Methods, devices, storage media, and processors for predicting user churn

Abstract

This invention discloses a method, device, storage medium, and processor for predicting user churn, comprising: acquiring attribute information of multiple user attributes, wherein the multiple attributes are attributes included in an attribute set; and predicting user churn based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset. This invention solves the technical problem of inaccurate prediction in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of user data mining, and in particular to a method, apparatus, and processor for constructing classifiers for classification and prediction. Background Technology

[0002] Many user-driven businesses (such as online games, e-commerce, telecommunications services, and WeChat) inevitably face the problem of user churn. User churn refers to users ceasing to use the service. For a business, this is a loss, especially when the number of churned users is particularly large, which can be a huge blow to its operations. Therefore, the ability to predict impending user churn before it occurs, and to take measures to retain users and extend their lifespan within the business, is a very attractive and important area of ​​research for businesses.

[0003] Currently, most mainstream methods for predicting user churn originate from the field of machine learning, such as decision trees, support vector machines, and Naive Bayes. These methods are relatively weak predictive approaches. To improve prediction accuracy, these methods can be used as basis functions, employing additive models for enhancement. Among these, boosting trees based on decision trees are an effective method, with XGBoost being a representative example.

[0004] When using a boosting tree model to predict user churn, the input is various user attributes and category information, and the output is the user churn probability. This method generates a specified number of decision trees sequentially based on these attributes. Each decision tree's generation depends on previously generated decision trees, with the aim of minimizing the difference between the model's predictions so far and the actual values.

[0005] However, in the process of generating each decision tree, the algorithm employs a greedy strategy to select the splitting attributes and corresponding splitting values ​​for each node in the tree. Therefore, the following situation may occur: given the previous nodes, examining each attribute individually will reveal some attributes (weak attributes) that are less helpful for prediction than other attributes (strong attributes). However, combining weak attributes may be more helpful for prediction than a single strong attribute, especially when the difference between strong and weak attributes is not significant. In such cases, the combination of weak attributes can be selected as the splitting attribute.

[0006] However, in existing boosting tree models, under a greedy strategy, these weak attributes are not selected as splitting attributes. This results in the user's attribute information not being fully utilized, thus affecting the accuracy of user churn prediction.

[0007] Furthermore, existing classification models also suffer from the problem of insufficient utilization of object attributes during classifier training. This leads to the technical issue of inaccurate classification results from models trained using existing techniques.

[0008] There are currently no effective solutions to the problems mentioned above, such as the inability to fully utilize user attribute information, which affects the accuracy of user churn prediction, and the inaccuracy of classification models trained by existing technologies. Summary of the Invention

[0009] To address the problems of existing technologies, embodiments of the present invention provide a method, apparatus, and processor for constructing a classifier for classification and prediction. The technical solution is as follows:

[0010] Firstly, a method for predicting user churn is provided, comprising: acquiring attribute information of multiple user attributes, wherein the multiple attributes are attributes included in an attribute set; and predicting user churn based on the attribute information using a classification model trained through machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0011] Secondly, a method for constructing a classification model is provided, comprising: acquiring a first training data set, and determining an attribute set including multiple attributes based on the data in the first training data set; training a first classifier model of the classification model using the first training data set through machine learning; determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; removing attributes from the attribute set in the first subset to obtain a second subset; combining at least some attributes in the second subset to form new attributes; and adding the new attributes to the attribute set.

[0012] Thirdly, a method for classifying objects is provided, comprising: obtaining attribute information of multiple attributes from object data, wherein the multiple attributes are attributes included in an attribute set; and classifying the objects based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0013] Fourthly, a storage medium is provided, the storage medium including a stored program, wherein, when the program is running, it controls the processor to execute any of the methods described above.

[0014] Fifthly, a processor is provided for running a program, wherein the program executes any of the methods described above during runtime.

[0015] Sixthly, a device for predicting user churn is provided, comprising: a transmission device; an input / output interface; and a processor. The transmission device and the input / output interface are used to acquire user data, and the processor runs a program. During program execution, the following processing steps are performed on data output from the transmission device and / or the input / output interface: acquiring attribute information of multiple user attributes, where the multiple attributes are attributes included in an attribute set; and based on the attribute information, using a classification model trained through machine learning, predicting user churn. The classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0016] In a seventh aspect, an apparatus for constructing a classification model is provided, the classification model including multiple classifier models. The apparatus includes: a transmission device; an input / output interface; and a processor. The transmission device and the input / output interface are used to acquire data, and the processor runs a program. During program execution, the following processing steps are performed on the data output from the transmission device and / or the input / output interface: acquiring a first training data set, and determining an attribute set including multiple attributes based on the data in the first training data set; training a first classifier model of the classification model using the first training data set; determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; removing attributes from the attribute set in the first subset to obtain a second subset; combining at least a portion of the attributes in the second subset to form a new attribute; and adding the new attribute to the attribute set.

[0017] Eighthly, a device for classifying objects is provided, comprising: a transmission device; an input / output interface; and a processor. The transmission device and the input / output interface are used to acquire object data, and the processor runs a program. During program execution, the following processing steps are performed on data output from the transmission device and / or the input / output interface: acquiring attribute information of multiple attributes from the object data, the multiple attributes being attributes included in an attribute set; and classifying the objects based on the attribute information using a classification model trained through machine learning, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0018] Ninthly, a device for predicting user churn is provided, comprising: a transmission device; an input / output interface; and a storage medium for storing a program. The transmission device and the input / output interface are used to acquire user data. During runtime, the program performs the following processing steps on the data output by the transmission device and / or the input / output interface: acquiring attribute information of multiple user attributes, where the multiple attributes are attributes included in an attribute set; and based on the attribute information, using a classification model trained through machine learning, predicting user churn. The classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of the attributes in the second subset.

