Model training, service allocation method, device and equipment based on user classification

Abstract

The embodiments of this specification provide a method, device and equipment for model training and business allocation based on user classification. The method includes: obtaining user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; determining the user category probability of the unlabeled data through the neighboring labeled data corresponding to the unlabeled data; constructing a neighbor similarity regularization feature based on the user category probability; using the user sample data to generate a classification regularization feature corresponding to at least two information categories; training a user classification model by combining the neighbor similarity regularization feature and the classification regularization feature; the user classification model is used to classify users to allocate corresponding services. The above method reduces the time and resources consumed by a large amount of labeled data, optimizes the generalization effect of the classification model, and improves the user's business processing experience.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of artificial intelligence technology, and more particularly to a service allocation method, device, and equipment. Background Art

[0002] With the development of various industries, the subdivision of business types is also increasing. These businesses can be businesses that provide corresponding services to users, or businesses that require users to process in a timely manner. Correspondingly, different types of users need to obtain different services. Therefore, judging the services that users may obtain based on the relevant information of users in advance, and preparing the data and resources corresponding to the corresponding services in advance, can effectively improve the efficiency of business processing in the subsequent process and improve the user experience.

[0003] At present, when predicting the services that users will obtain, a large amount of sample data is often obtained in advance, and the corresponding machine learning model is trained using the sample data, so that the trained machine learning model can be used to predict the services required by different users. However, after obtaining the sample data, it is often necessary to annotate the sample data. In order to ensure the accuracy of the model, the sample data generally has a large amount. Annotating the sample data not only consumes a lot of time and resources, but also requires a high level of professional knowledge from the annotator, which affects the actual training effect of the model. Therefore, there is an urgent need for a method that can quickly and accurately train the model to ensure the user's business prediction effect. Summary of the invention

[0004] The purpose of the embodiments of this specification is to provide a user classification-based model training, business allocation method, device and equipment to solve the problem of how to quickly and accurately train the model to ensure the user's business prediction effect.

[0005] To solve the above technical problems, an embodiment of the present specification provides a model training method based on user classification, including: obtaining user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; determining the user category probability of unlabeled data through neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; training a user classification model by combining the neighbor similarity regularization features and the classification regularization features; the user classification model is used to determine the user category according to the user information to assign the user the business corresponding to the user category.

[0006] The embodiment of the present specification also proposes a model training device based on user classification, including: a user sample data acquisition module, used to acquire user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; a user category probability determination module, used to determine the user category probability of unlabeled data through neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; a neighbor similarity regularization feature construction module, Used to construct a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories; a classification regularization feature generation module, used to generate classification regularization features corresponding to at least two information categories using the user sample data; the information category corresponds to the category of different features of the user sample data; a user classification model acquisition module, used to obtain a user classification model by integrating the neighbor similarity regularization feature and the classification regularization feature training; the user classification model is used to determine the user category according to the user information to assign the user the service corresponding to the user category.

[0007] The embodiment of this specification also proposes a business allocation device, including a memory and a processor; the memory is used to store computer program instructions; the processor is used to execute the computer program instructions to implement the following steps: obtaining user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; determining the user category probability of the unlabeled data through the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; comprehensively training the neighbor similarity regularization feature and the classification regularization feature to obtain a user classification model; the user classification model is used to determine the user classification according to the user information to allocate the user the business corresponding to the user classification.

[0008] In order to solve the above technical problems, the embodiments of this specification also propose a service allocation method based on user classification, including: obtaining user feature information of the target user; inputting the user feature information into a user classification model to obtain a user category; the user classification model is obtained in the following manner: after obtaining user sample data containing labeled data and unlabeled data, determining the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; comprehensively training the neighbor similarity regularization features and the classification regularization features to obtain a user classification model; and allocating the service corresponding to the user category to the target user.

[0009] The embodiment of this specification also proposes a service allocation device based on user classification, including: a user feature information acquisition module, used to acquire user feature information of a target user; a user category acquisition module, used to input the user feature information into a user classification model to obtain a user category; the user classification model is obtained in the following manner: after acquiring user sample data containing labeled data and unlabeled data, determining the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; training the user classification model by combining the neighbor similarity regularization features and the classification regularization features; a service allocation module, used to allocate the service corresponding to the user category to the target user.

