularization system can fuse an open source Fugue distributed technical architecture to decouple the model algorithm logic of the upper layer from the parallel computing implementation of the bottom layer, so that a user only needs to define a candidate gain model, does not need to pay attention to computing implementation details of the bottom layer, and ensures that the corresponding model can perform more efficient training and prediction in a distributed mode. In the implementation of the local code, the local code can be planted on a computing engine such as Spark and Dask to run only by a small amount of change, so that the decoupling of the algorithm logic and the concurrency frame is achieved, different corresponding codes are not required to be developed respectively due to various high concurrency frames, and the production cost is greatly saved.

And the validation module can perform one-key export validation based on platform presentation or directly perform one-key export validation based on generalized characteristics or service benefit evaluation in the model module. The one-key deriving validation comprises directly carrying out real-time strategies on all users of the product, or sampling validation, namely, online packing of experimental groups.

In the embodiment of the application, corresponding business targets and product data (such as searching product data, recommending product data, mobile phone application program product data and the like) can be combined, after a popularization system is adopted to carry out multi-aspect comparison and model selection of models, the models with the best gain effect are obtained, potential optimal benefit groups are defined, product improvement suggestions are provided pertinently, meanwhile, verification can be carried out by further combining a small flow experiment, the sample size of the small flow experiment can be enough for modeling, the whole flow can be shown in fig. 8, the first step is strategy online, the second step is gain model comparison, the third step is to determine the optimal gain effect groups based on gain model comparison, and the fourth step is to verify whether the small flow verification decision is pushed or not, wherein partial user use data can be adopted for verification.

The popularization system of the embodiment of the application integrates the training, analysis and evaluation of the gain model, supports the user gain analysis under the complex scene of the Internet product, integrates a plurality of gain models, can meet the gain analysis requirements of a plurality of scenes, adopts uniform evaluation indexes, can support the effect comparison and the visualized graph expression of a plurality of different models, and is integrated with a distributed technical architecture to realize high concurrency training, thereby effectively solving the problem of the performance bottleneck commonly existing in single-point computing.

According to embodiments of the present application, the present application also provides an electronic device, a readable storage medium and a computer program product.

FIG. 9 shows a schematic block diagram of an example electronic device 900 that may be used to implement an embodiment of the application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the applications described and/or claimed herein.

As shown in fig. 9, the apparatus 900 includes a computing unit 901 that can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 902 or a computer program loaded from a storage unit 908 into a RAM (Random Access Memory ) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other by a bus 904. An I/O (Input/Output) interface 905 is also connected to bus 904.

Various components in device 900 are connected to I/O interface 905, including: an input unit 906 such as a keyboard, a mouse, or the like; an output unit 907 such as various types of displays, speakers, and the like; a storage unit 908 such as a magnetic disk, an optical disk, or the like; and a communication unit 909 such as a network card, modem, wireless communication transceiver, or the like. The communication unit 909 allows the device 900 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunications networks.

The computing unit 901 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 901 include, but are not limited to, a CPU (Central Processing Unit ), GPU (Graphic Processing Units, graphics processing unit), various dedicated AI (Artificial Intelligence ) computing chips, various computing units running machine learning model algorithms, DSP (Digital Signal Processor ), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the respective methods and processes described above, such as a generalization method. For example, in some embodiments, the promotional method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 908. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device 900 via the ROM902 and/or the communication unit 909. When the computer program is loaded into the RAM 703 and executed by the computing unit 901, one or more steps of the promotion method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform the promotion method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit System, FPGA (Field Programmable Gate Array ), ASIC (Application-Specific Integrated Circuit, application-specific integrated circuit), ASSP (Application Specific Standard Product, special-purpose standard product), SOC (System On Chip ), CPLD (Complex Programmable Logic Device, complex programmable logic device), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present application may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of the present application, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory, erasable programmable read-Only Memory) or flash Memory, an optical fiber, a CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display ) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network ), WAN (Wide Area Network, wide area network), internet and blockchain networks.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service (Virtual Private Server, virtual special servers) are overcome. The server may also be a server of a distributed system or a server that incorporates a blockchain.

According to an embodiment of the present application, there is also provided a computer program product, which when executed by an instruction processor in the computer program product, performs the promotion method set forth in the above embodiment of the present application.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present application may be performed in parallel, sequentially, or in a different order, so long as the desired results of the technical solution disclosed in the present application can be achieved, and are not limited herein.

The above embodiments do not limit the scope of the present application. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of the present application.

