CN114611615A

CN114611615A - Object classification processing method and device, computer equipment and storage medium

Info

Publication number: CN114611615A
Application number: CN202210257064.4A
Authority: CN
Inventors: 钟子宏
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2022-03-16
Filing date: 2022-03-16
Publication date: 2022-06-10

Abstract

The application relates to an object classification processing method, an object classification processing device, computer equipment and a storage medium. Can be applied to the fields of maps and traffic, including: determining an object classification model of the classification accuracy to be analyzed; acquiring historical object characteristics of a plurality of objects in at least three target historical periods respectively; inputting the historical object characteristics of each object into an object classification model with classification accuracy to be analyzed for classification aiming at the historical object characteristics of each object in each historical period so as to predict the prediction category of each object in the next period of the historical period, and obtaining the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods; and determining the classification accuracy of the object classification model in the current period based on the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods. By adopting the method, the accuracy of classification accuracy can be improved.

Description

Object classification processing method and device, computer equipment and storage medium

Technical Field

The present application relates to the field of computer technologies, and in particular, to an object classification processing method and apparatus, a computer device, and a storage medium.

Background

With the development of computer technology, machine learning is applied more and more widely, for example, in the fields of transportation, games, and automatic driving. Taking the traffic field as an example, for a scene needing to be classified in the traffic field, a corresponding classification model can be obtained based on machine learning training, and classification is realized by using the classification model.

At present, because the classification accuracy is an important index of the classification model, the classification accuracy of the classification model generally needs to be analyzed, but the accuracy of the current classification model is not accurate enough, so that the accuracy of the classification accuracy is low.

Disclosure of Invention

In view of the above, it is necessary to provide an object classification processing method, an apparatus, a computer device, a computer readable storage medium, and a computer program product, which can improve the accuracy of classification.

In one aspect, the application provides an object classification processing method. The method comprises the following steps: determining an object classification model of the classification accuracy to be analyzed; acquiring historical object characteristics of a plurality of objects in at least three target historical periods respectively; the at least three target history cycles are at least three continuous history cycles selected from the previous cycle of the current cycle; inputting the historical object features of each object into the object classification model for classification aiming at the historical object features of each object in each target historical period so as to predict the prediction category of each object in the next period of the target historical period, and obtain the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods; the at least two adjacent history cycles are target history cycles closer to the current cycle of the at least three target history cycles; determining the classification accuracy of the object classification model in the current cycle based on the prediction category of each of the objects in the current cycle and the prediction categories of each of the objects in the at least two adjacent historical cycles.

On the other hand, the application also provides an object classification processing device. The device comprises: the model determining module is used for determining an object classification model of the classification accuracy to be analyzed; the characteristic acquisition module is used for acquiring historical object characteristics of a plurality of objects in at least three target historical periods respectively; the at least three target history cycles are at least three continuous history cycles selected from the previous cycle of the current cycle; a class obtaining module, configured to, for a historical object feature of each object in each historical period, input the historical object feature of each object into the object classification model for classification, so as to predict a prediction class of each object in a next period of the historical period, and obtain a prediction class of each object in the current period and prediction classes of each object in at least two adjacent historical periods; the at least two adjacent history cycles are target history cycles closer to the current cycle of the at least three target history cycles; an accuracy determination module for determining the classification accuracy of the object classification model in the current cycle based on the prediction category of each of the objects in the current cycle and the prediction categories of each of the objects in the at least two adjacent historical cycles.

In some embodiments, the at least two adjacent history cycles include a first history cycle adjacent to and prior to the current cycle and a second history cycle adjacent to and prior to the first history cycle; the accuracy determination module is further to: obtaining the prediction category of each object in the current period to obtain the current prediction category of each object; obtaining the prediction category of each object in the first history period to obtain the first prediction category of each object; obtaining the prediction category of each object in the second history period to obtain the second prediction category of each object; and determining the classification accuracy of the object classification model in the current period based on the current prediction category, the first prediction category and the second prediction category of each object.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a first class condition in the plurality of objects to obtain a first object number; the first category condition includes: the real category of the object in the current period is a first preset category, and the current prediction category of the object is the first preset category; counting the number of objects meeting a second category condition in the plurality of objects to obtain a second object number; the second category condition includes: the real category of the object in the second history period is the first preset category, the second prediction category of the object is the second preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the first preset category; determining a classification accuracy of the object classification model for the current cycle based on the first number of objects and the second number of objects.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a third category condition in the plurality of objects to obtain a third object number; the third category of conditions includes: the real category of the object in the second history period is the first preset category, the second prediction category of the object is the second preset category, the real category of the object in the first history period is the second preset category, and the first prediction category of the object is the first preset category; counting based on the first object quantity and the second object quantity to obtain a forward statistical value; the forward statistical value is in positive correlation with the first object quantity and the second object quantity; counting based on the first object quantity and the third object quantity to obtain a negative statistic value; the negative statistic value is in positive correlation with the first object quantity and the third object quantity; determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics; the classification accuracy and the positive statistic value form a positive correlation relationship, and the classification accuracy and the negative statistic value form a negative correlation relationship.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category condition includes: the real category of the object in the current period is the first preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the second preset category; counting based on the first object number, the second object number and the fourth object number to obtain a forward statistical value; the forward statistical value is in a negative correlation relation with the fourth object number.

In some embodiments, the negative statistics comprise a first negative statistics, the classification accuracy comprises a first classification accuracy; the accuracy determination module is further to: counting the number of objects meeting a fifth category condition in the plurality of objects to obtain a fifth object number; the fifth category condition includes: the real category of the object in the first history period is the first preset category, the first prediction category of the object is the second preset category, the real category of the object in the second history period is the first preset category, and the second prediction category of the object is the second preset category; counting based on the first object number, the third object number and the fifth object number to obtain a first negative statistic value; the first negative statistic value and the fifth object number form a positive correlation; determining a first classification accuracy of the object classification model for the current period based on the positive statistics and the first negative statistics.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category condition includes: the real category of the object in the current period is the first preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the second preset category; counting based on the first object number, the third object number, the fourth object number and the fifth object number to obtain a first negative statistic value; the first negative statistic is negatively correlated with the fourth number of objects.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a sixth category condition in the plurality of objects to obtain a sixth object number; the sixth category of conditions includes: the real category of the object in the current period is the second preset category, the current prediction category of the object is the first preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the second preset category; counting based on the first object number, the third object number, the fourth object number, the fifth object number and the sixth object number to obtain a first negative statistic value; the first negative statistic is in negative correlation with the sixth number of objects.

In some embodiments, the negative statistics comprise a second negative statistics, the classification accuracy comprises a second classification accuracy; the accuracy determination module is further to: counting based on the first object quantity, the second object quantity and the third object quantity to obtain a second negative-going statistical value; the second negative statistic value and the second object number form a positive correlation; the determining, based on the positive statistics and the negative statistics, the classification accuracy of the object classification model in the current period comprises: determining a second classification accuracy of the object classification model for the current period based on the positive statistics and the second negative statistics.

In some embodiments, the apparatus further comprises: the first model determining module is used for determining the object classification model with the classification accuracy to be analyzed as the object classification model to be trained under the condition that the classification accuracy is smaller than an accuracy threshold; the second model determining module is used for training the object classification model to be trained to obtain a new object classification model with classification accuracy to be analyzed, and returning to the step of inputting the historical object features of each object into the object classification model for classification until the classification accuracy reaches an accuracy threshold; and the third model determining module is used for determining the object classification model with the classification accuracy to be analyzed under the condition that the classification accuracy reaches the accuracy threshold as the trained object classification model.

In some embodiments, the second model determination module is further to: inputting the object characteristics of the training object in a first historical period into the object classification model to be trained for classification to obtain the prediction category of the training object in the current period; and adjusting parameters of the object classification model to be trained on the basis of the difference between the prediction category of the training object in the current period and the real category of the training object in the current period to obtain a new object classification model with classification accuracy to be analyzed.

In some embodiments, the current period is a current promotion period for a target service, and the classification category of the trained object classification model is any one of retention or churn; the apparatus is further configured to: acquiring object characteristics of a plurality of candidate objects in the current popularization period; inputting the object characteristics of each candidate object in the current promotion period into the trained object classification model for classification to obtain the prediction category of each candidate object in the next promotion period of the target service; selecting a prediction category as a reserved target object from each candidate object; and pushing promotion content related to the target service to the target object in the next promotion period.

On the other hand, the application also provides computer equipment. The computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor realizes the steps of the object classification processing method when executing the computer program.

In another aspect, the present application also provides a computer-readable storage medium. The computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the above-mentioned object classification processing method.

In another aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of the object classification processing method described above.

The object classification processing method, the device, the computer equipment, the storage medium and the computer program product determine an object classification model of classification accuracy to be analyzed, acquire historical object characteristics of a plurality of objects in at least three target historical periods respectively, the at least three target historical periods are at least three continuous historical periods selected from a previous period of a current period, input the historical object characteristics of each object into the object classification model for classification aiming at the historical object characteristics of each object in each historical period so as to predict the prediction category of each object in a next period of the historical periods, and obtain the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods, wherein the at least two adjacent historical periods are target historical periods which are closer to the current period in the at least three target historical periods, and determining the classification accuracy of the object classification model in the current period based on the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods. Therefore, the classification accuracy is determined by combining the prediction categories of the object in a plurality of continuous periods (including the current period), and the accuracy of the classification accuracy is improved.

Drawings

FIG. 1 is a diagram of an application environment of a method for object classification processing in some embodiments;

FIG. 2 is a flow diagram of a method for object classification processing in some embodiments;

FIG. 3 is a schematic illustration of various cycles in some embodiments;

FIG. 4 is a schematic diagram of counting a second number of objects in some embodiments;

FIG. 5 is a diagram that illustrates pushing promotional content to a target object in some embodiments;

FIG. 6 is a flow diagram of a method for object classification processing in some embodiments;

FIG. 7 is a flow diagram of a method for object classification processing in some embodiments;

FIG. 8 is a block diagram of an object classification processing apparatus in some embodiments;

FIG. 9 is a diagram of the internal structure of a computer device in some embodiments;

FIG. 10 is a diagram of the internal structure of a computer device in some embodiments.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The object classification processing method provided by the embodiment of the application can be applied to the application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104, or may be placed on the cloud or other server. A variety of applications may be installed on the terminal 102, such as an instant messaging application, a third party payment application, a video viewing application, or a vehicle service application. An applet may be embedded in an application installed on the terminal 102, such as an instant messaging application or a third party payment application, and the applet includes, but is not limited to, at least one of a ride applet, a take-away applet, or a vehicle service applet.

Specifically, the server 104 may determine an object classification model of the classification accuracy to be analyzed, obtain historical object features of a plurality of objects in at least three target history periods respectively, where the at least three target history periods are at least three consecutive history periods selected from a previous period of a current period, for the historical object feature of each object in each history period, input the historical object feature of each object into the object classification model for classification, so as to predict a prediction category of each object in a next period of the history periods, obtain the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent history periods, where the at least two adjacent history periods are the target history periods closer to the current period among the at least three target history periods, based on the prediction category of each object in the current period and the prediction categories of each object in the at least two adjacent history periods, and determining the classification accuracy of the object classification model in the current period. The server 104 may determine the object classification model with the classification accuracy greater than the accuracy threshold as a trained object classification model, and predict a class to which the candidate object belongs in a cycle next to the current cycle by using the trained object classification model. The server 104 may transmit the category to which the candidate object belongs in the next cycle of the current cycle to the terminal 102, or the server 104 may determine the target object from the respective candidate objects based on the category to which the candidate object belongs in the next cycle of the current cycle, and push the corresponding content to the terminal of the target object, for example, the terminal 102, in the next cycle of the current cycle.

