Model training method and device and service provider package recommendation method

Abstract

The invention discloses a model training method and device and an operator package recommendation method. The method comprises the following steps: acquiring historical operator package data and historical voice data of a target object; generating a first row vector corresponding to the historical service provider package data, and performing sentiment analysis on the historical voice data to obtain a second row vector; combining the first row vector and the second row vector to obtain a target matrix; inputting the target matrix into a generative adversarial network model; determining an emotional tendency reward according to the first row vector and the second row vector; and determining the target probability as an optimization parameter, updating the optimization parameter, and obtaining a trained generative adversarial network model if a loss function about the target probability meets a preset convergence condition. According to the method and the device, the technical problem that the recommendation accuracy of the service provider package is relatively low due to the fact that a related service provider package recommendation method is limited by a static model and the dynamic variability of a user demand is not fully considered is solved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a model training method, device and operator package recommendation method. Background Art

[0002] In the current telecommunications service market, there are many types of packages, covering different data traffic, call duration, number of SMS and other services, aiming to meet the needs of different user groups. However, the current telecom operator package recommendation system is difficult to provide accurate and personalized services due to its reliance on static models and simple rule systems. Although these methods can make preliminary recommendations based on users' basic information (such as age, gender, region, etc.) and package usage history, they do not fully consider the diversity and dynamic changes of individual user needs, and ignore the complexity of user behavior patterns. This leads to limited recommendation accuracy and user satisfaction, because static models find it difficult to capture users' real-time needs and emotional fluctuations, while simple statistical methods lack personalized and intelligent decision-making mechanisms.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The present application provides a model training method, device and operator package recommendation method to at least solve the technical problem of low recommendation accuracy for operator packages due to the fact that the relevant operator package recommendation method is limited to a static model and does not fully consider the dynamic variability of user needs.

[0005] According to one aspect of the present application, a model training method is provided, including: obtaining historical operator package data and historical voice data of a target object, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data; generating a first row vector corresponding to the historical operator package data, performing sentiment analysis on the usage demand information in the historical voice data, and obtaining a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages; merging the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package categories of the recommended packages, and the columns are used to represent The features corresponding to the package category; the target matrix is ​​input into the generative adversarial network model to obtain the target data with a similar distribution to the target matrix output by the generator in the generative adversarial network model, and the target probability that the target data output by the discriminator in the generative adversarial network model is the real data; according to the first row vector and the second row vector, the emotional tendency reward is determined, and according to the emotional tendency reward, the reward function is determined, the target matrix and the target data are determined as the state vector, the recommended package for the target object is determined as the action vector, and the objective function is determined according to the target probability state vector, the action vector and the reward function; the target probability is determined as the optimization parameter to maximize the objective function with the target, and the optimization parameter is updated. In the discriminator, if the loss function about the target probability meets the preset convergence condition, the generative adversarial network model that has completed the training is obtained.

[0006] Optionally, the objective function is determined according to the target probability state vector, the action vector and the reward function, including: determining several state vectors and action vector pairs under the current optimization parameters as observation values ​​to obtain an observation matrix including multiple observation values; determining the negative log marginal likelihood function according to the target probability, the observation matrix and the reward function, and using the negative log marginal likelihood function as the objective function; updating the optimization parameters, including: obtaining the model parameters in the multi-output student T process, wherein the model parameters include: degrees of freedom, mean function, covariance function, parameter matrix and optimization parameters; calculating the derivatives of the negative log marginal likelihood with respect to the model parameters in the multi-output student T process, and updating the model parameters in the opposite direction of the derivative.

[0007] Optionally, determining the emotional tendency reward based on the first row vector and the second row vector includes: determining the first weight based on the activity index of the historical operator package used by the target object; determining the second weight based on the package value index of the historical operator package used by the target object; determining the emotional tendency reward based on the first weight, the second weight, the first row vector and the second row vector.

[0008] Optionally, a reward function is determined based on the sentiment tendency reward, including: obtaining the target object's satisfaction index with the historical operator package used by the target object, the operator's revenue index with the historical operator package used by the target object, and the market adaptability index of the historical operator package used by the target object; and performing weighted summation of the sentiment tendency reward, satisfaction index, revenue index, and market adaptability index to obtain the reward function.

[0009] Optionally, updating the optimization parameters includes: obtaining a number of state vectors and action vector pairs under the current optimization parameters; calculating a reward function in each state vector and action vector, and updating the optimization parameters according to the calculation results until the mathematical expectation of the state vector and action vector pair is greater than a preset threshold.

[0010] Optionally, generating a first row vector corresponding to the historical operator package data includes: performing data cleaning on the historical operator package data, and interpolating and filling zeros on missing data in the historical operator package data after data cleaning to obtain initial data of multiple dimensions, wherein the multiple dimensions include: identification information, age information, gender information, regional information, package subsidy information, package usage information, call information outside the package, text message information outside the package, and traffic information outside the package of the target object; performing feature extraction on the initial data of multiple dimensions to obtain multiple initial features, and normalizing the multiple initial features to obtain multiple first features; and generating different types of recommended packages recommended to the target object based on the multiple first features to obtain a first row vector.

[0011] Optionally, sentiment analysis is performed on the usage demand information in the historical voice data to obtain a second row vector, including: converting the historical voice data into text data, performing feature extraction processing on the text data to obtain a text feature vector, wherein the feature extraction processing includes at least one of the following: word frequency statistics, grammatical analysis, and topic modeling; using a pre-trained sentiment analysis model to perform sentiment tendency analysis on the text feature vector to obtain a quantitative score of the target object's sentiment tendency towards different types of recommended packages; converting the quantitative score into a vector form to obtain a second row vector.

[0012] According to another aspect of the present application, a method for recommending an operator package is also provided, including: obtaining operator package data and voice data of a target object, wherein the operator package data is multi-dimensional information including attribute information and package usage information about the target object, and the voice data includes usage requirement information for the operator package data; inputting the vectorized operator package data and voice data into a trained generative adversarial network model to obtain a recommended package for the target object output by the trained generative adversarial network model, wherein the generative adversarial network model is obtained by training using the above model training method.

[0013] According to another aspect of the present application, a model training device is also provided, including: an acquisition module, used to acquire historical operator package data and historical voice data of a target object, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data; a generation module, used to generate a first row vector corresponding to the historical operator package data, perform sentiment analysis on the usage demand information in the historical voice data, and obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's emotional tendency information for different types of recommended packages; a merging module, used to merge the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to characterize the package categories of the recommended packages, and the columns are used to represent the target packages. The features corresponding to the package category are collected; an input module is used to input the target matrix into the generative adversarial network model to obtain the target data with a similar distribution to the target matrix output by the generator in the generative adversarial network model, and the target probability that the target data output by the discriminator in the generative adversarial network model is the real data; a determination module is used to determine the emotional tendency reward according to the first row vector and the second row vector, and determine the reward function according to the emotional tendency reward, determine the target matrix and the target data as the state vector, determine the recommended package for the target object as the action vector, and determine the objective function according to the target probability state vector, the action vector and the reward function; an update module is used to determine the target probability as an optimization parameter to maximize the objective function to update the optimization parameter. In the discriminator, if the loss function about the target probability meets the preset convergence condition, a generative adversarial network model that has completed training is obtained.

[0014] According to another aspect of the present application, a non-volatile storage medium is also provided, the storage medium includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above model training method.

[0015] According to another aspect of the present application, an electronic device is provided, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above model training method is executed when the program is running.

[0016] According to another aspect of the present application, a computer program is also provided, wherein the above model training method is implemented when the computer program is executed by a processor.

