Object classification method and apparatus, computer device and storage medium

Abstract

The application relates to an object classification method and device, computer equipment, a storage medium and a computer program product. Embodiments of the application can be applied to a vehicle-mounted scene. The method comprises: obtaining a plurality of training sample subsets obtained through random sampling; constructing a corresponding initial decision tree based on the training sample subsets, wherein a decision tree node of the initial decision tree is determined based on a randomly selected feature category from each feature category corresponding to the training samples; inputting the training samples in the training sample subsets into the corresponding initial decision tree to obtain initial prediction labels; adjusting the corresponding initial decision tree based on the training labels and the initial prediction labels corresponding to each training sample in the same training sample subset until a first convergence condition is met, to obtain target decision trees; and generating an object classification model based on the target decision trees, so that the prediction labels corresponding to target objects are obtained according to the prediction results of the target objects by the target decision trees, thereby improving the object classification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an object classification method and device, computer equipment, storage medium and computer program product. BACKGROUND

[0002] With the development of computer technology, more and more application programs emerge in the network. In order to maintain the network environment and protect the physical and mental health of the object, the operation permission of different types of objects in the same application program can be distinguished.

[0003] In the traditional technology, different types of objects are usually identified by experience rules. For example, for the same application program, the operation time of different types of objects is different, and the object operating the application program at a specific time is identified as a specific type of object corresponding to the specific time. However, the object classification method based on experience rules has great subjectivity and low classification accuracy. SUMMARY

[0004] Therefore, it is necessary to provide an object classification method, device, computer equipment, computer readable storage medium and computer program product capable of improving classification accuracy to solve the above technical problems.

[0005] The present application provides an object classification method. The method comprises:

[0006] obtaining a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set, the training sample subset comprises a training sample and a training label corresponding to the training sample, the training sample is obtained based on operation data of a training object in a target application, and the training label is used to determine the operation permission of the training object in the target application;

[0007] constructing a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset; the decision tree node of the initial decision tree is determined based on a randomly selected feature category from feature categories corresponding to each training feature contained in the training sample;

[0008] inputting the training sample in the training sample subset into the corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample;

[0009] adjusting the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until a first convergence condition is met, to obtain a target decision tree corresponding to each training sample subset;

[0010] generate an object classification model based on the target decision trees; the object classification model is configured to input target features corresponding to a target object into the target decision trees, and obtain a target prediction label corresponding to the target object based on prediction results of the target decision trees.

[0011] The application further provides an object classification device. The device comprises:

[0012] a training set acquisition module configured to acquire a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set, and comprises training samples and training labels corresponding to the training samples; the training samples are obtained based on operation data of a training object in a target application; and the training labels are used to determine operation permissions of the training object in the target application;

[0013] an initial decision tree construction module configured to construct decision trees based on the training sample subsets, and obtain initial decision trees corresponding to the training sample subsets respectively; a decision tree node of the initial decision trees is determined based on randomly selected feature categories from feature categories corresponding to each training feature contained in the training samples;

[0014] a decision tree prediction module configured to input the training samples in the training sample subsets into the corresponding initial decision trees, and obtain initial prediction labels corresponding to each training sample;

[0015] a target decision tree generation module configured to adjust the corresponding initial decision trees based on the training labels and the initial prediction labels corresponding to each training sample in a same training sample subset, until a first convergence condition is met, and obtain target decision trees corresponding to the training sample subsets respectively;

[0016] a model generation module configured to generate an object classification model based on the target decision trees; the object classification model is configured to input target features corresponding to a target object into the target decision trees, and obtain a target prediction label corresponding to the target object based on prediction results of the target decision trees.

[0017] The application provides an object classification method. The method comprises:

[0018] acquiring target features corresponding to a target object; the target features are obtained based on operation data of the target object in a target application;

[0019] inputting the target features into an object classification model, and obtaining a target prediction label corresponding to the target object; the target prediction label is used to determine operation permissions of the target object in the target application; and the target prediction label is obtained based on prediction results of each target decision tree in the object classification model;

[0020] The training process of the object classification model comprises the following steps:

[0021] obtaining a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set; constructing a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset; a decision tree node of the initial decision tree is determined from feature categories corresponding to each training feature contained in the training sample based on a randomly selected feature category; inputting a training sample in the training sample subset into the corresponding initial decision tree to obtain an initial predicted label corresponding to each training sample; adjusting the corresponding initial decision tree based on a training label and the initial predicted label corresponding to each training sample in the same training sample subset until a first convergence condition is met to obtain a target decision tree corresponding to each training sample subset; and generating an object classification model based on each target decision tree.

[0022] The application also provides an object classification device. The device comprises:

[0023] a data acquisition module configured to acquire a target feature corresponding to a target object, the target feature being obtained based on operation data of the target object in a target application;

[0024] a label prediction module configured to input the target feature into an object classification model to obtain a target predicted label corresponding to the target object; the target predicted label is used to determine an operation permission of the target object in the target application, and the target predicted label is obtained based on a prediction result of each target decision tree in the object classification model;

[0025] The training process of the object classification model comprises the following steps:

[0026] obtaining a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set; constructing a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset; a decision tree node of the initial decision tree is determined from feature categories corresponding to each training feature contained in the training sample based on a randomly selected feature category; inputting a training sample in the training sample subset into the corresponding initial decision tree to obtain an initial predicted label corresponding to each training sample; adjusting the corresponding initial decision tree based on a training label and the initial predicted label corresponding to each training sample in the same training sample subset until a first convergence condition is met to obtain a target decision tree corresponding to each training sample subset; and generating an object classification model based on each target decision tree.

[0027] A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the various data query methods described above when executing the computer program.

[0028] A computer-readable storage medium stores a computer program, and the computer program implements the steps of the various data query methods described above when executed by a processor.

[0029] A computer program product includes a computer program, and the computer program implements the steps of the various data query methods described above when executed by a processor.

[0030] The object classification method, device, computer device, storage medium, and computer program product described above, by obtaining a plurality of training sample subsets, each training sample subset is obtained by randomly sampling the same training sample set, the training sample subset includes a training sample and a training label corresponding to the training sample, the training sample is obtained based on operation data of a training object in a target application, and the training label is used to determine the operation permission of the training object in the target application. Based on the training sample subset, a decision tree is constructed to obtain an initial decision tree corresponding to each training sample subset. The decision tree node of the initial decision tree is determined based on a randomly selected feature category from the feature categories corresponding to each training feature included in the training sample. The training sample in the training sample subset is input into the corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample. Based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset, the corresponding initial decision tree is adjusted until a first convergence condition is met, and a target decision tree corresponding to each training sample subset is obtained. An object classification model is generated based on each target decision tree. Subsequently, a target feature corresponding to a target object can be input into the object classification model, and the object classification model outputs a prediction label corresponding to the target object based on the prediction results of each target decision tree. In this way, the object classification model is trained based on the training sample set, and the object is classified based on the object classification model, which can improve the classification accuracy and efficiency of the object. Moreover, the object classification model includes a plurality of target decision trees, and different training sample subsets can be obtained by randomly sampling the same training sample set. Different target decision trees can be trained based on different training sample subsets through random feature selection. The object classification model can output more accurate prediction labels based on the prediction results of different target decision trees, further improving the classification accuracy of the object. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 An application environment diagram of the object classification method in one embodiment;

[0032] Figure 2 A flowchart of the object classification method in one embodiment;

[0033] Figure 3 This is a schematic diagram of the process of training a decision tree in one embodiment;

[0034] Figure 4 This is a flowchart illustrating the object classification method in another embodiment;

[0035] Figure 5 This is a schematic diagram of the process for identifying game players in one embodiment;

[0036] Figure 6 This is a schematic diagram illustrating the training and application of an ensemble model in one embodiment;

[0037] Figure 7 This is a schematic diagram of model training in one embodiment;

[0038] Figure 8 This is a schematic diagram of player features in one embodiment;

[0039] Figure 9 This is a structural block diagram of an object classification device in one embodiment;

[0040] Figure 10 This is a structural block diagram of an object classification device in another embodiment;

[0041] Figure 11 This is a structural block diagram of the object classification device in another embodiment;

[0042] Figure 12 This is an internal structural diagram of a computer device in one embodiment;

[0043] Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] The object classification method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another server.

[0046] The terminal 102 and the server 104 can be used cooperatively to perform the object classification method provided in the embodiments of the present application. For example, the server obtains a plurality of training sample subsets from the terminal, each training sample subset being obtained by randomly sampling a same training sample set, the training sample subset including a training sample and a training label corresponding to the training sample, the training sample being obtained based on operation data of a training object in a target application, and the training label being used to determine an operation permission of the training object in the target application. The server constructs a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset, a decision tree node of the initial decision tree being determined based on a randomly selected feature category from feature categories corresponding to each training feature included in the training sample. The server inputs the training sample in the training sample subset into the corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample, and adjusts the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until a first convergence condition is met, to obtain a target decision tree corresponding to each training sample subset, and generates an object classification model based on each target decision tree. The server sends the object classification model to the terminal. The terminal obtains a target feature corresponding to a target object, inputs the target feature into each target decision tree in the object classification model, and obtains a prediction label corresponding to the target object based on a prediction result of each target decision tree in the object classification model, and the object classification model outputs the target prediction label. The terminal can determine an operation permission of the target object in the target application based on the target prediction label, so as to control the operation of the target object in the target application. The target feature is obtained based on operation data of the target object in the target application.

[0047] The terminal 102 and the server 104 can also be used independently to perform the object classification method provided in the embodiments of the present application. For example, the terminal trains an object classification model based on training data, and obtains a target prediction label corresponding to a target object based on the object classification model.

[0048] The terminal 102 can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle-mounted device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The terminal is installed with a target application. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers or a cloud server.

[0049] The embodiments of the present application can be applied to various scenes such as cloud technology, artificial intelligence, intelligent transportation, assisted driving, and games.

[0050] In one embodiment, as Figure 2As shown, a subject classification method is provided, which is applied to a computer device for example, which can be the terminal 102 or the server 104 in the above Figure 1 . Referring to Figure 2 , the subject classification method comprises the following steps:

[0051] In step S202, a plurality of training sample subsets are obtained; each training sample subset is obtained by randomly sampling a same training sample set, the training sample subset comprises a training sample and a training label corresponding to the training sample, the training sample is obtained based on operation data of a training subject in a target application, and the training label is used to determine the operation permission of the training subject in the target application.

[0052] The target application refers to an application program that needs to distinguish the operation permission of different types of subjects. The target application can be a game application, a video application, an e-commerce application, etc. For example, the target application can be a game application, and the operation permission of adult players and underage players in the game application is different, and the operation time of underage players in the game application is limited. The target application can also be a video application, and the operation permission of adults and minors in the video application is also different, and the video watching time and video watching type of minors in the video application can be limited.

[0053] The training sample is obtained based on the operation data of the training subject in the target application. One training sample can be composed of a plurality of operation features corresponding to one training subject. The operation feature is obtained by feature extraction on the operation data. The operation feature is used to reflect the use of the training subject in the target application. Each training sample has a corresponding training label. The training label is used to determine the operation permission of the training subject in the target application. For example, if the training label is a negative label, it means that the operation permission of the training subject in the target application is unrestricted, and if the training label is a positive label, it means that the operation permission of the training subject in the target application is restricted.

[0054] The training sample set comprises a plurality of training samples. Random sampling of the same training sample set can obtain a plurality of training sample subsets, each training sample subset can contain the same training sample, and can contain different training samples. For example, the training sample set can be randomly sampled with replacement, so that the same training sample can appear in different training sample subsets.