The terminal 102 may be, but not limited to, various desktop computers, notebook computers, smart phones, tablet computers, vehicle-mounted terminals, internet of things devices, and portable wearable devices, and the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart vehicle-mounted devices, and the like. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, and the like. The server 104 may be implemented as a stand-alone server or as a server cluster comprised of multiple servers.

The object classification processing method provided by the application can be applied to the map field, for example, an object classification model which is used for classifying users in the application software platform of the map class and has high classification accuracy can be obtained by using the object classification processing method provided by the application, so that the users in the application software platform of the map class are classified, and therefore more appropriate function services related to the map are provided for the users based on the classes of the users.

The object classification processing method provided by the application can be based on artificial intelligence, for example, the object classification model can be a neural network model. Among them, Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best result. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and the like.

Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

With the research and progress of artificial intelligence technology, the artificial intelligence technology is developed and researched in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, robots, smart medical services, smart customer service, internet of vehicles, automatic driving, smart traffic and the like.

The scheme provided by the embodiment of the application relates to the technologies such as artificial neural network of artificial intelligence, and is specifically explained by the following embodiment:

in some embodiments, as shown in fig. 2, an object classification processing method is provided, which may be executed by a terminal or a server, and may also be executed by both the terminal and the server, and is described by taking the application of the method to the server 104 in fig. 1 as an example, including the following steps:

step 202, determining an object classification model of the classification accuracy to be analyzed.

The object may be any one of living things including but not limited to at least one of natural human, animal or plant, or non-living things including but not limited to at least one of a table or a computer.

The object classification model is a model for classifying an object, i.e., for identifying a class of an object. The object classification model may be a neural network model, and the object classification model to be analyzed for classification accuracy may be trained or untrained. The object classification model may be a two-classification or multi-classification model, where multi-classification refers to at least three classifications, and the object classification model may be, for example, an LR (Logistic Regression) two-classification model. The logistic regression model transforms the linear regression model into a probabilistic prediction model by introducing a Sigmoid function into the linear regression model to map the continuous output values of the uncertain range of the linear regression into the range of (0, 1).

The object classification model may be matched to services provided by the application, and each service in the application may be matched with an object classification model. Services include, but are not limited to, services in a standalone application, which may also be referred to as a parent application, or services in a child application. The parent application program is an application program for bearing the child application program and provides a running environment for realizing the child application program. The parent application is a native application. A native application is an application that can run directly on an operating system. The child application can be run in the parent application. The child application runs in the running environment of the parent application. The parent application includes, but is not limited to, at least one of a social application, a dedicated application that specifically supports the child application, a file management application, a mail application, or a game application, etc.

The sub-application may be, for example, a travel servlet in instant messaging software. The travel service applet is an applet for providing a travel service including, but not limited to, at least one of a preferential fueling service, a car washing service, a car moving code service, or a designated driving service. Each service may have a matched object classification model, e.g., an offer service has a matched object classification model, and a car wash service also has a corresponding object classification model. The object may be a user registered in an application, for example, when the application is an application providing a travel service for an owner of a vehicle, the object may be the owner of the vehicle registered in the application.

The input of the object classification model is the object characteristics of the first period, and the output of the object classification model is the predicted class to which the object belongs in the second period. The first period and the second period are adjacent periods, and the first period is before the second period. The period may be, for example, a promotion period, where the promotion period is a period in which content is pushed to an object, and the promotion period may be, for example, a marketing period, and may also be other periods, which is not limited herein.

The service of the application program can have a plurality of marketing periods, for example, for the preferential fueling service, a plurality of marketing periods can be provided, and marketing is carried out in each marketing period. The object classification model may be a model of a current marketing cycle versus a category used to identify the object, and in particular, the object classification model may be used to predict a category to which the object belongs in a marketing cycle next to the current marketing cycle based on object characteristics of the object in the current marketing cycle. For example, for the current marketing period of the preferential fueling service, an object classification model matched with the current marketing period of the preferential fueling service can be trained, and the class to which the next marketing period of the object belongs in the current marketing period is identified by using the trained object classification model.

The object characteristics are used for reflecting the characteristics of the object, and the characteristics of the same object in different marketing periods can be different or the same. The object characteristics include at least one of object attribute characteristics or object interaction characteristics. The object attribute feature is a feature obtained based on object attribute data, the object attribute data includes at least one of gender, age, or region, and the object attribute feature includes at least one of gender feature, age feature, or region feature. The object interaction attribute feature is a feature obtained based on object interaction data, the object interaction data refers to data generated by an object in an application program, and the object interaction data may include at least one of active attribute data, recharge attribute data, coupon attribute data, and the like. The activity attribute data includes, but is not limited to, at least one of an active number of days, an active duration, a number of active functions, a number of days interval of a registration time from a current time, and the like. The load attribute data includes, but is not limited to, at least one of a load amount, a consumption amount, a number of loads days, a first load interval from a current time, and the like. The coupon attribute data includes, but is not limited to, at least one of a function click, coupon pickup information, coupon usage information, or coupon expiration information. The coupon pickup information includes at least one of a type, a number, or a value of the picked-up coupon, the coupon use information includes at least one of a type, a number, or a value of the used coupon, and the coupon expiration information includes at least one of a type, a number, or a value of the expired coupon. The coupon may include at least one of a gift bag or a gift certificate.

The application program may include a function model corresponding to each service, where the function model is used to provide a corresponding service, and for example, the function module corresponding to the preferential fueling service is a preferential fueling function module. The class of the object may be determined according to the interaction between the object and the function module. Taking the preferential fueling service as an example, and the marketing period of the preferential fueling service includes T-3, T-2, T-1 and T, where T-3, T-2, T-1 and T are 4 consecutive marketing periods, and T is the current period, as shown in fig. 3, a schematic diagram of each period is shown. In the preferential fueling service scenario, the categories for which the object classification model is used to identify can be set to include retention and churn. If the object logs in the preferential fueling function module in the T-1 period and does not log in the preferential fueling function module in the T period, the category of the object in the T period is lost. If the object logs in the preferential fueling function module in the T-1 period and also logs in the preferential fueling function module in the T period, the category of the object in the T period is retention. If the object logs in the preferential fueling function module in the T-2 period and does not log in the preferential fueling function module in the T-1 period, the category of the object in the T-1 period is lost. If the subject logs in the preferential fueling function module in the T-2 period and also logs in the preferential fueling function module in the T-1 period, the category of the subject in the T-1 period is the retention. If the object logs in the preferential fueling function module in the T-3 period and does not log in the preferential fueling function module in the T-2 period, the category of the object in the T-2 period is lost. If the object logs in the preferential fueling function module in the T-3 period and also logs in the preferential fueling function module in the T-2 period, the category of the object in the T-2 period is the retention.

Specifically, the server may obtain an object classification model to be trained, where the object classification model to be trained may be a trained model or an untrained model. The server can train the object classification model to be trained, and determine the trained object classification model as the object classification model with the classification accuracy to be analyzed.

In some embodiments, the server may obtain one or more object features of the object in the first history period, where a plurality refers to at least two, input the object features in the first history period into an object classification model to be trained, predict a class to which the object belongs in the current period, adjust parameters of the object classification model based on a difference between the predicted class of each object and a true class of the object in the first history period, and determine the object classification model with the adjusted parameters as the object classification model with classification accuracy to be analyzed. The first history period and the current period are 2 continuous periods, and the first history period is before the current period.

Step 204, acquiring historical object characteristics of a plurality of objects in at least three target historical periods respectively; the at least three target history cycles are at least three consecutive history cycles selected from a previous cycle of the current cycle.

The at least three target history cycles are at least three continuous history cycles selected from a previous cycle of the current cycle, the at least three target history cycles include a previous cycle of the current cycle, and the previous cycle of the current cycle refers to a cycle adjacent to and before the current cycle. For example, T-3, T-2, T-1 and T are 4 consecutive periods, and if the current period is T, T-3, T-2 and T-1 belong to the at least three target history periods. The history object feature is an object feature that an object has in a history period.

Specifically, the server may obtain the history object characteristics of the objects in each of three target history periods, where the three target history periods are three consecutive history periods selected from a previous period of a current period, for example, the current period is T, and the three target history periods are T-3, T-2, and T-1 in sequence, where T-1 is the previous period of the current period.

In some embodiments, the three target history periods are sequentially a third history period, a second history period and a first history period, the third history period is before and adjacent to the second history period, the second history period is before and adjacent to the first history period, the first history period is before and adjacent to the current period, for example, the current period is T, the third history period is T-3, the second history period is T-2, and the first history period is T-1. The server can obtain the object characteristics of each object in the first history period, the object characteristics in the second history period and the object characteristics in the third history period to obtain the respective history object characteristics of each object.

Step 206, inputting the historical object features of each object into an object classification model for classification according to the historical object features of each object in each target historical period so as to predict the prediction category of each object in the next period of the target historical period, and obtain the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods; the at least two adjacent history cycles are target history cycles closer to the current cycle of the at least three target history cycles.

The prediction category is a category identified by the object classification model. The at least two adjacent history cycles are the target history cycles closer to the current cycle in the at least three target history cycles. For example, the current period is T, the at least three target history periods are three target history periods, the three target history periods are T-3, T-2 and T-1, respectively, the at least two adjacent history periods are two adjacent history periods, and the two adjacent history periods are T-2 and T-1, respectively, because T-2 and T-1 are closer to T in T-3, T-2 and T-1.

Specifically, the at least three target history cycles include a third history cycle, a second history cycle, and a first history cycle, and the at least two adjacent history cycles include the second history cycle and the first history cycle. The server can determine the object characteristics of the object in the first history period as the first history object characteristics of the object, determine the object characteristics of the object in the second history period as the second history object characteristics of the object, and determine the object characteristics of the object in the third history period as the third history object characteristics of the object. The server can input the first historical object features of the object into the object classification model to identify the class, predict the class to which the object belongs in the current period, and determine the predicted class to which the object belongs in the current period as the current prediction class. The server may input the second historical object feature of the object into the object classification model to identify the class, predict the class to which the object belongs in the first historical period, and determine the predicted class to which the object belongs in the first historical period as the first prediction class. The server may input the third history object feature of the object into the object classification model to perform class identification, obtain a class to which the object belongs in the second history period, and determine the predicted class to which the object belongs in the second history period as the second prediction class. Since the first history cycle and the second history cycle belong to adjacent history cycles, the first prediction category and the second prediction category belong to prediction categories adjacent to the history cycles.

And step 208, determining the classification accuracy of the object classification model in the current period based on the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods.

The classification accuracy is used for reflecting the accuracy of the classification of the object classification model, namely the classification accuracy is used for reflecting the accuracy of the predicted class, and the higher the classification accuracy is, the higher the accuracy of the class predicted by the object classification model is.

