User data recognition model updating method, user data recognition method and device

Abstract

The application discloses a user data recognition model updating method, a user data recognition method and device. The user data recognition model updating method comprises the following steps: acquiring first training data and a first recognition model to be updated, wherein the first recognition model comprises a label recognition output layer; training the first recognition model according to the first training data to obtain a second recognition model and an evaluation value of a recognition capability of the second recognition model; when the evaluation value is greater than or equal to a preset threshold value, iteratively training the first recognition model according to a preset weight adjustment rule until the evaluation value is less than the preset threshold value, taking first migration weights of the second recognition model, which comprise the label recognition output layer, as target migration weights, and taking the second recognition model corresponding to the evaluation value less than the preset threshold value as a target recognition model. According to the embodiment of the application, the accuracy of the updated recognition model in recognizing user data can be improved.

Description

Technical Field

[0001] This application belongs to the field of data processing, and in particular relates to a user data identification model update method, a user data identification method and apparatus. Background Technology

[0002] In an environment where big data technology is becoming increasingly mature, artificial intelligence is becoming more widespread, and market competition is becoming increasingly fierce, making full use of data mining and machine learning technologies will help operators provide refined and intelligent services.

[0003] As a product of data mining, identification models play a crucial role in areas such as target customer segmentation, potential customer identification, and customer behavior prediction. However, rebuilding an identification model is time-consuming and requires significant computational and human resources. Currently, model transfer learning is commonly used, leveraging the parameters of the identification model from the old billing period to assist in building the model for the new billing period. However, during user data identification, this model transfer learning technique updates the identification model using training data. As the time difference between user data in the new billing period and user data in the old billing period increases, the distribution differences between the two billing periods also widen, leading to a gradual decrease in the predictive accuracy of the updated identification model. Summary of the Invention

[0004] This application provides a user data identification model update method, a user data identification method, and an apparatus, which can improve the accuracy of the updated identification model in identifying user data.

[0005] In a first aspect, embodiments of this application provide a user data identification model update method, including:

[0006] Acquire first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label.

[0007] The first recognition model is trained based on the first training data to obtain the second recognition model and the evaluation value of the recognition ability of the second recognition model. The second recognition model includes the first transfer weight of the label recognition output layer.

[0008] When the evaluation value is greater than or equal to the preset threshold, the first transfer weight and the first training data of the label recognition output layer in the second recognition model are adjusted according to the preset weight adjustment rules, and the process returns to the step "train the first recognition model according to the first training data to obtain the second recognition model". The first recognition model is iteratively trained until the evaluation value is less than the preset threshold. The first transfer weight of the second recognition model, including the label recognition output layer, is taken as the target transfer weight, and the second recognition model with the evaluation value less than the preset threshold is taken as the target recognition model.

[0009] In some possible implementations of the first aspect, the multiple first sample data include a first sample set and a second sample set, wherein the first sample set includes first sample data obtained in a first time period, and the second sample set includes first sample data obtained in a second time period, the second time period being the time following the first time period. The first recognition model is trained based on the first training data to obtain a second recognition model and an evaluation value of the recognition capability of the second recognition model, including:

[0010] The first recognition model is trained based on the first sample set to obtain the third recognition model, and the first recognition model is trained based on the second sample set to obtain the fourth recognition model.

[0011] Obtain the second transfer weights of the label recognition output layer in the third recognition model and the third transfer weights of the label recognition output layer in the fourth recognition model;

[0012] The first transfer weight of the label recognition output layer is determined based on the preset weight transfer function, the second transfer weight, and the third transfer weight;

[0013] The first recognition model is configured based on the first migration weight, and the second recognition model is obtained.

[0014] The first training data, including multiple first sample data, is identified according to the second identification model to obtain the first identification result set.

[0015] Based on the first recognition result set and the preset label corresponding to each first sample data, determine the evaluation value of the recognition capability of the second recognition model.

[0016] In some possible implementations of the first aspect, the first transfer weights and the first training data of the label recognition output layer in the second recognition model are adjusted according to a preset weight adjustment rule, including:

[0017] Based on the recognition results of each first sample data in the first recognition result set and the preset label corresponding to each first sample data, the first sample data whose recognition results are different from the preset label are used as the second sample data to obtain the second training dataset.

[0018] According to the preset weight adjustment rules, the first learning weight of each second sample data and the second learning weight of the preset label corresponding to the second sample data are adjusted respectively.