[0055] Specifically, the computer device can obtain a training sample set locally or from other terminals or servers, randomly sample the training sample set to obtain multiple training sample subsets, and then train a model based on each training sample subset to obtain an object classification model composed of multiple target decision trees. Subsequently, the computer device can classify any object based on the object classification model, thereby controlling the operation of the object in the target application, achieving the purpose of maintaining the network environment and protecting the object.

[0056] In step S204, a decision tree is constructed based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset. The decision tree nodes of the initial decision tree are determined based on randomly selected feature categories from the feature categories corresponding to each training feature included in the training sample.

[0057] The decision tree is a tree structure in which each internal node represents a test on an attribute or feature, each branch represents a test output, and each leaf node represents a class.

[0058] Specifically, the computer device can construct a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset. When constructing the decision tree, the computer device can generate a decision tree node based on a subset of features. When generating the decision tree node, the computer device can randomly select multiple feature categories as candidate feature categories from the feature categories corresponding to each training feature included in the training sample, and determine a target feature category from the candidate feature categories to generate the decision tree node. It can be understood that since the decision tree node is determined based on randomly selected feature categories, the initial decision tree corresponding to each training sample subset is different. In addition, each training sample subset is also different, and then the corresponding initial decision tree is trained based on the training sample subset to obtain a target decision tree, and the target decision trees corresponding to each training sample subset are also different. However, each target decision tree has a certain prediction ability, so the object classification model generated based on each target decision tree can have high accuracy and generalization performance.

[0059] In an embodiment, one feature category can be randomly selected from the candidate feature categories as the target feature category. The feature importance of each candidate feature category can also be calculated based on the training sample subset, and the target feature category can be determined from the candidate feature categories based on the feature importance. The feature importance is used to represent the influence of the feature on the prediction result, and can also be regarded as the classification ability of the feature. For example, the training sample subset can be divided into a first sample subset and a second sample subset, the target split point corresponding to each candidate feature category is determined based on the first sample subset, the split accuracy of each target split point is calculated based on the second sample subset, the classification accuracy is taken as the feature importance, and the candidate feature category with the largest feature importance is selected as the target feature category. It can be understood that the feature space can be divided into two regions based on any target split point, for example, the candidate feature category is the operation duration, the target split point is whether the operation duration is greater than 10 hours, the training samples with the operation duration greater than 10 hours are divided into region A, and the training samples with the operation duration less than or equal to 10 hours are divided into region B. There is a significant difference in the distribution of training samples corresponding to the two regions divided based on the target split point. There are more positive training samples in one region, and more negative training samples in the other region. The region with more positive training samples can be considered to correspond to the positive label, and the region with more negative training samples can be considered to correspond to the negative label. The training samples in the second sample subset are divided based on the target split point, the split label corresponding to the training sample is determined according to the region into which the training sample falls, and the split accuracy is calculated based on the real label and the split label corresponding to each training sample in the second sample subset. When determining the target split point, the split point with the greatest difference in the distribution of training samples corresponding to the two regions divided can be taken as the target split point. Of course, the computer device can also calculate the feature importance based on other custom formulas or algorithms.

[0060] In an embodiment, a plurality of feature categories are randomly selected from the feature categories as candidate feature categories, the optimal split feature is determined from the candidate feature categories based on the feature importance, the optimal split point corresponding to the optimal split feature is further determined, the feature space is divided into two regions based on the optimal split point corresponding to the optimal split feature, and each training sample in the training sample subset is distributed into the two regions according to the value of the optimal split feature. The above division process is repeated for each region until the division stopping condition is met, and a plurality of regions are obtained. The decision tree node corresponding to the optimal split point of the optimal split feature used in each space division is generated, and each decision tree node is connected in the division order to obtain an initial decision tree. In each division, a plurality of feature categories can be randomly selected from the remaining feature categories (feature categories that are not optimal split features) as candidate feature categories, or a plurality of feature categories can be randomly selected from all feature categories as candidate feature categories.

[0061] Step S206, inputting the training samples in the training sample subset into the corresponding initial decision tree to obtain the initial prediction label corresponding to each training sample.

[0062] Step S208, adjusting the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until the first convergence condition is met to obtain the target decision tree corresponding to each training sample subset respectively.

[0063] Specifically, the computer device can take the training sample as the input data of the initial decision tree and take the training label as the expected output of the initial decision tree, adjust the decision tree parameters of the initial decision tree through multiple iterations of training, and take the final training decision tree as the target decision tree.

[0064] After generating the initial decision tree, the computer device can input the training samples in each training sample subset into the corresponding initial decision tree to obtain the initial prediction label corresponding to each training sample in each training sample subset. After inputting the data into the initial decision tree, the initial decision tree can output the initial prediction label based on the decision tree nodes and node parameters passed by the input data. The computer device updates in reverse based on the difference between the training label and the initial prediction label corresponding to each training sample in the same training sample subset, adjusts the decision tree parameters of the initial decision tree corresponding to the training sample subset, obtains a new initial decision tree, inputs the training samples in the training sample subset into the new initial decision tree to obtain new initial prediction labels, updates in reverse based on the difference between the new initial prediction labels and the training labels, adjusts the decision tree parameters of the initial decision tree corresponding to the training sample subset again, and iteratively trains in this way. The adjustment target each time is to make the difference between the initial prediction label and the training label smaller and smaller until the first convergence condition is met to obtain the target decision tree corresponding to the training sample subset. Each training sample subset independently trains the corresponding initial decision tree, thereby finally obtaining the target decision tree corresponding to each training sample subset respectively.

[0065] The first convergence condition can be at least one of the difference between the initial prediction result and the training label being less than a preset difference, the number of iterations being less than a preset number, etc. Adjusting the decision tree parameters can be adjusting the decision tree form, for example, adjusting the split point of the decision tree, and adjusting the decision tree parameters can also be adjusting the node parameters of the decision tree nodes.

[0066] It can be understood that since the training sample subset and the initial decision tree form are different, any two target decision trees obtained through training are not completely the same, but any two target decision trees can output the prediction result corresponding to the same object.

[0067] In step S210, the object classification model is generated based on the target decision trees. The object classification model is configured to input the target feature of the target object into each target decision tree, and obtain the target prediction label of the target object based on the prediction results of the target decision trees.

[0068] The target object refers to an object to be classified or to have its operation permission determined. The target feature is obtained based on operation data of the target object in the target application.

[0069] Specifically, after the target decision trees are trained, the computer device can combine the target decision trees to form the object classification model. In subsequent model application, the computer device can obtain the target feature corresponding to the target object, input the target feature into the object classification model, and input the target feature into each target decision tree. After data processing by the decision trees, each target decision tree outputs a prediction result, and the object classification model outputs a target prediction label corresponding to the target object based on the prediction results of the target decision trees. For example, the object classification model can calculate the average of the prediction results, and obtain the target prediction label based on the average.

[0070] In one embodiment, the computer device can directly control the operation of the target object in the target application based on the target prediction label. For example, if the target prediction label is the target label, the operation of the target object in the target application is limited. The computer device can also determine the operation permission of the target object in the target application based on the target prediction label and other data, and then control the operation of the target object in the target application.

[0071] In the object classification method, a plurality of training sample subsets are obtained, each training sample subset is obtained by randomly sampling the same training sample set, the training sample subset includes training samples and training labels corresponding to the training samples, the training samples are obtained based on operation data of the training objects in the target application, the training labels are used to determine the operation permissions of the training objects in the target application, a decision tree is constructed based on the training sample subset, an initial decision tree corresponding to each training sample subset is obtained, a decision tree node of the initial decision tree is determined based on a randomly selected feature category from feature categories corresponding to each training feature included in the training samples, the training samples in the training sample subset are input into the corresponding initial decision tree, initial prediction labels corresponding to each training sample are obtained, the corresponding initial decision tree is adjusted based on the training labels and the initial prediction labels corresponding to each training sample in the same training sample subset until a first convergence condition is met, target decision trees corresponding to each training sample subset are obtained, and an object classification model is generated based on each target decision tree. Subsequently, a target feature corresponding to a target object can be input into the object classification model, and the object classification model outputs a prediction label corresponding to the target object based on prediction results of each target decision tree. In this way, the object classification model is trained based on the training sample set, and the object classification model is used to classify objects, which can improve the classification accuracy and efficiency of the objects. Moreover, the object classification model includes a plurality of target decision trees, different training sample subsets can be obtained by randomly sampling the same training sample set, different target decision trees can be trained based on different training sample subsets through random feature selection, and the object classification model can output more accurate prediction labels based on prediction results of different target decision trees, further improving the classification accuracy of the objects.

[0072] In one embodiment, the training samples include positive training samples and negative training samples, the operation permission corresponding to the positive training samples is less than the operation permission corresponding to the negative training samples, and the training object corresponding to the negative training samples includes at least one of a current time active object, a target time state object, a target platform active object, and a target time registration object. The current time active object refers to an object that is active in the target application for more than a first preset threshold in a current time period. The target time state object refers to an object that has a target state. The target platform active object refers to an object that is active in the target application for more than a second preset threshold through a target running platform. The target time registration object refers to an object that registers an account in the target application earlier than a target time.

[0073] The training samples include positive training samples and negative training samples. The training label corresponding to the positive training sample is a positive label, and the training label corresponding to the negative training sample is a negative label. The operation permission corresponding to the positive training sample is less than the operation permission corresponding to the negative training sample, that is, the operation permission of the object with the positive label in the target application is less than the operation permission of the object with the negative label in the target application. For example, the allowed operation duration of the object with the positive label in the target application is less than the allowed operation duration of the object with the negative label in the target application.

[0074] The current time active object refers to an object whose activity in the target application is greater than a first preset threshold in a current time period. The current time period refers to a currently used data statistical time period, which can be set according to actual needs. For example, the current time period can be one month in the past, and each object whose activity in the target application is greater than the first preset threshold in the one month in the past is obtained as a candidate object. The activity can be calculated according to a custom formula or algorithm, for example, the activity is calculated based on the operation duration, and the more the operation duration, the higher the activity. The first preset threshold can be set according to actual needs. It can be understood that the current time active object represents an object that is more active in the target application in the current time period, and can also be considered as an object that is less restricted in the target application in the current time period, that is, an object with higher operation permission.

[0075] The target time state object refers to an object whose time state is a target state. The time state can be determined according to the object age, for example, if the object age of the object is greater than a preset age, the corresponding time state is the target state. The time state can also be determined according to the guardian mark, for example, if the object has the guardian mark, the corresponding time state is the target state. The object with the guardian mark can manage and control the operation permission of a specific object, for example, for a game application, a parent with a growth guardian business mark can view the game time and consumption record of the child, and control the game time and game type of the child. It can be understood that the object whose time state is the target state can more freely and reasonably dispose of its own time, and can have higher operation permission in the target application. Of course, the time state can also be determined comprehensively according to the object age and the guardian mark, for example, if the object age of the object is greater than a preset age and the object has the guardian mark, the corresponding time state is the target state.

[0076] The target platform active object refers to an object whose activity in the target application is greater than a second preset threshold through a target running platform, that is, an object whose activity in the target application is greater than the second preset threshold by running the target application through the target running platform. The target running platform can refer to a client, for example, a computer terminal. The second preset threshold can be set according to actual needs.

[0077] The target time registration object refers to an object whose account registration time in the target application is earlier than the target time. It can be understood that the earlier the account registration time, the longer the target object uses the target application, and the target object should have higher operation permissions.