The classification accuracy in the current period is used for reflecting the accuracy of the class to which the object predicted by the object classification model belongs in the current period. The greater the classification accuracy in the current cycle, the higher the accuracy of the class to which the object classification model predicts the object belongs in the current cycle.

Specifically, the classification accuracy of the object classification model at the current cycle is affected by at least one of a predicted class or a true class of the object at the neighboring historical cycle. The server may determine a classification accuracy of the object classification model at the current cycle based on the current prediction class, the first prediction class, and the second prediction class of each object.

In some embodiments, for each object, the server may obtain a real category to which the object belongs in the current period, a real category to which the object belongs in the first history period, and a real category to which the object belongs in the second history period, and obtain the classification accuracy of the object classification model in the current period based on the current prediction category, the first prediction category, the second prediction category, and the number of the objects counted in each real category of each object.

In some embodiments, after obtaining the classification accuracy, the server may adjust parameters of the object classification model based on the classification accuracy, classify the object based on the object classification model after the parameter adjustment, for example, may continuously adjust the parameters of the object classification model until the classification accuracy of the object classification model is greater than or equal to the accuracy threshold, classify the object using the object classification model whose classification accuracy is greater than or equal to the accuracy threshold, for example, input the object features of the object in the current cycle into the object classification model, and predict the class to which the object belongs in the next cycle of the current cycle. The accuracy threshold may be preset, for example may be 90%.

In some embodiments, for a plurality of services provided in an application, for example, an offer service, a target object classification model may be generated for each service in the plurality of services by using the steps in the above embodiments, where the target object classification model refers to an object classification model with an accuracy greater than or equal to an accuracy threshold, and the server may store target object classification models corresponding to the respective services. The terminal may send a content promotion request for a specific object and a specific service in a specific application to a server, where a target object classification model corresponding to a feature service of the specific application is stored in the server, and the server may obtain, in response to the content promotion request, an object feature of the specific object in a current period, input the object feature of the specific object in the current period into the target object classification model for classification, obtain a category to which the specific object belongs in a next period of the current period, determine, based on the obtained category, content to be pushed to the specific object, and push the determined content to the specific object.

The object classification processing method comprises the steps of determining an object classification model of classification accuracy to be analyzed, obtaining historical object characteristics of a plurality of objects in at least three target historical periods respectively, wherein the at least three target historical periods are at least three continuous historical periods selected from a previous period of a current period, inputting the historical object characteristics of each object in each historical period into the object classification model for classification aiming at the historical object characteristics of each object in each historical period so as to predict the prediction category of each object in a next period of the historical periods, obtaining the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods, wherein the at least two adjacent historical periods are the target historical periods which are closer to the current period in the at least three target historical periods, and based on the prediction categories of each object in the current period and the prediction categories of each object in the at least two adjacent historical periods, and determining the classification accuracy of the object classification model in the current period. Therefore, the classification accuracy is determined by combining the prediction categories of the object in a plurality of continuous periods (including the current period), and the accuracy of the classification accuracy is improved.

In some embodiments, the at least two adjacent history cycles include a first history cycle adjacent to and prior to the current cycle and a second history cycle adjacent to and prior to the first history cycle; determining the classification accuracy of the object classification model in the current cycle based on the predicted class of each object in the current cycle and the predicted classes of each object in at least two adjacent historical cycles comprises: obtaining the prediction type of each object in the current period to obtain the current prediction type of each object; obtaining the prediction category of each object in a first history period to obtain the first prediction category of each object; obtaining the prediction category of each object in a second history period to obtain a second prediction category of each object; and determining the classification accuracy of the object classification model in the current period based on the current prediction category, the first prediction category and the second prediction category of each object.

The current prediction type of the object refers to the type to which the predicted object belongs in the current period. The first prediction type of the object refers to a type to which the predicted object belongs in the first history period, and the second prediction type of the object refers to a type to which the predicted object belongs in the second history period.

Specifically, the server may obtain the real category of each object in each target history period, count the number of objects based on the real category of each object and the predicted category, and determine the classification accuracy of the object classification model in the current period based on the counted number. The real category includes a first preset category and a second preset category, where the first preset category is different from the second preset category, for example, the first preset category is loss, and the second preset category is retention. The first preset category may be represented by 1, the second preset category may be represented by 0, that is, 1 may represent run-off, and 0 may represent retention.

In this embodiment, since the classification accuracy of the object classification model in the current period may be affected by the category to which the object in the history period adjacent to the current period belongs, the classification accuracy of the object classification model in the current period is determined based on the current prediction category, the first prediction category, and the second prediction category of each object, so that the accuracy of the classification accuracy of the object classification model in the current period is improved.

In some embodiments, determining the classification accuracy of the object classification model at the current cycle based on the current prediction class, the first prediction class, and the second prediction class of the respective object comprises: counting the number of objects meeting a first class condition in the plurality of objects to obtain a first object number; the first category conditions include: the real category of the object in the current period is a first preset category, and the current prediction category of the object is a first preset category; counting the number of objects meeting the second category condition in the plurality of objects to obtain a second object number; the second category of conditions includes: the real category of the object in the second history period is a first preset category, the second prediction category of the object is a second preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a first preset category; and determining the classification accuracy of the object classification model in the current period based on the first object number and the second object number.

Specifically, the server may perform statistical calculation based on the first number of objects and the second number of objects to obtain the classification accuracy of the object classification model in the current period. The statistical calculation includes at least one of addition, subtraction, multiplication, division, or the like.

In some embodiments, the server may select from the plurality of objects that satisfies: the real type of the object in the second history period is a first preset type, and the second prediction type of the object is a second preset type, so as to form a first object group, and the objects are selected from the plurality of objects, so that the following conditions are met: and forming a second object group by using the objects with the conditions that the real class of the objects in the first history period is a first preset class and the first prediction class of the objects is the first preset class, and counting the number of the same objects in the first object group and the second object group to obtain the number of the second objects. To explain with the first preset category as 1 and the second preset category as 0, as shown in fig. 4, the area a is each object whose real category is 1 and second prediction category is 0 in the second history period, that is, the area a is the first object group, the area B is each object whose real category is 1 and second prediction category is 1 in the first history period, that is, the area B is the second object group, and the area C is each object satisfying the second category condition, that is, the number of objects in the area C is the second object number.

In some embodiments, the server may count a number of objects satisfying a third category condition in the plurality of objects to obtain a third number of objects, where the third category condition includes: the real category of the object in the second history period is a first preset category, the second prediction category of the object is a second preset category, the real category of the object in the first history period is a second preset category, and the first prediction category of the object is a first preset category. The server can perform statistical calculation based on the first object number, the second object number and the third object number to obtain the classification accuracy of the object classification model in the current period. The statistical calculation includes at least one of addition, subtraction, multiplication, division, or the like.

In this embodiment, since the process of counting the number of the second objects is related to the categories of the objects in the first history period and the second history period, the classification accuracy of the object classification model in the current period is determined based on the number of the first objects and the number of the second objects, so that the influence of the categories of the objects in the first history period and the second history period is reflected in the classification accuracy, and the accuracy of the classification accuracy is improved.

In some embodiments, determining the classification accuracy of the object classification model at the current cycle based on the first number of objects and the second number of objects comprises: counting the number of objects meeting a third category condition in the plurality of objects to obtain a third object number; the third category of conditions includes: the real category of the object in the second history period is a first preset category, the second prediction category of the object is a second preset category, the real category of the object in the first history period is a second preset category, and the first prediction category of the object is a first preset category; counting based on the first object quantity and the second object quantity to obtain a forward statistical value; the forward statistical value is in positive correlation with the first object quantity and the second object quantity; counting based on the first object quantity and the third object quantity to obtain a negative statistical value; the negative statistic value is in positive correlation with the first object quantity and the third object quantity; determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics; the classification accuracy and the positive statistic value form a positive correlation relationship, and the classification accuracy and the negative statistic value form a negative correlation relationship.

Wherein, the forward statistical value is in positive correlation with the first object quantity and the second object quantity. The negative statistical value is in positive correlation with the number of the first objects and the number of the third objects. The classification accuracy and the positive statistic value form a positive correlation relationship, and the classification accuracy and the negative statistic value form a negative correlation relationship.

The positive correlation refers to: under the condition that other conditions are not changed, the changing directions of the two variables are the same, and when one variable changes from large to small, the other variable also changes from large to small. It is understood that a positive correlation herein means that the direction of change is consistent, but does not require that when one variable changes at all, another variable must also change. For example, it may be set that the variable b is 100 when the variable a is 10 to 20, and the variable b is 120 when the variable a is 20 to 30. Thus, the change directions of a and b are both such that when a is larger, b is also larger. But b may be unchanged in the range of 10 to 20 a. The negative correlation relationship refers to: under the condition that other conditions are not changed, the changing directions of the two variables are opposite, and when one variable is changed from large to small, the other variable is changed from small to large. It is understood that the negative correlation herein means that the direction of change is reversed, but it is not required that when one variable changes at all, the other variable must also change.

Specifically, the server may select from the plurality of objects that satisfies: and forming a third object group by the objects under the condition that the real type of the objects in the first history period is a second preset type and the first prediction type of the objects is a first preset type, and counting the number of the same objects in the first object group and the third object group to obtain the number of the third objects.

In some embodiments, the server may count a number of objects satisfying a seventh category condition among the plurality of objects, to obtain a seventh number of objects, where the seventh category condition includes: the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a second preset category. The server may perform statistics based on the first number of objects, the second number of objects, and the seventh number of objects, to obtain a forward statistical value. The forward statistical value and the seventh object number have positive correlation. The server may perform statistics based on the first number of objects, the third number of objects, and the seventh number of objects, to obtain a negative statistic value. The negative statistic value and the number of the seventh objects form a positive correlation relationship.

In some embodiments, the classification accuracy is positively correlated with positive statistics and the classification accuracy is negatively correlated with negative statistics. The server can calculate the ratio of the positive statistics value to the negative statistics value, and determine the classification accuracy of the object classification model in the current period according to the calculated ratio.

In this embodiment, since the process of counting the number of the third objects is related to the categories of the objects in the first history period and the second history period, the classification accuracy of the object classification model in the current period is determined based on the negative statistical value obtained by counting the number of the first objects and the number of the third objects and the positive statistical value and the negative statistical value, the influence of the categories of the objects in the first history period and the second history period is reflected, and the accuracy of the classification accuracy is improved.

In some embodiments, performing statistics based on the first number of objects and the second number of objects, and obtaining the forward statistical value comprises: counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category of conditions includes: the real category of the object in the current period is a first preset category, the real category of the object in the first historical period is a first preset category, and the first prediction category of the object is a second preset category; counting based on the first object number, the second object number and the fourth object number to obtain a forward statistical value; the forward statistical value is in a negative correlation with the fourth number of objects.

Specifically, the server may select from the plurality of objects that satisfies: and forming a fourth object group by the objects under the condition that the real class of the objects in the first history period is a first preset class and the first prediction class of the objects is a second preset class. The server may select an object whose real category is the first preset category in the current period from the plurality of objects to form a fifth object group, and the server may count the number of the same objects in the fourth object group and the fifth object group to obtain the number of the fourth objects. Wherein the server may select from the plurality of objects that satisfies: the real type of the object in the current period is a first preset type, and the current prediction type of the object is the object of the condition of the first preset type, so as to form a first sub-object group, and the objects are selected from the plurality of objects to satisfy the following conditions: and forming a second sub-object group by using the objects under the condition that the real class of the objects in the current period is a first preset class and the current prediction class of the objects is a second preset class, and combining the first sub-object group and the second sub-object group to obtain a fifth object group.