[0019] Secondly, embodiments of this application provide a user data identification method, including:

[0020] Obtain user data to be identified;

[0021] User data is input into the recognition model to obtain the recognition result of the user data. The weight of the label recognition output layer of the recognition model is the target migration weight, which is obtained according to the user data recognition model update method described in the first aspect or any implementable method of the first aspect.

[0022] In some possible implementations of the second aspect, the recognition results of user data are saved to a preset database for use in updating the trained user data recognition model after a preset time period, based on the user data and the recognition results of user data in the preset database within the preset time period.

[0023] Thirdly, embodiments of this application provide a user data identification model update apparatus, the apparatus comprising:

[0024] The acquisition module is used to acquire first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label.

[0025] The processing module is used to train the first recognition model based on the first training data to obtain a second recognition model and an evaluation value of the recognition capability of the second recognition model, wherein the second recognition model includes a first transfer weight of the label recognition output layer.

[0026] The processing module is further configured to, when the evaluation value is greater than or equal to a preset threshold, adjust the first transfer weight of the label recognition output layer in the second recognition model and the first training data respectively according to the preset weight adjustment rules, and return to the step "train the first recognition model according to the first training data to obtain the second recognition model", iteratively train the first recognition model until the evaluation value is less than the preset threshold, take the first transfer weight of the second recognition model including the label recognition output layer as the target transfer weight, and take the second recognition model corresponding to the evaluation value being less than the preset threshold as the target recognition model.

[0027] Fourthly, embodiments of this application provide a user data identification device, the device comprising:

[0028] The acquisition module is used to acquire user data to be identified;

[0029] The processing module is used to input user data into the recognition model to obtain the recognition result of the user data. The weight of the label recognition output layer of the recognition model is the target migration weight, which is obtained according to the user data recognition model update method described in the first aspect or any implementable method of the first aspect.

[0030] Fifthly, this application provides a user data identification device, the device comprising: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the user data identification model update method as described in the first aspect or any implementable embodiment of the first aspect, or executes the user data identification method as described in the second aspect or any implementable embodiment of the second aspect.

[0031] Sixthly, this application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the user data identification model update method as described in the first aspect or any implementable embodiment of the first aspect, or perform the user data identification method as described in the second aspect or any implementable embodiment of the second aspect.

[0032] In a seventh aspect, embodiments of this application provide a computer program product, wherein instructions in the computer program product, when executed by a processor of an electronic device, cause the electronic device to perform a user data identification model update method as described in the first aspect or any implementable of the first aspect, or to perform a user data identification method as described in the second aspect or any implementable of the second aspect.

[0033] The user data recognition model update method, user data recognition method, and apparatus of this application embodiment acquire first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label. Next, the first recognition model is trained according to the first training data to obtain a second recognition model and an evaluation value of the recognition ability of the second recognition model. The second recognition model includes a first transfer weight of the label recognition output layer. When the evaluation value is greater than or equal to a preset threshold, the first transfer weight of the label recognition output layer in the second recognition model and the first training data are adjusted according to a preset weight adjustment rule. The process returns to the step "train the first recognition model according to the first training data to obtain the second recognition model". The first recognition model is iteratively trained until the evaluation value is less than the preset threshold. The first transfer weight of the second recognition model, including the label recognition output layer, is taken as the target transfer weight. According to the embodiments of this application, the weight update mechanism of the label recognition output layer uses the real label results to update the model parameters, which can continuously optimize the parameters of the recognition model and improve the accuracy of the updated recognition model in recognizing user data. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating a user data identification model update method provided in an embodiment of this application;

[0036] Figure 2 This is a flowchart illustrating a user data identification method provided in an embodiment of this application;

[0037] Figure 3 This is a schematic diagram of the structure of a user data identification model updating device provided in an embodiment of this application;

[0038] Figure 4 This is a schematic diagram of the structure of a user data identification device provided in an embodiment of this application;

[0039] Figure 5 This is a schematic diagram of the structure of a user data identification device provided in an embodiment of this application. Detailed Implementation

[0040] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0041] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0042] It should be noted that the implementation of this application constitutes legal use. The acquisition, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0043] In an environment where big data technology is maturing, artificial intelligence is becoming increasingly prevalent, and market competition is intensifying, fully leveraging data mining and machine learning technologies is beneficial for operators to provide refined and intelligent services. Against this backdrop, combining the characteristics of operators' service products and the characteristics of the selected channels, and utilizing big data resources to establish identification models that recognize the services users need, is a crucial guarantee for providing accurate services to customers.

