Mobile package recommendation method and device, and storage medium

Abstract

The present application discloses a method and device for recommending mobile packages, and a storage medium. The method comprises: obtaining an initial data set; wherein the initial data set includes communication service data of a service object; classifying the initial data set according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple service types; wherein the preset classification strategy is used to distinguish service objects according to different service types; each service type corresponds to a group of sub-data sets; extracting training data and test data from a group of sub-data sets in the multiple groups of sub-data sets; constructing a target recommendation model corresponding to a service type based on the training data and test data; traversing all sub-data sets in the multiple groups of sub-data sets until multiple target recommendation models corresponding to multiple service types are obtained; and using the multiple target recommendation models to perform recommendation processing on the recommended objects. The present application can perform personalized package recommendations, achieving efficient and intelligent package recommendations.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method and device for recommending a mobile package, and a storage medium. Background Art

[0002] With the launch of fifth-generation mobile networks (5G), the services provided by operators will become more diverse. Customers are an important resource for operators' development. In order to retain current users while attracting more new users, competition among the three major operators is becoming increasingly fierce. They are promoting user consumption through extensive and wide-ranging marketing models and lowering prices.

[0003] However, different users have different consumption habits. Due to insufficient analysis of customer needs, the traditional extensive and wide-ranging marketing model will not only consume a lot of manpower and financial resources, resulting in high marketing costs and poor results, but will also reduce users' consumption experience and satisfaction, and fail to directly stimulate user consumption, thus leading to customer churn. It cannot meet the development trend of mobile business diversity under the current situation of the mutual integration of mobile communications and the Internet. Summary of the Invention

[0004] The present application provides a mobile package recommendation method, device and storage medium, which can perform personalized package recommendations, achieve efficient and intelligent package recommendations, and meet the development trend of mobile service diversity.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] In a first aspect, an embodiment of the present application provides a method for recommending a mobile package, the method comprising:

[0007] Acquire an initial data set; wherein the initial data set includes communication service data of the service object;

[0008] Classifying the initial data set according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types; wherein the preset classification strategy is used to distinguish the service objects according to different business types; and one business type corresponds to one group of sub-data sets;

[0009] extracting training data and test data from the set of sub-datasets among the plurality of sets of sub-datasets;

[0010] Constructing a target recommendation model corresponding to the business type according to the training data and the test data;

[0011] Traversing all sub-data sets in the multiple groups of sub-data sets until a plurality of target recommendation models corresponding to the multiple business types are obtained;

[0012] The multiple target recommendation models are used to perform recommendation processing on the recommended objects.

[0013] In the above solution, the initial data set is classified according to a preset classification strategy to obtain multiple sub-data sets corresponding to multiple business types, including:

[0014] Extracting the first consumption parameter and attribute parameter of the service object from the communication service data;

[0015] Based on the first consumption parameter and the attribute parameter, the initial data set is classified to obtain the multiple groups of sub-data sets corresponding to the multiple business types.

[0016] In the above solution, the classification processing of the initial data set based on the first consumption parameter and the attribute parameter to obtain the multiple sub-data sets corresponding to the multiple business types includes:

[0017] Obtain the corresponding relationship between attribute parameters and attribute weight values;

[0018] Based on the first consumption parameter, the attribute parameter, and the corresponding relationship between the attribute parameter and the attribute weight value, the initial data set is classified to obtain the multiple groups of sub-data sets.

[0019] It can be seen that in the present invention, the mobile package recommendation device can combine consumption parameters, attribute parameters, and the correspondence between different attribute parameters and different attribute weight coefficients to divide users of different business types, thereby improving the accuracy of the division and being able to accurately determine the business type corresponding to the user.

[0020] In the above solution, extracting training data and test data from the set of sub-datasets among the multiple sets of sub-datasets includes:

[0021] extracting first package parameters from the set of sub-data sets;

[0022] Determining, according to the first package parameter, a plurality of pre-package categories corresponding to the one service type;

[0023] Extracting learning data sets corresponding to the plurality of pre-package categories from the one sub-data set;

[0024] The learning data set is segmented according to a preset ratio to obtain the training data and the test data corresponding to the one business type.

[0025] In the above solution, constructing a target recommendation model corresponding to the business type based on the training data and the test data includes:

[0026] Determining a neural network structure according to a preset optimization algorithm; wherein the preset optimization algorithm is used to determine the initialization weights of the neural network structure;

[0027] Inputting the training data into the neural network structure for machine learning to obtain a pre-recommendation model;

[0028] The pre-recommendation model is corrected using the test data to obtain a target recommendation model.

[0029] In the above solution, inputting the training data into the neural network structure for machine learning to obtain a pre-recommendation model includes:

[0030] Extracting training consumption parameters and training package parameters from the training data;

[0031] The training consumption parameters and the training package parameters are input into the neural network structure to perform the machine learning to obtain the pre-recommendation model.