In some embodiments, the server may count a number of objects satisfying an eighth category condition in the plurality of objects, to obtain an eighth number of objects, where the eighth category condition includes: the real category of the object in the current period is a first preset category, the current prediction category of the object is a first preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a second preset category. The server may count the number of objects satisfying a ninth category condition in the plurality of objects, to obtain a ninth object number, where the ninth category condition includes: the real category of the object in the current period is a first preset category, the current prediction category of the object is a second preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a second preset category. The server may sum the eighth number of objects with the ninth number of objects to obtain a fourth number of objects.

In some embodiments, the server may perform statistics based on the first number of objects, the second number of objects, the fourth number of objects, and the seventh number of objects, resulting in a forward statistical value. Specifically, the server may sum the first number of objects, the seventh number of objects, and the second number of objects, and perform difference calculation between the sum result and the fourth number of objects to obtain a forward statistical value. For example, the forward statistical value S1 is calculated by: s1 ═ TP_t+FN_t-1)-(TP_t|t-1+FN_t|t-1)+TP_t-1|t-2. Wherein S1 represents a forward statistical value, TP_tIndicating the number of first objects, FN_t-1Indicates the seventh number of objects, (TP)_t|t-1+FN_t|t-1) Denotes the fourth number of objects, TP_t-1|t-2Representing the number of second objects, TP_t|t-1Indicates the eighth number of objects, FN_t|t-1Indicating the ninth number of objects.

In this embodiment, since the process of counting the number of the fourth objects is related to the category of the objects in the first history period, the forward statistical value is obtained by counting based on the number of the first objects, the number of the second objects, and the number of the fourth objects, and the accuracy of the classification accuracy obtained based on the forward statistical value is improved.

In some embodiments, the negative statistics comprise a first negative statistics, and the classification accuracy comprises a first classification accuracy; counting based on the first object number and the third object number to obtain a negative statistic value, wherein the counting comprises the following steps: counting the number of objects meeting a fifth category condition in the plurality of objects to obtain a fifth object number; the fifth category of conditions includes: the real category of the object in the first history period is a first preset category, the first prediction category of the object is a second preset category, the real category of the object in the second history period is the first preset category, and the second prediction category of the object is the second preset category; counting based on the first object number, the third object number and the fifth object number to obtain a first negative statistic value; the first negative statistic value and the number of the fifth objects form a positive correlation; determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics comprises: and determining the first classification accuracy of the object classification model in the current period based on the positive-going statistic and the first negative-going statistic.

Specifically, the negative statistics may include a first negative statistics, the classification accuracy includes a first classification accuracy, and the server may determine the first classification accuracy based on the positive statistics and the first negative statistics, the first classification accuracy being in a positive correlation with the positive statistics, and the first classification accuracy being in a negative correlation with the first negative statistics. For example, the server may calculate a ratio of the positive statistics to the first negative statistics, and determine the calculated ratio as the first classification accuracy. The server may perform a summation operation based on the first number of objects, the third number of objects, and the fifth number of objects to obtain a first negative statistic.

In some embodiments, the server may perform a statistical operation based on the first number of objects, the third number of objects, the fifth number of objects, and the seventh number of objects to obtain a first negative statistical value. The first negative statistic is positively correlated with the number of seventh objects. Specifically, the server may perform a summation operation on the first number of objects, the third number of objects, the fifth number of objects, and the seventh number of objects to obtain a first negative statistical value.

In some embodiments, the server may count the plurality of objects that satisfy: and obtaining the tenth object number according to the number of the objects under the condition that the real class of the objects in the first history period is the second preset class and the first prediction class of the objects is the first preset class. The server may perform summation operation on the first number of objects, the third number of objects, the fifth number of objects, the seventh number of objects, and the tenth number of objects to obtain the first negative statistical value.

In some embodiments, the server may count the plurality of objects that satisfy: and the real category of the object in the current period is a second preset category, and the current prediction category of the object is a first preset category, so as to obtain the quantity of the eleventh objects. The server may perform summation operation on the first number of objects, the third number of objects, the fifth number of objects, the seventh number of objects, the tenth number of objects, and the eleventh number of objects to obtain the first negative statistical value.

In this embodiment, since the process of counting the number of the fifth objects is related to the categories of the objects in the first history period and the second history period, the first negative statistical value is obtained by counting based on the number of the first objects, the number of the third objects, and the number of the fifth objects, and the accuracy of the first classification accuracy obtained based on the first negative statistical value is improved.

In some embodiments, counting is performed based on the first number of objects, the third number of objects, and the fifth number of objects, to obtain a first negative-going statistic value; the positively correlating the first negative statistics with the fifth number of objects comprises: counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category of conditions includes: the real type of the object in the current period is a first preset type, the real type of the object in the first historical period is a first preset type, and the first prediction type of the object is a second preset type; counting based on the first object number, the third object number, the fourth object number and the fifth object number to obtain a first negative statistic value; the first negative statistic is negatively correlated with the number of fourth objects.

Specifically, the server may perform summation operation on the first object number, the third object number, and the fifth object number, and perform difference calculation on a summation result and the fourth object number to obtain a first negative statistical value.

In some embodiments, the server may perform a summation operation on the first number of objects, the third number of objects, the fifth number of objects, the seventh number of objects, the tenth number of objects, and the eleventh number of objects, and perform a difference calculation on a result of the summation and the fourth number of objects to obtain a first negative statistical value.

In this embodiment, since the process of counting the number of the fourth objects is related to the category of the objects in the first history period, the first negative statistical value is obtained by counting based on the number of the first objects, the number of the third objects, the number of the fourth objects, and the number of the fifth objects, and the accuracy of the first classification accuracy obtained based on the first negative statistical value is improved.

In some embodiments, performing statistics based on the first number of objects, the third number of objects, the fourth number of objects, and the fifth number of objects, and obtaining a first negative statistic comprises: counting the number of objects meeting the sixth category condition in the plurality of objects to obtain the number of sixth objects; the sixth category of conditions includes: the real category of the object in the current period is a second preset category, the current prediction category of the object is a first preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a second preset category; counting based on the first object number, the third object number, the fourth object number, the fifth object number and the sixth object number to obtain a first negative statistic value; the first negative statistic is negatively correlated with the sixth number of objects.

Specifically, the server may perform a summation operation based on the first number of objects, the third number of objects, and the fifth number of objects, to obtain a first summation result. For example, the server may perform a summation operation on the first number of objects, the third number of objects, the fifth number of objects, the seventh number of objects, the tenth number of objects, and the eleventh number of objects to obtain a first summation result. The server may perform summation operation on the fourth object number and the sixth object number to obtain a second summation result, and perform difference calculation on the first summation result and the second summation result to obtain a first negative statistical value. For example, the first negative statistical value S2 is calculated by the formula: s2 ═ (TP)_t+FP_t+FN_t-1+FP_t-1)-(TP_t|t-1+FN_t|t-1+FP_t|t-1)+(FN_t-1|t-2+FP_t-1|t-2). Wherein S2 represents the first negative statistic, FP_tIndicates the eleventh number of objects, FP_t-1Indicates the tenth number of objects, FN_t-1|t-2Indicates the fifth number of objects, FP_t-1|t-2Representing the number of third objects, FP_t|t-1Indicating a sixth number of objects.

In this embodiment, since the process of counting the number of the sixth objects is related to the category of the objects in the first history period, the first negative statistical value is obtained by counting based on the first number of the objects, the third number of the objects, the fourth number of the objects, the fifth number of the objects, and the sixth number of the objects, and the accuracy of the first classification accuracy obtained based on the first negative statistical value is improved.

In some embodiments, the negative statistics comprise a second negative statistics, and the classification accuracy comprises a second classification accuracy; counting based on the first object number and the third object number to obtain a negative statistic value, wherein the counting comprises the following steps: counting based on the first object quantity, the second object quantity and the third object quantity to obtain a second negative-going statistical value; the second negative statistic value and the second object quantity form a positive correlation; determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics comprises: and determining a second classification accuracy of the object classification model in the current period based on the positive-going statistic and the second negative-going statistic.

Specifically, the second classification accuracy is negatively correlated with the second negative statistical value. The second negative statistics value and the second object number form a positive correlation, and the server may perform statistics based on the first object number, the second object number, and the third object number to obtain a second negative statistics value. For example, the server may perform a summation calculation based on the first number of objects, the second number of objects, and the third number of objects, resulting in a second negative statistic.

In some embodiments, the negative statistics include a second negative statistics, and the second negative statistics is in a negative correlation with the fourth number of objects, and the server may perform statistics based on the first number of objects, the second number of objects, the third number of objects, and the fourth number of objects to obtain the second negative statistics. Specifically, the server may perform summation calculation on the first object number, the second object number, and the third object number, and perform difference calculation on the summation result and the fourth object number to obtain a second negative statistical value.

In some embodiments, the second negative statistical value is in a negative correlation with the sixth number of objects, and the server may perform statistics based on the first number of objects, the second number of objects, the third number of objects, the fourth number of objects, and the sixth number of objects to obtain the second negative statistical value. Specifically, the server may perform summation calculation on the first object number, the second object number, and the third object number, perform difference calculation on the summation result and the fourth object number, and perform difference calculation on the difference calculation result and the sixth object number to obtain a second negative statistical value.

In some embodiments, the second negative statistics are positively correlated with the seventh number of objects. The server may perform statistics based on the first number of objects, the second number of objects, the third number of objects, the fourth number of objects, the sixth number of objects, and the seventh number of objects to obtain a second negative statistic value. Specifically, the server may perform summation calculation on the first object number, the second object number, the third object number, and the seventh object number, perform difference calculation on the summation result and the fourth object number, and perform difference calculation on the difference calculation result and the sixth object number to obtain a second negative statistical value.

In some embodiments, the second negative statistics are positively correlated with the tenth number of objects. The server may perform summation calculation on the first object number, the second object number, the third object number, the seventh object number, and the tenth object number, perform difference calculation on a result of the summation and the fourth object number, and perform difference calculation on a result of the difference calculation and the sixth object number to obtain a second negative statistical value.

In some embodiments, the server may count the plurality of objects that satisfy: the real class of the object in the current period is a first preset class, and the current prediction class of the object is a second preset classThe number of objects of other conditions is the twelfth number of objects. The second negative statistic has a positive correlation with the twelfth number of objects. The server may sum up the first object number, the second object number, the third object number, the seventh object number, the tenth object number, and the twelfth object number, perform difference calculation on the result of the sum and the fourth object number, and perform difference calculation on the result of the difference calculation and the sixth object number to obtain a second negative statistical value. For example, the second negative statistical value S3 is calculated by the formula: s3 ═ (TP)_t+FN_t+FN_t-1+FP_t-1)-(TP_t|t-1+FN_t|t-1+FP_t|t-1)+(TP_t-1|t-2+FP_t-1|t-2). Wherein S3 represents the second negative statistic, FN_tRepresenting the twelfth object quantity, FP_t-1Indicates the tenth object number, FP_t|t-1Indicates the sixth number of objects, FP_t-1|t-2Representing a third number of objects.