Claims

1. A promotional method comprising:

acquiring user use data corresponding to a target strategy;

the user use data are respectively input into a plurality of gain models to obtain a first prediction result output by each gain model, wherein the first prediction result comprises a prediction result corresponding to each user;

Determining the gain corresponding to each model according to the first prediction results corresponding to the gain models, and selecting the model with the highest gain from the gain models as a target gain model;

determining a target crowd corresponding to the target strategy according to the target gain model;

the method further comprises the steps of:

acquiring training data corresponding to a plurality of candidate gain models and the target strategy;

and training the plurality of candidate gain models by utilizing the training data respectively to obtain the plurality of gain models.

2. The method of claim 1, the selecting a target gain model from among the plurality of gain models comprising:

determining a first evaluation result corresponding to each gain model according to a first prediction result corresponding to each gain model;

and selecting the target gain model from among the gain models according to a first evaluation result corresponding to each gain model.

3. The method of claim 1, wherein the determining, according to the target gain model, a target crowd corresponding to the target policy comprises:

generating an explanatory tree diagram corresponding to the target gain model according to a first prediction result corresponding to the target gain model;

Determining target user characteristics corresponding to user gains larger than a threshold according to the user gains corresponding to the user characteristics in the explanatory tree diagram;

and determining the target crowd according to the target user characteristics.

4. The method of claim 1, wherein after said deriving said plurality of gain models, further comprising:

obtaining test data corresponding to the target strategy;

respectively inputting the test data into the gain models to obtain a second prediction result output by each gain model;

and evaluating the second prediction results corresponding to each gain model by using the same evaluation index so as to determine the second evaluation results corresponding to each gain model.

5. The method of claim 4, wherein the method further comprises:

and displaying second evaluation results of one or more gain models in the plurality of gain models.

6. The method of claim 1, wherein the obtaining a plurality of candidate gain models comprises:

acquiring a plurality of initial gain models;

training the plurality of initial gain models respectively to obtain a plurality of trained gain models;

the plurality of candidate gain models is selected from among the plurality of trained gain models.

7. The method of claim 1, wherein after the determining the target crowd corresponding to the target policy according to the target gain model, further comprising:

and promoting the target strategy to the target crowd.

8. The method of claim 1, wherein the plurality of gain models are modeled differently.

9. A promotional system comprising: a platform module and a model module, wherein,

the model module is used for acquiring user use data corresponding to the target strategy sent by the platform module, inputting the user use data into a plurality of gain models respectively to acquire a first prediction result output by each gain model, and sending the first prediction result output by each gain model to the platform module, wherein the first prediction result comprises a prediction result corresponding to each user;

the platform module is used for acquiring user use data corresponding to a target strategy, sending the user use data to the model module, acquiring a first prediction result output by each gain model sent by the model module, determining the gain size corresponding to each model according to the first prediction results respectively corresponding to the gain models, selecting a model with the highest gain from the gain models as a target gain model, and determining a target crowd corresponding to the target strategy according to the target gain model;

The system further comprises: a data module and an evaluation module, wherein,

the data module is used for acquiring training data corresponding to the target strategy and sending the training data to the model module;

the model module is used for acquiring the training data sent by the data module, acquiring a plurality of candidate gain models from the evaluation module, and respectively training the plurality of candidate gain models by utilizing the training data to obtain the plurality of gain models.

10. The system of claim 9, wherein,

the platform module is used for determining a first evaluation result corresponding to each gain model according to a first prediction result corresponding to each gain model, and selecting the target gain model from the gain models according to the first evaluation result corresponding to each gain model.

11. The system of claim 9, wherein,

the platform module is used for generating an explanatory tree diagram corresponding to the target gain model according to a first prediction result corresponding to the target gain model, determining target user characteristics corresponding to user gains larger than a threshold according to user gains corresponding to user characteristics in the explanatory tree diagram, and determining the target crowd according to the target user characteristics.

12. The system of claim 9, wherein,

the model module is used for acquiring test data corresponding to the target strategy, inputting the test data into the gain models respectively, acquiring a second prediction result output by each gain model, and sending the second prediction result output by each gain model to the evaluation module;

the evaluation module is configured to obtain a second prediction result output by each gain model sent by the model module, and evaluate, with the same evaluation index, the second prediction result corresponding to each gain model to determine a second evaluation result corresponding to each gain model.

13. The system of claim 12, wherein,

the platform module is further configured to display a second evaluation result of one or more gain models in the plurality of gain models.

14. The system of claim 9, wherein,

the evaluation module is used for acquiring a plurality of initial gain models, respectively training the plurality of initial gain models to obtain a plurality of trained gain models, and selecting the plurality of candidate gain models from the plurality of trained gain models.

15. The system of claim 9, further comprising: an validation module, wherein,

and the validation module is used for popularizing the target strategy to the target crowd.

16. The system of claim 9, wherein the plurality of gain models are modeled differently.

17. An electronic device, comprising:

at least one processor; and

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

18. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-8.