[0019] In a tenth aspect, an apparatus for constructing a classification model is provided, the classification model including multiple classifier models. The apparatus includes: a transmission device; an input / output interface; and a storage medium for storing a program. The transmission device and the input / output interface are used to acquire data, and the program, during runtime, performs the following processing steps on the data output by the transmission device and / or the input / output interface: acquiring a first training data set, and determining an attribute set including multiple attributes based on the data in the first training data set; training a first classifier model of the classification model using the first training data set; determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; removing attributes from the attribute set in the first subset to obtain a second subset; combining at least a portion of the attributes in the second subset to form a new attribute; and adding the new attribute to the attribute set.

[0020] Eleventhly, a device for classifying objects is provided, comprising: a transmission device; an input / output interface; and a storage medium for storing a program. The transmission device and the input / output interface are used to acquire object data. During runtime, the program performs the following processing steps on the data output by the transmission device and / or the input / output interface: acquiring attribute information of multiple attributes from the object data, the multiple attributes being attributes included in an attribute set; and classifying the objects based on the attribute information using a classification model trained through machine learning, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes, wherein the new attributes are formed by combining at least a portion of the attributes in the second subset.

[0021] In a twelfth aspect, a device for predicting user churn is provided, comprising: a processor; and a memory connected to the processor for providing instructions to the processor to perform the following processing steps: acquiring attribute information of multiple attributes of a user, the multiple attributes being attributes included in an attribute set; and predicting user churn based on the attribute information using a classification model trained by machine learning, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs a classification operation based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0022] In a thirteenth aspect, an apparatus for constructing a classification model is provided, the classification model including multiple classifier models, comprising: a processor; and a memory connected to the processor for providing the processor with instructions to perform the following processing steps: acquiring a first training data set, and determining an attribute set including multiple attributes based on data in the first training data set; training a first classifier model of the classification model using the first training data set; determining a first subset of the attribute set based on the first classifier model. The attributes in the first subset are those used by the first classifier model for classification; removing attributes from the attribute set in the first subset to obtain a second subset; combining at least a portion of the attributes in the second subset to form new attributes; and adding the new attributes to the attribute set.

[0023] In a fourteenth aspect, an apparatus for classifying objects is provided, comprising: a processor; and a memory connected to the processor for providing the processor with instructions to perform the following processing steps: obtaining attribute information of multiple attributes from object data, the multiple attributes being attributes included in an attribute set; and classifying the objects based on the attribute information using a classification model trained by machine learning, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs a classification operation based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0024] In a fifteenth aspect, a device for predicting user churn is provided, comprising: an acquisition module for acquiring attribute information of multiple attributes of a user, wherein the multiple attributes are attributes included in an attribute set; and a prediction module for predicting user churn based on the attribute information and using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and the first classifier model performs a classification operation based on the attributes in the first subset, and the second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0025] In a sixteenth aspect, an apparatus for constructing a classification model is provided, comprising: an attribute determination module for acquiring a first training data set and determining an attribute set including multiple attributes based on data in the first training data set; a training module for training a first classifier model of the classification model using the first training data set through machine learning; a first subset determination module for determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; a second subset determination module for removing attributes from the attribute set from the first subset to obtain a second subset; a combination module for combining at least a portion of the attributes in the second subset to form a new attribute; and an addition module for adding the new attribute to the attribute set.

[0026] In a seventeenth aspect, an apparatus for classifying objects is provided, comprising: an acquisition module for acquiring attribute information of multiple attributes from object data, wherein the multiple attributes are attributes included in an attribute set; and a classification module for classifying the objects based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0027] The beneficial effects of the technical solution provided by the embodiments of the present invention are as follows: Through the technical solution of the present invention, attributes not used by the first classifier (i.e., weak attributes) can be recombine as new attributes for training the classifier. This improves the classification model's ability to select attributes, extracting unselected weak attributes, and then utilizing the classifier model's ability to combine attributes to combine these weak attributes for use. This method not only controls the number of generated candidate weak attribute combinations but also ensures the quality of these combinations, utilizing weak attribute information in a fast and effective way, thereby improving prediction accuracy. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of a computer terminal 10 or mobile device 10 used to implement the methods provided in the embodiments of the present invention.

[0030] Figure 2 This is a schematic diagram illustrating an application scenario of the method provided in the embodiments of the present invention.

[0031] Figure 3 This is a flowchart of a method for predicting user churn provided in an embodiment of the present invention.

[0032] Figure 4 This is a schematic diagram of the classification model in the method provided in the embodiments of the present invention.

[0033] Figure 5This is a schematic diagram of the sample dataset used to train the classification model in the method provided in the embodiments of the present invention.

[0034] Figure 6 This is a schematic diagram of a classifier model described in the method provided in the embodiments of the present invention.

[0035] Figure 7 This is a schematic diagram of a classification tree described in the method provided in the embodiments of the present invention.

[0036] Figure 8 This is a flowchart of a method for constructing a classification model provided in an embodiment of the present invention.

[0037] Figure 9 This is a flowchart illustrating a specific example of a method for constructing a classifier provided in an embodiment of the present invention.

[0038] Figure 10 It is used for explanation Figure 9 A detailed flowchart of step S912.

[0039] Figure 11 This is a flowchart of a method for classifying objects provided in an embodiment of the present invention.

[0040] Figure 12A , Figure 12B and Figure 12C This is a block diagram of a device for predicting user churn provided in an embodiment of the present invention.

[0041] Figure 13A , Figure 13B and Figure 13C This is a block diagram of a device for constructing a classification model provided in an embodiment of the present invention.

[0042] Figure 14A , Figure 14B and Figure 14C This is a block diagram of a device for classifying objects provided in an embodiment of the present invention.

[0043] Figure 15 This is a block diagram of a device for predicting user churn provided in an embodiment of the present invention.

[0044] Figure 16 This is a block diagram of a device for constructing a classification model provided in an embodiment of the present invention.