[0044] As a product of data mining, identification models play a crucial role in areas such as target customer segmentation, potential customer identification, and customer behavior prediction. However, due to the characteristics of operator data, such as high noise levels, numerous missing data points, and rapid updates, user data exhibits distributional differences across different time periods after updates. For example, due to environmental factors, average data usage may decrease while voice usage may increase across all provinces in certain months. Therefore, it is difficult to directly and accurately identify these newly generated billing cycle data using identification models from older billing cycles.

[0045] Rebuilding a recognition model is time-consuming and requires significant computational and human resources. Currently, model transfer learning is commonly used, leveraging the recognition model parameters from the old billing period data to assist in building the model for the new billing period. However, during user data recognition, model transfer learning updates the recognition model using training data. As the time difference between user data in the new billing period and user data in the old billing period increases, the distribution differences between the two billing periods also widen. This leads to a gradual decrease in the prediction accuracy of the updated recognition model.

[0046] To address this, this application provides a user data recognition model update method, user data recognition method, and apparatus. The method involves acquiring first training data and a first recognition model to be updated. The first recognition model includes a label recognition output layer, which outputs the recognition result for the category of a preset label. Next, the first recognition model is trained based on the first training data to obtain a second recognition model and an evaluation value of the recognition capability of the second recognition model. The second recognition model includes a first transfer weight of the label recognition output layer. When the evaluation value is greater than or equal to a preset threshold, the first transfer weight of the label recognition output layer in the second recognition model and the first training data are adjusted according to a preset weight adjustment rule. The process then returns to the step "training the first recognition model based on the first training data to obtain the second recognition model," and iteratively trains the first recognition model until the evaluation value is less than the preset threshold. The first transfer weight of the second recognition model, including the label recognition output layer, is then used as the target transfer weight. According to this application, the weight update mechanism for the label recognition output layer uses real label results to update model parameters, enabling continuous optimization of the recognition model's parameters and improving the accuracy of the updated recognition model in recognizing user data.

[0047] The user data identification model update method provided in the embodiments of this application is described below with reference to the accompanying drawings. Figure 1 A flowchart illustrating a user data identification model update method according to an embodiment of this application is shown. Figure 1 As shown, the method may include the following steps:

[0048] Step 110: Obtain the first training data and the first recognition model to be updated.

[0049] The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label.

[0050] Specifically, the first sample data can be service data generated by users when using services provided by the operator. These services include, for example, internet data services and voice services, without specifying any particular service. Analyzing user service data can, for example, determine whether a user is a potential churner or a potential customer for a specific service.

[0051] Before updating and training the first identification model, the first identification model is an identification model trained based on historical data. For example, the first identification model can be an identification model for identifying whether a user is a potential churned user, and correspondingly, the categories of preset labels can include non-churned users or potential churned users.

[0052] Therefore, in order for the first recognition model to effectively identify newly generated user data, it is necessary to update the first recognition model.

[0053] Step 120: Train the first recognition model based on the first training data to obtain the second recognition model and the evaluation value of the recognition ability of the second recognition model.

[0054] The second recognition model includes the first transfer weights of the label recognition output layer.

[0055] The first training data includes multiple first sample data and a preset label corresponding to each first sample data. Therefore, after training the first recognition model on the first training data, the second recognition model is obtained.

[0056] The evaluation value of the recognition ability of the second recognition model can be, for example, the G-mean value, which can take into account both accuracy and recall.

[0057] In some embodiments, when the evaluation value is less than a preset threshold, the model update can be considered complete, thereby obtaining a second recognition model with the recognition accuracy included.

[0058] Step 130: When the evaluation value is greater than or equal to the preset threshold, the first transfer weight and the first training data of the label recognition output layer in the second recognition model are adjusted according to the preset weight adjustment rules, and the process returns to step 120 to iteratively train the first recognition model until the evaluation value is less than the preset threshold. The first transfer weight of the second recognition model, including the label recognition output layer, is taken as the target transfer weight, and the second recognition model with the evaluation value less than the preset threshold is taken as the target recognition model.

[0059] In some embodiments, the multiple first sample data include a first sample set and a second sample set, wherein the first sample set includes first sample data obtained in a first time period, and the second sample set includes first sample data obtained in a second time period, wherein the second time period is the time after the first time period.