[0010] The embodiment of this specification also proposes a service allocation device based on user classification, including a memory and a processor; the memory is used to store computer program instructions; the processor is used to execute the computer program instructions to implement the following steps: obtaining user feature information of a target user; inputting the user feature information into a user classification model to obtain a user category; the user classification model is obtained in the following manner: after obtaining user sample data containing labeled data and unlabeled data, determining the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; comprehensively training the neighbor similarity regularization feature and the classification regularization feature to obtain a user classification model; and allocating the service corresponding to the user category to the target user.

[0011] It can be seen from the technical solution provided by the above embodiments of this specification that after obtaining the user sample data, the embodiments of this specification only annotate part of the data, and then determine the user category probability and the neighbor similarity regularization feature in turn while considering the correlation between the labeled data and the unlabeled data, and combine the classification regularization feature corresponding to the information category to realize the training of the user classification model, so that the user classification model can be used to complete the classification of users, and then the corresponding services are assigned to users according to the user category. Through the above method, the time and resources consumed by a large amount of labeled data are reduced, the correlation between data is fully explored, the generalization effect of the classification model is optimized, and the user's business processing experience is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 This is a flow chart of a model training method based on user classification according to an embodiment of this specification;

[0014] Figure 2 A schematic diagram of a model training process according to an embodiment of this specification;

[0015] Figure 3 A schematic diagram of a service allocation process according to an embodiment of this specification;

[0016] Figure 4 This is a flow chart of a service allocation method based on user classification according to an embodiment of this specification;

[0017] Figure 5 This is a module diagram of a model training device based on user classification according to an embodiment of this specification;

[0018] Figure 6 This is a module diagram of a service allocation device based on user classification according to an embodiment of this specification;

[0019] Figure 7 This is a structural diagram of a model training device based on user classification according to an embodiment of this specification;

[0020] Figure 8 This is a structural diagram of a service allocation device based on user classification according to an embodiment of this specification. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of this specification to clearly and completely describe the technical solutions in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0022] In order to solve the above technical problems, firstly, a model training method based on user classification is introduced in the embodiment of this specification. The execution subject of the model training method based on user classification can be a model training device based on user classification. The model training device based on user classification includes but is not limited to a server, an industrial computer, a PC, etc. Figure 1 As shown, the user classification-based model training method may include the following specific implementation steps.

[0023] S110: Acquire user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record.

[0024] User sample data can be sample data used to train a model. The user sample data can be basic information of the user, such as education, gender, age, etc., or historical records of the user's business operations. Specifically, the user sample data can include labeled data and unlabeled data. In the field of machine learning, modeling and learning based on partially labeled data and partially unlabeled data is fast and accurate, and has good application value.

[0025] The labeled data is data marked with corresponding labels. The label may be a user category. Specifically, the user category may be set based on the user's business processing record. The user category is the result obtained by classifying the user in combination with the user's business processing situation. Specifically, the user category may include a positive category and a negative category, which respectively indicate a high historical business processing volume and a low historical business processing volume.

[0026] Preferably, in order to reduce the time consumed in labeling data, the labeled data may correspond to only a small proportion of the overall user sample data.

[0027] In some implementations, labeling the user sample data may be performed by first determining the test users corresponding to the user sample data after determining the part of the user sample data that needs to be labeled. After that, the business processing records of the test users within the preset test time may be obtained, and then labels may be added to the user sample data corresponding to the test users based on the business processing records. The preset test time may be set based on the needs of the actual application, for example, it may be set to 3 records, 7 days, one month, three months, etc.

[0028] Using a specific example to illustrate, assume that the added labels include positive labels and negative labels. A positive label can indicate that the user has a relatively frequent business processing record, and a negative label can indicate that the user has almost no business processing record in the recent period. Accordingly, the preset test time can be set to 3 months. After obtaining the business processing record of the test user within three months, if the test user has a business processing record within three months, a positive label is marked; if the test user has no business processing record within three months, a negative label is marked.

[0029] The above implementation is only a specific example designed in combination with label categories. In actual applications, other methods can be used to label data based on different types of labels, which will not be described in detail here.

[0030] S120: Determine the user category probability of the unlabeled data by using the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category.

[0031] After obtaining the labeled data and the unlabeled data, the user category probability of the unlabeled data can be determined using the labeled data. In the case where the label of the labeled data has identified the corresponding user category, the probability that the unlabeled data corresponds to each user category can be determined based on the similarity between the unlabeled data and the labeled data. The user category probability is used to represent the probability that the unlabeled data belongs to each user category.

[0032] Specifically, the user category probability may be determined by neighboring labeled data. The neighboring labeled data is data in the labeled data whose difference with the unlabeled data is less than a specified difference threshold. The specified difference threshold is used to indicate the maximum difference between data in the same category.