[0017] According to another aspect of the present application, a computer program product is also provided, which includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above model training method is implemented.

[0018] In the present application, historical operator package data and historical voice data of a target object are obtained, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data; a first row vector corresponding to the historical operator package data is generated, and sentiment analysis is performed on the usage demand information in the historical voice data to obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages; the first row vector and the second row vector are merged to obtain a target matrix, wherein the rows of the target matrix are used to characterize the package categories of the recommended packages, and the columns are used to characterize the features corresponding to the package categories; the target matrix is ​​input into a generative adversarial network model to obtain target data having a similar distribution to the target matrix output by a generator in the generative adversarial network model, and a generated The target data output by the discriminator in the adversarial network model is the target probability of the real data; the emotional tendency reward is determined according to the first row vector and the second row vector, and the reward function is determined according to the emotional tendency reward, the target matrix and the target data are determined as the state vector, the recommended package for the target object is determined as the action vector, and the target function is determined according to the target probability state vector, the action vector and the reward function; the target probability is determined as the optimization parameter to maximize the objective function and update the optimization parameter. In the discriminator, if the loss function of the target probability meets the preset convergence condition, the training of the generative adversarial network model is obtained, which achieves the purpose of fully considering the dynamic variability of user needs, thereby achieving the technical effect of improving the recommendation accuracy of operator packages, and then solving the technical problem of low recommendation accuracy of operator packages due to the fact that the relevant operator package recommendation method is limited to a static model and does not fully consider the dynamic variability of user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 is a flow chart of a model training method according to an embodiment of the present application;

[0021] Figure 2is a schematic diagram of a model training system according to an embodiment of the present application;

[0022] Figure 3 is a training and prediction diagram of a multi-output Student-T process according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a method for recommending operator packages according to an embodiment of the present application;

[0024] Figure 5 is a structural diagram of a model training device according to an embodiment of the present application;

[0025] Figure 6 It is a hardware structure block diagram of a computer terminal according to a model training method of an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] In the related art, traditional operator package recommendations are often limited to static models or simple rule-based recommendation systems. Therefore, the dynamic variability of user needs is not fully considered, resulting in low accuracy in recommending operator packages. In order to solve this problem, the present application provides a related solution in the embodiments, which is described in detail below.

[0029] According to an embodiment of the present application, a method embodiment of a model training method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 1 is a flow chart of a model training method according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0031] Step S101, obtaining historical operator package data and historical voice data of the target object, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data.

[0032] It should be noted that the operator package data includes the service combination of the operator subscribed by the target object, such as the user's monthly consumption, subsidy, package usage, Internet behavior, calling habits and SMS usage. Historical voice data includes the package usage needs, feedback and opinions expressed by users in communication with customer service.

[0033] Step S102, generate a first row vector corresponding to the historical operator package data, perform sentiment analysis on the usage demand information in the historical voice data, and obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages.

[0034] In the above step S102, the original package data is cleaned and feature extracted, and the missing data is interpolated and normalized to obtain the first row vector M corresponding to the package data based on the user's ID, age, gender, region, monthly consumption, subsidy, package usage, Internet behavior, calling habits, and SMS usage. j1 ,m j2 ,..,m jn ], where m j is the number of categories of packages recommended to user j, each m jj Represents the attribute value of the recommended package in category j.

[0035] Use sentiment analysis technology to process historical voice data and extract information about users' sentiment tendencies towards different package types, such as positive, neutral or negative emotions. Convert these sentiment tendencies into numerical components to form a vector, which is the second row vector. Each component corresponds to a package type, indicating the target object's sentiment tendency towards the package. Specifically, perform speech recognition on the communication voice data with the user, convert it into text, extract features from the text, perform sentiment analysis, and output the sentiment tendency matrix E = n*k, where n is the number of users and k is the number of package categories. For user i, the sentiment tendency vector can be expressed as E i =[e i1 ,e i2 ,…,e ik ], each e ij Indicates the intensity of user i’s sentiment towards package category j.

[0036] Step S103, merging the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package categories of the recommended packages, and the columns are used to represent the features corresponding to the package categories.

[0037] Specifically, the extracted first row vector (package data matrix) and the second row vector (emotional feature matrix) are combined into a matrix called the target matrix. The rows of the matrix represent the package category of the recommended package, and the columns are the features corresponding to the package category, including emotional tendency, usage frequency, consumption habits, etc.

[0038] Step S104, input the target matrix into the generative adversarial network model to obtain target data output by the generator in the generative adversarial network model having a similar distribution to the target matrix, and the target probability that the target data output by the discriminator in the generative adversarial network model is real data.

[0039] In step S104, the target matrix obtained in the above steps is used as input and sent to the generator part of the generative adversarial network (GAN) model. The generator generates new target data with a similar distribution to the target matrix based on the input data. These generated data are intended to simulate the preferences and usage of potential users for packages, thereby enriching the training data set of the model. At the same time, the discriminator part of the GAN model is used to evaluate the generated data to determine the probability that the generated data is real data. The output of the discriminator will be used for subsequent model optimization to improve the fidelity of the generated data.

[0040] Step S105, determine the emotional tendency reward based on the first row vector and the second row vector, and determine the reward function based on the emotional tendency reward, determine the target matrix and the target data as the state vector, determine the recommended package for the target object as the action vector, and determine the target function based on the target probability state vector, the action vector and the reward function.

[0041] In step S105, the emotional tendency R is added to the reward function sentiment (t,a), used to reward the model based on the user's emotional inclination towards the package. The stronger the inclination, the higher the reward score. sentiment (t,a) can be expressed as:

[0042]

[0043] Among them, w i is the weight of user i, determined based on the package activity, w j is the weight of package category j, which is determined based on the package value.

[0044] Therefore, the reward function R(t, a) is designed as follows:

[0045]

[0046] in, and λ sentiment is the weight coefficient, which is used to measure the importance of different factors; R user (t, a), R operator (t, a), R market (t, a) represent user satisfaction reward, operator revenue reward and market adaptability reward respectively.

[0047] The target matrix and target data are used as the state vector, the recommended package is used as the action vector, and the reward function determined by the emotional tendency reward is used to construct the objective function. The objective function is in the following form:

[0048] f(θ)=E[R(t,a)]=∑ t,a P(t,a|θ)*R(t,a)

[0049] Among them, R(t,a) is the reward function and P(t,a|θ) is the probability that the model takes action a in state t given the parameter θ.

[0050] Step S106, the target probability is determined as the optimization parameter to maximize the objective function, update the optimization parameter, and in the discriminator, if the loss function of the target probability meets the preset convergence condition, a trained generative adversarial network model is obtained.

[0051] In step S106, the target probability of the authenticity of the target data in the discriminator is determined as an optimization parameter. The adjustment of the optimization parameter is intended to maximize the objective function to improve the effectiveness and personalization of the recommendation strategy. The optimization parameters are updated using the gradient descent method or other optimization algorithms, such as particle swarm optimization, until the loss function of the target probability meets the preset convergence conditions to ensure that the model can stably learn and optimize the recommendation strategy. Once the optimization parameter update is completed and the loss function converges, the trained generative adversarial network model can be obtained.

[0052] Preferably, the Bayesian optimization algorithm can be used to update the optimization parameters, which specifically includes the following steps: 1. Sampling: Sample a series of state-action pairs (t, a) according to the current parameter θ. 2. Evaluation: Evaluate the reward function R(t, a) on each pair (t, a). 3. Update: Update the parameter θ according to the evaluation result so as to maximize the objective function f(θ).