[0078] In one embodiment, the performance of the model trained based on the positive training sample and various negative training samples is verified respectively through experiments, and the experimental results are as follows:

[0079] 1. Positive sample: real-name authentication of minors, negative sample: real-name authentication of adults

[0080] Table 1

[0081]

[0082]

[0083] Among them, the AUC value refers to the area surrounded by the ROC curve and the X axis, and the closer the value is to 1, the better the model optimization effect. The accuracy rate refers to the number of samples of the predicted correct category / total sample number. The recall rate refers to the proportion of the predicted correct proportion in all positive samples. The F1 value refers to the harmonic mean of the precision and recall rate. The false positive rate refers to the proportion of the predicted positive sample in the actual negative sample, for example, in the game application, the false positive rate can be the probability of being judged as a minor player although it is actually an adult player. The simulation proportion of minors refers to the proportion of the number of minors in the real-name authentication object.

[0084] 2. Positive sample: real-name authentication of minors, negative sample: real-name authentication of adults + adult parents marked by the growth guardian service

[0085] Table 2

[0086] AUC value 0.921 Accuracy 0.640 Recall 0.472 F1 value 0.487 False positive rate 6.15％ Proportion of simulated delivery of minors 5.32％

[0087] Among them, compared with Table 1, the training effect of Table 2 is better, the proportion of minors authentication increases by 129.3%, and the false positive rate decreases by 33.8%. Real-name authentication of adults and adult parents marked by the growth guardian service can be considered as target time state objects.

[0088] 3. Positive sample: real-name authentication of minors, negative sample: real-name authentication of adults + active objects in the past month

[0089] Table 3

[0090] AUC value 0.914 Accuracy 0.663 Recall 0.394 F1 value 0.513 False positive rate 5.92％ Proportion of simulated delivery of minors 4.70％

[0091] Among them, compared with Table 1, the training effect of Table 3 is better, the proportion of minors authentication increases by 102.6%, and the false positive rate decreases by 36.3%. The active object in the past month can be considered as the current time active object.

[0092] 4. Positive samples: real-name authentication minors, negative samples: real-name authentication adults + end game active objects + registered objects with earlier years

[0093] Table 4

[0094] AUC value 0.932 Accuracy 0.741 Recall 0.628 F1 value 0.620 False positive rate 4.71％ Proportion of simulated delivery of minors 10.67％

[0095] Compared with Table 1, the training effect of Table 4 is better, the proportion of minor authentication increases by 359.9%, and the false positive rate decreases by 49.3%. The end game active object can be considered as the target platform active object, and the registered object with earlier years can be considered as the target time registered object.

[0096] In the above embodiments, the negative training samples in the training sample set correspond to training objects that can include at least one of the current time active object, the target time state object, the target platform active object, and the target time registered object. Using the operation features of these training objects as negative training samples helps to ensure the training effect of the model and improve the accuracy of the model.

[0097] In one embodiment, the training features include at least one of the object attribute features, the operation interaction features, the operation duration features, the device login features, the registration time features, the running platform features, and the target association features between the training object and the target associated object. The target associated object is an associated object of the training object that has a target operation permission.

[0098] Among them, the object attribute feature is used to represent the attribute information of the object. The object attribute feature can include at least one attribute information of the object age, the object gender, the terminal identifier of the target application running terminal, the terminal model, the terminal running network, the terminal running area, and the terminal running operating system. The object attribute feature can specifically include various attribute information of the object before real-name authentication. The terminal running area refers to the geographical area where the terminal is located, for example, the geographical areas of first-tier, second-tier, and third-tier cities.

[0099] The operation interaction feature is obtained based on interaction operations of an object with other objects in the target application, and is used to represent the operation frequency and operation proficiency of the object in the target application. For example, if the target application is a game application, the interaction operation can be a confrontation operation, and the operation interaction feature can include data such as match win rate, match draw rate, match loss rate, mode match win rate, total number of matches, total number of match days, average number of matches per day (total number of matches / number of match days), average number of matches per weekday (total number of matches on weekdays / number of match days on weekdays, average number of matches per weekend and holiday (total number of matches on weekends and holidays / number of match days on weekends and holidays), proportion of average number of matches per weekend and holiday to total average number of matches, average number of matches per hour on weekends and holidays, and proportion of average number of matches per hour on weekends and holidays.

[0100] The operation duration feature is used to represent the operation duration of the object in the target application. The operation duration feature can include at least one attribute information in operation duration, average duration, and duration proportion. For example, the operation duration feature can include data such as average daily duration on weekdays, average daily duration on weekends and holidays, duration of each period on weekdays, duration of each period on weekends and holidays, proportion of morning / midday / afternoon / night / late night duration per day, and the like.

[0101] The device login feature is obtained based on the terminal and account used by the object to log in to the target application, and is used to represent the login frequency and login specificity of the terminal and account. The device login feature can include at least one data in the number of devices logged in historically, the number of natural persons bound to the account historically, the number of devices logged in by the account in a target time period, the number of accounts bound to the account by the natural person in a target time period, and whether the account is logged in on a suspicious device in a target time period. The target time period can be set according to actual needs, for example, the target time period is summer vacation. A suspicious device refers to a device that has logged in the target application by a target type object in history, for example, a device that has logged in a game by a minor in history.

[0102] The registration time feature is used to represent the registration time and registration duration of the object in the target application. The registration time feature can include data such as registration duration and registration year.

[0103] The running platform feature is used to represent the activity of the object operating the target application on different running platforms. The running platform can specifically include a client (e.g., a computer terminal) and a mobile terminal (e.g., a mobile phone terminal). The running platform feature can include data such as the number of active days on the client in a preset time period, the active duration on the client in a preset time period, the number of active days on the mobile terminal in a preset time period, and the active duration on the mobile terminal in a preset time period. The preset time period can be set according to actual needs, for example, the preset time period is one month.

[0104] The associated object of a certain object refers to a friend of the certain object in the target application. The target associated object refers to an associated object with a target operation permission in each associated object of the training object, and can also be considered as a target type friend of the training object in the target application. For example, the target associated object can be a minor friend of the training object in the target application. The target association feature between the training object and the target associated object is used to represent the importance of the target associated object to the training object. The target association feature can include target associated object quantity, target associated object proportion, and target associated object closeness sum, etc. The target associated object closeness is generated based on the communication information between the training object and the target associated object, and the communication information includes at least one of conversation information, article gifting information, and invitation information.

[0105] In one embodiment, the operation interaction feature, the operation duration feature, and the device login feature can be operation features corresponding to the target time period. For example, the operation interaction feature includes the win rate, the draw rate, and the loss rate of the game in July-August.

[0106] Specifically, the training feature can include feature information of at least one dimension, and specifically can include at least one of the object attribute feature, the operation interaction feature, the operation duration feature, the device login feature, the registration time feature, the running platform feature, and the target association feature. It can be understood that the training feature of each dimension can also include feature information of at least one category, for example, the target association feature can include target associated object quantity and target associated object proportion, which are two categories of feature information.

[0107] In one embodiment, the computer device can obtain the application log reported by the target application, and extract each feature information of the same object from the application log to obtain the training feature or the target feature.

[0108] In the above embodiments, the training feature includes various data, which helps to ensure the training effect of the model.

[0109] In one embodiment, the process of obtaining the training sample includes the following steps:

[0110] Feature extraction is performed on the operation data of the current training object in the target application to obtain a plurality of initial operation features; a plurality of target operation features are determined from each initial operation feature; each target operation feature is cross-featured to obtain a cross-operation feature; and a training sample corresponding to the current training object is obtained based on the initial operation feature and the cross-operation feature.

[0111] Wherein, the cross-features refer to synthetic features formed by combining individual features. Cross-features help to represent non-linear relationships.

[0112] Specifically, when acquiring training data of the model, the computer device can perform feature extraction on operation data of the training object in the target application to obtain a plurality of initial operation features, take the initial operation features as training features, and thus obtain training samples. In order to further improve the training effect of the model, the computer device can also select part of the operation features from each initial operation feature as target operation features, perform feature cross on the target operation features to obtain cross operation features, and take the initial operation data and the cross operation data as training features, respectively, and thus obtain training samples. When performing feature cross, each target operation feature can be combined two by two to perform feature cross, and thus at least one cross operation feature can be obtained.

[0113] The selection of the target operation features can be determined according to actual needs. For example, the computer device can randomly select a plurality of operation features from the initial operation features as target operation features. The computer device can also obtain operation features belonging to different dimensions from the initial operation features as target operation features. Performing feature cross on operation features of different dimensions can improve the amount of information expressed by the cross operation features. The computer device can also obtain a plurality of operation features with the largest difference between different types of objects from the initial operation features as target operation features, and the cross operation features obtained further help the model to distinguish different types of objects.

[0114] In one embodiment, the computer device can multiply the feature values of different target operation features to obtain cross operation features. The computer device can also divide the feature value range of the target operation features to obtain a plurality of feature value intervals, combine the feature value intervals of target operation features of different categories to obtain a plurality of combined intervals, obtain initial features based on each combined interval, and update the initial features based on the feature value interval to which the feature value of the target operation feature of different categories belongs to obtain cross operation features.

[0115] For example, the operation duration can be divided into 3 feature value intervals, A, B, and C, and the target associated object ratio can be divided into 2 feature value intervals, 1 and 2. Combining the feature value intervals obtains 6 combined intervals, A and 1, B and 1, C and 1, A and 2, B and 2, and C and 2. The initial features (0, 0, 0, 0, 0, 0) are generated according to the 6 combined intervals. If the operation duration of an object falls into the feature value interval A and the target associated object ratio falls into the feature value interval 2, the cross operation features obtained by updating the initial features based on the operation duration and the target associated object ratio are (0, 0, 0, 1, 0, 0).

[0116] In the above embodiments, the training samples are composed of original operation features and cross features obtained through feature crossing. Model training based on such data helps improve the accuracy of the model.

[0117] The model effects of model training with and without feature crossing are tested through experiments. The experimental results are shown in Table 5. As can be seen from Table 5, feature crossing helps ensure the training effect of the model and improve the prediction accuracy of the model.

[0118] Table 5

[0119]

[0120]

[0121] In one embodiment, a decision tree is constructed based on a training sample subset, and an initial decision tree corresponding to each training sample subset is obtained, including:

[0122] A plurality of candidate feature categories are randomly determined from each current feature category. In the current training sample subset, a first split coefficient corresponding to a feature bin is calculated based on a feature bin set corresponding to the candidate feature category, and a plurality of first split coefficients corresponding to each candidate feature category are obtained. A second split coefficient is determined based on each first split coefficient corresponding to the same candidate feature category, and a second split coefficient corresponding to each candidate feature category is obtained. A target feature category is determined from each candidate feature category based on the second split coefficient. A decision tree node is generated based on the target feature category. The current feature category is updated based on the target feature category, and the step of randomly determining a plurality of candidate feature categories from each current feature category is returned to be executed until a preset condition is met, and a plurality of decision tree nodes are obtained. An initial decision tree is generated based on each decision tree node.