[0033] In some embodiments, the difference between labeled data and unlabeled data may be determined by a K-nearest neighbor similarity algorithm. The idea of ​​the K-nearest neighbor similarity algorithm is that most of the k most similar samples in the feature space (i.e., the closest samples in the feature space) belong to a certain category, and by calculating the Euclidean distance between different sample data, the sample data can be classified according to the distance between different samples. The specific implementation method can be set based on the actual application situation, and the implementation process will not be repeated here.

[0034] After determining the neighboring labeled data of the unlabeled data, the number of the neighboring labeled data corresponding to different user categories can be counted to determine the user category probability. Assuming that the user categories include positive categories and negative categories, the corresponding user category probabilities also include positive category probabilities and negative category probabilities, then the positive category probability p + (x u ) = k + / k, where k + is the number of positive samples in the neighboring labeled data, k is the number of neighboring labeled data, and the negative category probability p - (x u ) = k - / k, where k - is the number of negative samples in the neighbor labeled data.

[0035] In some implementations, the user sample data may be preprocessed before obtaining the user category probability, including constructing the user sample data into original features corresponding to each user, and then completing the original features based on preset feature fields.

[0036] Since the data tables corresponding to the sample data can be obtained separately based on different data types, after obtaining the sample data, the data columns in different data tables can be aggregated according to the user identifier to obtain the original features corresponding to each user. Accordingly, in order to ensure that the original features can be effectively utilized in the subsequent process, the missing value columns in the original features can be completed. The preset feature fields can be fields corresponding to different completion rules. For example, for the missing values ​​of the numerical features in the original features, the preset feature fields can be completed with the value of "0"; for the missing values ​​of the non-numerical features, the preset feature fields can be completed with the value of "unknown". In actual applications, other preset feature fields can also be used to complete the original features as needed, which is not limited to the above examples and will not be repeated here.

[0037] In some implementations, the user sample data may be in the form of constructed features so that it can be better applied in a subsequent model training process, that is, the user sample data may be user information features.

[0038] When obtaining the user information features, the original features can be first constructed based on the collected user information. The specific method of constructing the original features can refer to the introduction in the above implementation mode, which will not be repeated here. Encode the category features in the original features, where the category features include at least one of education level and gender, and the encoding method can be One-Hot encoding.

[0039] Based on the different time features of user sample data, time series historical features and time series aggregation features can also be constructed.

[0040] The time series historical features can represent the features of user sample data in the corresponding time interval, which can be specifically used by F his =[feature time ,time=1,2,3,4,5,6], which can represent the features constructed using user sample data in different time intervals when the time periods are the first month, the second month, the third month, the fourth month, the fifth month, and the sixth month respectively.

[0041] The time series aggregation feature can represent the feature obtained based on the time interval and the method of obtaining the feature. Specifically, F agg =[f(feature) time ,time=1-3,1-6,1-9,1-12] to obtain the time series aggregation feature F agg , where f() takes Mean(), Max(), Min(), and Std() as the average value, and the time period can be the previous month, the previous three months, the previous six months, and the previous 12th month, respectively, so as to realize the aggregation of sample data features within a certain time interval based on a certain data collection method. The feature categories obtained in actual applications are not limited to the above specific examples, and will not be repeated here.

[0042] After obtaining the original features, time series historical features, and time series aggregated features, these features can be combined to obtain features for the subsequent model training process, thereby facilitating model training.

[0043] S130: constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories.

[0044] After obtaining the user category probability, a neighbor similarity regularization feature can be constructed based on the user category probability. The neighbor similarity regularization feature is feature data constructed under the consideration of the relationship between unlabeled data and labeled data, so as to fully learn the potential distribution information between unlabeled data and neighbor labeled data.

[0045] Specifically, when the user category includes a positive category and a negative category, the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - Negative category.

[0046] Preferably, when the user sample data is obtained based on different information categories, when constructing the neighbor similarity regularization feature, the neighbor similarity regularization feature can also be constructed for each information category. Specifically, in combination with the above formula, the subclassifier f can be a subclassifier corresponding to one of the information categories, and accordingly, each information category also corresponds to its own subclassifier. By respectively obtaining the neighbor similarity regularization features under different information categories, the differences in the corresponding data under different perspectives are maximized, the data discrimination is improved, and the training effect of the model is beneficial.