[0053] According to the above steps, historical operator package data and historical voice data of the target object are obtained, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data; a first row vector corresponding to the historical operator package data is generated, and sentiment analysis is performed on the usage demand information in the historical voice data to obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages; the first row vector and the second row vector are merged to obtain a target matrix, wherein the rows of the target matrix are used to characterize the package categories of the recommended packages, and the columns are used to characterize the features corresponding to the package categories; the target matrix is ​​input into a generative adversarial network model to obtain a generative adversarial network. The target data output by the generator in the network model has a similar distribution to the target matrix, and the target data output by the discriminator in the generative adversarial network model is the target probability of the real data; according to the first row vector and the second row vector, the emotional tendency reward is determined, and the reward function is determined according to the emotional tendency reward, the target matrix and the target data are determined as the state vector, the recommended package for the target object is determined as the action vector, and the target function is determined according to the target probability state vector, the action vector and the reward function; the target probability is determined as the optimization parameter to maximize the objective function to update the optimization parameter. In the discriminator, if the loss function of the target probability meets the preset convergence condition, the way to complete the training of the generative adversarial network model is obtained, which achieves the purpose of fully considering the dynamic variability of user needs, thereby achieving the technical effect of improving the accuracy of recommendation of operator packages.

[0054] The following Figure 1The steps shown are exemplary and explanatory.

[0055] According to some optional embodiments of the present application, the objective function is determined based on the target probability state vector, the action vector and the reward function, which can be achieved by the following method: several state vectors and action vector pairs under the current optimization parameters are determined as observation values ​​to obtain an observation matrix including multiple observation values; the negative log marginal likelihood function is determined based on the target probability, the observation matrix and the reward function, and the negative log marginal likelihood function is used as the objective function.

[0056] In addition, the optimization parameters can be updated by the following method: obtaining the model parameters in the multi-output student-T process, where the model parameters include: degrees of freedom, mean function, covariance function, parameter matrix and optimization parameters; calculating the derivatives of the negative log marginal likelihood with respect to the model parameters in the multi-output student-T process, and updating the model parameters in the opposite direction of the derivative.

[0057] Through the above steps, based on the user's historical package data, on-site voice data and personalized emotional tendencies, the package recommendation strategy can be continuously iterated and optimized through Bayesian optimization and MOSTP model, ultimately achieving the goal of improving recommendation accuracy and user satisfaction, and realizing a more intelligent and personalized package recommendation service.

[0058] According to other optional embodiments of the present application, determining the emotional tendency reward based on the first row vector and the second row vector can be achieved by the following method: determining the first weight based on the activity index of the historical operator package used by the target object; determining the second weight based on the package value index of the historical operator package used by the target object; determining the emotional tendency reward based on the first weight, the second weight, the first row vector and the second row vector.

[0059] In the above embodiment, first, the historical package data used by the target object is obtained from the operator database, including but not limited to the package type, monthly consumption, package usage frequency, etc. For each historical package, its activity index is calculated. The activity can be comprehensively evaluated based on factors such as the frequency, duration, and data consumption of the user using the package. For example, the activity index = the number of times the user uses the package / subscription duration.

[0060] Then, based on the activity index of each historical package, normalization techniques (such as min-max scaling) are used to map the index value to the [0,1] interval to ensure the comparability of the weights. The normalized activity index is used as the user's package preference weight, indicating the user's preference for high-activity packages, which is the first weight. The weight can be further fine-tuned based on user feedback, package satisfaction and other information to reflect actual user preferences.

[0061] Then, the second weight is determined based on the package value index of the historical operator packages used by the target object. Among them, the package value can be based on the ratio of the service content and cost provided, or the actual benefits obtained by the user after using the package (such as saved fees, additional services, etc.). For each historical package, calculate its value index. For example, package value index = actual user benefit / package cost. Similar to the activity index, the package value index is normalized to ensure the rationality and comparability of the weight. The normalized package value index is used as another dimension of package preference weight to reflect the user's preference for high-value packages, which is the second weight. The second weight can also be dynamically adjusted based on user feedback and market changes.

[0062] The emotional tendency reward R is calculated by the following formula sentiment :

[0063]

[0064] Among them, w i is the weight of user i, w j is the weight of package category j, e ij is the sentiment intensity of user i towards package category j (elements in the second row of the vector), m jj Represents the attribute value of the recommended package in category j (elements in the first row of the vector).

[0065] Through the above steps, the user's preference weights for different packages can be determined based on the user's historical package usage and sentiment analysis results, and then the sentiment tendency reward can be calculated. This process not only takes into account the objective indicators of user behavior (such as activity and package value), but also integrates the user's sentiment tendency, providing the model with more comprehensive and personalized feedback signals, which helps to improve the accuracy of package recommendations and user satisfaction.

[0066] In some optional embodiments of the present application, a reward function is determined based on an emotional tendency reward, which can be achieved by the following method: obtaining a satisfaction index of the target object with respect to the historical operator packages used by the target object, a revenue index of the operator with respect to the historical operator packages used by the target object, and a market adaptability index of the historical operator packages used by the target object; performing weighted summation of the emotional tendency reward, the satisfaction index, the revenue index, and the market adaptability index to obtain a reward function.

[0067] In the above embodiment, first, the historical package usage records of the target object (user) are collected from the operator's user behavior database, including package type, usage (such as data usage, call time, number of text messages), monthly consumption, subsidy, etc. At the same time, user feedback after each package change is recorded, such as whether to continue to use the same package, whether to upgrade or downgrade, and the frequency of use after the package change, etc. These data will be used to calculate the satisfaction index. According to the collected user feedback data, a quantitative index of satisfaction is defined. For example, the satisfaction can be calculated by the ratio of the number of positive comments in the questionnaire survey or user feedback to the total number of feedback, or the user's acceptance of the new package can be evaluated by analyzing the comparison between the frequency of use after the package change and before the change. Then, the relationship between the historical package data and the user's consumption behavior is analyzed to calculate the benefits obtained by the operator from the historical packages used by the target object. It can include direct benefits (such as monthly package fees) and indirect benefits (such as additional service fees in the package). At the same time, the market trends in the target object's area, the competitor's package strategy and the target object's package preference changes are considered to evaluate the market adaptability of the package. Secondly, the communication records with the target object are sentimentally analyzed to extract the user's positive or negative sentiment tendencies towards different package categories.

[0068] Finally, the reward function is obtained by weighted summing the emotional tendency reward, satisfaction index, profit index and market adaptability index. The reward function R(t, a) is designed as follows:

[0069]

[0070] in, and λ sentiment is the weight coefficient, which is used to measure the importance of different factors; R user (t, a), R operator (t, a), R market (t, a), R sentiment They represent user satisfaction reward, profit index, market adaptability index and emotional tendency reward respectively.

[0071] Through the above steps, we can accurately capture the needs of users, the revenue goals of operators, market dynamics and user emotional tendencies, so as to build a more personalized and effective package recommendation strategy. This not only improves user satisfaction, but also enhances the market competitiveness of operators.

[0072] As some optional embodiments of the present application, updating the optimization parameters can be achieved by the following method: obtaining several state vector and action vector pairs under the current optimization parameters; calculating the reward function in each state vector and action vector, and updating the optimization parameters according to the calculation results until the mathematical expectation of the state vector and action vector pair is greater than a preset threshold.