[0045] Figure 17 This is a block diagram of a device for classifying objects provided in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0047] According to embodiments of the present invention, methods for predicting user churn, constructing classification models, and classifying objects are also provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0048] It should be noted that the method embodiments provided in the present invention can be executed on a mobile terminal, computer terminal or similar computing device. Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing the methods provided in embodiments of the present invention is shown. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0049] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0050] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the above-mentioned application vulnerability detection method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0051] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0052] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0053] Figure 1 The hardware structure block diagram shown can serve as an exemplary block diagram not only for the aforementioned computer terminal 10 (or mobile device), but also as an exemplary block diagram for the aforementioned server.

[0054] in, Figure 2 A schematic diagram illustrating a specific application scenario of an embodiment of the present invention is shown. (Reference) Figure 2 As shown, users can use the client of an application (such as an online game, e-commerce, telecommunications service, WeChat, etc.) on computer terminal 201. Furthermore, computer terminal 201 can access server 202 through the client, and server 202 can retrieve information about the customer from database 203. In this embodiment, the method described can run on computer terminal 201, or on server 202 or database 203.

[0055] Under the aforementioned operating environment, one aspect of the embodiments of this application provides, as follows: Figure 3 The method shown is for predicting user churn. Figure 3 A flowchart illustrating a method for predicting user churn according to one aspect of an embodiment of this application is shown, with reference to Figure 3 As shown, the method includes:

[0056] S302: Obtain attribute information for multiple user attributes, where the multiple attributes are attributes included in an attribute set; and

[0057] S304: Based on attribute information, use a classification model trained by machine learning to predict user churn. The classification model includes a first classifier model and a second classifier model.

[0058] The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are distinct. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. These new attributes are formed by combining at least a subset of attributes from the second subset.

[0059] Therefore, the method described above, based on user attribute information, uses a classification model trained through machine learning to predict user churn. The selected attributes are drawn from an attribute set, which may include at least one of the following: user's age, user's gender, user's geographic location, user's occupation, total duration of user's service usage, duration of the user's most recent service usage, time elapsed since the user's most recent service usage, user's total spending, and user's most recent spending.

[0060] However, existing classification models, due to the use of greedy algorithms during classifier training, can only utilize a subset of attributes (i.e., strong attributes, corresponding to the first subset mentioned above) for classification operations and user churn prediction. Consequently, the classifier model in the classification model cannot fully utilize user attributes for prediction when making churn predictions based on user attributes, resulting in inaccurate measurement results.

[0061] In the method described above in this embodiment, the classification model used includes not only a classifier model trained using attributes from the attribute set (i.e., the first classifier model), but also a second classifier model. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of attributes from a second subset (corresponding to weak attributes) that differ from the first subset (i.e., strong attributes). This allows the classification model to fully utilize user attributes for training, resulting in more accurate measurement results when predicting user churn.

[0062] Furthermore, as a specific example, the first classifier model and the second classifier model can be a classification tree model or a neural network-based classifier model. Moreover, the combination of at least a portion of the attributes in the second subset can be done in pairs or by combining any number of attributes.

[0063] Optionally, the first subset and the second subset are determined by the following operations: training a first classifier model based on the attributes in the attribute set; determining the first subset based on the attributes used for classification by the first classifier model; and removing the first subset from the attribute set to obtain the second subset.

[0064] Specifically, refer to Figure 4 As shown, by inputting user attribute information into classification model 400, a prediction result regarding user churn is obtained. Classification model 400 includes classifier model 410 (i.e., the first classifier model) and classifier model 420 (i.e., the second classifier model).

[0065] Assuming the user's attributes are used θ i Indicate, for example

[0066] θ 1 represents the user's age;

[0067] θ 2 indicates the user's gender;

[0068] θ 3 indicates the user's geographic location;

[0069] θ 4 indicates "the user's profession";

[0070] θ 5 represents the total duration of user engagement with the service;

[0071] θ 6 indicates "the duration of the user's most recent use of the service";

[0072] θ 7 represents "the time elapsed since the user last used the service";

[0073] θ 8 represents "the user's total payment amount";

[0074] θ 9 represents "the user's most recent payment amount".

[0075] User information can then be obtained using vectors ( θ 1, θ 2, θ 3, θ 4, θ 5, θ 6, θ 7, θ 8, θ 9) indicates.

[0076] in, Figure 5 This example dataset is shown for predicting user churn for a specific business. The dataset contains attribute information for n users, some of whom are churned and others are not.

[0077] Furthermore, the data and attribute set of the above sample dataset { θ 1, θ 2, θ 3, θ 4, θ 5, θ 6, θ 7, θ 8, θ The attributes in 9} are related.

[0078] To predict user churn, a classifier model 410 is trained based on the attributes in the aforementioned attribute set, resulting in a classification model 400. Due to the use of a greedy strategy and other reasons during the training of the classifier model 410, some attributes (i.e., weak attributes) are not used by the classifier model 410 for classification operations.

[0079] For example, classifier model 410 will base its classification on attributes. θ 1 (User's age) θ 2 (User's gender) θ 4 (User's profession) and θ 7 (The time elapsed since the user last used the service) (i.e., the first subset { θ 1, θ 2, θ 4, θ 7) Predict user churn.

[0080] However, other attributes { θ 3, θ 5, θ 6, θ 8, θ 9} (i.e., the second subset) becomes the weak attributes that were not selected by the classifier model 410 for classification. However, these weak attributes, when combined, still possess an effect no less powerful than strong attributes. Therefore, at least some attributes in the second subset are combined to form new attributes. For example... θ 6 and θ 9 are combined to form new attributes. c 1; θ 6 and θ 8 are combined to form new attributes. c 2; θ 5 and θ 6 are combined to form new attributes. c 3.

[0081] Therefore, based on the new attribute ( c 1, c 2, c 3) The sample dataset is added to the initial sample dataset, thus obtaining the attribute-based { θ 1, θ 2, θ 3, θ 4, θ 5, θ 6, θ 7, θ 8, θ 9, c 1, c 2, c The sample dataset 3} is used to train the classifier model 420.

[0082] Of course, the above process can be continued to form new attributes and update the sample dataset to train a new classifier model until the termination condition is met.