[0047] The information categories may be categories corresponding to different features of user sample data. In some embodiments, the information categories may include user basic information categories, business information categories, and transaction information categories. The user basic information category may be a category corresponding to the user's own information, such as user information such as education, age, and gender; the business information category may be a category to which the corresponding business belongs, such as in the financial field, it may be different types of business such as withdrawal business, transfer business, and inquiry business. The transaction information category can be used to represent the specific interaction information of the user when processing the business, such as business processing time, specific business processing process, and other information. In actual applications, other information categories can also be determined according to needs, and are not limited to the above examples, so they will not be repeated here.

[0048] S140: Generate classification regularization features corresponding to at least two information categories using the user sample data; the information categories correspond to categories of different features of the user sample data.

[0049] After constructing the nearest-neighbor similarity regularization feature, the user sample data can also be used to generate classification regularization features corresponding to different information categories. The classification regularization feature is a feature obtained after considering the mutual influence effect between features of different information categories.

[0050] In this embodiment, corresponding sub-classifiers are respectively provided for different information categories, and different sub-classifiers are used to classify users according to information of different information categories. In order to enable different sub-classifiers to achieve corresponding technical effects, the outputs of the corresponding sub-classifiers under different information categories are also different, so as to avoid the sub-classifiers from tending to be consistent and improve the generalization effect of the model. Accordingly, in the case where the outputs of each classifier are not completely consistent, it is also necessary to comprehensively consider the degree of influence of the results between different sub-classifiers, so as to achieve the optimization of the sub-classifiers. The classification regularization feature is a feature constructed to obtain the result differences of these sub-classifiers.

[0051] Specifically, we can use the formula Get the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the subclassifier corresponding to the user basic information category, f2 is the subclassifier corresponding to the business information category, and f3 is the subclassifier corresponding to the transaction information category. The subclassifier is used to determine the user category. The above formula is constructed when the information category is the basic information category, the business information category, and the transaction information category. In actual applications, the above formula can be adjusted accordingly based on the number of information categories and the specific types of information categories, which will not be repeated here.

[0052] S150: A user classification model is obtained by training the neighbor similarity regularization feature and the classification regularization feature; the user classification model is used to determine the user classification according to the user information to allocate the user a service corresponding to the user classification.

[0053] Since the neighbor similarity regularization feature and the classification regularization feature can both reflect the classification effect of the sub-classifier, the model corresponding to the sub-classifier can be optimized based on the values ​​of the neighbor similarity regularization feature and the classification regularization feature to obtain the final user classification model.

[0054] In some embodiments, the training process can be based on the calculation of the objective function based on the experience loss, L2 regularization loss, neighbor similarity regularization feature and classification regularization feature. The experience loss can be the loss obtained based on the labeled data, which is used to reflect the degree of fit of the subclassifier on the labeled data. The model can also be further optimized based on the size of the experience loss. The L2 regularization loss is used to limit the overfitting phenomenon caused by too many parameters and improve the generalization ability of the model.

[0055] The objective function can be a function used to evaluate the model effect. Specifically, it can be a function using the formula Calculate the objective function, where L is the objective function, R emp is the experience loss, R ns is the neighbor similarity regularization feature, R vm is the classification regularization feature, R reg is the L2 regularization loss, and α, β, and γ are hyperparameters.

[0056] After obtaining the result corresponding to the objective function, the sub-classifier can be optimized based on the value of the objective function. Specifically, the objective function can be combined to solve the corresponding optimization problem using the gradient descent method. Repeat the steps of obtaining the objective function corresponding to the sub-classifier and optimizing the sub-classifier based on the objective function until the preset number of iterations is reached or the difference in the loss value of the test results before and after optimization is less than the preset threshold, and the optimization of the sub-classifier is completed. The sub-classifier can be used to build the final user classification model. Based on the above example, when the information categories include basic information categories, business information categories, and transaction information categories, the optimized sub-classifiers are f1, f2, and f3, respectively, corresponding to the above information categories.

[0057] The user classification model may be a model obtained after comprehensive optimization of the sub-classifiers. Specifically, the formula Get the user classification model, where F(x) is the user classification model, f v is a subclassifier, x is user information, ω1 is the first user category, and ω2 is the second user category. The above examples are only formulas set when the subclassifier corresponds to the above three information categories. In actual applications, the corresponding formula for generating a user classification model can be determined according to the specific information category, which is not limited to the above examples and will not be repeated here.

[0058] In some implementations, after the user classification model is trained, the user classification model may be tested using test data. Specifically, the formula Test. By inputting the corresponding test data x and comparing the test results with the actual categories corresponding to the test data, the accuracy of the user classification model can be verified.