[0073] In the above embodiment, first, the parameters in the reinforcement learning model and the Bayesian optimization model are initialized, including the parameter sets of the generator G and the discriminator S, and the hyperparameters of the Bayesian optimization model. The state space T and action space A of the reinforcement learning model are defined, where T is a set of state vectors containing user characteristics and preference data, and A is a set of combined actions for recommending packages. Then, according to the current optimization parameter θ, a series of combinations of state vectors and action vectors (t, a) are sampled. A random strategy or a model-based strategy can be used, such as an ε-greedy strategy. Make sure that the state vector t contains the user feature matrix and the sentiment tendency matrix obtained from the data preprocessing module. At the same time, the action vector a should contain the recommended package combination under the reinforcement learning strategy. For each sampled state-action pair (t, a), the reward function R(t, a) is calculated, which includes multiple factors such as user satisfaction, operator revenue, market trends, and a special emotional tendency reward function R is added. sentiment .

[0074] Finally, the Bayesian optimization algorithm is used to update the parameter θ according to the calculated reward function value, in order to maximize the expected reward E[R(t,a)]. The above steps are repeated, and new state-action pairs are continuously sampled and rewards are calculated until the mathematical expectation of the state vector and action vector pair exceeds the preset threshold or reaches the predetermined number of iterations. In other words, a preset threshold of mathematical expectation is set, and when the mathematical expectation E[R(t,a)] of the state-action pair calculated by the reinforcement learning model or the Bayesian optimization algorithm is greater than this threshold, the optimization process is stopped.

[0075] After the optimization process is completed, the best package recommendation strategy is output. This strategy can provide the best package combination based on user characteristics and emotional tendencies to achieve the goal of improving user satisfaction and operator revenue. For example, in one iteration, the state vector t represents a user's historical package usage and emotional tendency data, and the action vector a represents the combination of recommended packages. Under the current optimization parameter θ, the state-action pair (t 1 ,a 1 ), calculate the reward function R(t 1 ,a 1 ) is 80 points, which is lower than the preset threshold of 85 points. Therefore, the Bayesian optimization algorithm is used to update θ and sample the next state-action pair (t 2 ,a 2 ) and calculate the reward function again. This process is iterated until E[R(t,a)] exceeds the threshold, indicating that the model has found a set of recommendation strategies that can satisfy user satisfaction.

[0076] Through the above implementation steps, combined with the Bayesian optimization algorithm of reinforcement learning and the generative adversarial network, the package recommendation strategy can be dynamically adjusted to improve the personalization level and efficiency of the recommendation, ultimately achieving accurate satisfaction of user needs and increasing operator revenue.

[0077] In some optional embodiments of the present application, generating a first row vector corresponding to historical operator package data can be achieved by the following method: performing data cleaning on the historical operator package data, and interpolating and filling missing data in the historical operator package data after data cleaning to obtain initial data of multiple dimensions, wherein the multiple dimensions include: identification information, age information, gender information, regional information, package subsidy information, package usage information, call information outside the package, text message information outside the package, and traffic information outside the package of the target object; performing feature extraction on the initial data of multiple dimensions to obtain multiple initial features, and normalizing the multiple initial features to obtain multiple first features; based on the multiple first features, generating different types of recommended packages recommended to the target object to obtain a first row vector.

[0078] In the above embodiment, first, the collected historical operator package data is quality checked to remove or correct those records that are obviously wrong or abnormal. This includes checking whether the data format is correct, whether there are logical errors (such as negative age), whether the data is complete, etc. Next, identify missing values ​​in the data. For example, in the telecommunications package data, some users do not provide complete personal information, or there are unrecorded consumption behaviors in the usage data. For missing data, an interpolation and zero filling strategy is used for processing. Specifically, if a feature (such as call time outside the package) is missing in some user records, then this part of the data will be filled with zero. It should be noted that zero filling is reasonable in telecommunications data because missing data can be interpreted as non-use or zero consumption of a specific service, which is actually a usage pattern. For example, if the user data lacks the usage of out-of-package traffic, it can be considered that the user has not used the out-of-package traffic during the statistical period, so the value of this field is set to zero. Starting from the cleaned data, an initial data set containing multiple dimensions is constructed. These dimensions include, but are not limited to: identification information of the target object (such as user ID), age information, gender information, region information, package subsidy information, package usage information, call information outside the package, SMS information outside the package, and traffic information outside the package. For each dimension, meaningful features are extracted from the data. For example, age information can be converted into age groups, region information can be encoded into region categories, and package subsidies and usage information can be converted into consumer behavior features. The purpose of feature extraction is to convert raw data into a form that can be understood and processed by the model, and it is also to reduce the dimension of the data and reduce the computational complexity. After data cleaning, missing data processing, feature extraction and normalization, multiple first features are obtained. These features will be organized into a vector, namely the first row vector. This vector will be used as the input of the GAN network model to generate personalized recommended packages.

[0079] Through the above steps, we can extract structured, complete and standardized user feature vectors from the original historical operator package data, laying a solid foundation for subsequent model training and personalized package recommendations. This data preprocessing method not only ensures the quality of the data, but also supports the training of complex models, thereby achieving more accurate and personalized package recommendation services.

[0080] As other optional embodiments of the present application, sentiment analysis is performed on the usage demand information in the historical voice data to obtain a second row vector, which can be achieved by the following method: converting the historical voice data into text data, performing feature extraction processing on the text data to obtain a text feature vector, wherein the feature extraction processing includes at least one of the following: word frequency statistics, grammatical analysis, and topic modeling; using a pre-trained sentiment analysis model to perform sentiment tendency analysis on the text feature vector to obtain a quantitative score of the target object's sentiment tendency towards different types of recommended packages; converting the quantitative score into a vector form to obtain a second row vector.

[0081] In the above embodiment, speech recognition technology, such as the automatic speech recognition (ASR) model in deep learning, is used to convert the collected historical speech data of the target object into text data. The audio signal is converted into a digital signal that can be processed by a computer, and the language content therein is recognized using the ASR model, and the corresponding text description is output. Feature extraction is performed on the converted text data to extract key information related to the package recommendation.

[0082] Feature extraction may include at least one of the following processing methods: Count the frequency of occurrence of keywords related to package usage, demand expression, preference description, etc. in the text. Word frequency information helps to understand the user's attention to specific package functions. Use natural language processing (NLP) technology to analyze the grammatical rules and sentence structure of the text to identify the user's evaluation tone, affirmative or negative expressions, etc. on the package. Syntactic analysis can help understand the user's emotional tendencies and attitudes. Use topic models, such as latent Dirichlet allocation (LDA) or non-negative matrix factorization (NMF), to perform topic analysis on the text and identify common topics discussed in the text, such as package value, service experience, network quality, etc. Topic modeling helps to summarize a wide range of text content into several key topics, which is convenient for subsequent sentiment tendency analysis.

[0083] Then, based on the results of word frequency statistics, grammatical analysis, and topic modeling, a text feature vector is constructed. Each component in this vector represents a feature (such as keyword frequency, topic weight, etc.), and its value indicates the importance of the feature in the text data. The vector can be constructed using the bag-of-words model, TF-IDF (term frequency-inverse document frequency), or other advanced representation methods (such as word2vec, BERT, etc.). Use a pre-trained sentiment analysis model, such as a deep learning-based model (such as LSTM, GRU, Transformer, etc.) to perform sentiment analysis on the text feature vector. The pre-trained model should be trained on a large amount of text data with sentiment annotations, and be able to identify and quantify the sentiment in the text, including positive, negative, or neutral sentiment. The sentiment analysis model will output a quantitative score of the target object's sentiment toward different types of recommended packages. The score range can be from -1 to 1, where -1 indicates strong negative sentiment, 0 indicates neutral, and 1 indicates strong positive sentiment.

[0084] Finally, the quantitative score of sentiment tendency is converted into a vector form, that is, the second row vector. The number of rows of this vector is equal to the number of target objects, and the number of columns is equal to the number of package types. Each element in the vector represents the strength of a specific user's sentiment tendency towards a certain package type.