[0060] For example, the first recognition model is updated according to a preset update cycle. The first sample data can be the sample data used before the last update training of the first recognition model, i.e., the first sample data obtained within a first time period. The second sample data is the business data generated by the user in the most recent update cycle, i.e., the first sample data obtained within a second time period. It can be understood that the second time period is the time following the first time period, and the first sample data within the second time period is usually less than the first sample data within the first time period.

[0061] Step 120 in this application may specifically include the following steps:

[0062] Step 121: Train the first recognition model based on the first sample set to obtain the third recognition model, and train the first recognition model based on the second sample set to obtain the fourth recognition model.

[0063] Specifically, the first recognition model can be a model for recognizing user data built based on a Weighted Transfer Extreme Learning Machine (WTELM). This first recognition model includes an input layer and a hidden layer. Before the first recognition model updates its training, the weights of the input layer are adjusted. and the weights of the hidden layer Perform random initialization, where, and The value range is [0, 1], and the number of hidden layer nodes in the first recognition model is L.

[0064] Step 122: Obtain the second transfer weight of the label recognition output layer in the third recognition model and the third transfer weight of the label recognition output layer in the fourth recognition model.

[0065] The first recognition model includes a label recognition output layer. After training the first recognition model according to the first sample set and the second sample set respectively, different transfer weights for label recognition output can be obtained, namely, the second transfer weight and the third transfer weight.

[0066] Step 123: Determine the first transfer weight of the label recognition output layer based on the preset weight transfer function, the second transfer weight, and the third transfer weight.

[0067] To enable the model to effectively identify user data generated within a recent period, during update training, the first sample data in the second time period can be used as the target, and the first sample data in the first time period can be used as the source data. The objective function can be defined as follows:

[0068]

[0069] in, The hidden layer output weights of the third recognition model. These are the output weights of the hidden layer of the fourth recognition model. This is the activation function matrix of the third recognition model. This is the activation function matrix of the fourth recognition model. The preset label for the first sample data within the first time period. These are the preset labels for the first sample data within the second time period. This is a preset balance coefficient between the first sample data in the first time period and the first sample data in the second time period.

[0070] The objective function also includes a model parameter approximation term. The approximation term can make and It tends to be consistent, reduces overfitting, improves transfer efficiency, and increases model update speed. This represents the empirical error of the preset labels in the first sample data. This represents the empirical error of the preset labels in the second sample data.

[0071] To simplify the objective function, after calculation, it can be simplified as shown in formula (1).

[0072] (1)

[0073] in, , , I is the identity matrix, L is the number of nodes in the hidden layer; N is actually n. . This represents the number of the first sample within the first time period. This represents the number of the first samples within the second time period.

[0074] For example, the preset labels for each first sample data within the first time period can be represented in matrix form: The preset labels for each first sample data point within the second time period can be represented in matrix form:

[0075] As a specific example, the second and third transfer weights are substituted into the formula (1) to determine the first transfer weight of the label recognition output layer.

[0076] Step 124: Configure the first recognition model according to the first migration weight to obtain the second recognition model.

[0077] Step 125: Based on the second recognition model, the first training data, including multiple first sample data, are identified respectively to obtain the first recognition result set.

[0078] Specifically, by identifying the first training data, which includes multiple first sample data, according to the second recognition model, the recognition result for each first sample data can be obtained, and the recognition result includes the recognition label for the first sample data.

[0079] Taking the identification of whether a user is a potential churner as an example, the preset labels for the first sample data are, for example, "non-churner" or "potential churner." The identification result includes the identification label of the first sample data as either "non-churner" or "potential churner."

[0080] Step 126: Determine the evaluation value of the recognition capability of the second recognition model based on the first recognition result set and the preset label corresponding to each first sample data.

[0081] When the recognition result matches the preset label, it indicates that the second recognition model correctly identified the first sample data; when the recognition result differs from the preset label, it indicates that the second recognition model incorrectly identified the first sample data. Therefore, based on the first recognition result set and the preset label corresponding to each first sample data, an evaluation value for the recognition capability of the second recognition model is determined. The evaluation value for recognition capability can, for example, be the G-mean value.