[0059] After the user classification model is obtained, the data to be analyzed can be input into the user classification model, and the corresponding service can be determined according to the output user classification result, so as to allocate the service to the user.

[0060] Combine the following Figure 2 and Figure 3 , a specific example is used to illustrate the process of the above method.

[0061] Figure 2 It is a flowchart corresponding to the model training process. Among them, after obtaining a small number of labeled samples and a large number of unlabeled samples, the experience loss can be calculated separately using the small number of labeled samples. In combination with the small number of labeled samples and the large number of unlabeled samples, the corresponding neighbor information is constructed from the basic information perspective, the transaction information perspective and the product information perspective, and then the neighbor information is used to determine the neighbor similarity regularization term under each perspective. In addition, by integrating the information from each perspective, the perspective diversity regularization term can also be determined. Combining the experience loss, perspective diversity regularization, neighbor similarity regularization term and L2 regularization, the training of the user classification model can be realized.

[0062] Figure 3 The figure is a schematic diagram corresponding to the specific training and testing process. After obtaining data from the data warehouse, the data is preprocessed, and after constructing the corresponding features based on feature engineering, these features are divided into training samples and test samples. The training samples are used to complete the training of the user classification model, and the test samples can be used to obtain the prediction results corresponding to the user classification model. Then, the accuracy of the model can be judged based on the prediction results to achieve model optimization.

[0063] Based on the introduction of the above embodiments and examples, it can be seen that after obtaining the user sample data, the method only annotates part of the data, and then determines the user category probability and the neighbor similarity regularization feature in turn while considering the correlation between the labeled data and the unlabeled data, and combines the classification regularization feature corresponding to the information category to realize the training of the user classification model, so that the user classification model can be used to complete the classification of users, and then the corresponding services are assigned to users according to the user category. Through the above method, the time and resources consumed by a large amount of labeled data are reduced, the correlation between data is fully explored, the generalization effect of the classification model is optimized, and the user's business processing experience is improved.

[0064] based on Figure 1The corresponding model training method based on user classification introduces a service allocation method based on user classification in the embodiment of this specification. The execution subject of the service allocation method based on user classification can be a service allocation device based on user classification. Figure 4 As shown, the service allocation method based on user classification may include the following specific implementation steps.

[0065] S410: Obtain user characteristic information of the target user.

[0066] The target user may be a user to whom a service needs to be assigned, or to whom a service needs to be predicted. User characteristic information is information corresponding to the target user. The user characteristic information may be set according to the requirements of the user classification model. For example, the user characteristic information may be the identity information of the user himself, or may be information of historical services handled by the user, etc., without limitation.

[0067] S420: Input the user feature information into a user classification model to obtain user categories; the user classification model is obtained in the following manner: after obtaining user sample data containing labeled data and unlabeled data, determine the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; construct a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories; generate classification regularization features corresponding to at least two information categories using the user sample data; the information category corresponds to the category of different features of the user sample data; and obtain the user classification model by training the neighbor similarity regularization features and the classification regularization features.

[0068] After obtaining the user characteristic information, the user characteristic information can be input into the user classification model. Since the user classification model can determine the category corresponding to the user according to the user characteristic information, the corresponding user category can be output. For a specific introduction to the user classification model and the method for obtaining it, please refer to Figure 1 The introduction in the corresponding embodiments will not be repeated here.

[0069] S430: Allocate the service corresponding to the user category to the target user.

[0070] Based on different user categories, corresponding services can be preset, and after the user category corresponding to the target user is determined, the service corresponding to the user category can be allocated to the target user.

[0071] The specific correspondence between the user category and the service can be directly specified by the management personnel, or can be obtained through training based on historical data. The specific acquisition method can be set based on the actual application situation, which will not be described in detail here.

[0072] based on Figure 1 The corresponding model training method based on user classification introduces a model training device based on user classification in the embodiment of this specification. Figure 5 As shown, the user classification-based model training device includes the following modules.

[0073] The user sample data acquisition module 510 is used to acquire user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record.

[0074] The user category probability determination module 520 is used to determine the user category probability of unlabeled data by using the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category.

[0075] The neighbor similarity regularized feature construction module 530 is used to construct a neighbor similarity regularized feature based on the user category probability; the neighbor similarity regularized feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories.

[0076] The classification regularization feature generation module 540 is used to generate classification regularization features corresponding to at least two information categories using the user sample data; the information categories correspond to categories of different features of the user sample data.

