Charging policy generation method, apparatus, device, storage medium, and product

Abstract

The application discloses a charging policy generation method and device, equipment, storage medium and product, relates to infrastructure and IT support technical field, and the charging policy generation method comprises: acquiring target user call data collected in real time and user element and charging dimension data stored in a cache database; inputting the target user call data, the user element and the charging dimension data into an updated preset charging policy generation model to perform feature extraction, and obtaining a user data vector and a user demand vector; determining an attention weight based on the user data vector and the preset charging policy generation model, generating a charging policy based on the attention weight, the user data vector and the user demand vector, and obtaining a target charging policy. The application can avoid the problem that in the related art, most of the technologies depend on simple rule setting or limited data analysis, and cannot respond to changes in user on-demand use in real time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrastructure and IT support, and particularly relates to a charging policy generation method and device, equipment, storage medium and product. BACKGROUND

[0002] With the rapid development of information technology, communication services have become an indispensable part of people's daily life. Traditional charging methods are usually based on fixed price models, such as charging by the number of minutes and the amount of traffic usage. However, this static charging mode is difficult to meet the increasingly diverse and personalized user needs, especially in the context of the mobile Internet era, users' consumption habits and service needs change rapidly, and the traditional charging method is too rigid to effectively reflect the real value contribution of users, and also limits the ability of operators to provide more personalized services.

[0003] In related technologies, some improved charging methods begin to try to combine the actual use of users, such as adjusting the charging standard by analyzing the behavior patterns of users, but these methods mostly rely on simple rule setting or limited data analysis, and cannot respond to changes in user's on-demand use in real time, and the corresponding charging rules are not suitable for the unique needs of different users, and lack sufficient flexibility. SUMMARY

[0004] The main purpose of the present application is to provide a charging policy generation method, which aims to solve the technical problem that by analyzing the behavior patterns of users to adjust the charging standard, but these methods mostly rely on simple rule setting or limited data analysis, and cannot respond to changes in user's on-demand use in real time, and the corresponding charging rules are not suitable for the unique needs of different users, and lack sufficient flexibility.

[0005] To achieve the above purpose, the present application provides a charging policy generation method, which comprises:

[0006] acquiring real-time collected target user call data and user elements and charging dimension data stored in a cache database;

[0007] inputting the target user call data, the user elements and the charging dimension data into an updated preset charging policy generation model for feature extraction, to obtain a user data vector and a user demand vector;

[0008] determining an attention weight based on the user data vector and the preset charging policy generation model, and generating a charging policy based on the attention weight, the user data vector and the user demand vector, to obtain a target charging policy.

[0009] In an embodiment, the step of generating the attention weight based on the user data vector and the preset charging policy generation model, generating the target charging policy based on the attention weight, the user data vector and the user demand vector comprises:

[0010] generating an attention score of each vector in the user data vector based on the preset parameter of the attention mechanism layer of the preset charging policy generation model and the user data vector, wherein the preset parameter is obtained based on an initial parameter optimization;

[0011] normalizing the attention score to obtain an attention weight;

[0012] combining the attention weight, the user data vector and the user demand feature vector through additive pooling to obtain a first data vector after pooling;

[0013] performing user consumption behavior learning on the first data vector to obtain a second data vector comprising an updated user consumption mode;

[0014] interacting all features in the second data vector to obtain a third data vector after interaction;

[0015] generating a target policy corresponding to the third data vector through a preset mapping relationship between a preset policy and a data vector.

[0016] In an embodiment, the charging dimension data comprises a first personalized dimension or a second personalized dimension, and the user element comprises a first user element or a second user element. The step of obtaining the real-time collected target user bill data and the user element and the charging dimension data stored in the cache database comprises:

[0017] if the user end does not initiate the user demand, obtaining a user usage record within a preset time, and performing information extraction on historical user bill data in the user usage record to obtain a first user element and determine a first personalized dimension obtained based on analysis of the user usage record;

[0018] if the user end initiates the user demand, determining a second personalized dimension based on the user demand, and obtaining the user usage record within the preset time, and performing information extraction on historical user bill data in the user usage record to obtain a second user element;

[0019] storing the user element and the charging dimension data into the cache database.

[0020] In an embodiment, the step of storing the user element and the charging dimension data into the cache database comprises:

[0021] scaling the numerical data in the historical user call data, the user elements and the billing dimension data to a preset range based on the mean and standard deviation, to obtain the historical user call data, the user elements and the billing dimension data after scaling;

[0022] scaling the numerical data in the historical user call data, the user elements and the billing dimension data to a preset range based on the mean and standard deviation, to obtain the historical user call data, the user elements and the billing dimension data after scaling;

[0023] obtaining a billing history strategy corresponding to the historical user call data, and taking the historical user call data, the user elements and the billing dimension data as sample data, and the historical strategy as a sample label corresponding to the sample data;

[0024] obtaining an initial billing strategy generation model;

[0025] inputting the sample data and the sample label into the initial billing strategy generation model for iterative training to obtain a preset billing strategy generation model meeting the accuracy requirement.

[0026] In an embodiment, the step of inputting the sample data and the sample label into the initial billing strategy generation model for iterative training to obtain a preset billing strategy generation model meeting the accuracy requirement comprises:

[0027] based on an input layer of the initial billing strategy generation model, converting the sample data into multi-dimensional user-related data, and outputting the user-related data to an embedding layer of the initial billing strategy generation model;

[0028] based on the embedding layer, analyzing user data and user demand in the user-related data, converting the analyzed user-related data into a preset base vector, and respectively splicing the preset base vector corresponding to the user data and the preset base vector corresponding to the user demand, and inputting the spliced user data vector and user demand vector into an attention mechanism layer of the initial billing strategy generation model;

[0029] based on the attention mechanism layer and initial parameters of the attention mechanism layer, combining the user data vector and the user demand feature vector through additive pooling, and inputting the weighted first sample vector obtained into a memory network layer of the initial billing strategy generation model;

[0030] perform user consumption behavior learning on the first sample vector based on the memory network layer and an initial weight matrix and an initial bias vector of the memory network layer, and input a second sample vector including an updated user consumption pattern after the learning to an interaction layer of the initial charging policy generation model;

[0031] based on the interaction layer, interact all features in the second sample vector, and input a third sample vector after the interaction to a decision output layer of the initial charging policy generation model;

[0032] based on the decision output layer, perform policy mapping on the third sample vector to obtain an initial charging policy;

[0033] based on the sample label and the initial charging policy, optimize the initial parameters, the initial weight matrix and the initial bias vector to obtain a preset charging policy generation model meeting an accuracy requirement after optimization.