[0123] Specifically, for the current training sample subset, the computer device can take each feature category as a current feature category, randomly determine a plurality of feature categories as candidate feature categories from each current feature category, calculate a first split coefficient corresponding to a feature bin based on a feature bin set corresponding to the candidate feature category, and obtain a plurality of first split coefficients corresponding to each candidate feature category respectively. It can be understood that the feature bin set corresponding to a candidate feature category includes a plurality of feature bins, and each feature bin has a corresponding first split coefficient. The computer device can calculate the first split coefficient based on a self-defined formula or algorithm. The first split coefficient can be used to determine a second split coefficient corresponding to a certain feature category, or can be used to determine an optimal split point corresponding to a certain feature category. The computer device determines a second split coefficient based on each first split coefficient corresponding to the same candidate feature category, and obtains a second split coefficient corresponding to each candidate feature category respectively. The second split coefficient is used to determine the optimal split feature. For example, if the first split coefficient is smaller, it indicates that the corresponding split mode has a higher degree of differentiation for different objects, then each first split coefficient corresponding to the same candidate feature category can be sorted from small to large, and the computer device can obtain the minimum value as the second split coefficient, or obtain the average value of a plurality of first split coefficients ranked in the front as the second split coefficient. Then, the computer device can determine a target feature category from each candidate feature category based on the second split coefficient. For example, if the second split coefficient is smaller, it indicates that the corresponding split mode has a higher degree of differentiation for different objects, and the feature importance of the corresponding feature category is higher, then the candidate feature category corresponding to the second split coefficient with the smallest value can be taken as the target feature category. The target feature category can be considered as the optimal split feature currently found. Further, the computer device generates a decision tree node based on the target feature category, and obtains a first decision tree node. After determining the optimal split feature, the computer device can generate a decision tree node based on any split point corresponding to the optimal split feature. In order to improve the convergence speed of the model, the computer device can further determine the optimal split point corresponding to the optimal split feature, and generate a decision tree node based on the optimal split point. For example, a target feature bin can be determined from the feature bin set corresponding to the optimal split feature based on the first split coefficient, and the optimal split point is determined based on the target feature bin.

[0124] In generating the first decision tree node, the computer device can adopt equal frequency binning, equal length binning, chi-square binning, custom binning, etc. to perform binning processing on each training feature corresponding to the same candidate feature category to obtain a feature binning set corresponding to the candidate feature category. Binning processing refers to grouping disordered features and assigning the disordered features to a plurality of ordered bins. Each feature binning obtained through equal frequency binning has the same number of features. Each feature binning obtained through equal length binning has the same length of feature range. Chi-square binning is a binning method based on chi-square test. Custom binning refers to a binning method of custom binning split points. The training features of each feature category need to be independently subjected to binning processing. For example, the training features include registration duration and target associated object quantity, the registration duration corresponding to each training object in the same training sample subset is subjected to binning processing to obtain a feature binning set corresponding to the registration duration, and the target associated object quantity corresponding to each training object in the same training sample subset is subjected to binning processing to obtain a feature binning set corresponding to the target associated object quantity. One feature binning set includes a plurality of feature bins.

[0125] Further, the computer device can divide the feature space into two regions based on the first decision tree node, and different regions correspond to different branches of the decision tree node. For example, the decision tree node is whether the operation duration is greater than 9 hours, and the operation duration greater than 9 hours is a decision tree branch, and the operation duration less than or equal to 9 hours is another decision tree branch. The computer device can select any branch, any region to split the decision tree node to generate a second decision tree node. In determining the second decision tree node, the computer device can update the current feature category based on the target feature category, take the feature categories other than the target feature category (i.e., the remaining feature categories) as the current feature categories, randomly determine a plurality of new candidate feature categories from each current feature category, re-calculate a plurality of first split coefficients corresponding to each candidate feature category, determine a second split coefficient corresponding to each candidate feature category based on the first split coefficient, determine the target feature category from each candidate feature category based on the second split coefficient, generate a decision tree node based on the target feature category, and obtain the second decision tree node. Wherein, in calculating the first split coefficient, each training object is divided into each region of the feature space according to the value of the training feature, and is divided into different decision tree branches, and the feature bin set corresponding to the remaining feature categories is updated according to the division result. For example, the first decision tree node is whether the operation duration is greater than 9 hours, and before updating, the feature bin A corresponding to the registration duration includes data of 10 training objects, and after updating, for the first region or the first branch, the feature bin A corresponding to the registration duration includes data of 6 training objects, and the operation duration of the 6 training objects is greater than 9 hours, and for the second region or the second branch, the feature bin A corresponding to the registration duration includes data of 4 training objects, and the operation duration of the 4 training objects is less than or equal to 9 hours. Of course, the binning process can also be re-performed according to the division result.

[0126] The computer device repeats the above steps to continuously randomly select a feature category from the remaining feature categories, determine a new optimal split feature from the randomly selected feature category to generate a new decision tree node, and so on, until a preset condition is met, and a plurality of decision tree nodes are obtained. Each decision tree node is connected in turn according to the generation order and the parent-child relationship, thereby obtaining an initial decision tree. For example, the decision tree node used to divide region A and the decision tree node used to refine region A have a parent-child relationship.

[0127] It can be understood that if no new decision tree node is generated under a certain decision tree branch based on the split coefficient, a leaf node can be directly connected under the decision tree branch. The node parameters corresponding to each decision tree node can be randomly initialized or generated based on the corresponding second split coefficient. The initial decision tree can be considered as an initialized decision tree, and the decision tree morphology and node parameters can be adjusted and changed through multiple iterations subsequently.

[0128] The preset condition can be at least one of a decision tree depth greater than a preset depth, a feature binning range corresponding to a decision tree node being greater than a preset binning range, and the like.

[0129] In the above embodiments, based on the first splitting coefficient and the second splitting coefficient, the optimal splitting feature can be quickly selected from the training operation features, so as to quickly generate the decision tree node, and then obtain the initial decision tree. Moreover, since the candidate feature categories are randomly selected, the initial decision trees corresponding to each training sample subset are different, and the target decision trees obtained through subsequent training are also different, but each target decision tree can predict the data of the same object.

[0130] In an embodiment, the binning processing can be implemented through the following steps:

[0131] For the current training sample subset, each training feature corresponding to the current candidate feature category is initially binned to obtain a plurality of candidate bins, and the candidate bins are taken as the current bins. Based on the training labels corresponding to each current bin, the binning merging coefficient corresponding to adjacent current bins is calculated. Based on the binning merging coefficient, the adjacent current bins are binned and merged to obtain a plurality of merged bins. The merged bins are taken as the current bins, and the step of calculating the binning merging coefficient corresponding to adjacent current bins based on the training labels corresponding to each current bin is executed until the end condition is met, to obtain a plurality of feature bins corresponding to the current candidate feature category. Based on each feature bin corresponding to the current candidate feature category, a feature binning set corresponding to the current candidate feature category is obtained.

[0132] Specifically, when performing binning processing, initial binning can be performed first, and then binning merging can be performed to gradually optimize the binning result, and finally the feature binning set is obtained. For any feature category, the computer device can perform initial binning on each training feature corresponding to the current candidate feature category to obtain a plurality of candidate bins. For example, each training feature corresponding to the same feature category is sorted in ascending order of feature value, a plurality of binning split points are randomly determined, the value range of the training feature is cut based on the binning split points to obtain a plurality of feature value intervals, and each training object is put into the corresponding candidate bin according to the feature value interval into which the training feature falls.

[0133] For any one feature category, the computer device takes the candidate bins as current bins, and calculates bin merging coefficients corresponding to adjacent current bins based on the training labels corresponding to the current bins. The bin merging coefficients are used to represent the similarity of data distribution in adjacent feature bins. For example, the label proportion corresponding to each training label can be calculated based on the number of labels of each training label in a feature bin, and the bin merging coefficient can be obtained based on the difference between the label proportions corresponding to the same training label in adjacent feature bins. The computer device performs bin merging on adjacent current bins based on the bin merging coefficients, merges at least one group of adjacent feature bins with similar data distribution, and keeps other adjacent feature bins unchanged, thereby obtaining a plurality of merged bins. For example, if the smaller the bin merging coefficient is, the more similar the data distribution of adjacent feature bins is, then in one round of data processing, the adjacent feature bins with the smallest bin merging coefficient can be merged to obtain a merged bin, and other feature bins are taken as merged bins respectively, thereby obtaining a plurality of merged bins. The computer device can take the merged bins as current bins, repeat the above steps, calculate new bin merging coefficients, and perform a new round of bin merging, and so on, until the second preset condition is met, and a plurality of feature bins corresponding to the current candidate feature category are obtained. Finally, each feature bin corresponding to the current candidate feature category forms a feature bin set corresponding to the current candidate feature category.

[0134] The second preset condition can be at least one of that the number of feature bins is less than or equal to a preset bin number, and each bin merging coefficient is greater than a preset coefficient. For example, if the preset bin number is 10, and the number of merged bins is reduced from 11 to 10 after a certain round of data processing, the bin merging can be stopped. If each bin merging coefficient calculated based on the number of labels corresponding to each merged bin after a certain round of data processing is greater than the preset coefficient, the bin merging can be stopped.

[0135] In one embodiment, the bin merging coefficient can be calculated by the following formula:

[0136]

[0137]

[0138] wherein A ij represents the number of labels corresponding to the jth training label in the ith bin for a certain feature category. E ij represents the expected probability of A ij . N represents the total number of labels of the training labels for a certain feature category, and N i represents the number of labels corresponding to the ith bin for a certain feature category, and C j represents the proportion of the jth training label in all training labels for a certain feature category.

[0139] In the above embodiments, the bin merging can group data with similar distributions in one feature bin, so that the final feature bin set has certain data differences between different feature bins, and such a feature bin set helps to improve the convergence speed of the model during model training.

[0140] In one embodiment, the feature bin set obtained by the above binning processing is applied to model training, and the trained model is verified by data to obtain a minor authentication ratio of 7.3% and a false positive rate of 5.9%. Among them, the minor authentication ratio refers to the proportion of the number of minors in the real-name authentication object.

[0141] In one embodiment, the feature bin result of the time length can be as shown in Table 6. The bin result of Table 6 is obtained by clustering analysis on the operation time length of a large number of objects. Applying such bin results to model training, the trained model is verified by data to obtain a minor authentication ratio of 6.4% and a false positive rate of 5.1%.

[0142] Table 6

[0143]

[0144]

[0145] In one embodiment, the feature bin result of the time length can be as shown in Table 7, and Table 7 is an equal-length binning, which aims to let the model learn the optimal split point for generating decision tree nodes by itself. Applying such bin results to model training, the trained model is verified by data to obtain a minor authentication ratio of 7.1% and a false positive rate of 4.7%.

[0146] Table 7

[0147]

[0148] In one embodiment, in the current training sample subset, the first split coefficient corresponding to the feature bin is calculated based on the feature bin set corresponding to the candidate feature category, to obtain a plurality of first split coefficients corresponding to each candidate feature category, including:

[0149] The current feature bin is determined from each feature bin corresponding to the current feature bin set, each feature bin corresponding to the current feature bin set is divided into a first type of bin and a second type of bin based on the current feature bin, the label proportion corresponding to the first type of bin and the second type of bin is obtained based on the total number of labels corresponding to the first type of bin and the second type of bin respectively, the label distribution coefficient corresponding to the first type of bin and the second type of bin is obtained based on the number of labels corresponding to each training label in the same type of bin and the total number of labels, and the first split coefficient corresponding to the candidate feature category corresponding to the current feature bin is obtained based on the label proportion and the label distribution coefficient corresponding to the first type of bin and the second type of bin.

[0150] Specifically, in the calculation of the first split coefficient, a certain feature bin corresponding to a certain feature category can be taken as a candidate split point, each training feature corresponding to the feature category is divided into two types of data based on the candidate split point, the label proportion and the label distribution coefficient corresponding to any one type of data are calculated, and the first split coefficient corresponding to the candidate split point is obtained by fusing the label proportion and the label distribution coefficient corresponding to each type of data.