[0082] After obtaining the evaluation value, the relationship between the evaluation value and the preset threshold can be determined. When the evaluation value is less than the preset threshold, the first migration weight of the second recognition model, including the label recognition output layer, is used as the target migration weight, and the second recognition model corresponding to the evaluation value being less than the preset threshold is used as the target recognition model.

[0083] When the evaluation value is greater than or equal to the preset threshold, the first transfer weight and the first training data of the label recognition output layer in the second recognition model are adjusted according to the preset weight adjustment rules, and the process returns to step 120 to iteratively train the first recognition model until the evaluation value is less than the preset threshold.

[0084] In some embodiments, the first transfer weights and the first training data of the label recognition output layer in the second recognition model are adjusted according to a preset weight adjustment rule, including:

[0085] Step 131: Based on the recognition results of each first sample data in the first recognition result set and the preset label corresponding to each first sample data, the first sample data whose recognition results are different from the preset label is used as the second sample data to obtain the second training dataset.

[0086] Based on the recognition result of each first sample data in the first recognition result set and the preset label corresponding to each first sample data, the first sample data with recognition errors in the k-th iteration can be found. For example, the first sample data with recognition errors can be represented as:

[0087] Step 132: According to the preset weight adjustment rules, adjust the first learning weight of each second sample data and the second learning weight of the preset label corresponding to the second sample data respectively.

[0088] The preset weight adjustment rules include: , where k is the number of iterations, and M is the number of first samples that were misidentified.

[0089] Meanwhile, in order to improve the learning amount of the second sample data, the contribution of the preset label is increased. As a specific example, the contribution of the preset label of the second sample data can be adjusted as follows:

[0090]

[0091] Where k is the number of iterations. Taking the identification of whether a user is a potential churned user as an example, the first label can represent a non-churned user, and the second label can represent a potential churned user.

[0092] The above-mentioned methods for adjusting the contribution of preset labels are merely illustrative and are not specific limitations on the corresponding adjustment methods.

[0093] According to the embodiments of this application, by continuously adjusting the weights of each first sample in each iteration of training, the first recognition model can adaptively adjust the weights according to the differences between the samples, and distinguish the contribution of each sample to the modeling, thereby performing further sample balancing at the algorithm level, and solving the problem of few and noisy first sample data in the second time period.

[0094] Based on the above inventive concept, this application also provides a user data identification method. Figure 2 A flowchart illustrating a user data identification model update method according to an embodiment of this application is shown. Figure 2 As shown, the method may include the following steps:

[0095] Step 210: Obtain the user data to be identified.

[0096] Step 220: Input the user data into the recognition model to obtain the recognition result of the user data, wherein the weights of the label recognition output layer of the recognition model are the target migration weights.

[0097] The target migration weight is obtained according to the user data identification model update method provided in the embodiments of this application.

[0098] In some embodiments, after the identification result of the user data is obtained, the identification result of the user data is saved to a preset database for use in updating the trained user data identification model based on the user data and the identification result of the user data in the preset database within the preset time period after a preset time period.

[0099] The preset time period can be determined according to the preset update cycle. After the preset time period is reached, the user data within the preset time period can be used as the first sample data for the second time period to update and train the user data recognition model.

[0100] According to the embodiments of this application, the user data recognition model is optimized by back-compromising the model using the real labels of user data within a preset time period output by the user data recognition model. This fills in the gaps in the training data of the user data recognition model during the preset time period, enabling adaptive model back-optimization of the user data recognition model and effectively improving the accuracy of the user data recognition model. This avoids the problem of decreasing migration efficiency as the time difference between user data in the new billing period and user data in the old billing period increases.

[0101] Based on the same inventive concept, this application also provides a user data identification model updating device 300 corresponding to the above-described user data identification model updating method. Specifically, in conjunction with... Figure 3 Please provide a detailed explanation.

[0102] Figure 3 This is a schematic diagram of the structure of a user data identification model updating device provided in an embodiment of this application, as shown below. Figure 3 As shown, the user data identification model update device 300 may include an acquisition module 310 and a processing module 320.

[0103] The acquisition module 310 is used to acquire first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label.

[0104] The processing module 320 is used to train the first recognition model based on the first training data to obtain the evaluation value of the recognition ability of the second recognition model and the second recognition model, wherein the second recognition model includes the first transfer weight of the label recognition output layer.