In some embodiments, the server may calculate a ratio of the positive statistics to the first negative statistics to obtain a first classification accuracy, e.g., a first classification accuracy P_t|t-1,t-2Is formula (1):

wherein, P_t|t-1,t-2Representing the first classification accuracy, P in equation (1)_t|t-1,t-2In order to reflect the proportion of the real objects with the first preset category in the prediction category under the influence of the first history period and the second history period in the current period, P in the formula (1) is used_t|t-1,t-2This may also be referred to as precision, i.e. the first classification accuracy may be precision, which is the proportion of samples with positive true classification in samples predicted to be positive. The prediction is that the prediction category is the first preset category.

In some embodiments, the server may calculate a ratio of the positive statistics to a second negative statistics to a second classification accuracyE.g. a second classification accuracy R_t|t-1,t-2Is formula (2):

wherein R is_t|t-1,t-2R in equation (2) representing the second classification accuracy_t|t-1,t-2Reflecting the proportion of the objects with the predicted category of the first preset category in the objects with the real category of the first preset category in the current period under the influence of the first history period and the second history period, so that R in the formula (2)_t|t-1,t-2It may also be referred to as a recall ratio, i.e., the second classification accuracy may be a recall ratio, which is a ratio of samples successfully predicted by the model among the true positive samples. True is exactly that the true category is the first preset category.

To illustrate the meaning of each parameter in the formula (1) and the formula (2), if the first preset category is an attrition label, the attrition label is represented by "1", the second preset category is a retention label, the retention label is represented by "0", and the label refers to a category, then:

TP_t(number of first objects): the statistical number of the intersection of the target set predicted to be "1" in the T period and the target set actually predicted to be "1" in the T period, that is, the TP_tThe number of objects of which the real category is "1" in the T period and the prediction category is "1" in the T period among the respective objects;

FN_t(twelfth object number): the statistical number of the intersection of the object set actually having "1" in the T period and the object set predicted to have "0" in the T period, that is, FN_tThe number of objects of which the true category in the T period is "1" and the predicted category in the T period is "0", among the respective objects;

FP_t(eleventh number of objects): the statistical number after the intersection of the object set with the T-stage actual "0" and the object set with the T-stage predicted "1", namely FP_tThe number of objects of which the true category in the T period is "0" and the predicted category in the T period is "1", among the respective objects;

FN_t-1(seventh object number): the statistical number of the intersection of the object set actually being '1' in the T-1 stage and the object set predicted to be '0' in the T-1 stage, namely FN_t-1The number of objects of which the real category in the T-1 period is "1" and the prediction category in the T-1 period is "0" among the objects;

FP_t-1(tenth object number): the statistical number of the intersection of the object set actually being '0' in the T-1 stage and the object set predicted to be '0' in the T-1 stage, namely FP_t-1The number of objects of which the true class is "0" in the T-1 period and the predicted class is "1" in the T-1 period, among the respective objects;

TP_t|t-1(eighth object number): the statistical number of the intersection between the T-1 actual 1 and the target set predicted to be 0 and the intersection between the T-1 actual 1 and the target set predicted to be 1, namely the TP_t|t-1The number of objects which simultaneously satisfy 4 conditions of a real class of "1" in a T-1 period, a prediction class of "0" in the T-1 period, a real class of 1 in the T period, and a prediction class of "1" in the T period, among the objects;

TP_t-1|t-2(second object number): the statistical number of the intersection between the T-2 actual 1 and the T-2 predicted 0 object set and the intersection between the T-1 actual 1 and the T-2 predicted 1 object set, namely TP_t-1|t-2The number of objects which simultaneously satisfy 4 conditions of real class of "1" in T-2 period, prediction class of "0" in T-2 period, real class of "1" in T-1 period, and prediction class of "1" in T-1 period, among the objects;

FN_t|t-1(ninth object number): the statistical number of the intersection of the target set which is actually '1' in the T-1 period and is predicted to be '0' in the T-1 period and the target set which is actually '1' and is predicted to be '0' in the T-1 period, namely FN_t|t-1For each object, the true category in T-1 period is "1", the prediction category in T-1 period is "0", and the true category in T period is "1 ", the number of objects of 4 conditions that the prediction category in the T period is" 0 ";

FN_t-1|t-2(fifth object number): the statistical number of the intersection between the T-2 actual 1 and the T-2 predicted 0 object set and the intersection between the T-1 actual 1 and the T-2 predicted 0 object set, that is, FN_t-1|t-2The number of objects which simultaneously satisfy 4 conditions of real class of "1" in T-2 period, prediction class of "0" in T-2 period, real class of "1" in T-1 period, and prediction class of "0" in T-1 period, among the objects;

FP_t|t-1(sixth object number): the statistical number of the intersection of the T-1 actual time of "1" and the T-1 predicted object set of "0", and the intersection of the T-1 actual time of "0" and the T-1 predicted object set of "1", that is, FP_t|t-1The number of objects satisfying 4 conditions of a true category of "1" in a T-1 period, a prediction category of "0" in the T-1 period, a true category of "0" in the T period, and a prediction category of "1" in the T period at the same time among the objects.

FP_t-1|t-2(number of third objects): the statistical number of the intersection between the T-2 actual 1 and the T-2 predicted 0 object set and the intersection between the T-1 actual 0 and the T-2 predicted 1 object set, namely FP_t-1|t-2The number of objects satisfying 4 conditions of a true category of "1" in a T-2 period, a prediction category of "0" in a T-2 period, a true category of "0" in a T-1 period, and a prediction category of "1" in a T-1 period at the same time among the objects.

In this embodiment, the second classification accuracy of the object classification model in the current period is determined based on the positive statistics and the second negative statistics, and since the process of obtaining the second negative statistics relates to the categories of the objects in the first history period and the second history period, the influence of the categories of the objects in the first history period and the second history period on the second classification accuracy is reflected, and the accuracy of the second classification accuracy is improved.

In some embodiments, the method further comprises: determining an object classification model with classification accuracy to be analyzed as an object classification model to be trained under the condition that the classification accuracy is smaller than an accuracy threshold; training the object classification model to be trained to obtain a new object classification model with classification accuracy to be analyzed, and returning to the step of inputting the historical object characteristics of each object into the object classification model for classification until the classification accuracy reaches an accuracy threshold; and determining the object classification model with the classification accuracy reaching the classification accuracy to be analyzed under the condition of an accuracy threshold value as a trained object classification model.

Wherein the accuracy threshold may be preset or set as desired, for example 85% or 90%. The object classification model to be trained is a model that needs to be trained or further trained.

Specifically, after obtaining the classification accuracy of the object classification model in the current cycle, the server may compare the classification accuracy with an accuracy threshold, and when it is determined that the classification accuracy is less than the accuracy threshold, perform further training on the object classification model, for example, the object classification model with the classification accuracy to be analyzed may be determined as the object classification model to be trained, the object classification model to be trained is trained, after training, the historical object features of each object in each historical cycle are returned, and the historical object features of each object are input into the object classification model for classification until the classification accuracy is equal to or greater than the accuracy threshold. And when the classification accuracy is equal to or greater than the accuracy threshold, determining the object classification model as a trained object classification model. Wherein, the classification accuracy may include a first classification accuracy and a second classification accuracy, the classification accuracy being smaller than the accuracy threshold, meaning that at least one of the first classification accuracy and the second classification accuracy is smaller than the accuracy threshold. For example, if the first classification accuracy is precision and the second classification accuracy is recall, the classification accuracy is less than the accuracy threshold, which means that at least one of precision and recall is less than the accuracy threshold.

In some embodiments, after obtaining the trained object classification model, the server may predict the class of the object in the next cycle of the current cycle by using the trained object classification model. For example, the object features of the object in the current period may be obtained, the object features in the current period are input into a trained object classification model, and the class to which the object belongs in the next period of the current period is predicted.

In this embodiment, when the classification accuracy is smaller than the accuracy threshold, the object classification model with the classification accuracy to be analyzed is determined as the object classification model to be trained, the object classification model to be trained is trained to obtain a new object classification model with the classification accuracy to be analyzed, and the object classification model with the classification accuracy larger than the accuracy threshold is determined as the trained object classification model, so that the classification accuracy of the object classification model is improved.

In some embodiments, training the object classification model to be trained to obtain a new object classification model with classification accuracy to be analyzed includes: inputting the object characteristics of the training object in the first historical period into an object classification model to be trained for classification to obtain the prediction category of the training object in the current period; and adjusting parameters of the object classification model to be trained based on the difference between the prediction class of the training object in the current period and the real class of the training object in the current period to obtain a new object classification model with the classification accuracy to be analyzed.

The training object is an object to which the object features used by the training object classification model belong. The number of training subjects may be one or more.

Specifically, the server may input the object features of the training object in the first history period into the object classification model to be trained, obtain the prediction category of the training object in the current period, adjust the parameters of the object classification model to be trained based on the difference between the prediction category of the training object in the current period and the real category of the training object in the current period, and determine the object classification model after the parameters are adjusted as the new object classification model with the classification accuracy to be analyzed.

In some embodiments, a server may obtain a set of target objects, the set of target objects including a plurality of objects, and may divide the set of target objects into a set of training objects and a set of test objects. The target object set may be randomly divided, for example, the target object set may be randomly divided into a training object set and a test object set according to a specific ratio. The specific ratio refers to the ratio between the number of training subjects and the number of test subjects. The specific ratio may be set as needed or preset, for example, the specific ratio of the number of training subjects to the number of test subjects is a (1-a), where a represents the ratio between the number of training subjects and the total number, and the total number is the number of subjects included in the target subject set. For example, if a is 0.8, the specific ratio is 0.8:0.2, i.e., 8: 2.

In some embodiments, the server may input the object features of the training object in the first history period into the object classification model to be trained, obtain the predicted class of the training object in the current period, adjust the parameters of the object classification model to be trained based on the difference between the predicted class of the training object in the current period and the real class of the training object in the current period, obtain the object classification model after adjusting the parameters, obtain the object features of the plurality of test objects in the first history period into the object classification model after adjusting the parameters, obtain the predicted class of each test object in the plurality of test objects in the current period, count the percentage of the test objects in the plurality of test objects whose predicted classes are consistent with the real classes, suspend the training when the calculated percentage is greater than a preset percentage (e.g., 90%), determine the object classification model after adjusting the parameters as the new object classification model to be analyzed with classification accuracy, otherwise, training is continued.

In this embodiment, because the similarity between the object feature of the current period and the object feature of the first history period is relatively high, the model is trained by using the object feature of the history period closest to the current period, so that the trained object classification model is suitable for predicting the category of the next period based on the object feature of the current period.

In some embodiments, the current period is a current promotion period for the target service, and the classification category of the trained object classification model is any one of retention or loss; the method further comprises the following steps: acquiring object characteristics of a plurality of candidate objects in a current popularization period; inputting the object characteristics of each candidate object in the current promotion period into a trained object classification model for classification to obtain the prediction category of each candidate object in the next promotion period of the target service; selecting a prediction category as a reserved target object from each candidate object; and in the next promotion period, pushing promotion content related to the target service to the target object.