[0085] Through the above steps, historical voice data can be converted into text form, key features can be extracted using text processing technology, and then the pre-trained sentiment analysis model can be used to quantify the user's sentiment inclination towards different types of recommended packages, and finally the sentiment inclination vector can be constructed. This process not only takes into account the user's actual usage data, but also incorporates information on the user's emotions and attitudes, providing a more comprehensive data foundation for subsequent personalized recommendation services.

[0086] In some optional embodiments, a method for recommending an operator package is also provided, the method comprising the following steps: obtaining operator package data and voice data of a target object, wherein the operator package data is multi-dimensional information including attribute information and package usage information about the target object, and the voice data includes usage requirement information for the operator package data; inputting the vectorized operator package data and voice data into a trained generative adversarial network model to obtain a recommended package for the target object output by the trained generative adversarial network model, wherein the generative adversarial network model is obtained by training using the above model training method.

[0087] In the above embodiment, firstly, the detailed package usage records of the target object are extracted from the database of the telecom operator, including but not limited to user ID, age, gender, region, historical package type, monthly consumption quota, traffic usage, voice call minutes, number of text messages, etc. The recordings of the target object's communication with the operator's customer service are obtained. These recordings contain the user's specific needs for package services, feedback on usage experience, and suggestions for improvement of existing packages. The collected package data and voice data are preprocessed, including noise removal, missing value filling, outlier detection, etc., to ensure data quality.

[0088] Then, speech recognition technology is used to convert the voice data into text format, and feature extraction is performed on the textual user demand information, including word frequency statistics, grammatical analysis, topic modeling, etc., to identify key points of user concern, such as traffic demand, package price sensitivity, etc.

[0089] The collected operator package data is converted into a numerical vector, where each dimension represents a specific attribute, such as age, consumption amount, etc. Use appropriate encoding methods, such as one-hot encoding, label encoding, or normalization, to ensure that the data format is compatible with the model. Use the pre-trained sentiment analysis model to perform sentiment analysis on the converted text data and output the user's positive or negative sentiment score for the package service. Convert the results of the sentiment analysis into a vector format, with each user corresponding to a sentiment vector, where each dimension represents the intensity of the sentiment tendency for a specific package type.

[0090] The vector data obtained in the above steps are integrated to form a comprehensive vector containing user attribute information, package usage information and emotional tendency as the input of the generative adversarial network model. The trained generative adversarial network (GAN) model is called, and the integrated comprehensive vector is used as input to the generator part of the GAN model. The generator part outputs one or more recommended package sequences for the target object, including recommendations for different package combinations. The generated package sequence will be verified by the discriminator of the GAN model to ensure its consistency with the real user data, thereby improving the accuracy and authenticity of personalized recommendations. The final personalized recommended package is presented to the target object in a clear and understandable form, which can be through SMS, email, APP push, etc., to ensure that users can easily understand the recommended content.

[0091] Through the above specific implementation steps, this application can combine the target object's operator package data and voice demand information, use the generative adversarial network model to output highly personalized recommended packages, and effectively improve user satisfaction and the market competitiveness of telecom operators. This process fully considers the diversity of user needs and incorporates sentiment analysis to make recommendations more humane and accurate.

[0092] Figure 2 is a schematic diagram of a model training system according to an embodiment of the present application, such as Figure 2 As shown, the system includes the following modules:

[0093] The data preprocessing module is used to collect users' package usage data and on-site voice data, perform cleaning, feature extraction and sentiment analysis, and build user feature matrix and sentiment tendency matrix.

[0094] The GAN network model includes a generator G and a discriminator S. The generator G is used to receive a random vector z as input and output simulated user data; the discriminator S is used to receive real user data and data output by the generator and output the true probability of the data.

[0095] Reinforcement learning and Bayesian models, including the design of state space, action space and reward function. The Bayesian optimization model based on reinforcement learning is used to construct the objective function based on the user's voice data and the generated preference data, and maximize the solution to obtain the optimal package recommendation strategy. The design of the reward function takes into account the correlation between user satisfaction, operator revenue and packages, and also adds the user's emotional tendency function to the reward function. The reward function R is designed to guide the training and optimization of the model.

[0096] The model optimization module adjusts the parameters of the generator G and the discriminator D so that the generator G can generate simulated data that meets the user's preferences, and the discriminator D can accurately distinguish between real data and simulated data. The reinforcement learning model optimizes the package recommendation strategy by learning the reward function R.

[0097] Combine the following Figure 2 The model training method proposed in this application is described in detail.

[0098] Step S201, collect and pre-process the model input data, wherein the input data includes the user's package usage data and on-site voice data.

[0099] Specifically, the original package data is cleaned and features are extracted, and the missing data is interpolated and normalized to obtain the package data matrix M = [m j1 ,m j2 ,..,m jn ], where m j is the number of categories of packages recommended to user j, each m jjRepresents the attribute value of the recommended package in category j. Perform speech recognition on the voice data of on-site communication with users, convert it into text, extract features from the text, perform sentiment analysis, and output the sentiment tendency matrix E = n*k, where n is the number of users and k is the number of package categories. For user i, the sentiment tendency vector can be expressed as E j =[e i1 ,e i2 ,…,e ik ], each e jj Indicates the intensity of user i’s sentiment towards package category j.

[0100] The extracted package data matrix and sentiment feature matrix are combined to obtain input data D, where the rows in D represent the user's package data and the columns represent a feature in the package data, and finally the input data D is obtained.

[0101] Step S202, constructing and training a GAN network model.

[0102] The user preference model is modeled by constructing a generative adversarial network (GAN) model. The GAN network consists of two parts, namely the generator G and the discriminator S. The generator G is responsible for generating virtual user data similar to the real user data, and the discriminator S is responsible for judging and evaluating the similarity between the generated data and the real data.

[0103] Specifically, the generator in the network structure of the generator G can be expressed as G = G(Z; θ G ), where z represents the random vector z and the data x drawn from the real data that follows the Gaussian distribution, θ G is the parameter set of the generator. With set Z as the input of the generator, in this scheme, the generator uses three fully connected layers and one output layer, that is, G(z) = tanh(W 3 *ReLU(W 2 *ReLU(W 1 *z))), where W 1 ,W 2 ,W 3 is the weight matrix, θ G It is composed of a weight matrix. ReLU and tanh are activation functions used to introduce nonlinear characteristics. The output of the generator G is a vector. After the output vector passes through the tanh activation function, its value will be limited to the range of [-1,1]. The loss function of the generator can optionally use LG=E x [MSE(x,x G )]+E_z[MSE(D(G(z)),1)], where MSE(x,x G ) is the mean square error loss function, which is used to measure the generated data x GThe difference between the real data x; MSE(D(G(z)),1) is the mean square error loss function, which measures the discriminator D's performance on the generated data x G The difference between the evaluation of and the evaluation of the real data x. Output virtual user data G(x G )=x G , x G The same dimension as D. After inputting the training data, the generator parameters θ are optimized by minimizing the loss function LG G , optional loss function minimization algorithms such as gradient descent method, particle swarm and other swarm intelligence optimization algorithms.

[0104] In constructing the network structure of the discriminator S, the real data D and the output data G(x G ) is used as input data X, and the output of the discriminator is the probability P of the authenticity of the input data. S can be expressed as S(X)=1 / (1+exp(-s(W 4 *x)))W 4 is the weight matrix, X is the output x of the generator G G Or real data D. By designing the loss function LS of S, LS = -E[log(S(S))+log(1-S(G(z)))], the loss function is minimized and optimized to obtain the output data of S.