[0151] The computer device can take any one of the feature bins corresponding to the current feature bin set as the current feature bin, divide the feature bins corresponding to the current feature bin set into a first type of bin and a second type of bin based on the current feature bin. For example, the feature bins corresponding to the same candidate feature category are sorted in ascending order of feature values, the feature bins arranged before the reference feature bin are taken as the first type of bin, and the feature bins arranged after the reference feature bin and the reference feature bin are taken as the second type of bin. Then, for the first type of bin or the second type of bin, the computer device counts the label quantity of the training labels corresponding to each feature falling into the current type of bin to obtain the total label quantity corresponding to the first type of bin and the second type of bin, calculates the label proportion based on the total label quantity corresponding to the first type of bin and the second type of bin respectively to obtain the label proportion corresponding to the first type of bin and the second type of bin respectively. For example, the total label quantity corresponding to the first type of bin is 5, and the total label quantity corresponding to the second type of bin is 10, so the label proportion corresponding to the first type of bin is 5 / 15, and the label proportion corresponding to the second type of bin is 10 / 15. Further, for the first type of bin or the second type of bin, the computer device counts the label quantity corresponding to various training labels in the current type of bin, calculates the label distribution coefficient based on the label quantity corresponding to various training labels in the current type of bin and the total label quantity corresponding to the current type of bin to obtain the label distribution coefficient corresponding to the first type of bin and the second type of bin respectively. Finally, the first split coefficient corresponding to the reference feature bin is calculated based on the label proportion and the label distribution coefficient corresponding to the first type of bin and the second type of bin. For example, the label proportion and the label distribution coefficient corresponding to the same type of bin are fused to obtain the initial fusion result corresponding to the first type of bin and the second type of bin, and the initial fusion results corresponding to the first type of bin and the second type of bin are fused to obtain the first split coefficient.

[0152] The next feature bin in the current feature bin set is taken as the current feature bin, and the above steps are repeated to calculate the corresponding first split coefficient. In this way, the first split coefficient corresponding to each feature bin in the current feature bin set can be calculated.

[0153] In one embodiment, the first split coefficient can be calculated by the following formula:

[0154]

[0155]

[0156] wherein Gini(D, A) represents the first split coefficient corresponding to the feature bin A. D1 represents the first type of bin, D2 represents the second type of bin, and D represents the feature bin set corresponding to a certain feature category. D can be divided into D1 and D2 according to the feature bin A. represents the label proportion corresponding to the first type of bin, represents the label proportion corresponding to the second type of bin. Gini(D1) represents the label distribution coefficient corresponding to the first type of bin, and Gini(D2) represents the label distribution coefficient corresponding to the second type of bin. k represents the label proportion corresponding to the kth training label in D i In D, the label proportion corresponding to the kth training label, for example, the first type of bin includes 100 features, and 40 features correspond to the first label. That is, the total number of labels of the first type of bin is 100, the number of labels corresponding to the first label is 40, and the label proportion corresponding to the first label is 40 / 100 = 0.4.

[0157] In the above embodiment, the first split coefficient can be quickly calculated based on the label proportion and the label distribution coefficient corresponding to the first type of bin and the second type of bin. The first split coefficient can represent the distinguishing ability of the corresponding split point for different objects and different labels.

[0158] In one embodiment, the second split coefficient is determined based on the first split coefficient corresponding to the same candidate feature category, and the second split coefficient corresponding to each candidate feature category is obtained. The target feature category is determined from the candidate feature categories based on the second split coefficient, including:

[0159] From the first split coefficient corresponding to the same candidate feature category, the first split coefficient with the smallest value is obtained as the second split coefficient. The candidate feature category corresponding to the second split coefficient with the smallest value is obtained as the target feature category.

[0160] Specifically, the smaller the first split coefficient, the higher the distinguishing degree of the corresponding split mode for different objects. Therefore, the computer device can obtain the first split coefficient with the smallest value from the first split coefficient corresponding to the same candidate feature category as the second split coefficient, and obtain the second split coefficient corresponding to each category. Similarly, the smaller the second split coefficient, the higher the feature importance. Therefore, the computer device can obtain the candidate feature category corresponding to the second split coefficient with the smallest value as the target feature category, and obtain the optimal split feature.

[0161] In one embodiment, the decision tree node is generated based on the target feature category, including:

[0162] From the first split coefficient corresponding to the target feature category, the feature bin corresponding to the first split coefficient with the smallest value is obtained as the target feature bin. The decision tree node is generated based on the target feature bin.

[0163] Specifically, the computer device can determine an optimal split point from a plurality of split points corresponding to the target feature category, and generate the decision tree node based on the optimal split point. First, from each first split coefficient corresponding to the target feature category, the computer device can obtain a feature bin corresponding to the first split coefficient with the smallest value as a target feature bin, determine an optimal split point based on the target feature bin to generate the decision tree node. The optimal split point can be any one of the feature value ranges corresponding to the target feature bin, for example, the median value.

[0164] In the above embodiment, from each first split coefficient corresponding to the target feature category, the feature bin corresponding to the first split coefficient with the smallest value is obtained as the target feature bin, the decision tree node is generated based on the target feature bin, and the decision tree node is generated based on the optimal split point of the optimal split feature, which can accelerate the convergence speed of the model.

[0165] In one embodiment, the training samples include positive training samples and negative training samples, and the operation permission corresponding to the positive training samples is smaller than the operation permission corresponding to the negative training samples. As shown in Figure 3 Based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset, the initial decision tree corresponding to the training sample is adjusted until the first convergence condition is met, and the target decision tree corresponding to each training sample subset is obtained, including:

[0166] Step S302, based on the training label and the initial prediction label corresponding to each training sample in the current training sample subset, the initial decision tree corresponding to the current training sample subset is adjusted until the first convergence condition is met, and the intermediate decision tree corresponding to the current training sample subset is obtained.

[0167] Specifically, in order to further improve the training effect of the model, the computer device can also optimize the positive training samples. First, based on the original positive training samples and negative training samples, the initial decision tree is trained to obtain an intermediate decision tree, and then the training samples in the negative training samples that are predicted as positive labels by the intermediate decision tree are divided into positive training samples, and the intermediate decision tree is further trained to obtain a target decision tree. In this way, the positive samples are increased, and the problem of sample imbalance can be solved.

[0168] Therefore, the computer device can first train the initial decision tree based on each training sample in the current training sample subset, and during the training, the initial decision tree corresponding to the current training sample subset is adjusted based on the training label and the initial prediction label corresponding to each training sample until the first convergence condition is met, and the intermediate decision tree corresponding to the current training sample subset is obtained.

[0169] Step S304, input each training sample in the current training sample subset into the intermediate decision tree to obtain an intermediate prediction label corresponding to each training sample.

[0170] Step S306, updating the training label corresponding to the negative training sample with the intermediate prediction label as a positive label to a positive label.

[0171] Step S308, adjusting the intermediate decision tree based on the intermediate prediction label and the updated training label corresponding to each training sample until the second convergence condition is met, to obtain the target decision tree corresponding to the current training sample subset.

[0172] Specifically, the computer device inputs each training sample in the current training sample subset into the intermediate decision tree to obtain the intermediate prediction label corresponding to each training sample, updates the training label corresponding to the negative training sample with the intermediate prediction label as a positive label to a positive label, trains the intermediate decision tree based on the new positive training sample and negative training sample to obtain the target decision tree. When training the intermediate decision tree, the difference between the intermediate prediction label and the updated training label corresponding to each training sample is updated by back propagation, the decision tree parameters of the intermediate decision tree are adjusted, and iterative training is performed until the second convergence condition is met, to obtain the target decision tree corresponding to the current training sample subset.

[0173] The second convergence condition can be at least one of that the difference between the intermediate prediction result and the updated training label is less than a preset difference, the number of iterations is less than a preset number, etc.

[0174] In an embodiment, the performance of the model trained based on the positive training sample optimization scheme is verified by experiments, and the experimental results are as follows:

[0175] Table 8

[0176] AUC value 0.869 Accuracy 0.690 Recall 0.348 F1 value 0.461 False positive rate 8.71％ Proportion of simulated delivery of minors 3.16％

[0177] The positive samples include real-name authentication minors and real-name authentication adults predicted as minors by the intermediate decision tree, and the negative samples include real-name authentication adults. Compared with Table 1, the training effect of Table 8 is better, the proportion of minor authentication increases by 34.8%, and the false positive rate decreases by 6.5%. Real-name authentication adults predicted as minors by the intermediate decision tree can be considered as minors using adult-registered accounts to operate in the target application.

[0178] In the above embodiments, the initial decision tree is first trained based on the original positive training sample and negative training sample to obtain the intermediate decision tree, then the training sample in the negative training sample predicted as a positive label by the intermediate decision tree is classified into the positive training sample, and the intermediate decision tree is further trained to obtain the target decision tree. The target decision tree trained in this way can identify suspicious objects using accounts registered by normal objects in the target application.

[0179] In an embodiment, as Figure 4As shown, a subject classification method is provided, which is applied to a computer device for example, which can be the terminal 102 or the server 104 in the above Figure 1 . Referring to Figure 4 , the subject classification method comprises the following steps:

[0180] In step S402, target features corresponding to a target subject are obtained, which are obtained based on operation data of the target subject in a target application.

[0181] In step S404, the target features are input into a subject classification model to obtain a target prediction label corresponding to the target subject; the target prediction label is used to determine the operation permission of the target subject in the target application, and the target prediction label is obtained based on prediction results of each target decision tree in the subject classification model.

[0182] The training process of the subject classification model comprises the following steps: obtaining a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set; constructing a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset; a decision tree node of the initial decision tree is determined based on a randomly selected feature category from feature categories corresponding to each training feature contained in the training sample; inputting training samples in the training sample subset into the corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample; adjusting the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until a first convergence condition is met to obtain a target decision tree corresponding to each training sample subset; and generating the subject classification model based on each target decision tree.

[0183] It can be understood that the specific training process of the subject classification model can refer to the content of each related embodiment described above, which will not be repeated here. Similar to the training features, the target features are obtained based on the operation data of the target subject in the target application.

[0184] Specifically, the computer device can classify the target subject based on the trained subject classification model to determine the operation permission of the target subject in the target application. The computer device can obtain target features corresponding to the target subject, input the target features into the subject classification model, the target features are input into each target decision tree, and after data processing of the decision tree, each target decision tree outputs a prediction result, and the subject classification model finally outputs a target prediction label corresponding to the target subject based on the prediction results of each target decision tree.

[0185] In the object classification method, the target feature corresponding to the target object is input into an object classification model generated by a plurality of target decision trees, and the object classification model outputs a target prediction label corresponding to the target object based on the prediction results of the target decision trees. Based on the object classification model, the classification accuracy and efficiency of the object can be improved. Moreover, the object classification model includes a plurality of target decision trees, random sampling of the same training sample set can obtain different training sample subsets, different target decision trees can be trained based on different training sample subsets through random feature selection, and the object classification model can output more accurate prediction labels based on the prediction results of different target decision trees, further improving the classification accuracy of the object.

[0186] In one embodiment, the target feature is input into the object classification model to obtain a target prediction label corresponding to the target object, including:

[0187] The target feature is input into each target decision tree to obtain a prediction result corresponding to each target decision tree, respectively; each prediction result is normalized to obtain a normalized result, respectively; and the normalized results are statistically analyzed to obtain the target prediction label.

[0188] The normalization processing standardizes the prediction result in a preset range, so as to distinguish the prediction results corresponding to different types of objects.