[0034] In an embodiment, the step of inputting the sample data and the sample label to the initial charging policy generation model for iterative training to obtain the preset charging policy generation model meeting the accuracy requirement further includes:

[0035] storing the preset charging policy generation model in a shared memory library;

[0036] based on a preset timing time, performing model updating on the preset charging policy generation model stored in the shared memory library to obtain an updated preset charging policy generation model.

[0037] In addition, to achieve the above object, the application further provides a charging policy generation device, which comprises:

[0038] an acquisition module configured to acquire target user call data collected in real time and user element and charging dimension data stored in a cache database;

[0039] an extraction module configured to input the target user call data, the user element and the charging dimension data to the updated preset charging policy generation model for feature extraction to obtain a user data vector and a user demand vector;

[0040] a generation module configured to determine an attention weight based on the user data vector and the preset charging policy generation model, and generate a charging policy based on the attention weight, the user data vector and the user demand vector to obtain a target charging policy.

[0041] In addition, to achieve the above object, the present application further provides a charging policy generation device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the charging policy generation method as described above.

[0042] In addition, to achieve the above object, the present application further provides a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the charging policy generation method as described above.

[0043] In addition, to achieve the above object, the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the charging policy generation method as described above.

[0044] The one or more technical solutions provided by the present application have at least the following technical effects:

[0045] Compared with the related art, by analyzing the behavior mode of the user to adjust the charging standard, but these methods mostly depend on simple rule setting or limited data analysis, and cannot respond to the changes in the user's on-demand use in real time, and the corresponding charging rules are not suitable for the unique needs of different users, and lack sufficient flexibility. Compared with the related art, by acquiring the target user call data collected in real time and the user factors and charging dimension data stored in the cache database; by the updated preset charging policy generation model, the target user call data, the user factors and the charging dimension data are subjected to feature extraction to obtain a user data vector and a user demand vector, based on the user data vector, an attention weight is determined, based on the attention weight, the user data vector and the user demand vector, a charging policy is generated to obtain a target charging policy. It can be understood that, by the updated preset charging policy generation model, the target user call data, the user factors and the charging dimension data are subjected to feature extraction to obtain a user data vector and a user demand vector, and the user demand vector is introduced into the attention mechanism to generate a charging policy matched with the user data vector and the user demand vector, which can avoid the problem that in the related art, most of the methods depend on simple rule setting or limited data analysis, and cannot respond to the changes in the user's on-demand use in real time. BRIEF DESCRIPTION OF DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those of ordinary skill in the art, the other drawings can also be obtained based on these drawings without any creative work.

[0048] Figure 1 The flowchart provided by the charging policy generation method embodiment one of the present application;

[0049] Figure 2 The charging policy matching flowchart of the charging policy generation method of the present application;

[0050] Figure 3 The computing power charging formula rule graph of the charging policy generation method of the present application;

[0051] Figure 4 The flowchart provided by the charging policy generation method embodiment two of the present application;

[0052] Figure 5 The model structure graph of the charging policy generation method of the present application;

[0053] Figure 6 The module structure schematic diagram of the charging policy generation device of the present application embodiment;

[0054] Figure 7 The device structure schematic diagram of the hardware running environment involved in the charging policy generation method in the present application embodiment.

[0055] The object implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings in combination with the embodiments. DETAILED DESCRIPTION

[0056] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application, and are not used to limit the present application.

[0057] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings in the specification and specific embodiments.

[0058] The main solution of the present application embodiment is: obtaining the target user call data collected in real time and the user element and charging dimension data stored in the cache database; inputting the target user call data, the user element and the charging dimension data into the updated preset charging policy generation model to extract features, obtaining a user data vector and a user demand vector; determining an attention weight based on the user data vector and the preset charging policy generation model, generating a charging policy based on the attention weight, the user data vector and the user demand vector, and obtaining a target charging policy.

[0059] In the related art, some improved charging methods begin to try to combine the actual use of the user, such as adjusting the charging standard by analyzing the behavior mode of the user, but these methods mostly rely on simple rule setting or limited data analysis, and cannot respond to the changes of the user's on-demand use in real time, and the corresponding charging rules are not suitable for the unique needs of different users, and lack sufficient flexibility.

[0060] The application generates a user data vector and a user demand vector by extracting features from the target user bill data, user factors and charging dimension data according to the updated preset charging strategy generation model, and introduces the user demand vector into the attention mechanism to generate a charging strategy matched with the user data vector and the user demand vector, which can avoid the problem that the related art mostly relies on simple rule setting or limited data analysis and cannot respond to the changes of the user's on-demand use in real time.

[0061] It should be noted that the execution subject of the embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions. The charging strategy generation device is taken as an example to illustrate the embodiment and the following embodiments.

[0062] Based on this, the application embodiment provides a charging strategy generation method, which refers to Figure 1 , Figure 1 The flowchart of the first embodiment of the charging strategy generation method of the application is shown.