[0077] The user classification model acquisition module 550 is used to obtain a user classification model by integrating the neighbor similarity regularization features and classification regularization features; the user classification model is used to determine the user classification according to the user information to allocate the user the service corresponding to the user classification.

[0078] based on Figure 4 The corresponding service allocation method based on user classification introduces a service allocation device based on user classification in the embodiment of this specification. Figure 6 As shown, the service allocation device based on user classification includes the following modules.

[0079] The user characteristic information acquisition module 610 is used to acquire the user characteristic information of the target user.

[0080] The user category acquisition module 620 is used to input the user feature information into the user classification model to obtain the user category; the user classification model is obtained in the following manner: after obtaining user sample data containing labeled data and unlabeled data, the user category probability is determined according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; based on the user category probability, a neighbor similarity regularization feature is constructed; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; the user sample data is used to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; the user classification model is obtained by training the neighbor similarity regularization features and the classification regularization features.

[0081] The service allocation module 630 is configured to allocate the service corresponding to the user category to the target user.

[0082] based on Figure 1 Corresponding to the user classification-based model training method, the embodiment of this specification provides a user classification-based model training device. Figure 7 As shown, the user classification-based model training device may include a memory and a processor.

[0083] In this embodiment, the memory may be implemented in any appropriate manner. For example, the memory may be a read-only memory, a mechanical hard disk, a solid-state hard disk, or a USB flash drive, etc. The memory may be used to store computer program instructions.

[0084] In this embodiment, the processor may be implemented in any suitable manner. For example, the processor may take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. The processor can execute the computer program instructions to implement the following steps: obtaining user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; determining the user category probability of the unlabeled data through the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; training a user classification model by combining the neighbor similarity regularization features and the classification regularization features; the user classification model is used to determine the user category based on the user information to assign the user the business corresponding to the user category.

[0085] based on Figure 4 Corresponding to the service allocation method based on user classification, the embodiment of this specification provides a service allocation device based on user classification. Figure 8 As shown, the service allocation device based on user classification may include a memory and a processor.

[0086] In this embodiment, the memory may be implemented in any appropriate manner. For example, the memory may be a read-only memory, a mechanical hard disk, a solid-state hard disk, or a USB flash drive, etc. The memory may be used to store computer program instructions.

[0087] In this embodiment, the processor may be implemented in any suitable manner. For example, the processor may take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. The processor can execute the computer program instructions to implement the following steps: obtaining user feature information of the target user; inputting the user feature information into a user classification model to obtain a user category; the user classification model is obtained in the following manner: after obtaining user sample data containing labeled data and unlabeled data, determining the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; comprehensively training the neighbor similarity regularization features and the classification regularization features to obtain a user classification model; and allocating the business corresponding to the user category to the target user.

[0088] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog2. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0089] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0090] It can be known from the above description of the implementation mode that the technicians in this field can clearly understand that the present specification can be implemented by means of software plus the necessary first hardware platform. Based on such an understanding, the technical solution of the present specification can be essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present specification or some parts of the embodiments.

[0091] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0092] This specification can be used in many first or special computer system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0093] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0094] Although the present specification is described through embodiments, those skilled in the art will appreciate that there are many modifications and changes to the present specification without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A model training method based on user classification, characterized in that: include: Acquire user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; Determine the user category probability of the unlabeled data by using the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; Constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories; Generating classification regularization features corresponding to at least two information categories using the user sample data; The information categories correspond to categories of different characteristics of user sample data; A user classification model is obtained by training the neighbor similarity regularization feature and the classification regularization feature; the user classification model is used to determine the user classification according to the user information to allocate the user the service corresponding to the user classification; The user categories include positive categories and negative categories; The constructing of neighbor similarity regularization features based on the user category probability includes: Using the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - It is the negative category; The step of generating classification regularization features corresponding to at least two information categories using the user sample data includes: when the number of information categories is 3, using the formula Get the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the sub-classifier corresponding to the first information category, f2 is the sub-classifier corresponding to the second information category, and f3 is the sub-classifier corresponding to the third information category, and the sub-classifiers are used to determine the user category.

2. The method according to claim 1, characterized in that Before determining the user category probability of the unlabeled data by using the neighboring labeled data corresponding to the unlabeled data, the method further includes: The neighbor labeled data corresponding to the unlabeled data is obtained based on the K nearest neighbor similarity algorithm.

3. The method according to claim 1, characterized in that The neighbor similarity regularization feature includes features corresponding to each information category respectively.