[0189] Specifically, when the object classification model is applied, the computer device can input the target feature corresponding to the target object into the object classification model, the target feature is input into each target decision tree, and the data processing of the decision tree is performed. Each target decision tree outputs a prediction result, each prediction result is normalized, for example, the prediction result is normalized by a softmax function to obtain a normalized result, and finally the normalized results are statistically analyzed to obtain the target prediction label. For example, the average value of each normalized result is calculated, the target prediction label is obtained based on the average value; the prediction label corresponding to each normalized result is determined, and the label type with the largest number is selected from each prediction label as the target prediction label.

[0190] In one embodiment, the prediction result is normalized by the following formula:

[0191]

[0192] wherein p i represents the normalized result corresponding to the i th target decision tree, which can also be referred to as a prediction probability. exp represents a sigmoid function. i (x) represents the prediction result of the i th target decision tree. k(x) represents the prediction result of the kth target decision tree, and K represents the number of target decision trees.

[0193] In the above embodiment, the target features are respectively input into each target decision tree to obtain the prediction results corresponding to each target decision tree, each prediction result is normalized to obtain a normalized result, and statistical analysis of each normalized result can obtain an accurate target prediction label.

[0194] In one embodiment, the method further comprises:

[0195] When the target prediction label is a target label, a subject authentication request is generated, the subject authentication request is sent to a target terminal corresponding to a target subject, authentication information returned by the target terminal according to the subject authentication request is obtained, an authentication result of the target subject is determined based on the authentication information, and when the authentication result is authentication failure, the operation of the target subject in the target application is limited.

[0196] The target label is used to indicate that the model predicts that a certain subject belongs to a target type of subject, for example, the target label can be a positive label. The operation permission of the target type of subject in the target application is limited. The subject authentication request is used to determine the identity of the target subject to further determine whether the target subject is a target type of subject.

[0197] Specifically, in order to improve the classification accuracy of the target object, the computer device can comprehensively determine the operation permission of the target object in the target application based on the target prediction label and the authentication information. If it is determined based on the object classification model that the target prediction label corresponding to the target object is not the target label, it can be considered that the possibility of the target object belonging to the target type object is low, and at this time the operation of the target object in the target application can be directly unrestricted. If it is determined based on the object classification model that the target prediction label corresponding to the target object is the target label, it can be considered that the possibility of the target object belonging to the target type object is high, and in order to further determine whether the target object is the target type object, the computer device can generate an object authentication request and send it to the target terminal corresponding to the target object. The target terminal can collect personal information of the target object as authentication information according to the object authentication request, for example, collect at least one of face information, certificate information, communication contact information, etc. of the target object as authentication information, and feed back the authentication information to the computer device. The computer device can determine the identity information of the target object based on the authentication information to obtain the authentication result of the target object, for example, match the authentication information and the registration information to obtain the authentication result. If the authentication result shows that the target object is not the target type object, it is determined that the authentication result is authentication passed, and if the authentication result shows that the target object is the target type object, it is determined that the authentication result is authentication failed. Therefore, if the authentication result is authentication failed, it is determined that the target object is the target type object, and the operation of the target object in the target application needs to be restricted, for example, the operation time of the target object in the target application is limited, the virtual resource transfer share (for example, game currency use, recharge, etc.) of the target object in the target application is limited, etc.

[0198] In one embodiment, the authentication information can be face information, and the real-name authentication of the target object is performed based on the face information to determine whether the target object is a minor. If the face information and the real-name identity information in the account registration information match, it indicates that the target object is an adult, and it is determined that the authentication result is authentication passed, and if the face information and the real-name identity information in the account registration information do not match, it is determined that the authentication result is authentication failed, and it can be considered that the target object is a minor. The minor uses an adult's account in the target application, and the operation of the minor in the target application needs to be restricted.

[0199] In one embodiment, the prediction label can be represented by a probability. If the prediction label is greater than a preset probability, it is determined that the prediction label is the target label. The preset probability can be set according to actual needs, for example, set to 0.5.

[0200] In the above embodiments, whether the target object is the target type object is preliminarily determined based on the object classification model, if the preliminary determination result shows that the target object is not the target type object, subsequent data processing is not needed, resources are effectively saved, if the preliminary determination result shows that the target object is the target type object, whether the target object is the target type object is further determined according to the authentication information of the target object. If the authentication result again shows that the target object is the target type object, it is finally determined that the target object is the target type object, and the operation of the target object in the target application is limited. Through the double operation of model prediction and information authentication, the classification accuracy of the object can be further improved.

[0201] In a specific embodiment, the object classification method of the present application can be applied to a game application. According to the national protection policy for minors, the operation of minors in the game application is limited, for example, minors can only play games from Friday to Sunday or on legal holidays within 1 hour from 20:00 to 21:00 every day, and cannot play games in other time periods. Referring to Figure 5 In the object classification method of the present application, whether the game player playing games in a special time period (for example, a holiday) is a suspicious minor is identified by a high-risk model (i.e., an object classification model), if the identification result shows that the game player is not a suspicious minor, the game player is not limited to play games, if the identification result shows that the game player is a suspicious minor, the suspicious minor is face-popped in a non-game time period, and the information collected through the face pop is subjected to real-name authentication. If the real-name authentication is passed, the game player can be considered as an adult, and the game player is not limited to play games, if the real-name authentication is not passed, the game player can be considered as a minor, and the game player is limited to play games.

[0202] Referring to Figure 6 The integrated model for finding suspicious minor players is trained based on the positive and negative samples and the player features corresponding to each sample, and the finally trained integrated model can be referred to as a high-risk model. The player features of a to-be-identified player are input into the high-risk model, the high-risk model predicts the probability that the to-be-identified player belongs to a suspicious minor player, and if the prediction probability is greater than or equal to 0.5, the to-be-identified player is determined to be a suspicious minor player.

[0203] The training data of the model: the player features are derived from the game reporting log, specifically including basic attribute data (i.e. player portrait attributes), game data, time length data, suspicious device data, minor friend data, registration and game category data. The basic attribute data includes real name, age, real name, gender, real name, terminal, real name, model, real name, network, real name, regional classification (first-tier, second-tier, third-tier cities, etc.), real name, operating system, etc. The game data includes daily average time length on weekdays, daily average time length on weekends and holidays, time length of each period on weekdays, time length of each period on weekends and holidays, time length ratio of each day morning / noon / afternoon / night / deep night, etc. The time length data includes daily average time length on weekdays, daily average time length on weekends and holidays, time length of each period on weekdays, time length of each period on weekends and holidays, time length ratio of each day morning / noon / afternoon / night / deep night, etc. Suspicious device data includes the number of devices logged in since history, the number of natural persons bound since history, the number of account numbers logged in by the device, the number of account numbers bound by the natural person, whether logged in on a suspicious device (referring to a device with a minor login), etc. The minor friend data includes the number of minor friends, the proportion of minor friends, the total sum of the intimacy of minor friends, etc. The registration and game category data includes the year of the book, the earliest registration year, the number of active days of PC games, the active time length of PC games, the number of active days of mobile games, the active time length of mobile games, etc. Reference Figure 7 The features can be mined from various dimensions to determine the player features for model training.

[0204] The training process of the model: reference Figure 8 During training, a training sample set is obtained, and the training sample set is randomly sampled to generate multiple training sample subsets, D1, D2,..., Di. Classification training is performed based on the training sample subsets to obtain target decision trees corresponding to each training sample subset, which can also be referred to as classifiers, C1, C2,..., Ci. Each target decision tree constitutes an object classification model, which can also be referred to as a strong classifier. In application, the target features corresponding to the target object are input into the object classification model, and the object classification model, the prediction results of each target decision tree are voted, and the prediction result with more votes is taken as the target prediction label.

[0205] Expression of high-risk model:

[0206] Wherein, x represents the feature variable input into the model, i.e. the training sample, f m (x) represents the target decision tree, i.e. the model function calculated for x, and M represents the number of decision trees.

[0207] The various important parameters used in model training: random number seed (algorithm needs to specify the starting point of iteration when multiple rounds of iteration are performed): 34; number of decision trees: 100; purity (indicates the likelihood that a randomly selected sample is correctly classified in the model): Gini coefficient; maximum tree depth (indicates the distance between leaf nodes and root nodes, and the maximum tree depth is the critical point for stopping decision tree iteration, when the depth of the decision tree reaches the maximum tree depth, the decision tree will stop splitting): 8; maximum binning number of features: 32; validation set ratio (when constructing the model, the data set will be divided into a training set and a validation set, the training set data is used to construct the model, and the validation set data is used to verify the accuracy of the model): 20%.

[0208] Model training samples: positive samples can use at least one of the following: real-name authentication minors, real-name authentication adults predicted by intermediate decision trees as minors; negative samples can use at least one of the following: real-name authentication adults, adult parents with growth guardianship business marks, active players in the past month, active players of client games, and players registered for a long time.

[0209] The validation effect of the high-risk model based on the validation set is shown in Table 9.

[0210] Table 9

[0211] AUC value 0.932 Accuracy 0.741 Recall 0.628 F1 value 0.620 False positive rate 4.71％ Proportion of simulated delivery of minors 10.67％

[0212] Further, the effect of the high-risk model after being put into online use is shown in Table 10.

[0213] Table 10

[0214]

[0215] As can be seen from Table 10, in the face popup effect of the high-risk model from December 1 to 7, the daily average minor authentication ratio is 7.6%, which is 6.61 times that of the overall market, and the popup interception rate is 1.14 times that of the overall market. Among them, the popup interception rate = (the number of objects that do not pass the face in the popup / the number of popup objects) * 100%.

[0216] In the above embodiments, the integrated model algorithm is used for modeling by combining the basic attribute data, game data, time data, suspicious device data, minor friend data, registration and game category data, and the like. The model training parameters and training samples are optimized to train an integrated model. The integrated model accurately and efficiently identifies which players are minors in the game, and then restricts minors from playing games to prevent minors from becoming addicted to games.

[0217] It can be understood that the object classification method of the present application can also be applied to other applications, such as video applications, e-commerce applications, etc.

[0218] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.

[0219] Based on the same inventive concept, the embodiments of the present application also provide an object classification device for implementing the object classification method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more object classification device embodiments provided below can refer to the limitations of the object classification method described above, which will not be repeated here.

[0220] In one embodiment, as shown in Figure 9 An object classification device 900 is provided, comprising a training set acquisition module 902, an initial decision tree construction module 904, a decision tree prediction module 906, a target decision tree generation module 908, and a model generation module 910, wherein:

[0221] The training set acquisition module 902 is configured to acquire a plurality of training sample subsets. Each training sample subset is obtained by randomly sampling a same training sample set. Each training sample subset includes a training sample and a training label corresponding to the training sample. The training sample is obtained based on operation data of a training object in a target application. The training label is used to determine the operation permission of the training object in the target application.

[0222] The initial decision tree construction module 904 is configured to construct a decision tree based on each training sample subset, to obtain an initial decision tree corresponding to each training sample subset. A decision tree node of the initial decision tree is determined based on a randomly selected feature category from feature categories corresponding to each training feature included in the training sample.

[0223] The decision tree prediction module 906 is configured to input the training sample in each training sample subset into the corresponding initial decision tree, to obtain an initial prediction label corresponding to each training sample.

[0224] The target decision tree generation module 908 is configured to adjust the initial decision tree corresponding to each training sample in the training sample subset based on the training label and the initial prediction label corresponding to each training sample in the training sample subset, until a first convergence condition is met, to obtain a target decision tree corresponding to each training sample subset.

[0225] The model generation module 910 is configured to generate an object classification model based on the target decision trees. The object classification model is configured to input the target feature of the target object into each target decision tree, and obtain a target prediction label corresponding to the target object based on the prediction result of each target decision tree.