[0063] In the embodiment, the charging strategy matching flowchart is provided, and the charging strategy generation method comprises steps S100-S300: Figure 2 , Figure 2

[0064] Step S100, acquiring the target user bill data collected in real time and the user factors and charging dimension data stored in the cache database;

[0065] It should be noted that the real-time subject of the embodiment is a charging strategy generation device. The charging strategy generation device is provided with a cache database for storing user factors and charging dimension data. The target user bill includes data such as user geographic location, service type and usage. The user factors include user type, user credit rating and historical consumption level, etc. The charging dimension data includes pricing dimension, control dimension, personalized dimension and charging dimension, wherein each dimension in the charging dimension data can be represented by corresponding code.

[0066] ​It can be understood that the charging policy generation device obtains the real-time collected target user call data based on an API interface, and obtains user element and charging dimension data from a cache database, wherein the API interface mainly provides real-time prediction capability for real-time collection of user call data for daily newly added subscriptions or special requirement users.

[0067] In step S200, the target user call data, the user element and the charging dimension data are input into the updated preset charging policy generation model for feature extraction to obtain a user data vector and a user demand vector.

[0068] It should be noted that the user demand vector is determined based on personalized dimensions. The preset charging policy generation model is a model that is updated at a fixed time. The charging policy generation device extracts features of the target user call data, the user element and the charging dimension data by using the One-Hot Encoding method in the preset charging policy generation model to obtain the user data vector and the user demand vector determined based on personalized dimensions.

[0069] Specifically, the model input data format information is as follows:

[0070] {service_usage: [“BeiJing”, 111712, 100], user_profile: [0, 601, 0.5], dimension_features: [6000008, 6000008002]}.

[0071] The meanings of different dimensions in the above example are as follows:

[0072] {service_usage: [user geographic location, service type, usage amount], user_profile: [user type, user credit rating, historical consumption level], dimension_features: [charging dimension code, charging pricing code]}.

[0073] Further, the charging policy generation device converts the user geographic location “BeiJing” into an integer code 110 by using the One-Hot Encoding method, and outputs other features as they are. The converted data format is as follows:

[0074] [110, 111712, 100, 0, 601, 0.5, 6000008, 6000008002, [110, 111712, 100, 0.5]].

[0075] Step S300, determining an attention weight based on the user data vector and the preset charging policy generation model, and generating a target charging policy based on the attention weight, the user data vector and the user demand vector.

[0076] It can be understood that the target charging policy is an optimized charging policy obtained based on real-time target user bills. The charging policy generation device determines the attention weight of the model based on the user data vector, and generates an optimized target charging policy obtained based on real-time target user bills based on the obtained attention weight, the user data vector and the user demand vector.

[0077] In a possible implementation, the step of determining an attention weight based on the user data vector and the preset charging policy generation model, and generating a target charging policy based on the attention weight, the user data vector and the user demand vector includes the following steps:

[0078] Based on the preset parameters of the attention mechanism layer of the preset charging policy generation model and the user data vector, an attention score of each vector in the user data vector is generated, wherein the preset parameters are obtained based on initial parameters.

[0079] It should be noted that the charging policy generation device calculates the attention score of each feature vector in the user data vector based on the preset parameters of the attention mechanism layer of the preset charging policy generation model.

[0080] Specifically, the charging policy generation device calculates the attention score of each feature vector h i in the user data vector H, wherein h i is the i-th feature representation in the user data vector H, and the output e ti is the attention score of the feature vector of the user data vector H. Let s t be the parameter of the attention mechanism layer, which is a parameter obtained by model training, and the output e ti is the attention score of the feature vector of the user data vector H. The calculation formula is as follows:

[0081]

[0082] The attention score is normalized to obtain an attention weight.

[0083] It can be understood that the attention weight is a probability distribution of the attention score. The charging policy generation device normalizes the attention score of the user data vector, calculates its probability distribution, and assigns a corresponding attention weight to each feature based on the probability distribution.

[0084] Specifically, the charging policy generation device combines the output attention score e ti The softmax function is used for normalization processing, and the probability distribution is calculated. Each feature is given a corresponding weight based on the probability distribution, and the value is between 0 and 1. The attention weight vector a is output t The basic principle of the softmax function is to convert a set of numerical values into a probability distribution, and each numerical value is converted into a relative probability, and the sum of the probabilities is equal to 1. The formula is as follows:

[0085] a t = softmax(e t )

[0086] The attention weight, the user data vector and the user demand feature vector are combined through additive pooling to obtain a first data vector after pooling;

[0087] It should be noted that the charging policy generation device combines the attention weight, the user data vector and the user demand feature vector through additive pooling to obtain a first data vector after pooling.

[0088] Specifically, the charging policy generation device combines the attention weight vector a t and the user demand feature vector q through additive pooling processing, thereby obtaining a vector based on the dynamic change of user demand. The output vector c t after pooling, and the calculation formula is as follows:

[0089]

[0090] Wherein, c t is the output vector after pooling; h i is the i-th feature in the user data vector H; b1 and b2 are the bias vectors of the current layer, which are optimized and updated by the back propagation gradient descent method, and are used for comparison with the features in the input data.

[0091] The first data vector is subjected to user consumption behavior learning to obtain a second data vector including an updated user consumption mode;

[0092] It can be understood that the second data vector is a new user data vector that takes into account the user demand and contains an updated user consumption mode. The charging policy generation device performs user consumption behavior learning on the first data vector to obtain a new user data vector including an updated user consumption mode after learning.

[0093] All features in the second data vector are interacted to obtain a third data vector after interaction;

[0094] It should be noted that the charging policy generation device performs second-order interaction between all features of the new user data vector, that is, calculates the element-wise product of the user data features to obtain a third data vector after interaction.

[0095] Through the preset mapping relationship between the preset policy and the data vector, a target policy corresponding to the third data vector is generated.

[0096] It can be understood that the preset mapping relationship is pre-stored and used to represent the mapping relationship between the policy and the vector. The target policy is the policy corresponding to the third data vector. The charging policy generation device maps and finds the policy corresponding to the third data vector through the preset mapping relationship to obtain the target policy corresponding to the third data vector.