[0105] The generative adversarial network model is trained by alternately training the generator G and the discriminator S, minimizing the generator's loss function LG and the discriminator's loss function LS. After inputting the training data, the model parameters are optimized by minimizing the loss function LG. Optional algorithms for minimizing the loss function include gradient descent, particle swarm, and other swarm intelligence optimization algorithms.

[0106] Step S203, reinforcement learning and Bayesian optimization model design.

[0107] First, define the state space T (containing the processed D and the generated data x G ),T can be expressed as T={t 1 ,t 2 ,...,t n}, where t i is the i-th state vector, n is the total number of states; the action space A (recommended package combination), A can be expressed as A = {a 1 ,a 2 ,...,a m}, where a j is the jth action, m is the total number of actions;

[0108] The design of the reward function R(t, a) can affect the output effect of the model, and thus affect the probability of users accepting the package after recommendation. In this embodiment, the reward function includes but is not limited to: user satisfaction, operator revenue, package combination, market trend, package survival time, etc.; in particular, the emotional tendency reward function R is added to the reward function sentiment (t,a), used to reward the model based on the user's emotional inclination towards the package. The stronger the inclination, the higher the reward score. sentiment (t,a) can be expressed as:

[0109]

[0110] Among them, w i is the weight of user i, determined based on the package activity, w j is the weight of package category j, which is determined based on the package value.

[0111] Therefore, the reward function R(t, a) is designed as follows:

[0112]

[0113] in, and λ sentiment is the weight coefficient, which is used to measure the importance of different factors; R user (t, a), R operator (t, a), R market (t, a) represent user satisfaction reward, operator revenue reward and market adaptability reward respectively.

[0114] Step S204: model optimization and parameter adjustment.

[0115] Further construct the Bayesian optimization algorithm model, and transform the real data D and the generated data G(x G ), the reward function R matrix is ​​used as input data, where the real data D and the generated data G(x G ) is the training input, the reward function R matrix is ​​the training output, and the design space T' is the state space T after randomization. The Bayesian optimization model can effectively seek the global optimal solution of the black box objective function f in the unchanged design space. By updating the prior knowledge of the objective function and obtaining the corresponding observation value, the posterior distribution closer to f is updated, so as to quickly find the optimal solution. It can be expressed as:

[0116] In the Bayesian optimization model of this scheme, the optimal solution is solved by maximizing the objective function, and the objective function f is designed as follows:

[0117] f(θ)=E[R(t,a)]=∑ t,aP(t,a|θ)*R(t,a)

[0118] Where E[] represents the expected operation, R(t,a) is the reward function, and P(t,a|θ) is the probability of taking action a in state t when the parameter θ is given. The parameter is the hyperparameter of the Bayesian optimization algorithm.

[0119] The Bayesian optimization algorithm is executed in the following steps: 1. Sampling: Sample a series of state-action pairs (t, a) according to the current parameters θ. 2. Evaluation: Evaluate the reward function R(t, a) on each pair (t, a). 3. Update: Update the parameters θ according to the evaluation results in order to maximize the objective function f(θ).

[0120] Under an optional Bayesian optimization model, a multi-output Student-T process model is used to approximate the objective function value.

[0121] Multi Output Student's T Process (MOSTP) is an extension of the single output Student's T process. From the definition, we can see that if f is a multivariate Student's T process on X, the parameter freedom v>2, and the vector-valued mean function u satisfies Covariance function (also called kernel function) k: satisfies And the semi-positive definite parameter matrix Ω∈R d×d , we can then define that any finite set of vector-valued variables has a Student's T distribution with a joint matrix variable:

[0122]

[0123] where f,u∈R d is a row vector whose components are functions M∈R n×d , where M ij =μ j (x i ),Σ∈R n×n , where Σ ij = k(x i ,x j ).

[0124] MOSTP can be expressed as: f~MOSTP(v,u,k,Ω). Based on the definition of multivariate student T process, the MOSTP model is as follows:

[0125] Given n pairs of observations x i ∈R p ,y i ∈R 1×d , assuming:

[0126] f~MOSTP(v,u,k′,Ω),v>2

[0127] y i =f(x i ),i=1,…,n

[0128] Where v is the degree of freedom of MOSTP, so its predicted distribution can be obtained as follows:

[0129]

[0130] in:

[0131]

[0132]

[0133] From the expectation and covariance of the matrix variable Student's T distribution, the predicted mean and covariance are given by:

[0134]

[0135]

[0136] In the MOSTP model, the observations follow a matrix variable Student's T distribution. The negative log marginal likelihood function is:

[0137]

[0138] Therefore, the parameters of MOSTP include an additional parameter - the degree of freedom ν. The negative log marginal likelihood is relative to the parameter ν, θ i ,φ ij , The derivative of is as follows:

[0139]

[0140] Among them, U=K′+YΩ -1 Y T , τ=v+n-1, ψn() is the function lnΓ n (·) The derivative with respect to ν.

[0141] Given a training sample, MOSTP is trained. Given a test sample, MOSTP can output the predicted mean and predicted variance of the multiple outputs corresponding to the test sample. The input sample is a matrix variable, and the output value is a matrix variable. Finally, the parameters of the trained model are optimized, the reward function parameters are adjusted, and the optimal result is output.

[0142] Figure 3is a training and prediction diagram of a multi-output student T process according to an embodiment of the present application, such as Figure 3 As shown in Figure 1, the process specifically includes two stages: training and prediction. Specifically, the features extracted from the user's historical package data and on-site voice data are preprocessed and represented in matrix or vector form. Then they are sent to the MOSTP model for processing. The model can capture and learn the complex relationship between input data and output data, especially when dealing with high dimensions and multiple outputs. Through model training, its internal parameters (such as parameter degrees of freedom ν, covariance matrix, etc.) are adjusted to optimize the accuracy and reliability of the prediction. After being processed by the MOSTP model, not only the predicted value can be output, but also the uncertainty estimate of the prediction result, that is, the prediction variance, can be provided. The prediction result is also output in the form of a matrix or vector, which is used in subsequent reinforcement learning and Bayesian optimization algorithms to build a more accurate package recommendation strategy.

[0143] Figure 4 is a schematic diagram of a method for recommending operator packages according to an embodiment of the present application, such as Figure 4 As shown in the figure, firstly, the historical package data of the target user is extracted from the database of the telecom operator, including but not limited to user ID, age, gender, region, historical package type, monthly consumption quota, traffic usage, voice call minutes, number of text messages, etc. And the user's on-site voice data, including but not limited to the specific needs of the package service, user experience feedback and suggestions for improvement of the existing package. The collected package data and voice data are preprocessed, including noise removal, missing value filling, outlier detection, etc., to ensure data quality.

[0144] Then, speech recognition technology is used to convert speech data into text format, and feature extraction is performed on the textual user demand information, including word frequency statistics, grammatical analysis, topic modeling, etc., to identify key points of user concern, such as traffic demand, package price sensitivity, etc. The collected operator package data is then converted into a numerical vector, where each dimension represents a specific attribute, such as age, consumption amount, etc. Use appropriate encoding methods, such as unique hot encoding, label encoding, or normalization to ensure that the data format is compatible with the model. Use a pre-trained sentiment analysis model to perform sentiment tendency analysis on the converted text data and output the user's positive or negative sentiment score for the package service. The results of the sentiment analysis are converted into a vector format, with each user corresponding to a sentiment tendency vector, where each dimension represents the intensity of the sentiment tendency for a specific package type.