[0083] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0084] Continue to refer to Figure 4 As shown, in order to base on the second subset { θ 3, θ 5, θ 6, θ 8, θThe new attribute is determined based on the attribute of 9} (i.e., the weak attribute). Then, training is performed based on the weak attribute (i.e., training using the sample dataset associated with the aforementioned weak attribute) to obtain classifier model 430 (i.e., the third classifier model). Furthermore, at least a portion of the attributes in the aforementioned second subset are combined according to classifier model 430 to form the new attribute. However, those skilled in the art will understand that forming a new attribute based on classifier model 430 is merely one method for forming a new attribute based on attributes in the second subset. In fact, there are other methods for forming a new attribute based on attributes in the second subset. For example, a portion of attributes can be randomly selected from the second subset and combined to form a new attribute.

[0085] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree. As a specific example of a classification tree, the first classifier model, the second classifier model, and the third classifier model may include at least one XGBoost boosting tree. Of course, the first classifier model, the second classifier model, and the third classifier model may also include other types of classification trees.

[0086] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0087] Figure 6 A schematic diagram of classifier model 430 is shown. (Reference) Figure 6 As shown, classifier model 430 includes multiple classification tree models 431, 432, 433, 434, 435... The classification trees in classifier model 430 all utilize the second subset { θ 3, θ 5, θ 6, θ 8, θ The attributes in '9' (i.e., weak attributes) were used for training. Therefore, the classification trees described above are all based on a subset of attributes from the second subset.

[0088] Furthermore, Figure 7 A schematic diagram of classification tree 431 is shown. Although no schematic diagrams of other classification trees are given in this embodiment, those skilled in the art can conceive of other classification trees 432, 433, 434, 435, etc., based on the description of classification tree 431.

[0089] like Figure 7As shown, classification tree 431 consists of 5 nodes N1 to N5 and 7 paths. The endpoint of each path represents the probability of user churn along that path. P1 to P7 As an example, the predetermined number of attributes mentioned above can be two adjacent attributes. Therefore, in Figure 7 In this context, we can determine the attribute combinations formed by two adjacent attributes along the path, including: ( θ 5, θ 6), ( θ 6, θ 8), ( θ 5, θ 8) and ( θ 5, θ 9). Therefore, the above four attribute combinations can be counted from classification tree 431, and each attribute combination appears once.

[0090] The same operation is performed on all classification trees in classifier model 430 to count the attribute combinations appearing in the entire classifier model 430 and the frequency of each attribute combination appearing in classifier model 430. Based on this, attribute combinations with a frequency higher than a predetermined frequency are selected to form new attributes. For example, the top three most frequent attribute combinations can be selected to form new attributes. c 1. c 2 and c 3.

[0091] In other words, optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model 430 includes: obtaining the number of occurrences of each attribute combination in each classification tree 431, 432, 433, 434, 435...; and adding the number of occurrences of each attribute combination in each classification tree 431, 432, 433, 434, 435... to obtain the number of occurrences of each attribute combination in the third classifier model 430.

[0092] Optionally, the operation of obtaining the number of times each attribute combination appears in each classification tree 431, 432, 433, 434, 435... includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears in the classification tree.

[0093] In summary, the classification model used in this embodiment includes not only a classifier model trained using attributes from the attribute set (i.e., the first classifier model), but also a second classifier model. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of attributes from a second subset that differs from the first subset. This allows the classification model to fully utilize user attributes for training, resulting in more accurate measurement results when predicting user churn.

[0094] It's important to note that this method can be used not only for user churn prediction but also in many other scenarios requiring forecasting. Examples include predicting whether a user will purchase a product in e-commerce, whether a user will click on an ad in an advertising system, and whether a user can become a new WeChat user, among others.

[0095] Furthermore, refer to Figure 8 As shown, according to another aspect of this embodiment, a method for constructing a classification model is also provided. Figure 8 This is a flowchart of a method for constructing a classification model according to another aspect of this embodiment. (See reference) Figure 8 As shown, the method includes:

[0096] S802: Obtain the first training data set, and determine the attribute set including multiple attributes based on the data in the first training data set;

[0097] S804: The first classifier model that trains the classification model using the first training dataset through machine learning;

[0098] S806: Determine a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification;

[0099] S808: Remove the attributes from the first subset of the attribute set to obtain the second subset;

[0100] S810: Combine at least a portion of the attributes in the second subset to form a new attribute; and

[0101] S812: Add the new property to the property set.

[0102] The method described above trains a classification model using machine learning based on the object's attribute information. However, in existing technologies, the classifier model trained using greedy algorithms or similar methods can only utilize a subset of attributes (the first subset mentioned above) for classification, thus failing to fully leverage the object's attributes for prediction and resulting in inaccurate classification results.

[0103] In the method described above in this embodiment, after training a classifier model (i.e., the first classifier model) using attributes from the attribute set of the training dataset, at least a portion of attributes from a second subset outside the first subset are combined to form new attributes. These new attributes are then added to the attribute set to facilitate training a classifier model for the classification model (i.e., the second classifier model). This allows the training process of the classification model to fully utilize the object's attributes, thereby obtaining correct classification results. Therefore, the method described above in this embodiment can solve the technical problem of inaccurate classification in existing classification models.

[0104] Furthermore, as a specific example, the first classifier model and the second classifier model can be a classification tree model or a neural network-based classifier model. Moreover, the combination of at least a portion of the attributes in the second subset can be done in pairs or by combining any number of attributes.

[0105] Optionally, the method also includes: training a second classifier model of the classification model using machine learning based on the set of attributes with the new attributes added.

[0106] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0107] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0108] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times as new attributes.

[0109] Optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model includes: obtaining the number of occurrences of each attribute combination on each classification tree; and adding the number of occurrences of each attribute combination on each classification tree to obtain the number of occurrences of each attribute combination in the third classifier model.