4. The method according to claim 1, characterized in that The user classification model is obtained by integrating the neighbor similarity regularization features and the classification regularization features, including: Obtain empirical loss based on labeled data; Calculate the objective function based on empirical loss, L2 regularization loss, neighbor similarity regularization feature and classification regularization feature; The subclassifier is optimized according to the objective function; the subclassifier is used to determine the user category corresponding to the user sample data under different information categories; the subclassifier is also used to construct the neighbor similarity regularization feature and the classification regularization feature; The user classification model is obtained by comprehensively optimizing the sub-classifiers.

5. The method according to claim 4, characterized in that The optimizing the sub-classifier according to the objective function comprises: The steps of obtaining the objective function corresponding to the sub-classifier and optimizing the sub-classifier based on the objective function are repeatedly performed until a preset number of iterations is reached or the difference in loss values ​​of the test results before and after optimization is less than a preset threshold.

6. The method according to claim 4, characterized in that The objective function is calculated based on the empirical loss, L2 regularization loss, neighbor similarity regularization feature and classification regularization feature, including: Using the formula Calculate the objective function, where L is the objective function, R emp is the experience loss, R ns is the neighbor similarity regularization feature, R vm is the classification regularization feature, R reg is the L2 regularization loss, and α, β, and γ are hyperparameters.

7. The method according to claim 4, characterized in that The user classification model obtained by the comprehensive optimized sub-classifier includes: Using the formula Get the user classification model, where F(x) is the user classification model, f v is a sub-classifier, x is user information, ω1 is the first user category, and ω2 is the second user category.

8. The method according to claim 1, characterized in that The information category includes at least one of a user basic information category, a business information category, and a transaction information category.

9. The method according to claim 1, characterized in that Before determining the user category probability of the unlabeled data by using the neighboring labeled data corresponding to the unlabeled data, the method further includes: The user sample data is preprocessed; the preprocessing includes: constructing original features corresponding to each user based on the user sample data; and completing the original features based on preset feature fields.

10. The method according to claim 1, characterized in that The user sample data includes user information features; the acquiring of the user sample data includes: Constructing original features based on the user sample data; Encoding the category features in the original features; the category features include at least one of education level and gender; Construct time series historical features through user sample data in different time intervals; Use the average, maximum, minimum, and standard deviation of user sample data in time distribution to build time series aggregation features; The original features, time series historical features, and time series aggregate features are combined to obtain user information features.

11. The method according to claim 1, characterized in that Before obtaining the user sample data, the method further includes: Determine the test users corresponding to some user sample data; Obtaining the business processing records of the test user within a preset test time; A label is added to the user sample data corresponding to the test user based on the business processing record.

12. A model training device based on user classification, characterized in that: include: A user sample data acquisition module is used to acquire user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; A user category probability determination module is used to determine the user category probability of the unlabeled data by using the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes the labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; A neighbor similarity regularization feature construction module, used to construct a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between unlabeled data and labeled data of different user categories; A classification regularization feature generation module, used to generate classification regularization features corresponding to at least two information categories using the user sample data; The information categories correspond to categories of different characteristics of user sample data; A user classification model acquisition module is used to obtain a user classification model by training the neighbor similarity regularization feature and the classification regularization feature; the user classification model is used to determine the user classification according to the user information to allocate the user a service corresponding to the user classification; The user categories include positive categories and negative categories; the neighbor similarity regularization feature construction module is specifically used for: Using the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - It is the negative category; The classification regularization feature generation module is specifically used to: when the number of information categories is 3, use the formula Get the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the sub-classifier corresponding to the first information category, f2 is the sub-classifier corresponding to the second information category, and f3 is the sub-classifier corresponding to the third information category, and the sub-classifiers are used to determine the user category.

13. A model training device based on user classification, comprising a memory and a processor; The memory is used to store computer program instructions; The processor is used to execute the computer program instructions to implement the following steps: obtaining user sample data; the user sample data includes labeled data and unlabeled data; the labeled data corresponds to a user category based on the user's business processing record; determining the user category probability of the unlabeled data through the neighboring labeled data corresponding to the unlabeled data; the neighboring labeled data includes labeled data whose difference with the unlabeled data is less than a specified gap threshold; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; The information categories correspond to categories of different features of user sample data; a user classification model is obtained by training the neighbor similarity regularization features and the classification regularization features; the user classification model is used to determine the user classification according to the user information to allocate the user the service corresponding to the user classification; The user categories include positive categories and negative categories; The constructing of neighbor similarity regularization features based on the user category probability includes: Using the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - It is the negative category; The step of generating classification regularization features corresponding to at least two information categories using the user sample data includes: when the number of information categories is 3, using the formula Get the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the sub-classifier corresponding to the first information category, f2 is the sub-classifier corresponding to the second information category, and f3 is the sub-classifier corresponding to the third information category, and the sub-classifiers are used to determine the user category.