[0145] The vector data obtained in the above steps is input into the operator package recommendation system for integration to form a comprehensive vector containing user attribute information, package usage information and emotional tendency as the input of the generative adversarial network model. The trained generative adversarial network (GAN) model is called, and the integrated comprehensive vector is used as input to the generator part of the GAN model. The generator part outputs one or more recommended package sequences for the target object, including recommendations for different package combinations. The generated package sequence will be verified by the discriminator of the GAN model to ensure its consistency with the real user data, thereby improving the accuracy and authenticity of personalized recommendations.

[0146] The final personalized recommendation package should be presented to the target audience in a clear and understandable form, including the recommended package, the user's possible acceptance level, the recommendation script template, etc., to ensure that the user can easily understand the recommendation content.

[0147] Figure 5 is a structural diagram of a model training device according to an embodiment of the present application, such as Figure 5 As shown, the device comprises:

[0148] The acquisition module 51 is used to acquire the historical operator package data and historical voice data of the target object, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data.

[0149] The generation module 52 is used to generate a first row vector corresponding to the historical operator package data, perform sentiment analysis on the usage demand information in the historical voice data, and obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages.

[0150] The merging module 53 is used to merge the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package categories of the recommended packages, and the columns are used to represent the features corresponding to the package categories.

[0151] The input module 54 is used to input the target matrix into the generative adversarial network model to obtain the target data output by the generator in the generative adversarial network model having a similar distribution to the target matrix, and the target probability that the target data output by the discriminator in the generative adversarial network model is the real data.

[0152] The determination module 55 is used to determine the emotional tendency reward based on the first row vector and the second row vector, and determine the reward function based on the emotional tendency reward, determine the target matrix and the target data as the state vector, determine the recommended package for the target object as the action vector, and determine the target function based on the target probability state vector, the action vector and the reward function.

[0153] The updating module 56 is used to determine the target probability as the optimization parameter to maximize the objective function and update the optimization parameter. In the discriminator, if the loss function of the target probability satisfies the preset convergence condition, a trained generative adversarial network model is obtained.

[0154] Optionally, the determination module 55 is also used to perform the following steps: determining several state vectors and action vector pairs under the current optimization parameters as observation values ​​to obtain an observation matrix including multiple observation values; determining the negative log marginal likelihood function according to the target probability, the observation matrix and the reward function, and using the negative log marginal likelihood function as the objective function.

[0155] Optionally, the update module 56 is also used to perform the following steps: obtaining model parameters in a multi-output student-T process, wherein the model parameters include degrees of freedom, mean function, covariance function, parameter matrix, and optimization parameters; calculating the derivatives of the negative log marginal likelihood with respect to the model parameters in the multi-output student-T process, and updating the model parameters in the opposite direction of the derivatives.

[0156] Optionally, the generation module 52 is also used to perform the following steps: determine a first weight based on an activity index of a historical operator package used by the target object; determine a second weight based on a package value index of a historical operator package used by the target object; and determine an emotional tendency reward based on the first weight, the second weight, the first row vector, and the second row vector.

[0157] Optionally, the merging module 53 is also used to perform the following steps: obtaining the target object's satisfaction index with the historical operator package used by the target object, the operator's revenue index with the historical operator package used by the target object, and the market adaptability index of the historical operator package used by the target object; performing weighted summation on the emotional tendency reward, satisfaction index, revenue index, and market adaptability index to obtain a reward function.

[0158] Optionally, the update module 56 is also used to perform the following steps: obtain several state vectors and action vector pairs under the current optimization parameters; calculate the reward function in each state vector and action vector, and update the optimization parameters according to the calculation results until the mathematical expectation of the state vector and action vector pair is greater than a preset threshold.

[0159] Optionally, the generation module 52 is also used to perform the following steps: performing data cleaning on the historical operator package data, and interpolating and filling the missing data in the historical operator package data after the data cleaning to obtain initial data of multiple dimensions, wherein the multiple dimensions include: identification information, age information, gender information, regional information, package subsidy information, package usage information, call information outside the package, SMS information outside the package, and traffic information outside the package of the target object; performing feature extraction on the initial data of multiple dimensions to obtain multiple initial features, and normalizing the multiple initial features to obtain multiple first features; based on the multiple first features, generating different types of recommended packages recommended to the target object to obtain a first row vector.

[0160] Optionally, the generation module 52 is also used to execute the following steps: converting the historical voice data into text data, performing feature extraction processing on the text data to obtain a text feature vector, wherein the feature extraction processing includes at least one of the following: word frequency statistics, grammatical analysis, and topic modeling; using a pre-trained sentiment analysis model to perform sentiment tendency analysis on the text feature vector to obtain a quantitative score of the target object's sentiment tendency towards different types of recommended packages; converting the quantitative score into a vector form to obtain a second row vector.

[0161] The model training device is also used to perform the following steps: obtaining the operator package data and voice data of the target object, wherein the operator package data is multi-dimensional information including attribute information and package usage information about the target object, and the voice data includes usage requirement information for the operator package data; inputting the vectorized operator package data and voice data into the trained generative adversarial network model to obtain a recommended package for the target object output by the trained generative adversarial network model, wherein the generative adversarial network model is obtained by training with the above model training method.

[0162] It should be noted that the above Figure 5 The modules in the embodiment may be program modules (e.g., a set of program instructions for implementing a specific function) or hardware modules. For the latter, they may be expressed in the following forms, but are not limited thereto: the expression form of the above modules is a processor, or the functions of the above modules are implemented by a processor.

[0163] It should be noted that Figure 5 The preferred implementation of the illustrated embodiment can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0164] Figure 6 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a model training method. Figure 6As shown, the computer terminal 60 may include one or more (602a, 602b, ..., 602n are used to illustrate) processors 602 (the processor 602 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 604 for storing data, and a transmission module 606 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 6 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 6 More or fewer components as shown, or with Figure 6 Different configurations are shown.

[0165] It should be noted that the one or more processors 602 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 60. As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0166] The memory 604 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the model training method in the embodiment of the present application. The processor 602 executes various functional applications and data processing by running the software programs and modules stored in the memory 604, that is, realizing the above-mentioned model training method. The memory 604 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 604 may further include a memory remotely arranged relative to the processor 602, and these remote memories may be connected to the computer terminal 60 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0167] The transmission module 606 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 60. In one example, the transmission module 606 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 606 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0168] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 60 .

[0169] It should be noted that, in some optional embodiments, the above Figure 6 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 6 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0170] It should be noted that Figure 6 The computer terminal shown is used to execute Figure 1 The model training method shown, therefore the relevant explanations in the execution method of the above commands are also applicable to the electronic device and will not be repeated here.

[0171] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above model training method.

[0172] A program for a non-volatile storage medium to perform the following functions: obtaining historical operator package data and historical voice data of a target object, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data; generating a first row vector corresponding to the historical operator package data, performing sentiment analysis on the usage demand information in the historical voice data, and obtaining a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages; merging the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package categories of the recommended packages, and the columns are used to represent the package categories corresponding features; input the target matrix into the generative adversarial network model to obtain the target data with a similar distribution to the target matrix output by the generator in the generative adversarial network model, and the target probability that the target data output by the discriminator in the generative adversarial network model is the real data; determine the emotional tendency reward according to the first row vector and the second row vector, and determine the reward function according to the emotional tendency reward, determine the target matrix and the target data as the state vector, determine the recommended package for the target object as the action vector, and determine the objective function according to the target probability state vector, the action vector and the reward function; determine the target probability as the optimization parameter to maximize the objective function with the target, update the optimization parameter, and in the discriminator, if the loss function about the target probability meets the preset convergence condition, obtain the generative adversarial network model that has completed the training.

[0173] An embodiment of the present application also provides an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above model training method is executed when the program is running.