[0110] Optionally, the operation of obtaining the number of times each attribute combination appears on each classification tree includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0111] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0112] The inventors discovered in their practical work that the same problem exists not only in predicting user churn but also in other applications using classifiers. Therefore, based on the method described above, after training a classifier model using attributes from the attribute set of the training dataset (i.e., the first classifier model), at least a portion of attributes from a second subset outside the first subset are combined to form new attributes. These new attributes are then used to train a new classifier model (i.e., the second classifier model). This allows the training process of the classification model to fully utilize the object's attributes, thereby obtaining correct classification results. Therefore, the method described above in this embodiment can solve the technical problem of inaccurate classification in existing classification models.

[0113] To facilitate a better understanding of the embodiments of the present invention, the following uses the prediction of user churn as an example to illustrate the method of constructing a classification model in the embodiments of the present invention. The classification model used is a boosting tree model (preferably XGBoost), and the attribute combinations that generate new attributes are attribute combinations in pairs.

[0114] refer to Figure 9 As shown, the method includes:

[0115] S902: Obtain the original model training set;

[0116] S904: Use the boosting tree algorithm to train on the model training set to obtain a boosting tree model (corresponding to the first classifier model).

[0117] S906: Determine whether the termination condition is met. If the result is "No", proceed to step S908. If the result is "Yes", end.

[0118] S908: In the model training set, delete the attributes used in the boosting tree model trained in S904 (corresponding to the first subset) to generate a weak attribute training set (corresponding to the second subset).

[0119] S910: Use the boosting tree algorithm to train on the weak attribute training set to obtain a boosting tree model (corresponding to the third classifier model).

[0120] S912: Find the pairwise combinations of weak attributes in the boosting tree model trained in S910, form new attributes, add them to the model training set to generate a new model training set, and then return to S904 for training.

[0121] The “termination condition” in step S906 includes: reaching the number of iterations, or the difference between the AUC of the newly generated boosting tree model in step 2) on the test set and the AUC of the boosting tree model generated in step 2) in the previous iteration on the same test set is lower than a given threshold.

[0122] Furthermore, the process for finding pairwise combinations of weak attributes in step S912 is referenced. Figure 10 As shown, it includes:

[0123] S1002: In each tree of the boosting tree model trained in step S910, count the number of times two adjacent attributes appear together on each path;

[0124] S1004: Add up the number of times two adjacent attributes appear together on all paths to get the number of times two adjacent attributes appear together in each tree.

[0125] S1006: Add up the number of times two adjacent attributes appear together in each tree to get the number of times two adjacent attributes appear together in the boosted tree.

[0126] S1008: Sort the attributes that appear together in the lifting tree in descending order of frequency. Select the top n attributes that appear together most frequently. For each pair of adjacent attributes, generate a weak attribute combination. This weak attribute combination is the combination of the two adjacent attributes. Then proceed to step S904.

[0127] Therefore, by using the above methods, we can enhance the attribute selection ability of the tree model to extract the weak attributes that were not selected, and then leverage the attribute combination ability of the classification model to combine these weak attributes for use. This method not only controls the number of candidate weak attribute combinations generated but also ensures the quality of these combinations, utilizing weak attribute information in a fast and effective way, thereby improving prediction accuracy.

[0128] also, Figure 11 A flowchart illustrating a method for classifying objects according to another aspect of this embodiment is shown. (Reference) Figure 11 As shown, the method includes:

[0129] S1102: Retrieve attribute information for multiple attributes from object data, where the multiple attributes are attributes included in an attribute set; and

[0130] S1104: Based on attribute information, objects are classified using a classification model trained by machine learning. The classification model includes a first classifier model and a second classifier model.

[0131] The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are different from each other. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0132] As mentioned earlier, in the existing technology, when training the classifier model in the classification model, due to the use of greedy algorithms and other reasons, the trained classifier model can only use a portion of the attributes in the attribute set (i.e., the first subset mentioned above) for classification operations, thus failing to fully utilize the attributes of the object for prediction, resulting in inaccurate classification results.

[0133] In the method described above in this embodiment, the classification model used includes not only a classifier model trained using attributes from the attribute set (i.e., the first classifier model), but also a second classifier model. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of attributes from a second subset that differs from the first subset. This allows the classification model to fully utilize the object's attributes for training, resulting in more accurate measurement results during prediction.

[0134] Furthermore, as a specific example, the first classifier model and the second classifier model can be a classification tree model or a neural network-based classifier model. Moreover, the combination of at least a portion of the attributes in the second subset can be done in pairs or by combining any number of attributes.

[0135] Optionally, the first subset and the second subset are determined by the following operations: training a first classifier model based on the attributes in the attribute set; determining the first subset based on the attributes used for classification by the first classifier model; and removing the first subset from the attribute set to obtain the second subset.

[0136] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0137] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0138] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0139] Optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model includes: obtaining the number of occurrences of each attribute combination on each classification tree; and adding the number of occurrences of each attribute combination on each classification tree to obtain the number of occurrences of each attribute combination in the third classifier model.

[0140] Optionally, the operation of obtaining the number of times each attribute combination appears on each classification tree includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0141] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0142] Therefore, in the method described above in this embodiment, the classification model used includes not only a classifier model trained using attributes from the attribute set (i.e., the first classifier model), but also a second classifier model. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of attributes from a second subset that differs from the first subset. This allows the classification model to fully utilize the object's attributes for training, resulting in more accurate measurement results during prediction.

[0143] In addition, refer to Figure 1 As shown, this embodiment also provides a storage medium (e.g., Figure 1 (The memory shown). The storage medium includes a stored program, wherein, when the program is executed, the control processor 102 performs any of the methods described above.

[0144] Optionally, the method executed by the program includes: obtaining attribute information of multiple user attributes, wherein the multiple attributes are attributes included in an attribute set; and predicting user churn based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are distinct. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0145] Optionally, the method performed by the program includes: acquiring a first training data set, and determining an attribute set including multiple attributes based on the data in the first training data set; training a first classifier model of the classification model using machine learning with the first training data set; determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; removing attributes from the attribute set in the first subset to obtain a second subset; combining at least some attributes in the second subset to form new attributes; and training a second classifier model of the classification model using machine learning based on the attribute set after adding the new attributes.