[0097] In particular, after obtaining the target policy, the charging policy generation device can further predict future bills under different policies based on various different policies and historical bills. Since the charging policy output by the model is a computing power and charging formula rule, it is represented as a charging formula. According to the formula, the user's current consumption amount can be calculated, as shown in Figure 3 , Figure 3 A computing power and charging formula rule graph is provided.

[0098] In this embodiment, the charging policy generation device generates a user data vector and a user demand feature vector based on the constructed intelligent dynamic charging model, through user bills, user factors, and charging dimension data, and introduces a user demand feature vector to optimize the attention mechanism, and generates a target charging policy that matches the user data and user demand.

[0099] The present application provides a charging policy generation method, as shown in Figure 4 , Figure 4 The present application provides a charging policy generation method, as shown in

[0100] In a feasible embodiment, the steps of obtaining real-time collected target user bill data and stored user factors and charging dimension data in the cache database include steps A100-A300:

[0101] Step A100, if the user end does not initiate user demand, obtaining user usage records within a preset time, and extracting information through historical user bill data in the user usage records to obtain user factors and determine a first personalized dimension analyzed based on the user usage records;

[0102] It should be noted that the first personalized dimension is the user demand determined by the charging policy generation device, which can be various possible user demands in the charging policy generation device by default, or can accept the user demand on the user side. If the corresponding client on the user side does not initiate the request of the personalized service of the user demand, the judgment of the user demand will be automatically judged by the charging policy generation device to identify the consumption mode and preference of the user, and to determine which type of charging mode the user is more likely to accept, and to take the charging mode as the first personalized dimension, and to obtain the user elements by extracting information from the historical user bill data. Due to the special nature of the user elements, the user elements of the same user are basically unchanged.

[0103] Among them, the user demand includes the following typical demands:

[0104] a) fixed payment demand represented by annual package and monthly package;

[0105] b) pay-per-use demand for services or resources such as actual consumption of calls, traffic, etc. without prepayment of fixed fees;

[0106] c) peak-valley charging demand for setting different charging standards in different peak and valley periods;

[0107] d) customized charging demand for paying only for the used service function modules of the personalized customized service function modules.

[0108] It can be understood that the above a) to d) four kinds of user demands are only used as examples, and the charging policy generation device can provide various new demands in the process of providing services, and collect demands from the user side and finally launch products and services that meet different demands.

[0109] In particular, when the user signs a contract with the service party corresponding to the charging policy generation device, the bill that has occurred will inevitably point to one of the above-mentioned a) to d) four typical user demands, and the user demand in the personalized dimension can be determined based on the bill that has occurred after the user signs a contract with the service party (for example, one of the above-mentioned a) to d) four typical user demands).

[0110] Step A200, if the user end initiates the user demand, determine the second personalized dimension based on the user demand, and obtain the user usage record in a preset time, and extract information from the historical user bill data in the user usage record to obtain user elements;

[0111] It can be understood that the second personalized dimension is determined based on the user demand initiated by the user side corresponding to the user terminal. If the charging policy generation device receives a request for personalized service of the user demand initiated by the user terminal, it will store it in the data object indicated by the personalized dimension, wherein the personalized dimension and the remaining dimension are both encoded.

[0112] In particular, in order to avoid frequent modification of the existing encoding, each encoding can point to a specific data object. Since the personalized dimension indicates the user demand initiated by the user side, assuming that the user demand feature vector initiated by the user side is no longer [1, 0, 0, 0] but [0, 1, 0, 0] or even [0, 1, 1, 1], the user can be provided with the selection of new personalized services based on the user demand actively initiated by the user and the user bill generated, and the intelligent charging policy output by the model. The specific selection right is in the user.

[0113] Specifically, the charging policy generation device can accurately identify those users who pay fixed monthly or annual fees, and thus apply the corresponding charging policy, for example, the charging mode of the fixed monthly / yearly fee dimension, to meet the user's demand for stable cost. The following specific examples show the specific implementation steps of several typical user demands, but the present application is not limited to the following several user demands:

[0114] User demand 1: fixed budget, expecting stable cost

[0115] The charging policy generation device determines whether the user pays a fixed monthly or annual fee mainly by analyzing the user's historical consumption data and behavior patterns through the following steps:

[0116] a) Data collection and preprocessing: collect the user's usage records in the past period (the last three months), including but not limited to bill data, service type, usage amount, and combine with user element data such as user type, credit rating, historical consumption level information;

[0117] b) Set the judgment threshold: based on the above features, set the judgment threshold, and the same or similar amount of consumption records that appear continuously for several times in the same month or annual cycle can be used as an indicator of paying a fixed monthly or annual fee;

[0118] The charging policy generation device calculates the user's consumption amount in a certain time window, and uses statistical methods (standard deviation) to evaluate the fluctuation degree of the consumption amount. If the consumption amount fluctuates less, it indicates that the consumption amount is relatively stable.

[0119] c) Model training and prediction: use the intelligent dynamic charging model to train the model based on the labeled user consumption behavior data set; after training is completed, the model can predict whether the user pays a fixed monthly or annual fee according to the features of the user's consumption records;

[0120] Further, through the above steps, the billing policy generation device can accurately identify those users who pay a fixed monthly or annual fee, so as to apply the corresponding billing policy, the billing mode of the fixed fee dimension of the monthly / yearly package, to meet the user's demand for stable cost.

[0121] User demand 2: Pay-per-use, flexible cost control

[0122] The model obtains the pay-per-use dimension billing policy according to the user demand, and the user only pays for the actual consumed service or resource without prepayment of fixed fees. The user only pays for the actual consumed service or resource, and the judgment is made by the following method:

[0123] a) Extract key information from the user's recent 3 months of historical consumption data, including but not limited to service type, usage, time; these data are usually from the user's bill record, reflecting the user's actual consumption of various services at different time points;

[0124] b) Analyze the user's consumption pattern, including consumption frequency, usage fluctuation, and consumption period; for users who pay per use, their consumption records often show irregular consumption behavior, and each consumption record is directly related to the actual usage of service or resource, without fixed periodic large payment records; Specifically, the consumption frequency is measured by calculating the number of transactions within a certain time window, and the usage fluctuation is evaluated by statistical method (standard deviation).