The promotion period is a period in which the content is pushed to the object, and the promotion period may be, for example, a marketing period. The target service may be any service including, but not limited to, at least one of a premium fueling service, a car wash service, a vehicle moving code service, or a designated driving service. The trained object classification model is used to identify whether an object is surviving or losing. The candidate object may be an arbitrary object. The candidate objects may be the same as or different from the objects used in determining the classification accuracy of the object classification model for the current cycle. When the target service is a service proposed in the target application, the candidate object may be a user registered on the target application. The target application may be any application program. The next marketing period refers to a marketing period adjacent to and subsequent to the current marketing period. The predicted category of the candidate object is either retention or loss. For the T period, retention indicates: the object logs in the preferential refueling function module in the T-1 period, and does not log in the preferential refueling function module in the T period, and the loss represents that: the object logs in the preferential fueling function module in the T-1 period, and also logs in the preferential fueling function module in the T period. The target object is a candidate object whose prediction category is retention. The promotion content includes, but is not limited to, any one of an advertisement or a virtual resource, the virtual resource includes, but is not limited to, at least one of a coupon or a red packet, etc. When the promotion period is the marketing period, the promotion content may also be referred to as marketing content.

Specifically, for each candidate object, the server may input the object features of the current marketing period into a trained object classification model for classification, determine the classified category as a prediction category of the candidate object in a next marketing period of the target service, and after obtaining the prediction category of each candidate object, the server may determine the prediction category as a remaining candidate object and determine the prediction category as a remaining candidate object from each candidate object. In the next marketing period, the server can push promotion content related to the target service to the target object, for example, for the preferential fueling service, a coupon for fueling the vehicle can be pushed to the target object. As shown in fig. 5, a schematic diagram of transmitting promotion content to a terminal of a target object is shown.

In the embodiment, the object characteristics of each candidate object in the current marketing period are input into the trained object classification model for classification, the prediction category of each candidate object in the next marketing period of the target service is obtained, the prediction category is selected from each candidate object as the reserved target object, and the promotion content related to the target service is pushed to the target object in the next marketing period.

The application also provides an application scene, and the application scene applies the object classification processing method.

As shown in fig. 6, specifically, the application of the object classification processing method in the application scenario is as follows:

step 602, for each object of the plurality of objects, obtaining an object feature of the object in a first marketing period of the target service to obtain a first object feature, obtaining an object feature of the object in a second marketing period of the target service to obtain a second object feature, obtaining an object feature of the object in a third marketing period of the target service to obtain a third object feature, obtaining an object feature of the object in a current marketing period of the target service to obtain a current object feature.

The first marketing period, the second marketing period and the third marketing period are historical marketing periods. The current marketing period, the first marketing period, the second marketing period and the third marketing period are continuous periods. The first marketing period is adjacent to and before the current marketing period. The second marketing period is adjacent to and before the first marketing period. The third marketing period is adjacent to and before the second marketing period. The time length of each marketing period can be the same or different, for example, the time length of each marketing period is 7 days, the third marketing period is 1 month and 1 day to 1 month and 7 days, the second marketing period is 1 month and 8 days to 1 month and 14 days, the first marketing period is 1 month and 15 days to 1 month and 21 days, and the current marketing period is 1 month and 22 days to 1 month and 28 days.

Step 604, obtaining a classification model of the object to be trained, and determining a training object from each object.

Step 606, inputting the first object features of the training objects into the object classification model to be trained for classification, obtaining the prediction category of the objects in the current marketing period, obtaining the training prediction category of the objects, and adjusting the model parameters of the object classification model to be trained based on the training prediction category of the objects, so as to obtain the object classification model with the classification accuracy to be analyzed.

Step 608, inputting the first object features of each object into the object classification model with the classification accuracy to be analyzed for classification, obtaining the prediction category of each object in the current marketing period, obtaining the current prediction category of each object, inputting the second object features of each object into the object classification model with the classification accuracy to be analyzed for classification, obtaining the prediction category of each object in the first marketing period, obtaining the first prediction category of each object, inputting the third object features of each object into the object classification model with the classification accuracy to be analyzed for classification, obtaining the prediction category of each object in the second marketing period, and obtaining the second prediction category of each object.

Step 610, counting the number of objects meeting the first category condition in the plurality of objects to obtain a first object number; the first category conditions include: the real category of the object in the current period is a first preset category, and the current prediction category of the object is a first preset category; counting the number of objects meeting the second category condition in the plurality of objects to obtain a second object number; the second category of conditions includes: the real category of the object in the second history period is a first preset category, the second prediction category of the object is a second preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a first preset category; and determining the classification accuracy of the object classification model in the current period based on the first object number and the second object number.

Step 612, determining whether the classification accuracy reaches an accuracy threshold, if not, performing step 614, and if so, performing step 616.

And 614, determining the object classification model with the classification accuracy to be analyzed as the object classification model to be trained, and returning to 606.

And 616, determining the object classification model with the classification accuracy to be analyzed as a trained object classification model, and inputting the current object characteristics of the object into the trained object classification model for classification to obtain the prediction category of the object in the next marketing period of the current marketing period.

Step 618, selecting the prediction category as the target object to be saved from the objects.

And step 620, pushing marketing content related to the target service to the target object in the next marketing period of the current marketing period.

In this embodiment, the classification accuracy of the object classification model is determined by using the classes of the objects in two history periods adjacent to the current marketing period (i.e., the first history period and the second history period), and the object classification model is further trained based on the classification accuracy, so that the classification accuracy of the trained object classification model is improved.

The object classification processing method can be applied to a scene related to preferential fueling service in travel service, wherein in the scene, the object classification model is a two-classification model, the category is a loss early warning scene label, and the loss early warning scene label is any one of a loss label or a retention label, as shown in fig. 7.

In the data processing stage, vehicle owner log data (the vehicle owner log data are independently used after consent of a vehicle owner is obtained) are obtained, and the data are respectively processed into vehicle owner scene labels and vehicle owner characteristic data. Wherein the tags are constructed as follows:

for tags of T cycle: if the vehicle owner logs in the preferential fueling function module in the T-1 period and does not log in the preferential fueling function module in the T period, the vehicle owner is indicated as a lost vehicle owner of the preferential fueling module in the T period, and the loss early warning scene label Y of the vehicle owner in the T period is marked_tLabeled 1; if the vehicle owner logs in the preferential fueling function module in the T-1 period and also logs in the preferential fueling function module in the T period, the vehicle owner is shown as a reserved vehicle owner of the preferential fueling module in the T period, and the loss early warning scene label Y of the vehicle owner in the T period is labeled_tLabeled 0;

for the T-1 stage tags: if the vehicle owner logs in the preferential fueling function module in the T-2 period and does not log in the preferential fueling function module in the T-1 period, the vehicle owner is a lost vehicle owner of the preferential fueling function module in the T-1 period, and a loss early warning scene label Y of the vehicle owner in the T-1 period is marked_t-1Labeled 1; if the vehicle owner logs in the preferential fueling function module in the T-2 stage and also logs in the preferential fueling function module in the T-1 stage, the vehicle owner is a reserved vehicle owner of the preferential fueling function module in the T-1 stage, and the vehicle owner is marked with a loss early warning scene label Y in the T-1 period_t-1Labeled 0;

for T-2 cycle tags: if the vehicle owner logs in the preferential fueling function module in the T-3 period and does not log in the preferential fueling function module in the T-2 period, the vehicle owner is a lost vehicle owner of the preferential fueling function module in the T-2 period, and a loss early warning scene label Y of the vehicle owner in the T-2 period is marked_t-2Labeled 1; if the vehicle owner logs in the preferential fueling function module in the T-3 period and also logs in the preferential fueling function module in the T-2 period, the vehicle owner is a reserved vehicle owner of the preferential fueling function module in the T-2 period, and a loss early warning scene label Y of the vehicle owner in the T-2 period is marked_t-2The flag is 0. Wherein, T-3, T-2, T-1 and T are continuous periods, and T is the current period.

The characteristics are constructed as follows:

processing the owner log data to obtain owner characteristics (feature) X of the owner in the periods T-3, T-2, T-1 and T_t-iI |, 0,1,2,3 }. Wherein, X_t-3Indicating owner characteristics, X, of the owner in stage T-3_t-2Indicating owner characteristics, X, of the owner in the T-2 th stage_t-1Indicates owner characteristics, X, of the owner in the T-1 th stage_tIndicating the owner's characteristics of the owner during the tth period.

In the sample construction stage: will the car owner's character X of the T-1 stage_t-1And car owner label Y in T period_tVehicle owner sample data S is constructed according to vehicle owner identification (userid) matching_tThe sample data is divided into training samples and test samples, and the sample data (including the training samples and the test samples) is divided into sparse features and dense features. The sparse features are subjected to one-hot encoding (onehot encoding) processing, and the dense features are subjected to PCA (principal component analysis) decorrelation processing, normalization (normalization) processing, feature discretization processing, and the like. The principal component analysis method is a data dimension reduction algorithm, and mainly thinks that n-dimensional features are mapped to k dimensions, wherein the k dimensions are brand-new orthogonal features and are also called principal components, and the k-dimensional features are reconstructed on the basis of the original n-dimensional features. The main idea of onehot coding is: an N-bit status register is used to encode N states, each state being represented by its own independent register bit and only one bit being active at any one time. For example: sex characteristics: [ "Man" and "woman"]The corresponding onehot codes are: male sexual intercourse>10; for women>01. Coupon type feature: [ "No. 92", "No. 95", "No. 98"]The corresponding onehot codes are: 92->100, respectively; 95->010; no. 98->001; coupon discount amount characteristics: [ "15 yuan", "25 yuan", "30 yuan [)Yuan (Chinese character) "and" 50 Yuan (Chinese character) "]The corresponding onehot codes are: 15-element->1000, parts by weight; 25-element->0100; 30 yuan->0010; 50 yuan->0001. When a sample is [ "Man", "95", "15 yuan"]The result of the complete feature digitized onehot encoding is: [1,0,0,1,0,1,0,0,1]. Randomly cutting the processed sparse features, the processed dense features and the classification labels of the car owners into training samples S according to a certain proportion_t ^trainAnd test specimen S_t ^test. Construction of a prediction sample construction: feature X using T periods_tAnd the prediction samples are taken as prediction samples, and the prediction samples are distinguished into sparse features and dense features. Wherein, the sparse type characteristic is subjected to onehot processing, and the dense type characteristic is subjected to PCA decorrelation processing, normalization (standardization) processing, characteristic discretization processing and the like. Constructing a T-1 stage full-scale sample: vehicle owner characteristic X of input T-2 stage_t-2And car owner label Y of T-1 stage_t-1Constructing full-scale sample data S of the vehicle owner in the T-1 stage according to the matching of the vehicle owner identification_t-1. Constructing a T-2 stage full-scale sample: inputting the owner characteristic X of the T-3 stage_t-3And car owner label Y of T-2 stage_t-2Constructing full-scale sample data S of the vehicle owner in the T-2 stage according to the matching of the vehicle owner identification_t-2。

And (3) a model training and testing stage: inputting a training sample S_t ^trainAnd test specimen S_t ^test. And (3) performing model training and testing on training and testing samples of each scene (such as a scene of preferential fueling service) by adopting a two-classification model, and if T-stage evaluation indexes (indexes such as recall ratio, precision ratio, AUC (area Under cut) and the like) achieve an evaluation effect, respectively storing the model weight vector W of the model.