[0146] Optionally, the method executed by the program includes: obtaining attribute information of multiple attributes from object data, wherein the multiple attributes are attributes included in an attribute set; and classifying the object based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are different from each other. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0147] Optionally, the first subset and the second subset are determined by the following operations: training a first classifier model based on the attributes in the attribute set; determining the first subset based on the attributes used for classification by the first classifier model; and removing the first subset from the attribute set to obtain the second subset.

[0148] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0149] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0150] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0151] Optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model includes: obtaining the number of occurrences of each attribute combination on each classification tree; and adding the number of occurrences of each attribute combination on each classification tree to obtain the number of occurrences of each attribute combination in the third classifier model.

[0152] Optionally, the operation of obtaining the number of times each attribute combination appears on each classification tree includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0153] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0154] In addition, refer to Figure 12A , Figure 12B and Figure 12C As shown, this embodiment also provides devices 120A, 120B, and 120C for predicting user churn. Among them... Figure 12A , Figure 12B and Figure 12C The devices shown all follow similar inventive concepts, differing only slightly in their design structure. For more detailed structural information, please refer to... Figure 1 The terminal computer terminal 10 (or mobile device 10) shown is shown.

[0155] refer to Figure 12A As shown, device 120A includes: a transmission device; an input / output interface; and a processor. The transmission device and the input / output interface are used to acquire user data, and the processor runs a program. During program execution, the following processing steps are performed on data output from the transmission device and / or the input / output interface: acquiring attribute information of multiple user attributes, the multiple attributes being attributes included in an attribute set; and based on the attribute information, using a classification model trained through machine learning, predicting user churn for the user, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0156] refer to Figure 12BAs shown, device 120B includes: a transmission device; an input / output interface; and a processor. The transmission device and input / output interface are used to acquire user data, and the processor runs a program. During program execution, the following processing steps are performed on data output from the transmission device and / or the input / output interface: acquiring attribute information of multiple user attributes, where the multiple attributes are attributes included in an attribute set; and based on the attribute information, using a classification model trained through machine learning, predicting user churn. The classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0157] refer to Figure 12C As shown, device 120C includes: a processor; and a memory connected to the processor for providing the processor with instructions to perform the following processing steps: obtaining attribute information of multiple attributes from object data, the multiple attributes being attributes included in an attribute set; and classifying the object based on the attribute information using a classification model trained by machine learning, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0158] Optionally, the first subset and the second subset are determined by the following operations: training a first classifier model based on the attributes in the attribute set; determining the first subset based on the attributes used for classification by the first classifier model; and removing the first subset from the attribute set to obtain the second subset.

[0159] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0160] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0161] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0162] Optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model includes: obtaining the number of occurrences of each attribute combination on each classification tree; and adding the number of occurrences of each attribute combination on each classification tree to obtain the number of occurrences of each attribute combination in the third classifier model.

[0163] Optionally, the operation of obtaining the number of times each attribute combination appears on each classification tree includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0164] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0165] Therefore, in the device for predicting user churn described above in this embodiment, the classification model used includes not only a classifier model trained using attributes from the attribute set (i.e., the first classifier model), but also a second classifier model. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of attributes from a second subset that differs from the first subset. This allows the classification model to fully utilize the object's attributes for training, resulting in more accurate measurement results during prediction.

[0166] In addition, refer to Figure 13A , Figure 13B and Figure 13C As shown, this embodiment also provides devices 130A, 130B, and 130C for constructing classification models. Among them... Figure 13A , Figure 13B and Figure 13C The devices shown all follow similar inventive concepts, differing only slightly in their design structure. For more detailed structural information, please refer to... Figure 1 The terminal computer terminal 10 (or mobile device 10) shown is shown.

[0167] refer to Figure 13A The device 130A includes: a transmission device; an input / output interface; and a processor. The transmission device and the input / output interface are used to acquire data, and the processor runs a program. During program execution, the following processing steps are performed on the data output from the transmission device and / or the input / output interface: acquiring a first training data set, and determining an attribute set including multiple attributes based on the data in the first training data set; training a first classifier model using the first training data set; determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; removing attributes from the attribute set from the first subset to obtain a second subset; combining at least a portion of the attributes in the second subset to form a new attribute; and adding the new attribute to the attribute set.

[0168] refer to Figure 13B As shown, device 130B includes: a transmission device; an input / output interface; and a storage medium for storing a program. The transmission device and the input / output interface are used to acquire data. During runtime, the program performs the following processing steps on the data output by the transmission device and / or the input / output interface: acquiring a first training data set and determining an attribute set including multiple attributes based on the data in the first training data set; training a first classifier model using the first training data set; determining a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; removing attributes from the attribute set from the first subset to obtain a second subset; combining at least a portion of the attributes in the second subset to form a new attribute; and adding the new attribute to the attribute set.

[0169] refer to Figure 13C As shown, device 130C includes: a processor; and a memory connected to the processor for providing the processor with instructions to perform the following processing steps: acquiring a first training data set, and determining an attribute set including multiple attributes based on data in the first training data set; training a first classifier model for a classification model using the first training data set; determining a first subset of the attribute set based on the first classifier model. The attributes in the first subset are those used by the first classifier model for classification; removing attributes from the attribute set from the first subset to obtain a second subset; combining at least a portion of the attributes in the second subset to form new attributes; and adding the new attributes to the attribute set.

[0170] Optionally, the processing steps also include training a second classifier model of the classification model using machine learning based on the set of attributes with the added new attributes.

[0171] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0172] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0173] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times as new attributes.

[0174] Optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model includes: obtaining the number of occurrences of each attribute combination on each classification tree; and adding the number of occurrences of each attribute combination on each classification tree to obtain the number of occurrences of each attribute combination in the third classifier model.