[0125] c) Based on the user's consumption pattern, set the index to judge whether the user pays per use, and the user's consumption record shows small, frequent transactions without obvious periodic large expenditures, which is highly consistent with the characteristics of pay-per-use;

[0126] Set the threshold of small transactions, such as less than 50 yuan or 100 yuan of consumption;

[0127] Define the threshold of frequent transactions, such as more than 10 times of transactions within a month;

[0128] Check if there is a periodic large expenditure, such as a large consumption on a fixed date every month.

[0129] d) Use machine learning model to predict whether the user tends to pay per use based on the historical consumption data trained by the model; the model input includes the user's past usage behavior and consumption characteristics, and the output is the prediction result of the user's pay-per-use pattern.

[0130] Step A300, store the user elements and billing dimension data to the cache database.

[0131] It should be noted that after obtaining the user elements and the billing dimension, the billing policy generation device stores the above data in a cache database for subsequent calling.

[0132] In a possible implementation, the step of storing the user elements and the billing dimension data into the cache database comprises:

[0133] calculating the mean and the standard deviation of the numerical data in the historical user call data, the user elements and the billing dimension data in the historical user call data, the user elements and the billing dimension data;

[0134] It can be understood that when performing the standardization scaling, the billing policy generation device needs to calculate the mean and the standard deviation of each user feature in the entire user historical use data information.

[0135] Specifically, there is a feature matrix k, wherein each row represents a sample and each column represents a feature. For each feature, the billing policy generation device calculates the mean and the standard deviation of the feature in the entire data set.

[0136] Based on the mean and the standard deviation, the numerical data is scaled to a preset range to obtain the scaled historical user call data, the user elements and the billing dimension data;

[0137] It should be noted that the preset range is [0, 1]. For the numerical data, in order to realize the standardization scaling of the data, the billing policy generation device needs to determine the scaling of the data to [0, 1], so as to scale all numerical feature values to between 0 and 1.

[0138] Specifically, the billing policy generation device applies a standardization scaling formula to the numerical data to scale all values to the target range. The standardization scaling formula is as follows:

[0139]

[0140] wherein k is the value of the original feature, mean(k) is the mean of the feature set, σ(k) is the standard deviation of the feature set, and m is the converted value.

[0141] obtaining a billing history strategy corresponding to the historical user call data, and taking the historical user call data, the user elements and the billing dimension data as sample data, and the historical strategy as a sample label corresponding to the sample data;

[0142] It can be understood that the sample data is user call data, user elements and billing dimension data of the last three months. The billing policy generation device obtains the user call data, user elements and billing dimension data of the last three months as sample data, and obtains the billing history strategy corresponding to the historical user call data as a sample label.

[0143] An initial billing policy generation model is obtained.

[0144] It should be noted that the initial billing policy generation model is an intelligent dynamic fee management model (IDFM). The billing policy generation device obtains the intelligent dynamic fee management model, and refers to Figure 5 , Figure 5 A model structure diagram is provided.

[0145] The sample data and the sample label are input into the initial billing policy generation model for iterative training to obtain a preset billing policy generation model meeting the accuracy requirement.

[0146] It can be understood that the initial billing policy generation model is trained based on the sample data with the labeled sample label to obtain a preset billing policy generation model meeting the accuracy requirement.

[0147] Specifically, the billing policy generation device divides the sample data into a training set, a validation set and a test set. According to the data, 70% of the data is used for training, 15% of the data is used for validation, and 15% of the data is used for testing. The difference between the test set and the validation set is that the purposes are different: the validation set is mainly used to adjust the model parameters and select the model, and helps to determine the complexity of the model to prevent overfitting. The test set is used for final evaluation of the performance of the model, and the purpose is to evaluate the generalization ability of the model on unknown data.

[0148] In a feasible implementation, the step of inputting the sample data and the sample label into the initial billing policy generation model for iterative training to obtain a preset billing policy generation model meeting the accuracy requirement includes the following steps:

[0149] Based on the input layer of the initial billing policy generation model, the sample data is converted into multi-dimensional user-related data, and the user-related data is output to the embedding layer of the initial billing policy generation model.

[0150] It should be noted that the user-related data includes basic information and behavior data of the user, and the information related to charging includes consumption amount, consumption frequency, user credit level and user geographic location. The charging policy generation device receives the user call, user elements and charging dimensions through the input layer of the initial charging policy generation model, and outputs multi-dimensional user-related data, wherein the user-related data includes basic information and behavior data of the user, and the information related to charging includes consumption amount, consumption frequency, user credit level and user geographic location.

[0151] Based on the embedding layer, the user data and user demand in the user-related data are analyzed, the analyzed user-related data are converted into preset binary vectors, the preset binary vector corresponding to the user data and the preset binary vector corresponding to the user demand are spliced respectively, and the spliced user data vector and user demand vector are input into the attention mechanism layer of the initial charging policy generation model.

[0152] It can be understood that the preset binary is binary. The charging policy generation device analyzes the user data and user demand in the multi-dimensional user-related data output by the input layer through the embedding layer of the initial charging policy generation model, and then uses the One-Hot Encoding method to perform feature splicing on the analyzed multi-dimensional user-related data to splice the user data vector and user demand feature vector. The charging policy generation device not only bases on the user call, but also bases on the user demand, thereby providing personalized service, flexible and accurate charging, and better predictive charging policy. Therefore, the input layer receives the user call, user elements and charging dimensions, the personalized dimensions included in the charging dimensions indicate the user demand, and then the user data vector and user demand feature vector are spliced through the embedding layer.