[0226] The object classification device described above can improve the classification accuracy and efficiency of the object based on the object classification model trained based on the training sample set. In addition, the object classification model includes a plurality of target decision trees. Different training sample subsets can be obtained by randomly sampling the same training sample set. Different target decision trees can be trained based on the different training sample subsets through random feature selection. The object classification model can output more accurate prediction labels based on the prediction results of the different target decision trees, further improving the classification accuracy of the object.

[0227] In one embodiment, the training samples include positive training samples and negative training samples, the operation permission corresponding to the positive training samples is less than the operation permission corresponding to the negative training samples, and the training object corresponding to the negative training sample includes at least one of a current time active object, a target time state object, a target platform active object, and a target time registration object. The current time active object refers to an object with an activity greater than a first preset threshold in the target application in a current time period. The target time state object refers to an object with a target state. The target platform active object refers to an object with an activity greater than a second preset threshold in the target application through a target running platform. The target time registration object refers to an object with an account registration time earlier than a target time in the target application.

[0228] In one embodiment, the training features include at least one of an object attribute feature of the training object, an operation interaction feature, an operation duration feature, a device login feature, a registration time feature, a running platform feature, and a target association feature between the training object and a target associated object. The target associated object is an associated object of the training object that has a target operation permission.

[0229] In one embodiment, the training set acquisition module is further configured to perform feature extraction on the operation data of the current training object in the target application to obtain a plurality of initial operation features, determine a plurality of target operation features from the initial operation features, perform feature intersection on the target operation features to obtain an intersection operation feature, and obtain the training sample corresponding to the current training object based on the initial operation features and the intersection operation feature.

[0230] In one embodiment, the initial decision tree construction module is further configured to randomly determine a plurality of candidate feature categories from the plurality of current feature categories; calculate a first split coefficient corresponding to each feature bin in the set of feature bins corresponding to the candidate feature categories based on the set of feature bins corresponding to the candidate feature categories in the current training sample subset, to obtain a plurality of first split coefficients respectively corresponding to the candidate feature categories; determine a second split coefficient based on each first split coefficient corresponding to the same candidate feature category, to obtain a second split coefficient respectively corresponding to each candidate feature category, and determine a target feature category from the plurality of candidate feature categories based on the second split coefficient; generate a decision tree node based on the target feature category; update the current feature category based on the target feature category, and return to the step of randomly determining a plurality of candidate feature categories from the plurality of current feature categories until a preset condition is met, to obtain a plurality of decision tree nodes; and generate an initial decision tree based on the plurality of decision tree nodes.

[0231] In one embodiment, the initial decision tree construction module is further configured to determine a current feature bin from each feature bin corresponding to the set of current feature bins, divide each feature bin corresponding to the set of current feature bins into a first type of bin and a second type of bin based on the current feature bin; obtain a label proportion respectively corresponding to the first type of bin and the second type of bin based on a total number of labels respectively corresponding to the first type of bin and the second type of bin; obtain a label distribution coefficient respectively corresponding to the first type of bin and the second type of bin based on a number of labels corresponding to each training label in the same type of bin and the total number of labels; and obtain a first split coefficient corresponding to a candidate feature category corresponding to the current feature bin based on the label proportion and the label distribution coefficient corresponding to the first type of bin and the second type of bin.

[0232] In one embodiment, the initial decision tree construction module is further configured to obtain a first split coefficient with the smallest value as the second split coefficient from each first split coefficient corresponding to the same candidate feature category; and obtain a candidate feature category corresponding to the second split coefficient with the smallest value as the target feature category.

[0233] In one embodiment, the initial decision tree construction module is further configured to obtain a feature bin corresponding to the first split coefficient with the smallest value as a target feature bin from each first split coefficient corresponding to the target feature category; and generate a decision tree node based on the target feature bin.

[0234] In one embodiment, the training samples include positive training samples and negative training samples, where the operation permissions corresponding to positive training samples are less than those corresponding to negative training samples. The model generation module is further configured to: adjust the initial decision tree corresponding to the current training sample subset based on the training labels and initial predicted labels corresponding to each training sample in the current training sample subset until a first convergence condition is met, thus obtaining an intermediate decision tree corresponding to the current training sample subset; input each training sample in the current training sample subset into the intermediate decision tree to obtain intermediate predicted labels corresponding to each training sample; update the training labels corresponding to negative training samples with positive intermediate predicted labels to positive labels; and adjust the intermediate decision tree based on the intermediate predicted labels and updated training labels corresponding to each training sample until a second convergence condition is met, thus obtaining a target decision tree corresponding to the current training sample subset.

[0235] In one embodiment, such as Figure 10 As shown, an object classification device 1000 is provided, including: a data acquisition module 1002 and a label prediction module 1004, wherein:

[0236] The data acquisition module 1002 is used to acquire the target features corresponding to the target object. The target features are obtained based on the operation data of the target object in the target application.

[0237] The label prediction module 1004 is used to input the target features into the object classification model to obtain the target prediction label corresponding to the target object. The target prediction label is used to determine the operation permissions of the target object in the target application. The target prediction label is obtained based on the prediction results of each target decision tree in the object classification model.

[0238] The training process for an object classification model includes the following steps:

[0239] Obtain multiple training sample subsets; each training sample subset is obtained by randomly sampling the same training sample set; construct decision trees based on the training sample subsets to obtain the initial decision trees corresponding to each training sample subset; the decision tree nodes of the initial decision trees are determined based on randomly selected feature categories from the feature categories corresponding to each training feature contained in the training samples; input the training samples in the training sample subsets into the corresponding initial decision trees to obtain the initial predicted labels corresponding to each training sample; adjust the corresponding initial decision trees based on the training labels and initial predicted labels corresponding to each training sample in the same training sample subset until the first convergence condition is met to obtain the target decision trees corresponding to each training sample subset; generate object classification models based on each target decision tree.

[0240] The object classification apparatus classifies the object based on the object classification model, and can improve the classification accuracy and efficiency of the object. Moreover, the object classification model includes multiple target decision trees, different training sample subsets can be obtained by randomly sampling the same training sample set, different target decision trees can be trained by random feature selection based on the different training sample subsets, and the object classification model can output more accurate prediction labels based on the prediction results of the different target decision trees, thereby further improving the classification accuracy of the object.

[0241] In one embodiment, the label prediction module is further configured to input the target features into each target decision tree respectively to obtain a prediction result corresponding to each target decision tree respectively; perform normalization processing on each prediction result respectively to obtain a normalized result; and perform statistical analysis on each normalized result to obtain the target prediction label.

[0242] In one embodiment, as shown in FIG. 10, the object classification apparatus 1000 further includes: Figure 11

[0243] The object authentication model 1006 is configured to, when the target prediction label is the target label, generate an object authentication request, send the object authentication request to a target terminal corresponding to the target object, obtain authentication information returned by the target terminal according to the object authentication request, determine an authentication result of the target object based on the authentication information, and restrict operation of the target object in the target application when the authentication result is authentication failure.

[0244] Each module in the object classification apparatus can be implemented by software, hardware, or a combination thereof, in whole or in part. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations corresponding to each module.

[0245] In one embodiment, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in FIG. 11. Figure 12 ​As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores training samples, object classification models, and other data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When executed by the processor, the computer program implements an object classification method.

[0246] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 13 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements an object classification method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0247] Those skilled in the art will understand that Figure 12 , 13The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0248] In an embodiment, a computer device is also provided, including a memory and a processor. The memory stores a computer program. The processor executes the computer program to implement the steps in the above method embodiments.

[0249] In an embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0250] In an embodiment, a computer program product or a computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium. The processor executes the computer instructions to cause the computer device to perform the steps in the above method embodiments.

[0251] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0252] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0253] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0254] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method of object classification, characterized by, The method comprises: obtaining a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set, and the training sample subset comprises a training sample and a training label corresponding to the training sample, the training sample is obtained based on operation data of a training object in a target application, and the training label is used to determine an operation permission of the training object in the target application; the training sample comprises a positive training sample and a negative training sample, the operation permission corresponding to the positive training sample is smaller than the operation permission corresponding to the negative training sample, and the training object corresponding to the negative training sample comprises at least one of a current time active object, a target time state object, a target platform active object and a target time registration object; the current time active object refers to an object with an activity greater than a first preset threshold in the target application in a current time period; the target time state object refers to an object with a target state; the target platform active object refers to an object with an activity greater than a second preset threshold in the target application through a target running platform; and the target time registration object refers to an object with an account registration time earlier than a target time in the target application; constructing a decision tree based on the training sample subset to obtain an initial decision tree corresponding to each training sample subset; a decision tree node of the initial decision tree is determined based on a randomly selected feature category from feature categories corresponding to each training feature contained in the training sample; the training feature comprises at least one of an object attribute feature of the training object, an operation interaction feature, an operation time length feature, a device login feature, a registration time feature, a running platform feature and a target association feature between the training object and a target associated object; and the target associated object is an associated object with a target operation permission among each associated object of the training object; inputting the training sample in the training sample subset into the corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample; adjusting the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until a first convergence condition is met to obtain a target decision tree corresponding to each training sample subset; generating an object classification model based on each target decision tree; the object classification model is used to input a target feature corresponding to a target object into the target decision tree, and obtain a target prediction label corresponding to the target object based on a prediction result of the target decision tree.

2. The method of claim 1, wherein, The obtaining process of the training sample comprises the following steps: performing feature extraction on operation data of a current training object in the target application to obtain a plurality of initial operation features; determining a plurality of target operation features from each initial operation feature; performing feature cross on each target operation feature to obtain a cross operation feature; obtaining a training sample corresponding to the current training object based on the initial operation feature and the cross operation feature.

3. The method of claim 1, wherein, The method comprises: randomly determining a plurality of candidate feature categories from each current feature category; In the current training sample subset, the first split coefficient corresponding to the feature bin is calculated based on the feature bin set corresponding to the candidate feature category, and a plurality of first split coefficients corresponding to each candidate feature category are obtained; Determine the second split coefficient based on each first split coefficient corresponding to the same candidate feature category, obtain the second split coefficient corresponding to each candidate feature category, and determine the target feature category from the candidate feature categories based on the second split coefficient; Generate a decision tree node based on the target feature category; Update the current feature category based on the target feature category, and return to execute the step of randomly determining a plurality of candidate feature categories from each current feature category until a preset condition is met, and obtain a plurality of decision tree nodes; Generate the initial decision tree based on each decision tree node.

4. The method of claim 3, wherein, In the current training sample subset, the first split coefficient corresponding to the feature bin is calculated based on the feature bin set corresponding to the candidate feature category, and a plurality of first split coefficients corresponding to each candidate feature category are obtained; Determine the current feature bin from each feature bin corresponding to the current feature bin set, and divide each feature bin corresponding to the current feature bin set into a first type bin and a second type bin based on the current feature bin; Obtain the label proportion corresponding to the first type bin and the second type bin based on the total number of labels corresponding to the first type bin and the second type bin, respectively; Obtain the label distribution coefficient corresponding to the first type bin and the second type bin based on the total number of labels corresponding to the first type bin and the second type bin, respectively. Obtain the first split coefficient corresponding to the candidate feature category corresponding to the current feature bin based on the label proportion and the label distribution coefficient corresponding to the first type bin and the second type bin.

5. The method of claim 3, wherein, Determine the second split coefficient based on each first split coefficient corresponding to the same candidate feature category, obtain the second split coefficient corresponding to each candidate feature category, and determine the target feature category from the candidate feature categories based on the second split coefficient; From the first split coefficient corresponding to the same candidate feature category, obtain the first split coefficient with the smallest value as the second split coefficient; Obtain the candidate feature category corresponding to the second split coefficient with the smallest value as the target feature category.