[0175] Optionally, the operation of obtaining the number of times each attribute combination appears on each classification tree includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0176] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0177] Therefore, by using the aforementioned device, after training a classifier model (i.e., the first classifier model) using attributes from the attribute set of the training dataset, at least a portion of attributes from a second subset outside the first subset can be combined to form new attributes. These new attributes are then used to train a new classifier model (i.e., the second classifier model). This allows the training process of the classification model to fully utilize the object's attributes, thereby obtaining correct classification results. Therefore, the method described above in this embodiment can solve the technical problem of inaccurate classification in existing classification models.

[0178] In addition, refer to Figure 14A , Figure 14B and Figure 14CAs shown, this embodiment also provides devices 140A, 140B, and 140C for classifying objects. Among them... Figure 14A , Figure 14B and Figure 14C The devices shown all follow similar inventive concepts, differing only slightly in their design structure. For more detailed structural information, please refer to... Figure 1 The terminal computer terminal 10 (or mobile device 10) shown is shown.

[0179] refer to Figure 14A The device 140A includes: a transmission device; an input / output interface; and a processor. The transmission device and the input / output interface are used to acquire object data, and the processor runs a program. During program execution, the following processing steps are performed on data output from the transmission device and / or the input / output interface: acquiring attribute information of multiple attributes from the object data, where the multiple attributes are attributes included in an attribute set; and classifying the object based on the attribute information using a classification model trained through machine learning, where the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0180] refer to Figure 14B Device 140B includes: a transmission device; an input / output interface; and a storage medium for storing a program. The transmission device and the input / output interface are used to acquire object data. During runtime, the program performs the following processing steps on the data output by the transmission device and / or the input / output interface: acquiring attribute information of multiple attributes from the object data, where the multiple attributes are attributes included in an attribute set; and classifying the object based on the attribute information using a classification model trained through machine learning, where the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of the attributes in the second subset.

[0181] refer to Figure 14CThe device 140C includes: a processor; and a memory connected to the processor for providing the processor with instructions to perform the following processing steps: obtaining attribute information of multiple attributes from object data, the multiple attributes being attributes included in an attribute set; and classifying the object based on the attribute information using a classification model trained by machine learning, the classification model including a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are distinct. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0182] Optionally, the first subset and the second subset are determined by the following operations: training a first classifier model based on the attributes in the attribute set; determining the first subset based on the attributes used for classification by the first classifier model; and removing the first subset from the attribute set to obtain the second subset.

[0183] Optionally, the operation of forming a new attribute includes: training a third classifier model based on the attributes in the second subset; and combining at least a portion of the attributes in the second subset based on the third classifier model to form a new attribute.

[0184] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0185] Optionally, the operation of forming a new attribute includes: counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0186] Optionally, the operation of counting the number of occurrences of each attribute combination in the third classifier model includes: obtaining the number of occurrences of each attribute combination on each classification tree; and adding the number of occurrences of each attribute combination on each classification tree to obtain the number of occurrences of each attribute combination in the third classifier model.

[0187] Optionally, the operation of obtaining the number of times each attribute combination appears on each classification tree includes performing the following operations on each classification tree: counting the number of times each attribute combination appears on each path of the classification tree; and summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0188] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0189] In addition, refer to Figure 15 As shown, according to one aspect of an embodiment of the present invention, a device 150 for predicting user churn is provided, comprising: an acquisition module 1501, configured to acquire attribute information of multiple attributes of a user, wherein the multiple attributes are attributes included in an attribute set; and a prediction module 1502, configured to predict user churn based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and the first classifier model performs a classification operation based on the attributes in the first subset, and the second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0190] Optionally, the device 150 includes: a first unit for training a first classifier model based on attributes in an attribute set; a second unit for determining a first subset based on the attributes used for classification by the first classifier model; and a third unit for removing the first subset from the attribute set to obtain a second subset.

[0191] Optionally, the device 150 includes: a fourth unit for training a third classifier model based on attributes in the second subset; and a fifth unit for combining at least a portion of the attributes in the second subset based on the third classifier model to form new attributes.

[0192] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0193] Optionally, the fourth unit includes: a first device for counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and a second device for selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0194] Optionally, the first device includes: a first component for obtaining the number of times each attribute combination appears in each classification tree; and a second component for summing the number of times each attribute combination appears in each classification tree to obtain the number of times each attribute combination appears in the third classifier model.

[0195] Optionally, the first component includes: a first sub-component for counting the number of times each attribute combination appears on each path of the classification tree; and a second sub-component for summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0196] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0197] In addition, refer to Figure 16 As shown, another aspect of the present invention provides an apparatus 160 for constructing a classification model, comprising: an attribute determination module 1601, configured to acquire a first training data set and determine an attribute set including multiple attributes based on data in the first training data set; a training module 1602, configured to train a first classifier model of the classification model using the first training data set through machine learning; a first subset determination module 1603, configured to determine a first subset of the attribute set based on the first classifier model, wherein the attributes in the first subset are the attributes used by the first classifier model for classification; a second subset determination module 1604, configured to remove attributes from the attribute set in the first subset to obtain a second subset; a combination module 1605, configured to combine at least a portion of the attributes in the second subset to form a new attribute; and an addition module 1606, configured to add the new attribute to the attribute set.

[0198] Optionally, the device 160 includes: a first unit for training a third classifier model based on attributes in a second subset; and a second unit for combining at least a portion of the attributes in the second subset based on the third classifier model to form new attributes.

[0199] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0200] Optionally, the second unit includes: a first device for counting the number of occurrences of each attribute combination in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and a second device for selecting attribute combinations that occur more than a predetermined number of times as new attributes.

[0201] Optionally, the first device includes: a first component for obtaining the number of times each attribute combination appears in each classification tree; and a second component for summing the number of times each attribute combination appears in each classification tree to obtain the number of times each attribute combination appears in the third classifier model.