[0153] Based on the attention mechanism layer and the initial parameters of the attention mechanism layer, the user data vector and the user demand feature vector are combined through additive pooling, and the weighted first sample vector is input into the memory network layer of the initial charging policy generation model.

[0154] It should be noted that the charging policy generation device combines the user data vector and the user demand feature vector through additive pooling based on the initial parameters to obtain the weighted first sample vector.

[0155] Based on the memory network layer and the initial weight matrix and initial bias vector of the memory network layer, the first sample vector is subjected to user consumption behavior learning, and the second sample vector including updated user consumption mode after learning is input into the interaction layer of the initial charging policy generation model.

[0156] It can be understood that the memory network layer is an LSTM layer (Long Short-Term Memory Layer). The billing policy generation device learns the user consumption behavior pattern step by step through the forgetting gate calculation, the input gate calculation, the parameter update of the current layer, and the output gate calculation in the long short-term memory neural network on the weighted user data feature vector output by the attention mechanism layer through the LSTM layer of the initial billing policy generation model, which is not only used for predicting future consumption behavior, but also used for generating: a new user data vector x considering user demand and containing updated user consumption patterns, wherein: x t-1 represents the updated user data vector at t-1 time, W represents the weight matrix of the current layer, b represents the bias vector of the current layer, which are updated by the back propagation gradient descent method, C t-1 represents the context vector at t-1 time, represents the storage of long-term dependency information, h t represents the input user data feature vector at t time. At t time, the corresponding calculation formula is as follows:

[0157] x,C t =LSTM(x t-1 ,C t-1 ,h t ;W,b)

[0158] Based on the interaction layer, all features in the second sample vector are interacted, and the third sample vector after interaction is input to the decision output layer of the initial billing policy generation model.

[0159] It should be noted that the billing policy generation device performs second-order interaction between all features of the new user data vector x output by the LSTM layer through the interaction layer of the initial billing policy generation model, that is, calculates the element-by-element product of user data features. The output of this layer is a user data vector z containing second-order cross terms between features. The specific formula is as follows:

[0160]

[0161] wherein v is the weight factor of the current layer, w i is the weight matrix of the i-th feature, w j is the weight matrix of the j-th feature, and the initial values are all set to 0.9, y is the second-order cross of all features of the user data, and represents element multiplication. is an activation function, and w a is the weight matrix of the current layer. Parameters v, w i and w aThe update is optimized by a back propagation gradient descent method, and the user data vector x input to the interaction layer (it can be understood that the user data vector at this time has passed through the attention mechanism layer and the LSTM layer of the previous stage, x contains n features, and the input dimension of the interaction layer is n, and the output dimension of the interaction layer is n+C(n, 2)=n+n(n-1) / 2.

[0162] If the input x vector is:

[0163] x=[1 10,1 11 12,1 00,0.5 01,0.5 00,0 00 00 08,0 00 00 00 80 02,[1 10,1 11 12,1 00,0.5 01]];

[0164] The output vector z is as follows:

[0165] z=[1 10,1 11 12,1 00,0.5 01,0.5 00,0 00 00 08,0 00 00 00 80 02,x1·x2,x1·x3,...,x7·

[0166] x8]。

[0167] Based on the decision output layer, the third sample vector is mapped to obtain an initial charging strategy;

[0168] It can be understood that the charging strategy generation device maps and integrates the user data vector z output by the interaction layer and containing the second-order cross terms between features through a fully connected layer (Fully Connected Layer) plus a softmax activation function through the decision output layer (Decision Layer) of the initial charging strategy generation model, and outputs the final model prediction result. The charging strategy output by the decision output layer is f, and the corresponding formula of f is as follows:

[0169] f=softmax(Vz+b′)

[0170] Wherein, V, b' are the parameter factors of the layer, the value range is [0, 1], and the initial values of the two are 0.5, which are optimized and updated by a back propagation gradient descent method.

[0171] Based on the sample label and the initial charging strategy, the initial parameters, initial weight matrix and initial bias vector are optimized to obtain a preset charging strategy generation model meeting the accuracy requirement after optimization.

[0172] It should be noted that the loss function of the initial charging policy generation model of the charging policy generation device uses the mean square error (MSE) function and the Adam optimizer, sets 0.01 as the initial learning rate, uses the adaptive learning rate optimizer to automatically adjust the learning rate of each parameter according to the historical gradient information, sets the batch value to 20 according to the data set (2000000 data), randomly divides the data set in each batch according to the proportion of 7:1.5:1.5, that is, 1400000 training sets, test sets and validation sets each with 300000, iterates 10000 times, uses the training set to calculate the predicted value in each round of iteration, updates the model parameters through back propagation, and at the same time, the validation set is used to monitor the performance of the model to avoid overfitting. Finally, the test data set is used to give an independent evaluation of the expected performance of the model in actual application, and finally an effective and reliable trained intelligent dynamic charging model is obtained.

[0173] In a feasible implementation, after the step of inputting the sample data and the sample label into the initial charging policy generation model for iterative training to obtain a preset charging policy generation model meeting the accuracy requirement, the method further includes:

[0174] storing the preset charging policy generation model in a shared memory library;

[0175] It can be understood that the shared memory library can improve the access speed and efficiency of the model, and in turn, quickly predict and adjust the user demand and charging dimension. The charging policy generation device loads the generated model into the shared memory library.

[0176] Based on a preset timing time, the preset charging policy generation model stored in the shared memory library is updated to obtain an updated preset charging policy generation model.

[0177] It should be noted that since new users who need to be retrained due to daily subscription or special requirements may not be included in the prediction results or the prediction results may not be the latest, it may lead to the matching of the obtained charging policy not being optimal. Therefore, the charging policy generation device will be based on a preset timing time (for example, 0:30) to train and update the latest charging policy generation model in the system every day.