14. A service allocation method based on user classification, characterized in that: include: Obtain user characteristic information of target users; Inputting the user characteristic information into a user classification model to obtain a user category; The user classification model is obtained in the following manner: after obtaining user sample data including labeled data and unlabeled data, determining user category probabilities according to the labeled data and the unlabeled data; the user category probabilities include probabilities that the unlabeled data belongs to each user category; constructing neighbor similarity regularization features based on the user category probabilities; the neighbor similarity regularization features include data constructed based on the correspondence between the unlabeled data and labeled data of different user categories; and generating classification regularization features corresponding to at least two information categories using the user sample data; The information categories correspond to categories of different features of user sample data; the user classification model is obtained by training the neighbor similarity regularization features and classification regularization features; Allocating the service corresponding to the user category to the target user; The user categories include positive categories and negative categories; The constructing of neighbor similarity regularization features based on the user category probability includes: Using the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - It is the negative category; The step of generating classification regularization features corresponding to at least two information categories using the user sample data includes: when the number of information categories is 3, using the formula Get the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the sub-classifier corresponding to the first information category, f2 is the sub-classifier corresponding to the second information category, and f3 is the sub-classifier corresponding to the third information category, and the sub-classifiers are used to determine the user category.

15. A service allocation device based on user classification, characterized in that: include: A user feature information acquisition module, used to acquire user feature information of a target user; A user category acquisition module, used for inputting the user feature information into a user classification model to obtain a user category; the user classification model is obtained in the following manner: after obtaining user sample data including labeled data and unlabeled data, determining the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; The information categories correspond to categories of different features of user sample data; the user classification model is obtained by training the neighbor similarity regularization features and classification regularization features; A service allocation module, used for allocating services corresponding to the user category to the target user; The user category includes a positive category and a negative category; when the user category acquisition module constructs a neighbor similarity regularization feature based on the user category probability, it is specifically used to: Using the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - It is the negative category; When the user category acquisition module generates classification regularization features corresponding to at least two information categories using the user sample data, it is specifically used to: when the number of information categories is 3, use the formula Get the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the sub-classifier corresponding to the first information category, f2 is the sub-classifier corresponding to the second information category, and f3 is the sub-classifier corresponding to the third information category, and the sub-classifiers are used to determine the user category.

16. A service allocation device based on user classification, comprising a memory and a processor; The memory is used to store computer program instructions; The processor is used to execute the computer program instructions to implement the following steps: obtaining user feature information of a target user; inputting the user feature information into a user classification model to obtain a user category; the user classification model is obtained in the following manner: after obtaining user sample data containing labeled data and unlabeled data, determining the user category probability according to the labeled data and the unlabeled data; the user category probability includes the probability that the unlabeled data belongs to each user category; constructing a neighbor similarity regularization feature based on the user category probability; the neighbor similarity regularization feature includes data constructed based on the correspondence between the unlabeled data and the labeled data of different user categories; using the user sample data to generate classification regularization features corresponding to at least two information categories; the information category corresponds to the category of different features of the user sample data; comprehensively training the neighbor similarity regularization feature and the classification regularization feature to obtain a user classification model; allocating the business corresponding to the user category to the target user; The user categories include positive categories and negative categories; and constructing neighbor similarity regularization features based on the user category probabilities includes: Using the formula Determine the neighbor similarity regularization feature, where R ns is the neighbor similarity regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, p + (x u ) is the user category probability of the unlabeled data corresponding to the positive category, p - (x u ) is the user category probability of the unlabeled data corresponding to the negative category, is the classification result obtained after the sample data is input into the sub-classifier, and the sub-classifier is used to determine the user category, ω + is the positive category, ω - It is the negative category; The step of generating classification regularization features corresponding to at least two information categories using the user sample data includes: when the number of information categories is 3, using the formula Take the classification regularization feature, where R vm is the classification regularization feature, X U is unlabeled data, |X U | is the number of unlabeled data, f1 is the sub-classifier corresponding to the first information category, f2 is the sub-classifier corresponding to the second information category, and f3 is the sub-classifier corresponding to the third information category, and the sub-classifiers are used to determine the user category.

Images