[0202] Optionally, the first component includes: a first sub-component for counting the number of times each attribute combination appears on each path of the classification tree; and a second sub-component for summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0203] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0204] In addition, refer to Figure 17 As shown, according to one aspect of an embodiment of the present invention, a device 170 for predicting user churn is provided, comprising: an acquisition module 1701, configured to acquire attribute information of multiple attributes from object data, wherein the multiple attributes are attributes included in an attribute set; and a classification module 1702, configured to classify the object based on the attribute information using a classification model trained by machine learning, wherein the classification model includes a first classifier model and a second classifier model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set, and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset.

[0205] Optionally, the device 170 includes: a first unit for training a first classifier model based on attributes in an attribute set; a second unit for determining a first subset based on the attributes used for classification by the first classifier model; and a third unit for removing the first subset from the attribute set to obtain a second subset.

[0206] Optionally, the device 170 includes: a fourth unit for training a third classifier model based on attributes in the second subset; and a fifth unit for combining at least a portion of the attributes in the second subset based on the third classifier model to form new attributes.

[0207] Optionally, the first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

[0208] Optionally, the fourth unit includes: a first device for counting the number of times each attribute combination appears in the third classifier model, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and a second device for selecting attribute combinations that appear more than a predetermined number of times to form a new attribute.

[0209] Optionally, the first device includes: a first component for obtaining the number of times each attribute combination appears in each classification tree; and a second component for summing the number of times each attribute combination appears in each classification tree to obtain the number of times each attribute combination appears in the third classifier model.

[0210] Optionally, the first component includes: a first sub-component for counting the number of times each attribute combination appears on each path of the classification tree; and a second sub-component for summing the number of times each attribute combination appears on each path to obtain the number of times each attribute combination appears on the classification tree.

[0211] Optionally, the multiple attributes in the attribute set include at least one of the following: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service use, duration of user's most recent service use, time elapsed since user's most recent service use, user's total payment amount, and user's most recent payment amount.

[0212] Therefore, in the device described above in this embodiment, the classification model used includes not only a classifier model trained using attributes from the attribute set (i.e., the first classifier model), but also a second classifier model. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of attributes from a second subset that differs from the first subset. This allows the classification model to fully utilize the object's attributes for training, resulting in more accurate measurement results during prediction.

[0213] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0214] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0215] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0216] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0217] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0218] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0219] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting user churn, characterized in that, include: Obtain attribute information for multiple attributes of a user, wherein the multiple attributes are attributes included in an attribute set; as well as Based on the attribute information, a classification model trained through machine learning is used to predict user churn. The classification model includes a first classifier model and a second classifier model. Classification operations are performed on each of the first and second classifier models to obtain the prediction result output by the classification model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are mutually exclusive. The first classifier model is trained based on the attributes in the attribute set. The first classifier model performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes. The new attributes are formed by combining at least a portion of the attributes in the second subset. The samples used for training the first and second classifiers both include attribute sets corresponding to churned users and attribute sets corresponding to non-churned users. The multiple attributes in the attribute set include: user's age, user's gender, user's geographical location, user's occupation, user's total service usage time, user's most recent service usage time, the time since the user's most recent service usage, user's total payment amount, and user's most recent payment amount.

2. The method according to claim 1, characterized in that, The first subset and the second subset are determined through the following operations: Train the first classifier model based on the attributes in the attribute set; The first subset is determined based on the attributes used by the first classifier model for classification. And remove the first subset from the attribute set to obtain the second subset.

3. The method according to claim 2, characterized in that, The operations for forming the new attribute include: Train a third classifier model based on the attributes in the second subset; and The new attribute is formed by combining at least some of the attributes in the second subset according to the third classifier model.

4. The method according to claim 3, characterized in that, The first classifier model, the second classifier model, and the third classifier model each include at least one classification tree.

5. The method according to claim 4, characterized in that, The operations for forming the new attribute include: The frequency of each attribute combination in the third classifier model is counted, wherein the attribute combination is a combination of a predetermined number of adjacent attributes on the path of the classification tree of the third classifier model; and attribute combinations that appear more than a predetermined number of times are selected to form the new attribute.

6. The method according to claim 5, characterized in that, The operation of counting the frequency of each attribute combination in the third classifier model includes: Obtain the number of times each attribute combination appears in each classification tree; and The number of times each attribute combination appears on each classification tree is added together to obtain the number of times each attribute combination appears in the third classifier model.

7. The method according to claim 6, characterized in that, The operation to obtain the number of times each attribute combination appears in each classification tree includes performing the following operations on each classification tree: Count the number of times each attribute combination appears on every path in the classification tree; as well as The number of times each attribute combination appears on each path is added together to obtain the number of times each attribute combination appears on the classification tree.

8. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the processor to perform the method according to any one of claims 1 to 7.

9. A processor, characterized in that, The processor is used to run a program, wherein the program executes the method according to any one of claims 1 to 7 when it runs.

10. A device for predicting user churn, characterized in that, include: The acquisition module is used to acquire attribute information of multiple attributes of the user, wherein the multiple attributes are attributes included in the attribute set; as well as The prediction module is used to predict user churn based on the attribute information and using a classification model trained by machine learning. The classification model includes a first classifier model and a second classifier model. Classification operations are performed on the first and second classifier models respectively to obtain the prediction result output by the classification model. The attribute set includes a first subset and a second subset, and the attributes included in the first subset and the second subset are different from each other. The first classifier model is trained based on the attributes in the attribute set and performs classification operations based on the attributes in the first subset. The second classifier model is trained based on the attribute set after adding new attributes, where the new attributes are formed by combining at least a portion of the attributes in the second subset. The samples used for training the first and second classifiers both include attribute sets corresponding to churned users and attribute sets corresponding to non-churned users. The multiple attributes in the attribute set include: user's age, user's gender, user's geographical location, user's occupation, total duration of user's service usage, duration of user's most recent service usage, time elapsed since the user's most recent service usage, user's total payment amount, and user's most recent payment amount.

Images