[0105] The processing module 320 is further configured to, when the evaluation value is greater than or equal to a preset threshold, adjust the first transfer weight and the first training data of the label recognition output layer in the second recognition model according to the preset weight adjustment rules, and return to the step "train the first recognition model according to the first training data to obtain the second recognition model", iteratively train the first recognition model until the evaluation value is less than the preset threshold, take the first transfer weight of the second recognition model including the label recognition output layer as the target transfer weight, and take the second recognition model with the evaluation value less than the preset threshold as the target recognition model.

[0106] In some embodiments, the multiple first sample data include a first sample set and a second sample set, wherein the first sample set includes first sample data obtained in a first time period, and the second sample set includes first sample data obtained in a second time period, wherein the second time period is the time after the first time period;

[0107] The processing module 320 is also used to train the first recognition model based on the first sample set to obtain the third recognition model, and to train the first recognition model based on the second sample set to obtain the fourth recognition model;

[0108] The acquisition module 310 is also used to acquire the second transfer weight of the label recognition output layer in the third recognition model and the third transfer weight of the label recognition output layer in the fourth recognition model.

[0109] The processing module 320 is also used to determine the first transfer weight of the label recognition output layer according to the preset weight transfer function, the second transfer weight and the third transfer weight;

[0110] The processing module 320 is also used to configure the first recognition model according to the first migration weight to obtain the second recognition model;

[0111] The processing module 320 is further configured to identify the first training data, including multiple first sample data, according to the second recognition model to obtain a first recognition result set;

[0112] The processing module 320 is also used to determine the evaluation value of the recognition capability of the second recognition model based on the first recognition result set and the preset label corresponding to each first sample data.

[0113] In some embodiments, the processing module 320 is further configured to, based on the recognition result of each first sample data in the first recognition result set and the preset label corresponding to each first sample data, take the first sample data whose recognition result is different from the preset label as the second sample data to obtain the second training dataset;

[0114] The processing module 320 is also used to adjust the first learning weight of each second sample data and the second learning weight of the second sample data corresponding to the preset label according to the preset weight adjustment rules.

[0115] It is understood that the user data identification model update device 300 in this application embodiment can correspond to the execution subject of the user data identification model update method provided in this application embodiment. Specific details regarding the operation and / or functions of each module / unit of the user data identification model update device 300 can be found in the above-described embodiments of this application. Figure 1 The description of the relevant part of the user data identification model update method is omitted here for the sake of brevity.

[0116] The user data recognition model update device of this application embodiment acquires first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label. Next, the first recognition model is trained according to the first training data to obtain a second recognition model and an evaluation value of the recognition ability of the second recognition model. The second recognition model includes a first transfer weight of the label recognition output layer. When the evaluation value is greater than or equal to a preset threshold, the first transfer weight of the label recognition output layer in the second recognition model and the first training data are adjusted according to a preset weight adjustment rule. The process returns to the step "train the first recognition model according to the first training data to obtain the second recognition model". The first recognition model is iteratively trained until the evaluation value is less than the preset threshold. The first transfer weight of the second recognition model, including the label recognition output layer, is taken as the target transfer weight. According to the embodiments of this application, the weight update mechanism of the label recognition output layer uses the real label results to update the model parameters, which can continuously optimize the parameters of the recognition model and improve the accuracy of the updated recognition model in recognizing user data.

[0117] Based on the same inventive concept, this application also provides a user data identification device 400 corresponding to the above-described user data identification method. (Specifically combined with...) Figure 4 Please provide a detailed explanation.

[0118] Figure 4 This is a schematic diagram of the structure of a user data identification device provided in an embodiment of this application, as shown below. Figure 4 As shown, the user data identification device 400 may include an acquisition module 410 and a processing module 420.

[0119] Module 410 is used to acquire user data to be identified;

[0120] The processing module 420 is used to input user data into the recognition model to obtain the recognition result of the user data. The weight of the label recognition output layer of the recognition model is the target migration weight, which is obtained according to the user data recognition model update method described in the embodiments of this application.

[0121] In some embodiments, the user data identification device further includes:

[0122] The saving module is used to save the recognition results of user data to a preset database, so that after a preset time period, the trained user data recognition model can be updated based on the user data and the recognition results of user data in the preset database within the preset time period.

[0123] It is understood that the user data identification device 400 in this application embodiment can correspond to the execution subject of the user data identification method provided in this application embodiment. Specific details regarding the operation and / or function of each module / unit of the user data identification device 400 can be found in the above-described embodiments of this application. Figure 2 The description of the relevant part of the user data identification method will not be repeated here for the sake of brevity.