And obtaining an evaluation label: inputting a model weight vector W and inputting a T period full-scale sample S_t. Adopting a binary classification algorithm and dividing labels according to a threshold value of 0.5 (wherein, the probability is more than or equal to 0.5 and is recorded as 1, and the probability is less than 0.5 and is recorded as 0), and obtaining a model evaluation index sequence E of the T-period full-scale sample_t. Similarly, the full sample S in the T-1 period is input_t-1Adopting a binary classification algorithm and dividing labels according to a threshold value of 0.5 to obtain a model evaluation index sequence E of the T-1 period full-scale sample_t-1. Transfusion systemFull sample S in T-2 phase_t-2Adopting a binary classification algorithm and dividing labels according to a threshold value of 0.5 to obtain a model evaluation index sequence E of the T-2 period full-scale sample_t-2. The model evaluation index sequence comprises tags predicted by the model for each sample.

And a second-order recursive confusion matrix construction stage, wherein the second-order recursive confusion matrix is constructed and comprises the formula (1) and the formula (2). Wherein the confusion matrix: a specific matrix for presenting the performance visualization effect of a supervised learning algorithm summarizes records in a data set according to two criteria of real category and classification judgment made by a classification model, wherein each column represents a predicted value, and each row represents an actual category. Recursive confusion matrix: the calculation indexes (recall, precision) in the confusion matrix at time t are influenced by the confusion matrix at time t-1. Second order recursive confusion matrix: the calculation indexes (recall, precision) in the confusion matrix at time t are affected by the confusion matrix at time t-1 and time t-2.

And (3) a model evaluation stage: inputting a model evaluation index sequence E of T-period full-scale samples_tModel evaluation index sequence E of full-scale samples in T-1 stage_t-1Model evaluation index sequence E of full-scale samples in T-2 stage_t-2User tag data Y in T period_tT-1 phase user tag data Y_t-1T-2 phase user tag data Y_t-1. Substituting into formulas (1) and (2) to obtain recall ratio R of T phase under the influence of T-2 and T-1 phases_t|t-1,t-2Precision ratio P_t|t-1,t-2And the model evaluation effect is obtained again. If the model does not reach the target effect (general experience, recall ratio is more than or equal to 90%, precision ratio is more than or equal to 85%), repeating the steps 3-6 (namely, the model training test stage to the model evaluation stage) until the model reaches the target effect.

Model prediction stage, inputting prediction sample X_tAnd a model W of the model training test phase. Substituting prediction sample X by using a binary classification algorithm_tObtaining a prediction probability by the model W, and dividing the labels according to a threshold value of 0.5, wherein the probability is more than or equal to 0.5 and is marked as 1; less than 0.5 is noted as 0. Complete the whole model trainingAnd (6) measuring.

And in the preferential fueling vehicle owner recommending stage, inputting the vehicle owner identification marked as 1 obtained by model prediction in the model predicting stage, issuing a preferential fueling coupon to the part of the vehicle owner identification by taking the travel service applet as a channel, and informing the vehicle owner of issuing information in a short message or instant communication mode.

The operation activities often have the condition of superposition of multi-period effects under the same operation activity, taking a fueling activity as an example, the operation strategy sets the use period of the coupon to be effective within 21 days, the activity period to be 7 days, and each activity period carries out continuous operation activities (namely, after the 7-day activity is finished, the operation activity of the next activity period is started, and the effective use period of the coupon at each period is 21 days), the valid period of the coupon at the T period comprises the activity periods at the T-1 period and the T-2 period, so that the operation activities at the T-1 period and the T-2 period generate positive influence on the operation activities at the current period (T), sample data at the T-1 period and the T-2 period often influence the model effect of the current period (T) and influence the evaluation of the current model by the sample data at the current period (T), the effect of model prediction is affected and the contribution of the current operation activity to the current model effect cannot be accurately reflected. In the embodiment of the application, the second-order recursive confusion matrix is adopted in model evaluation, so that the influence of T-1 and T-2 operation activity data on the model can be more accurately reflected and removed, and the accurate evaluation method of the model effect is realized. For the marketing activities of the internet of vehicles for preferential fueling, a mode of issuing preferential fueling coupons or gift exchange for vehicle owners meeting the conditions is often adopted. However, because the effect of multi-phase activities is often superposed in the same operation activity, namely the model effect of the T phase is influenced by the T-1 phase and the T-2 phase, the probability calculation models (such as the formula (1) and the formula (2)) of the T phase recall ratio and the precision ratio under the influence of the T-1 phase and the T-2 phase are obtained through strict mathematical derivation, and the influence on the model effect is realized, so that the effect of the current operation activity can be more accurately attributed to the current operation strategy by the accurate evaluation method of the model effect. The formula (1) and the formula (2) are provided on the basis of a second-order recursive confusion matrix, and the calculation formula obtained through strict data reasoning can be realized only by creative labor.

The object classification processing method provided by the application can also be applied to model effect evaluation of superposition of multiple service scenes and multiple activity periods, can effectively distinguish the model effect of each service scene and each activity period, and can accurately reflect the effect of each service scene and each activity period model. The object classification processing method can be combined with marketing recommendation activities in various scenes, can carry various machine learning algorithms and deep learning algorithms, can be suitable for various activity scenes, and has good scene expansibility. The object classification processing method provided by the application can be popularized and applied to business modules of preferential refueling, preferential car washing, car moving code business, designated driving business and the like of travel services.

It should be understood that, although the steps in the flowcharts related to the embodiments are shown in sequence as indicated by the arrows, the steps are not necessarily executed in sequence as indicated by the arrows. The steps are not limited to being performed in the exact order illustrated and, unless explicitly stated herein, may be performed in other orders. Moreover, at least a part of the steps in the flowcharts related to the above embodiments may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of performing the steps or stages is not necessarily sequential, but may be performed alternately or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the application also provides an object classification processing device for realizing the object classification processing method. The implementation scheme for solving the problem provided by the apparatus is similar to the implementation scheme described in the above method, so specific limitations in one or more embodiments of the object classification processing apparatus provided below may refer to the limitations on the object classification processing method in the foregoing, and details are not described herein again.

In some embodiments, as shown in fig. 8, there is provided an object classification processing apparatus including: a model determination module 802, a feature acquisition module 804, a category derivation module 806, and an accuracy determination module 808, wherein:

a model determining module 802, configured to determine an object classification model of the classification accuracy to be analyzed;

a feature obtaining module 804, configured to obtain historical object features of the multiple objects in at least three target historical periods respectively; at least three target history cycles, which are at least three continuous history cycles selected from the previous cycle of the current cycle;

a category obtaining module 806, configured to, for a historical object feature of each object in each historical period, input the historical object feature of each object into an object classification model for classification, so as to predict a prediction category of each object in a next period of the historical period, and obtain a prediction category of each object in a current period and prediction categories of each object in at least two adjacent historical periods; the at least two adjacent history cycles are target history cycles which are closer to the current cycle in the at least three target history cycles;

an accuracy determination module 808, configured to determine a classification accuracy of the object classification model in the current cycle based on the prediction category of each object in the current cycle and the prediction categories of each object in at least two adjacent historical cycles.

In some embodiments, the at least two adjacent history cycles include a first history cycle adjacent to and prior to the current cycle and a second history cycle adjacent to and prior to the first history cycle; the accuracy determination module is further to: obtaining the prediction type of each object in the current period to obtain the current prediction type of each object; obtaining the prediction category of each object in a first history period to obtain the first prediction category of each object; obtaining the prediction category of each object in a second history period to obtain a second prediction category of each object; and determining the classification accuracy of the object classification model in the current period based on the current prediction category, the first prediction category and the second prediction category of each object.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a first class condition in the plurality of objects to obtain a first object number; the first category conditions include: the real category of the object in the current period is a first preset category, and the current prediction category of the object is a first preset category; counting the number of objects meeting the second category condition in the plurality of objects to obtain a second object number; the second category of conditions includes: the real category of the object in the second history period is a first preset category, the second prediction category of the object is a second preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a first preset category; and determining the classification accuracy of the object classification model in the current period based on the first object number and the second object number.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a third category condition in the plurality of objects to obtain a third object number; the third category of conditions includes: the real category of the object in the second history period is a first preset category, the second prediction category of the object is a second preset category, the real category of the object in the first history period is a second preset category, and the first prediction category of the object is a first preset category; counting based on the first object quantity and the second object quantity to obtain a forward statistical value; the forward statistical value is in positive correlation with the first object quantity and the second object quantity; counting based on the first object quantity and the third object quantity to obtain a negative statistical value; the negative statistic value is in positive correlation with the first object quantity and the third object quantity; determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics; the classification accuracy and the positive statistical value form a positive correlation relationship, and the classification accuracy and the negative statistical value form a negative correlation relationship.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category of conditions includes: the real category of the object in the current period is a first preset category, the real category of the object in the first historical period is a first preset category, and the first prediction category of the object is a second preset category; counting based on the first object number, the second object number and the fourth object number to obtain a forward statistical value; the forward statistical value is in a negative correlation with the fourth number of objects.

In some embodiments, the negative statistics comprise a first negative statistics, and the classification accuracy comprises a first classification accuracy; the accuracy determination module is further to: counting the number of objects meeting a fifth category condition in the plurality of objects to obtain a fifth object number; the fifth category of conditions includes: the real category of the object in the first history period is a first preset category, the first prediction category of the object is a second preset category, the real category of the object in the second history period is the first preset category, and the second prediction category of the object is the second preset category; counting based on the first object number, the third object number and the fifth object number to obtain a first negative statistic value; the first negative statistic value and the number of the fifth objects form a positive correlation; and determining the first classification accuracy of the object classification model in the current period based on the positive-going statistic and the first negative-going statistic.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category of conditions includes: the real category of the object in the current period is a first preset category, the real category of the object in the first historical period is a first preset category, and the first prediction category of the object is a second preset category; counting based on the first object number, the third object number, the fourth object number and the fifth object number to obtain a first negative statistic value; the first negative statistic is negatively correlated with the number of fourth objects.

In some embodiments, the accuracy determination module is further to: counting the number of objects meeting the sixth category condition in the plurality of objects to obtain the number of sixth objects; the sixth category of conditions includes: the real category of the object in the current period is a second preset category, the current prediction category of the object is a first preset category, the real category of the object in the first history period is a first preset category, and the first prediction category of the object is a second preset category; counting based on the first object number, the third object number, the fourth object number, the fifth object number and the sixth object number to obtain a first negative statistic value; the first negative statistic is negatively correlated with the sixth number of objects.

In some embodiments, the negative statistics comprise a second negative statistics, and the classification accuracy comprises a second classification accuracy; the accuracy determination module is further to: counting based on the first object quantity, the second object quantity and the third object quantity to obtain a second negative-going statistical value; the second negative statistic value and the second object quantity form a positive correlation; determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics comprises: and determining a second classification accuracy of the object classification model in the current period based on the positive-going statistic and the second negative-going statistic.