[0174] The processor is used to run a program that performs the following functions: obtaining historical operator package data and historical voice data of a target object, wherein the historical operator package data is multi-dimensional information including attribute information about the target object and historical package usage information, and the historical voice data includes usage demand information for the operator package data; generating a first row vector corresponding to the historical operator package data, performing sentiment analysis on the usage demand information in the historical voice data, and obtaining a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages; merging the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package categories of the recommended packages, and the columns are used to represent the package categories corresponding features; input the target matrix into the generative adversarial network model to obtain the target data with a similar distribution to the target matrix output by the generator in the generative adversarial network model, and the target probability that the target data output by the discriminator in the generative adversarial network model is the real data; determine the emotional tendency reward according to the first row vector and the second row vector, and determine the reward function according to the emotional tendency reward, determine the target matrix and the target data as the state vector, determine the recommended package for the target object as the action vector, and determine the objective function according to the target probability state vector, the action vector and the reward function; determine the target probability as the optimization parameter to maximize the objective function with the target, update the optimization parameter, and in the discriminator, if the loss function about the target probability meets the preset convergence condition, obtain the generative adversarial network model that has completed the training.

[0175] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0176] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0177] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0178] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0179] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0180] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0181] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0182] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A model training method, characterized in that: include: Acquire historical operator package data and historical voice data of the target object, wherein the historical operator package data is multi-dimensional information including attribute information and historical package usage information about the target object, and the historical voice data includes usage demand information for the operator package data; Generate a first row vector corresponding to the historical operator package data, perform sentiment analysis on the usage demand information in the historical voice data, and obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are the target object's sentiment tendency information for different types of recommended packages; Merging the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package categories of the recommended packages, and the columns are used to represent the features corresponding to the package categories; Inputting the target matrix into a generative adversarial network model, obtaining target data output by a generator in the generative adversarial network model having a similar distribution to the target matrix, and a target probability that the target data output by a discriminator in the generative adversarial network model is real data; Determine an emotional tendency reward according to the first row vector and the second row vector, determine a reward function according to the emotional tendency reward, determine the target matrix and the target data as a state vector, determine a recommended package for the target object as an action vector, and determine an objective function according to the target probability, the state vector, the action vector, and the reward function; The target probability is determined as an optimization parameter to maximize the objective function, and the optimization parameter is updated. In the discriminator, if the loss function of the target probability satisfies a preset convergence condition, a generative adversarial network model that has completed training is obtained.

2. The method according to claim 1, characterized in that Determining an objective function according to the target probability, the state vector, the action vector, and the reward function includes: Determine a plurality of state vectors and action vector pairs under the current optimization parameters as observation values, and obtain an observation matrix including a plurality of observation values; determine a negative logarithmic marginal likelihood function according to the target probability, the observation matrix, and the reward function, and use the negative logarithmic marginal likelihood function as the objective function; Updating the optimization parameters includes: Obtaining model parameters in a multi-output Student-T process, wherein the model parameters include: degrees of freedom, mean function, covariance function, parameter matrix, and the optimization parameters; The derivatives of the negative log marginal likelihood with respect to the model parameters in the multi-output Student-T process are calculated, and the model parameters are updated along the inverse direction of the derivatives.

3. The method according to claim 1, characterized in that Determining a sentiment tendency reward according to the first row vector and the second row vector includes: Determining a first weight according to an activity index of a historical operator package used by the target object; Determining a second weight according to a package value index of a historical operator package used by the target object; An emotional tendency reward is determined according to the first weight, the second weight, the first row vector, and the second row vector.

4. The method according to claim 1 or 3, characterized in that: According to the emotional tendency reward, a reward function is determined, including: Acquire a satisfaction index of the target object with respect to a historical operator package used by the target object, a revenue index of the operator with respect to the historical operator package used by the target object, and a market adaptability index of the historical operator package used by the target object; The reward function is obtained by performing a weighted summation on the emotional tendency reward, the satisfaction index, the revenue index and the market adaptability index.

5. The method according to claim 1, characterized in that Updating the optimization parameters includes: Obtain several state vectors and action vector pairs under the current optimization parameters; calculate the reward function in each state vector and action vector, and update the optimization parameters according to the calculation results until the mathematical expectation of the state vector and action vector pair is greater than a preset threshold.

6. The method according to claim 1, characterized in that Generating the first row vector corresponding to the historical operator package data, including: Cleaning the historical operator package data, and performing interpolation and zero filling processing on missing data in the historical operator package data after data cleaning to obtain initial data of multiple dimensions, wherein the multiple dimensions include: identification information, age information, gender information, region information, package subsidy information, package usage information, call information outside the package, SMS information outside the package, and traffic information outside the package of the target object; Feature extraction is performed on the initial data of multiple dimensions to obtain multiple initial features, and the multiple initial features are normalized to obtain multiple first features; based on the multiple first features, different types of recommended packages recommended to the target object are generated to obtain the first row vector.

7. The method according to claim 1, characterized in that Perform sentiment analysis on the usage demand information in the historical voice data to obtain a second row vector, including: Converting the historical speech data into text data, performing feature extraction processing on the text data to obtain a text feature vector, wherein the feature extraction processing includes at least one of the following: word frequency statistics, grammatical analysis, and topic modeling; Use a pre-trained sentiment analysis model to perform sentiment analysis on the text feature vector to obtain a quantitative score of the target object's sentiment toward different types of recommended packages; convert the quantitative score into a vector form to obtain the second row vector.

8. A method for recommending operator packages, characterized in that: include: Acquire operator package data and voice data of the target object, wherein the operator package data is multi-dimensional information including attribute information and package usage information about the target object, and the voice data includes usage demand information for the operator package data; The vectorized operator package data and voice data are input into a trained generative adversarial network model to obtain a recommended package for the target object output by the trained generative adversarial network model, wherein the generative adversarial network model is trained by the model training method described in any one of claims 1 to 7.

9. A model training device, characterized in that: include: an acquisition module, configured to acquire historical operator package data and historical voice data of a target object, wherein the historical operator package data is multi-dimensional information including attribute information and historical package usage information about the target object, and the historical voice data includes usage demand information for the operator package data; A generating module, configured to generate a first row vector corresponding to the historical operator package data, and perform sentiment analysis on the usage demand information in the historical voice data to obtain a second row vector, wherein the components in the first row vector are different types of recommended packages for the target object, and the components in the second row vector are sentiment tendency information of the target object for different types of recommended packages; a merging module, configured to merge the first row vector and the second row vector to obtain a target matrix, wherein the rows of the target matrix are used to represent the package category of the recommended package, and the columns are used to represent the features corresponding to the package category; An input module, used to input the target matrix into a generative adversarial network model, to obtain target data output by a generator in the generative adversarial network model having a similar distribution to the target matrix, and a target probability that the target data output by a discriminator in the generative adversarial network model is real data; A determination module, configured to determine an emotional tendency reward according to the first row vector and the second row vector, determine a reward function according to the emotional tendency reward, determine the target matrix and the target data as a state vector, determine a recommended package for the target object as an action vector, and determine an objective function according to the target probability, the state vector, the action vector, and the reward function; An updating module is used to determine the target probability as an optimization parameter to maximize the objective function and update the optimization parameter. In the discriminator, if the loss function of the target probability satisfies a preset convergence condition, a trained generative adversarial network model is obtained.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the model training method described in any one of claims 1 to 7 and the operator package recommendation method described in claim 8.

11. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein when the program is run, the model training method described in any one of claims 1 to 7 and the operator package recommendation method described in claim 8 are executed.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the model training method described in any one of claims 1 to 7 and the operator package recommendation method described in claim 8.

Images