[0124] According to the embodiments of this application, the user data recognition model is optimized by back-compromising the model using the real labels of user data within a preset time period output by the user data recognition model. This fills in the gaps in the training data of the user data recognition model during the preset time period, enabling adaptive model back-optimization of the user data recognition model and effectively improving the accuracy of the user data recognition model. This avoids the problem of decreasing migration efficiency as the time difference between user data in the new billing period and user data in the old billing period increases.

[0125] Figure 5 A schematic diagram of the structure of a user data identification device according to an embodiment of this application is shown. Figure 5 As shown, the device may include a processor 501 and a memory 502 storing computer program instructions.

[0126] Specifically, the processor 501 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0127] Memory 502 may include mass storage for information or instructions. For example, and not limitingly, memory 502 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 502 may include removable or non-removable (or fixed) media, or memory 502 may be non-volatile solid-state memory. Memory 502 may be internal or external to a user data identification device.

[0128] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0129] The processor 501 reads and executes the computer program instructions stored in the memory 502 to implement the method described in the embodiments of this application and achieve the corresponding technical effects achieved by executing the method in the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0130] In one example, the user data identification device may further include a communication interface 503 and a bus 510. Wherein, as... Figure 5 As shown, the processor 501, memory 502, and communication interface 503 are connected through bus 510 and complete communication with each other.

[0131] The communication interface 503 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0132] Bus 510 includes hardware, software, or both, that couples components of an online information flow metering device together. For example, and not as a limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 510 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0133] The user data identification device can execute the user data identification model update method and user data identification method in the embodiments of this application, thereby achieving the corresponding technical effects of the user data identification model update method and user data identification method described in the embodiments of this application.

[0134] Furthermore, in conjunction with the user data identification model update method and user data identification method in the above embodiments, this application embodiment can provide a readable storage medium for implementation. This readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any one of the user data identification model update methods and user data identification methods in the above embodiments. Examples of readable storage media can be non-transitory machine-readable media, such as electronic circuits, semiconductor memory devices, read-only memory (ROM), floppy disks, compact disc read-only memory (CD-ROM), optical discs, hard disks, etc.

[0135] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0136] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (ROM), flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0137] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0138] Furthermore, by combining the user data identification model update method and user data identification method and apparatus in the above embodiments, as well as the readable storage medium, this application embodiment can provide a computer program product for implementation. When the instructions in the computer program product are executed by the processor of an electronic device, the electronic device causes the electronic device to execute any one of the user data identification model update method and user data identification method in the above embodiments.

[0139] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0140] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for updating a user data identification model, characterized in that, include: The system acquires first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label. The first recognition model includes a recognition model for identifying whether a user is a potential churned user. The categories of the preset label include non-churned users or potential churned users. The first sample data includes service data generated by the user when using the service provided by the operator. The first recognition model is trained based on the first training data to obtain a second recognition model and an evaluation value of the recognition capability of the second recognition model, wherein the second recognition model includes the first transfer weight of the label recognition output layer. When the evaluation value is greater than or equal to a preset threshold, the first transfer weight of the label recognition output layer in the second recognition model and the first training data are adjusted according to the preset weight adjustment rules, and the process returns to the step "train the first recognition model according to the first training data to obtain the second recognition model". The first recognition model is iteratively trained until the evaluation value is less than the preset threshold. The first transfer weight of the second recognition model, including the label recognition output layer, is taken as the target transfer weight, and the second recognition model corresponding to the evaluation value being less than the preset threshold is taken as the target recognition model. The plurality of first sample data includes a first sample set and a second sample set, wherein the first sample set includes first sample data obtained in a first time period, and the second sample set includes first sample data obtained in a second time period, wherein the second time period is the time after the first time period; The first recognition model is trained based on the first training data to obtain a second recognition model and an evaluation value of the recognition capability of the second recognition model, including: The first recognition model is trained based on the first sample set to obtain the third recognition model, and the first recognition model is trained based on the second sample set to obtain the fourth recognition model. Obtain the second transfer weight of the label recognition output layer in the third recognition model and the third transfer weight of the label recognition output layer in the fourth recognition model; The first transfer weight of the label recognition output layer is determined based on the preset weight transfer function, the second transfer weight, and the third transfer weight; The first identification model is configured according to the first migration weight to obtain the second identification model; Based on the second recognition model, an evaluation value of the recognition capability of the second recognition model is obtained.