In some embodiments, the apparatus further comprises: the first model determining module is used for determining the object classification model with the classification accuracy to be analyzed as the object classification model to be trained under the condition that the classification accuracy is smaller than an accuracy threshold; the second model determining module is used for training the object classification model to be trained to obtain a new object classification model with classification accuracy to be analyzed, and the step of inputting the historical object characteristics of each object into the object classification model for classification is returned until the classification accuracy reaches the accuracy threshold; and the third model determining module is used for determining the object classification model with the classification accuracy to be analyzed under the condition that the classification accuracy reaches the accuracy threshold as the trained object classification model.

In some embodiments, the second model determination module is further to: inputting the object characteristics of the training object in the first historical period into an object classification model to be trained for classification to obtain the prediction category of the training object in the current period; and adjusting parameters of the object classification model to be trained based on the difference between the prediction class of the training object in the current period and the real class of the training object in the current period to obtain a new object classification model with the classification accuracy to be analyzed.

In some embodiments, the current period is a current promotion period for the target service, and the classification category of the trained object classification model is any one of retention or loss; the apparatus is also configured to: acquiring object characteristics of a plurality of candidate objects in a current popularization period; inputting the object characteristics of each candidate object in the current promotion period into a trained object classification model for classification to obtain the prediction category of each candidate object in the next promotion period of the target service; selecting a prediction category as a reserved target object from each candidate object; and in the next promotion period, pushing promotion content related to the target service to the target object.

The modules in the object classification processing device may be wholly or partially implemented by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In some embodiments, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 9. The computer device includes a processor, a memory, an Input/Output interface (I/O for short), and a communication interface. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing data involved in the object classification processing method. The input/output interface of the computer device is used for exchanging information between the processor and an external device. The communication interface of the computer device is used for connecting and communicating with an external terminal through a network. The computer program is executed by a processor to implement an object classification processing method.

In some embodiments, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 10. The computer apparatus includes a processor, a memory, an input/output interface, a communication interface, a display unit, and an input device. The processor, the memory and the input/output interface are connected by a system bus, and the communication interface, the display unit and the input device are connected by the input/output interface to the system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The input/output interface of the computer device is used for exchanging information between the processor and an external device. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless communication can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement an object classification processing method. The display unit of the computer equipment is used for forming a visual and visible picture, and can be a display screen, a projection device or a virtual reality imaging device, the display screen can be a liquid crystal display screen or an electronic ink display screen, the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

Those skilled in the art will appreciate that the configurations shown in fig. 9 and 10 are merely block diagrams of some configurations relevant to the present disclosure, and do not constitute a limitation on the computing devices to which the present disclosure may be applied, and that a particular computing device may include more or less components than those shown, or combine certain components, or have a different arrangement of components.

In some embodiments, a computer device is provided, comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the object classification processing method described above when executing the computer program.

In some embodiments, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, realizes the steps of the above-mentioned object classification processing method.

In some embodiments, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps in the object classification processing method described above.

It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the relevant laws and regulations and standards of the relevant country and region. For example, in the present application, the log data of the owner is all desensitized to use when the owner's consent is obtained separately.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), Magnetic Random Access Memory (MRAM), Ferroelectric Random Access Memory (FRAM), Phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others. The databases referred to in various embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the various embodiments provided herein may be, without limitation, general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, or the like.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. An object classification processing method, characterized in that the method comprises:

determining an object classification model of the classification accuracy to be analyzed;

acquiring historical object characteristics of a plurality of objects in at least three target historical periods respectively; the at least three target history cycles are at least three continuous history cycles selected from the previous cycle of the current cycle;

inputting the historical object features of each object into the object classification model for classification aiming at the historical object features of each object in each target historical period so as to predict the prediction category of each object in the next period of the target historical period, and obtain the prediction category of each object in the current period and the prediction categories of each object in at least two adjacent historical periods; the at least two adjacent history cycles are target history cycles closer to the current cycle of the at least three target history cycles;

determining the classification accuracy of the object classification model in the current cycle based on the prediction category of each of the objects in the current cycle and the prediction categories of each of the objects in the at least two adjacent historical cycles.

2. The method of claim 1, wherein the at least two adjacent history cycles comprise a first history cycle adjacent to and prior to the current cycle and a second history cycle adjacent to and prior to the first history cycle;

said determining a classification accuracy of said object classification model in said current cycle based on said predicted class of each said object in said current cycle and said predicted classes of each said object in said at least two adjacent historical cycles comprises:

obtaining the prediction category of each object in the current period to obtain the current prediction category of each object;

obtaining the prediction category of each object in the first history period to obtain the first prediction category of each object;

obtaining the prediction category of each object in the second history period to obtain the second prediction category of each object;

and determining the classification accuracy of the object classification model in the current period based on the current prediction category, the first prediction category and the second prediction category of each object.

3. The method of claim 2, wherein determining the classification accuracy of the object classification model in the current cycle based on the current prediction class, the first prediction class, and the second prediction class of each of the objects comprises:

counting the number of objects meeting a first class condition in the plurality of objects to obtain a first object number; the first category condition includes: the real category of the object in the current period is a first preset category, and the current prediction category of the object is the first preset category;

counting the number of objects meeting a second category condition in the plurality of objects to obtain a second object number; the second category condition includes: the real category of the object in the second history period is the first preset category, the second prediction category of the object is the second preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the first preset category;

determining a classification accuracy of the object classification model for the current cycle based on the first number of objects and the second number of objects.

4. The method of claim 3, wherein the determining the classification accuracy of the object classification model for the current cycle based on the first number of objects and the second number of objects comprises:

counting the number of objects meeting a third category condition in the plurality of objects to obtain a third object number; the third category of conditions includes: the real category of the object in the second history period is the first preset category, the second prediction category of the object is the second preset category, the real category of the object in the first history period is the second preset category, and the first prediction category of the object is the first preset category;

counting based on the first object quantity and the second object quantity to obtain a forward statistical value; the forward statistical value is in positive correlation with the first object quantity and the second object quantity;

counting based on the first object quantity and the third object quantity to obtain a negative statistic value; the negative statistic value is in positive correlation with the first object quantity and the third object quantity;

determining the classification accuracy of the object classification model in the current period based on the positive statistics and the negative statistics; the classification accuracy and the positive statistic value form a positive correlation relationship, and the classification accuracy and the negative statistic value form a negative correlation relationship.

5. The method of claim 4, wherein the counting based on the first number of objects and the second number of objects to obtain a forward statistic comprises:

counting the number of objects meeting a fourth category condition in the plurality of objects to obtain a fourth object number; the fourth category condition includes: the real category of the object in the current period is the first preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the second preset category;

counting based on the first object number, the second object number and the fourth object number to obtain a forward statistical value; the forward statistical value is in a negative correlation relation with the fourth object number.

6. The method of claim 4, wherein the negative statistics comprise a first negative statistics, and wherein the classification accuracy comprises a first classification accuracy; the counting based on the first object number and the third object number to obtain a negative statistic value comprises:

counting the number of objects meeting a fifth category condition in the plurality of objects to obtain a fifth object number; the fifth category condition includes: the real category of the object in the first history period is the first preset category, the first prediction category of the object is the second preset category, the real category of the object in the second history period is the first preset category, and the second prediction category of the object is the second preset category;

counting based on the first object number, the third object number and the fifth object number to obtain a first negative statistic value; the first negative statistic value and the fifth object number form a positive correlation;

the determining, based on the positive statistics and the negative statistics, the classification accuracy of the object classification model in the current period comprises:

determining a first classification accuracy of the object classification model for the current period based on the positive statistics and the first negative statistics.

7. The method of claim 6, wherein the counting based on the first number of objects, the third number of objects, and the fifth number of objects results in a first negative statistic; the positively correlating the first negative statistics with the fifth number of objects comprises:

counting based on the first object number, the third object number, the fourth object number and the fifth object number to obtain a first negative statistic value; the first negative statistic is negatively correlated with the fourth number of objects.

8. The method of claim 7, wherein the counting based on the first number of objects, the third number of objects, the fourth number of objects, and the fifth number of objects to obtain a first negative statistic comprises:

counting the number of objects meeting a sixth category condition in the plurality of objects to obtain a sixth object number; the sixth category of conditions includes: the real category of the object in the current period is the second preset category, the current prediction category of the object is the first preset category, the real category of the object in the first history period is the first preset category, and the first prediction category of the object is the second preset category;

counting based on the first object number, the third object number, the fourth object number, the fifth object number and the sixth object number to obtain a first negative statistic value; the first negative statistic is in negative correlation with the sixth number of objects.

9. The method of claim 4, wherein the negative statistics comprise a second negative statistics, and wherein the classification accuracy comprises a second classification accuracy; the counting based on the first object number and the third object number to obtain a negative statistic value comprises:

counting based on the first object quantity, the second object quantity and the third object quantity to obtain a second negative-going statistical value; the second negative statistic value and the second object number form a positive correlation;

determining a second classification accuracy of the object classification model for the current period based on the positive statistics and the second negative statistics.

10. The method of claim 1, further comprising:

determining the object classification model with the classification accuracy to be analyzed as an object classification model to be trained under the condition that the classification accuracy is smaller than an accuracy threshold;

training the object classification model to be trained to obtain a new object classification model with classification accuracy to be analyzed, and returning to the step of inputting the historical object characteristics of each object into the object classification model for classification until the classification accuracy reaches an accuracy threshold;

and determining the object classification model with the classification accuracy reaching the classification accuracy to be analyzed under the condition of an accuracy threshold value as a trained object classification model.

11. The method of claim 10, wherein the training of the object classification model to be trained to obtain a new object classification model with classification accuracy to be analyzed comprises:

inputting the object characteristics of the training object in a first historical period into the object classification model to be trained for classification to obtain the prediction category of the training object in the current period;

and adjusting parameters of the object classification model to be trained based on the difference between the prediction class of the training object in the current period and the real class of the training object in the current period to obtain a new object classification model with classification accuracy to be analyzed.

12. The method of claim 10, wherein the current period is a current promotion period for a target service, and the classification category of the trained object classification model is any one of retention or loss; the method further comprises the following steps:

acquiring object characteristics of a plurality of candidate objects in the current popularization period;

inputting the object characteristics of each candidate object in the current promotion period into the trained object classification model for classification to obtain the prediction category of each candidate object in the next promotion period of the target service;

selecting a prediction category as a reserved target object from each candidate object;

and pushing promotion content related to the target service to the target object in the next promotion period.

13. An object classification processing apparatus, characterized in that the apparatus comprises:

the model determining module is used for determining an object classification model of the classification accuracy to be analyzed;

the characteristic acquisition module is used for acquiring historical object characteristics of a plurality of objects in at least three target historical periods respectively; the at least three target history cycles are at least three continuous history cycles selected from the previous cycle of the current cycle;

a class obtaining module, configured to, for a historical object feature of each object in each historical period, input the historical object feature of each object into the object classification model for classification, so as to predict a prediction class of each object in a next period of the historical period, and obtain a prediction class of each object in the current period and prediction classes of each object in at least two adjacent historical periods; the at least two adjacent history cycles are target history cycles closer to the current cycle of the at least three target history cycles;

an accuracy determination module for determining the classification accuracy of the object classification model in the current cycle based on the prediction category of each of the objects in the current cycle and the prediction categories of each of the objects in the at least two adjacent historical cycles.

14. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 12.

15. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 12.

16. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of any one of claims 1 to 12 when executed by a processor.