6. The method of claim 3, wherein, Based on the target feature category, generate a decision tree node, including: From the first split coefficient corresponding to the target feature category, obtain the feature bin corresponding to the first split coefficient with the smallest value as the target feature bin; Generate a decision tree node based on the target feature bin.

7. The method according to any one of claims 1 to 6, characterized in that, The training sample includes positive training samples and negative training samples, and the operation permission corresponding to the positive training sample is less than the operation permission corresponding to the negative training sample; Adjust the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until the first convergence condition is met, and obtain the target decision tree corresponding to each training sample subset, including: adjust the initial decision tree corresponding to the current training sample subset based on the training label corresponding to each training sample in the current training sample subset and the initial prediction label, until a first convergence condition is met, to obtain an intermediate decision tree corresponding to the current training sample subset; input each training sample in the current training sample subset into the intermediate decision tree to obtain an intermediate prediction label corresponding to each training sample; update the training label corresponding to a negative training sample with an intermediate prediction label as a positive label to a positive label; adjust the intermediate decision tree based on the intermediate prediction label corresponding to each training sample and the updated training label, until a second convergence condition is met, to obtain a target decision tree corresponding to the current training sample subset.

8. A method of classifying an object, characterized by, The method comprises: obtaining a target feature corresponding to a target object, the target feature being obtained based on operation data of the target object in a target application; inputting the target feature into an object classification model to obtain a target prediction label corresponding to the target object; the target prediction label is used to determine the operation permission of the target object in the target application, and the target prediction label is obtained based on the prediction result of each target decision tree in the object classification model; The training process of the object classification model comprises the following steps: obtaining a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set; constructing a decision tree based on a training sample subset to obtain an initial decision tree corresponding to each training sample subset; a decision tree node of the initial decision tree is determined based on a randomly selected feature category from a feature category corresponding to each training feature contained in a training sample; the training sample comprises a positive training sample and a negative training sample, the operation permission corresponding to the positive training sample is less than the operation permission corresponding to the negative training sample, and the training object corresponding to the negative training sample comprises at least one of a current time active object, a target time state object, a target platform active object and a target time registered object; the current time active object refers to an object with an activity greater than a first preset threshold in the target application in a current time period; the target time state object refers to an object with a target state; the target platform active object refers to an object with an activity greater than a second preset threshold in the target application through a target running platform; and the target time registered object refers to an object with an account registration time earlier than a target time in the target application; the training feature comprises at least one of an object attribute feature of a training object, an operation interaction feature, an operation time length feature, a device login feature, a registration time feature, a running platform feature, and a target association feature between the training object and a target associated object; the target associated object is an associated object with a target operation permission among each associated object of the training object; inputting the training sample in the training sample subset into the corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample; adjust the initial decision tree corresponding to each training sample in the same training sample subset based on the training label and the initial prediction label corresponding to each training sample, until a first convergence condition is met, to obtain a target decision tree corresponding to each training sample subset; generate an object classification model based on each target decision tree.

9. The method of claim 8, wherein, The method further comprises: when the target prediction label is a target label, generating an object authentication request and sending the object authentication request to a target terminal corresponding to the target object; obtaining authentication information returned by the target terminal according to the object authentication request, and determining an authentication result of the target object based on the authentication information; when the authentication result is not passed, limiting the operation of the target object in the target application.

10. The method of claim 8, wherein, The device comprises: a training set acquisition module configured to acquire a plurality of training sample subsets; each training sample subset is obtained by randomly sampling a same training sample set, and comprises a training sample and a training label corresponding to the training sample; the training sample is obtained based on operation data of a training object in a target application, and the training label is used to determine an operation permission of the training object in the target application; the training sample comprises a positive training sample and a negative training sample; the operation permission corresponding to the positive training sample is smaller than the operation permission corresponding to the negative training sample; the training object corresponding to the negative training sample comprises at least one of a current time active object, a target time state object, a target platform active object, and a target time registered object; the current time active object refers to an object with an activity greater than a first preset threshold in the target application in a current time period; the target time state object refers to an object with a target state; the target platform active object refers to an object with an activity greater than a second preset threshold in the target application through a target running platform; and the target time registered object refers to an object with an account registration time earlier than a target time in the target application; an initial decision tree construction module configured to construct a decision tree based on the training sample subset, to obtain an initial decision tree corresponding to each training sample subset; a decision tree node of the initial decision tree is determined based on a randomly selected feature category from a feature category corresponding to each training feature contained in the training sample; the training feature comprises at least one of an object attribute feature of a training object, an operation interaction feature, an operation duration feature, a device login feature, a registration time feature, a running platform feature, and a target association feature between the training object and a target associated object; and the target associated object is an associated object with a target operation permission among each associated object of the training object. ​ 11. An object classification apparatus characterized by comprising: ​ ​ ​ The decision tree prediction module is configured to input each training sample in the training sample subset into a corresponding initial decision tree to obtain an initial prediction label corresponding to each training sample; The target decision tree generation module is configured to adjust the corresponding initial decision tree based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset until a first convergence condition is met, to obtain a target decision tree corresponding to each training sample subset; The model generation module is configured to generate an object classification model based on each target decision tree; the object classification model is configured to input a target feature corresponding to a target object into each target decision tree, and obtain a target prediction label corresponding to the target object based on a prediction result of each target decision tree.

12. The object classification apparatus according to claim 11, characterized by The training set acquisition module is further configured to perform feature extraction on operation data of the current training object in the target application to obtain a plurality of initial operation features; and determine a plurality of target operation features from each initial operation feature. The feature intersection module is configured to perform feature intersection on each target operation feature to obtain an intersection operation feature. The training sample generation module is configured to obtain a training sample corresponding to the current training object based on the initial operation feature and the intersection operation feature.

13. The object classification apparatus according to claim 11, wherein The initial decision tree construction module is further configured to randomly determine a plurality of candidate feature categories from each current feature category; calculate a first split coefficient corresponding to a feature bin in a feature bin set corresponding to a candidate feature category based on the feature bin in the feature bin set corresponding to the candidate feature category in the current training sample subset to obtain a plurality of first split coefficients corresponding to each candidate feature category; determine a second split coefficient based on each first split coefficient corresponding to a same candidate feature category to obtain a second split coefficient corresponding to each candidate feature category, and determine a target feature category from the candidate feature categories based on the second split coefficient. The decision tree node generation module is configured to generate a decision tree node based on the target feature category. The current feature category updating module is configured to update the current feature category based on the target feature category, return to the step of randomly determining a plurality of candidate feature categories from each current feature category, and execute until a preset condition is met to obtain a plurality of decision tree nodes; and generate the initial decision tree based on each decision tree node.

14. The object classification apparatus according to claim 13, wherein The initial decision tree construction module is further configured to determine a current feature bin from each feature bin corresponding to a current feature bin set, divide each feature bin corresponding to the current feature bin set into a first type of bin and a second type of bin based on the current feature bin; obtain a label proportion corresponding to the first type of bin and the second type of bin based on a total number of labels corresponding to the first type of bin and the second type of bin, respectively; obtain a label distribution coefficient corresponding to the first type of bin and the second type of bin based on a label quantity corresponding to each training label and a total number of labels in a same type of bin; and obtain a first split coefficient corresponding to a candidate feature category corresponding to the current feature bin based on the label proportion and the label distribution coefficient corresponding to the first type of bin and the second type of bin.

15. The object classification apparatus according to claim 13, wherein The initial decision tree construction module is further configured to obtain a first split coefficient with the smallest value from the first split coefficients corresponding to the same candidate feature category as the second split coefficient, and obtain the candidate feature category corresponding to the second split coefficient with the smallest value as the target feature category.

16. The object classification apparatus according to claim 13, wherein The initial decision tree construction module is further configured to obtain a feature bin corresponding to the first split coefficient with the smallest value from the first split coefficients corresponding to the target feature category as the target feature bin. A decision tree node is generated based on the target feature bin.

17. The object classification apparatus according to any one of claims 11 to 16, characterized by The training samples include positive training samples and negative training samples, and the operation permission corresponding to the positive training samples is smaller than the operation permission corresponding to the negative training samples. The model generation module is further configured to adjust the initial decision tree corresponding to the current training sample subset based on the training label and the initial prediction label corresponding to each training sample in the current training sample subset until a first convergence condition is met, to obtain an intermediate decision tree corresponding to the current training sample subset.

18. An object classification apparatus characterized by comprising: Each training sample in the current training sample subset is input into the intermediate decision tree to obtain an intermediate prediction label corresponding to each training sample. The training label corresponding to a negative training sample with a positive intermediate prediction label is updated to a positive label. The intermediate decision tree is adjusted based on the intermediate prediction label and the updated training label corresponding to each training sample until a second convergence condition is met, to obtain a target decision tree corresponding to the current training sample subset. The apparatus comprises: A data acquisition module configured to acquire target features corresponding to a target object, the target features being obtained based on operation data of the target object in a target application; A label prediction module configured to input the target features into an object classification model to obtain a target prediction label corresponding to the target object, the target prediction label being used to determine the operation permission of the target object in the target application, and the target prediction label being obtained based on prediction results of each target decision tree in the object classification model. The training process of the object classification model comprises the following steps: A plurality of training sample subsets are acquired, each training sample subset being obtained by randomly sampling a same training sample set; A plurality of training sample subsets are acquired, each training sample subset being obtained by randomly sampling a same training sample set; The decision tree is constructed based on the training sample subset, and an initial decision tree corresponding to each training sample subset is obtained; a decision tree node of the initial decision tree is determined based on a randomly selected feature category from feature categories corresponding to each training feature contained in the training sample; the training sample includes positive training samples and negative training samples, the operation permission corresponding to the positive training sample is smaller than the operation permission corresponding to the negative training sample, the training object corresponding to the negative training sample includes at least one of a current time active object, a target time state object, a target platform active object, and a target time registration object, the current time active object refers to an object with an activity degree greater than a first preset threshold in the target application in a current time period, the target time state object refers to an object with a target state, the target platform active object refers to an object with an activity degree greater than a second preset threshold in the target application through a target running platform, and the target time registration object refers to an object with an account registration time earlier than a target time in the target application; the training feature includes at least one of an object attribute feature of a training object, an operation interaction feature, an operation time length feature, a device login feature, a registration time feature, a running platform feature, and a target association feature between the training object and a target associated object, the target associated object is an associated object with a target operation permission among each associated object of the training object; The training sample in the training sample subset is input into the corresponding initial decision tree, and an initial prediction label corresponding to each training sample is obtained; Based on the training label and the initial prediction label corresponding to each training sample in the same training sample subset, the corresponding initial decision tree is adjusted until a first convergence condition is met, and a target decision tree corresponding to each training sample subset is obtained; An object classification model is generated based on each target decision tree.

19. The object classification apparatus according to claim 18, wherein, The label prediction module is further configured to input the target feature into each target decision tree to obtain a prediction result corresponding to each target decision tree, respectively; normalize each prediction result to obtain a normalized result; and perform statistical analysis on the normalized results to obtain the target prediction label.

20. The object classification apparatus according to claim 18, wherein The device further includes an object authentication model configured to generate an object authentication request when the target prediction label is a target label, and send the object authentication request to a target terminal corresponding to the target object; Obtain authentication information returned by the target terminal according to the object authentication request, and determine an authentication result of the target object based on the authentication information; when the authentication result is not passed, limit the operation of the target object in the target application. 21.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-20. The processor executes the computer program to implement the method of any one of claims 1-8.

22. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-8.

23. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-8.

Images