[0178] In particular, the charging policy generation model will package the trained model into a Docker image, deploy it as a microservice using Kubernetes or similar container orchestration tools, implement high-performance API services, reduce model size through model quantization and pruning, and improve inference speed. At the same time, cache high-frequency requests or results to reduce unnecessary repeated calculations.

[0179] Specifically, the specific indicators of real-time prediction by the model mainly focus on the performance, response speed, and stability of the model. Here are several key real-time prediction indicators:

[0180] Response Time: The time taken from receiving input data to producing a prediction output. In real-time prediction scenarios, low response time is crucial as it directly impacts user experience and system performance.

[0181] Throughput: Describes the number of requests a model can handle per unit of time. For high-concurrency real-time systems, high throughput means the model can efficiently handle a large number of simultaneous requests.

[0182] Model Stability: The consistency and predictability of a model's output, especially when faced with changing data distributions. The model should be able to maintain stable performance under different inputs.

[0183] Scalability: Tests the performance of the model and system architecture when handling increasing workloads. Good scalability means the system can maintain stable service levels as demand grows.

[0184] Specifically, the billing strategy generation model uses evaluation indicators to assess the performance of the classification model, with the following specific steps:

[0185] a) Performance Monitoring and Evaluation

[0186] After the model is deployed, its performance in the production environment is continuously monitored, including but not limited to accuracy, error rate, response time, and other key indicators. At the same time, A / B testing is implemented to compare the performance of new and old models or different model versions, ensuring that updates bring positive impacts.

[0187] b) Data Feedback Loop

[0188] Collect new data: Continuously collect new user behavior data and billing results, and fine-tune the existing model based on the results to reflect the latest data features.

[0189] c) Incremental Learning

[0190] Use new data or a small portion of historical data to update the model, reducing computational resource consumption and speeding up iteration.

[0191] In this embodiment, the billing strategy generation device can analyze user behavior data in real time and capture changes in user demand in a timely manner by establishing a pre-set billing strategy generation model and regularly training the model, allowing for dynamic adjustment of billing strategies.

[0192] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the charging policy generation method of the present application. More forms of simple transformation based on this technical concept are within the protection scope of the present application.

[0193] The present application also provides a charging policy generation device, please refer to Figure 6 , the charging policy generation device comprises:

[0194] The acquisition module 10 is configured to acquire the target user call data collected in real time and the user element and charging dimension data stored in the cache database.

[0195] The extraction module 20 is configured to input the target user call data, the user element and the charging dimension data into the updated preset charging policy generation model to perform feature extraction, so as to obtain a user data vector and a user demand vector.

[0196] The generation module 30 is configured to determine an attention weight based on the user data vector and the preset charging policy generation model, and generate a charging policy based on the attention weight, the user data vector and the user demand vector, so as to obtain a target charging policy.

[0197] The charging policy generation device provided by the present application adopts the charging policy generation method in the above embodiments, and can solve the technical problem of charging policy generation. Compared with the prior art, the charging policy generation device provided by the present application has the same beneficial effects as the charging policy generation method provided by the above embodiments, and other technical features in the charging policy generation device are the same as the features disclosed in the above embodiments, which will not be repeated here.

[0198] The present application provides a charging policy generation device, the charging policy generation device comprising: at least one processor; and a memory communicatively connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the charging policy generation method in Embodiment I.

[0199] Reference will now be made to Figure 7 , which shows a structural schematic diagram of a charging policy generation device suitable for implementing the embodiments of the present application. The charging policy generation device in the embodiments of the present application can include but is not limited to mobile terminals such as mobile phones, notebook computers, tablet computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like.Figure 7 The billing policy generation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0200] like Figure 7 As shown, the billing policy generation device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the billing policy generation device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the charging policy generation device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows charging policy generation devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0201] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0202] The charging policy generation device provided by the present application adopts the charging policy generation method in the above embodiments, and can solve the technical problem of charging policy generation. Compared with the prior art, the charging policy generation device provided by the present application has the same beneficial effects as the charging policy generation method provided by the above embodiments, and other technical features in the charging policy generation device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0203] It should be understood that parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0204] The above is merely specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0205] The present application provides a computer readable storage medium having computer readable program instructions (i.e. computer programs) stored thereon, the computer readable program instructions being used to execute the charging policy generation method in the above embodiments.

[0206] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system or device, or any combination thereof. More specific examples of the computer readable storage medium can include, but are not limited to: an electrical connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to: electrical wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination thereof.

[0207] The computer readable storage medium can be included in the charging policy generation device, or can exist separately and not be assembled into the charging policy generation device.

[0208] The computer readable storage medium carries one or more programs, and when the one or more programs are executed by the charging policy generation device, the charging policy generation device is caused to: acquire real-time collected target user call data and user element and charging dimension data stored in a cache database; input the target user call data, the user element and the charging dimension data into an updated preset charging policy generation model to perform feature extraction, to obtain a user data vector and a user demand vector; determine an attention weight based on the user data vector and the preset charging policy generation model, and generate a charging policy based on the attention weight, the user data vector and the user demand vector, to obtain a target charging policy.

[0209] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0210] The flowcharts and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may be executed in the reverse order, depending on the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or combinations of hardware and software.

[0211] The modules involved in the embodiments of the present application can be implemented in software or in hardware. In some cases, the names of the modules do not limit the modules themselves.

[0212] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer program) for executing the above-mentioned charging policy generation method, and can solve the technical problem of charging policy generation. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the charging policy generation method provided by the above-mentioned embodiments, which will not be repeated here.

[0213] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the charging policy generation method as described above.

[0214] The computer program product provided by the present application can solve the technical problem of charging policy generation. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the charging policy generation method provided by the above-mentioned embodiments, which will not be repeated here.