2. The method according to claim 1, characterized in that, Based on the second recognition model, an evaluation value of the recognition capability of the second recognition model is obtained, including: The first training data, including multiple first sample data, is identified according to the second identification model to obtain a first identification result set. Based on the first recognition result set and the preset label corresponding to each first sample data, the evaluation value of the recognition capability of the second recognition model is determined.

3. The method according to claim 2, characterized in that, The step of adjusting the first transfer weights of the label recognition output layer in the second recognition model and the first training data according to the preset weight adjustment rules includes: Based on the recognition result of each first sample data in the first recognition result set and the preset label corresponding to each first sample data, the first sample data whose recognition result is different from the preset label is used as the second sample data to obtain the second training dataset; According to the preset weight adjustment rules, the first learning weight of each second sample data and the second learning weight of the preset label corresponding to the second sample data are adjusted respectively.

4. A user data identification method, characterized in that, include: Obtain user data to be identified; The user data is input into the recognition model to obtain the recognition result of the user data, wherein the weight of the label recognition output layer of the recognition model is the target migration weight, and the target migration weight is obtained by the user data recognition model update method according to any one of claims 1-3.

5. The method according to claim 4, characterized in that, The method further includes: The recognition results of the user data are saved to a preset database for use in updating the trained user data recognition model after a preset time period, based on the user data in the preset database within the preset time period and the recognition results of the user data.

6. A user data identification model update device, characterized in that, The device includes: The acquisition module is used to acquire first training data and a first recognition model to be updated. The first training data includes multiple first sample data and a preset label corresponding to each first sample data. The preset label has multiple categories. The first recognition model includes a label recognition output layer, which is used to output the recognition result of the category of the preset label. The first recognition model includes a recognition model for identifying whether a user is a potential churned user. The categories of the preset label include non-churned users or potential churned users. The first sample data includes service data generated by the user when using the service provided by the operator. The processing module is used to train the first recognition model based on the first training data to obtain a second recognition model and an evaluation value of the recognition capability of the second recognition model, wherein the second recognition model includes a first transfer weight of the label recognition output layer. The processing module is further configured to, when the evaluation value is greater than or equal to a preset threshold, adjust the first transfer weight of the label recognition output layer in the second recognition model and the first training data respectively according to the preset weight adjustment rules, and return to the step "train the first recognition model according to the first training data to obtain the second recognition model", iteratively train the first recognition model until the evaluation value is less than the preset threshold, take the first transfer weight of the second recognition model including the label recognition output layer as the target transfer weight, and take the second recognition model corresponding to the evaluation value being less than the preset threshold as the target recognition model; The plurality of first sample data includes a first sample set and a second sample set, wherein the first sample set includes first sample data obtained in a first time period, and the second sample set includes first sample data obtained in a second time period, wherein the second time period is the time after the first time period; The processing module is further configured to: train the first recognition model based on the first sample set to obtain a third recognition model; train the first recognition model based on the second sample set to obtain a fourth recognition model; obtain the second transfer weight of the label recognition output layer in the third recognition model and the third transfer weight of the label recognition output layer in the fourth recognition model; determine the first transfer weight of the label recognition output layer based on a preset weight transfer function, the second transfer weight, and the third transfer weight; configure the first recognition model based on the first transfer weight to obtain a second recognition model; and obtain an evaluation value of the recognition capability of the second recognition model based on the second recognition model.

7. A user data identification device, characterized in that, The device includes: The acquisition module is used to acquire user data to be identified; The processing module is used to input the user data into the recognition model to obtain the recognition result of the user data, wherein the weight of the label recognition output layer of the recognition model is the target migration weight, and the target migration weight is obtained by the user data recognition model update method according to any one of claims 1-3.

8. A user data identification model update device, characterized in that, The device includes: a processor, and a memory storing computer program instructions; The processor reads and executes the computer program instructions to implement the user data identification model update method as described in any one of claims 1-3 or the user data identification method as described in any one of claims 4-5.

9. A readable storage medium, characterized in that, The readable storage medium stores computer program instructions, which, when executed by a processor, implement the user data identification model update method as described in any one of claims 1-3 or the user data identification method as described in any one of claims 4-5.

10. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the user data identification model update method as described in any one of claims 1-3 or the user data identification method as described in any one of claims 4-5.

Images