[0215] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A charging policy generation method characterized by comprising: The charging policy generation method comprises: obtaining real-time collected target user call data, and user elements and charging dimension data stored in a cache database; inputting the target user call data, the user elements and the charging dimension data into an updated preset charging policy generation model for feature extraction to obtain a user data vector and a user demand vector, wherein the charging dimension data comprises a personalized dimension, the user demand vector is determined based on the personalized dimension, and the personalized dimension comprises a first personalized dimension or a second personalized dimension; determining an attention weight based on the user data vector and the preset charging policy generation model, generating a charging policy based on the attention weight, the user data vector and the user demand vector to obtain a target charging policy; the step of obtaining real-time collected target user call data, and user elements and charging dimension data stored in a cache database comprises: if a user end does not initiate a user demand, obtaining user usage records within a preset time, and extracting information from historical user call data in the user usage records to obtain user elements and determine a first personalized dimension obtained based on analysis of the user usage records; if the user end initiates the user demand, determining a second personalized dimension based on the user demand, and obtaining the user usage records within the preset time, and extracting information from historical user call data in the user usage records to obtain user elements; storing the user elements and the charging dimension data to the cache database.

2. The charging policy generation method of claim 1, wherein, The step of determining an attention weight based on the user data vector and the preset charging policy generation model, and generating a charging policy based on the attention weight, the user data vector and the user demand vector to obtain a target charging policy comprises: generating an attention score of each vector in the user data vector based on preset parameters of an attention mechanism layer of the preset charging policy generation model and the user data vector, wherein the preset parameters are obtained based on optimization of initial parameters; normalizing the attention score to obtain an attention weight; combining the attention weight, the user data vector and the user demand feature vector through additive pooling to obtain a first data vector after pooling; performing user consumption behavior learning on the first data vector to obtain a second data vector comprising an updated user consumption mode; interacting all features in the second data vector to obtain a third data vector after interaction; generating a target strategy corresponding to the third data vector through a preset mapping relationship between a preset strategy and a data vector.

3. The charging policy generation method of claim 1, wherein, The step of storing the user elements and the charging dimension data to the cache database comprises: calculating the mean and standard deviation of numerical data in the historical user call data, the user elements and the charging dimension data in the historical user call data, the user elements and the charging dimension data. scaling the numerical data to a preset range based on the mean value and the standard deviation, to obtain scaled historical user call data, user elements and billing dimension data; obtaining a billing history strategy corresponding to the historical user call data, and taking the historical user call data, the user elements and the billing dimension data as sample data, and the historical strategy as a sample label corresponding to the sample data; obtaining an initial billing strategy generation model; inputting the sample data and the sample label into the initial billing strategy generation model for iterative training to obtain a preset billing strategy generation model meeting accuracy requirements.

4. The charging policy generation method of claim 3, wherein, The step of inputting the sample data and the sample label into the initial billing strategy generation model for iterative training to obtain a preset billing strategy generation model meeting accuracy requirements comprises: based on the input layer of the initial billing strategy generation model, converting the sample data into multi-dimensional user-related data, and outputting the user-related data to the embedding layer of the initial billing strategy generation model; based on the embedding layer, analyzing user data and user demand in the user-related data, converting the analyzed user-related data into a preset radix vector, and respectively splicing the preset radix vector corresponding to the user data and the preset radix vector corresponding to the user demand, and inputting the spliced user data vector and user demand vector into the attention mechanism layer of the initial billing strategy generation model; based on the attention mechanism layer and the initial parameters of the attention mechanism layer, combining the user data vector and the user demand feature vector through additive pooling, and inputting the weighted first sample vector obtained into the memory network layer of the initial billing strategy generation model; based on the memory network layer and the initial weight matrix and initial bias vector of the memory network layer, performing user consumption behavior learning on the first sample vector, and inputting the second sample vector learned including the updated user consumption mode into the interaction layer of the initial billing strategy generation model; based on the interaction layer, interacting all features in the second sample vector, and inputting the third sample vector obtained after interaction into the decision output layer of the initial billing strategy generation model; based on the decision output layer, performing strategy mapping on the third sample vector to obtain an initial billing strategy; based on the sample label and the initial billing strategy, optimizing the initial parameters, the initial weight matrix and the initial bias vector to obtain an optimized preset billing strategy generation model meeting accuracy requirements.

5. The charging policy generation method of claim 3, wherein, The step of inputting the sample data and the sample label into the initial billing strategy generation model for iterative training to obtain a preset billing strategy generation model meeting accuracy requirements comprises: storing the preset billing strategy generation model in a shared memory library; based on a preset timing time, updating the preset billing strategy generation model stored in the shared memory library to obtain an updated preset billing strategy generation model.

6. A charging policy generating apparatus characterized by comprising: The device comprises: An acquisition module is configured to acquire real-time collected target user call data and user elements and billing dimension data stored in a cache database; An extraction module is configured to input the target user call data, the user elements and the billing dimension data into an updated preset billing policy generation model to perform feature extraction, so as to obtain a user data vector and a user demand vector, wherein the billing dimension data comprises personalized dimensions, the user demand vector is determined based on the personalized dimensions, and the personalized dimensions comprise a first personalized dimension or a second personalized dimension; A generation module is configured to determine an attention weight based on the user data vector and the preset billing policy generation model, and generate a billing policy based on the attention weight, the user data vector and the user demand vector, so as to obtain a target billing policy. The acquisition module is further configured to acquire user usage records in a preset time period and perform information extraction on historical user call data in the user usage records to obtain user elements and determine a first personalized dimension obtained based on analysis of the user usage records, if the user end does not initiate a user demand; and determine a second personalized dimension based on the user demand, acquire the user usage records in the preset time period, perform information extraction on historical user call data in the user usage records to obtain user elements, if the user end initiates the user demand; and store the user elements and the billing dimension data to the cache database.

7. A charging policy generating apparatus characterized by comprising: The device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the billing policy generation method according to any one of claims 1 to 5.

8. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the billing policy generation method according to any one of claims 1 to 5.

9. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program is executed by the processor to implement the steps of the billing policy generation method according to any one of claims 1 to 5.

Images