[0032] In the above solution, the correction processing of the pre-recommendation model using the test data to obtain the target recommendation model includes:

[0033] Extracting test consumption parameters and test package parameters from the test data;

[0034] Inputting the test consumption parameters and the test package parameters into the pre-recommendation model to obtain a recommended test result;

[0035] The pre-recommendation model is corrected according to the recommendation test result to obtain the target recommendation model.

[0036] In the above solution, performing the correction process on the pre-recommendation model according to the recommendation test result to obtain the target recommendation model includes:

[0037] Determining whether the recommendation test result meets a preset recommendation threshold; wherein the preset recommendation threshold is used to reflect the recommendation accuracy of the pre-recommendation model;

[0038] If it is determined that the recommendation test result does not meet the preset recommendation threshold, the correction process is performed on the pre-recommendation model based on the recommendation test result to obtain the target recommendation model.

[0039] It can be seen that in the present invention, the mobile package recommendation device can use a preset optimization algorithm to optimize the weights of the neural network structure, overcome the defect that the neural network structure is prone to falling into local minima, and improve the accuracy of the neural network, so that the prediction accuracy of the mobile package recommendation model established based on the neural network is higher.

[0040] In the above solution, the method of performing recommendation processing on the recommended objects using the multiple target recommendation models includes:

[0041] Obtain target communication service data of the recommended object;

[0042] Determining the target business type corresponding to the object to be recommended based on the preset classification strategy;

[0043] Extracting target consumption parameters and target package parameters from the target communication service data;

[0044] Inputting the target consumption parameter and the target package parameter into the target recommendation model corresponding to the target service type for recommendation processing;

[0045] Obtain recommendation results, and send the recommendation results to the object to be recommended.

[0046] In the above solution, determining the target service type corresponding to the service object to be recommended based on the preset classification strategy includes:

[0047] extracting a second consumption parameter and a target attribute parameter from the target communication service data;

[0048] The target business type corresponding to the object to be recommended is determined according to the second consumption parameter, the target attribute parameter, and the corresponding relationship between the attribute parameter and the attribute weight value.

[0049] It can be seen that in the present invention, the mobile package recommendation device can accurately determine the type of user based on the consumption parameters and attribute parameters of the unknown user, and can accurately predict the package type used by the user according to the package recommendation model corresponding to the service type.

[0050] In a second aspect, an embodiment of the present application provides a mobile package recommendation device, the mobile package recommendation device comprising: an acquisition unit, a classification unit, an extraction unit, a creation unit, and a recommendation unit.

[0051] The acquisition unit is configured to acquire an initial data set, wherein the initial data set includes communication service data of the service object;

[0052] The classification unit is configured to classify the initial data set according to a preset classification strategy to obtain a plurality of sub-data sets corresponding to a plurality of business types; wherein the preset classification strategy is configured to distinguish the service objects according to different business types; and each business type corresponds to a group of sub-data sets;

[0053] The extraction unit is configured to extract training data and test data from the set of sub-datasets in the plurality of sets of sub-datasets;

[0054] The establishing unit is configured to construct a target recommendation model corresponding to the business type based on the training data and the test data;

[0055] The acquisition unit is further configured to traverse all sub-data sets in the multiple groups of sub-data sets until a plurality of target recommendation models corresponding to the multiple business types are obtained;

[0056] The recommendation unit is used to perform recommendation processing on the recommended objects using the multiple target recommendation models.

[0057] In a third aspect, an embodiment of the present application provides a mobile package recommendation device, which includes a processor and a memory storing instructions executable by the processor. When the instructions are executed by the processor, the mobile package recommendation method described above is implemented.

[0058] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a program stored thereon, which is applied to a mobile package recommendation device. When the program is executed by a processor, the mobile package recommendation method as described above is implemented.

[0059] The present application provides a mobile package recommendation method and device, and a storage medium, wherein the mobile package recommendation device can obtain an initial data set; wherein the initial data set includes communication service data of a service object; the initial data set is classified and processed according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple service types; wherein the preset classification strategy is used to distinguish service objects according to different service types; one service type corresponds to one group of sub-data sets; training data and test data are extracted from one group of sub-data sets in the multiple groups of sub-data sets; a target recommendation model corresponding to one service type is constructed based on the training data and the test data; all sub-data sets in the multiple groups of sub-data sets are traversed until multiple target recommendation models corresponding to multiple service types are obtained; and multiple target recommendation models are used to perform recommendation processing on the recommended objects. In other words, in the present application, the mobile package recommendation device can build different target recommendation models based on different service types, and then use different target recommendation models to perform package recommendation processing on the recommended objects of different service types, thereby being able to perform personalized package recommendations, achieving efficient and intelligent package recommendations, and meeting the development trend of mobile service diversity. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the structure of an exemplary Hopfield feedback neural network composed of three neurons proposed in an embodiment of the present application;

[0061] Figure 2 Schematic diagram of the composition structure of the mobile package recommendation device proposed in this application embodiment Figure 1 ;

[0062] Figure 3 Schematic diagram of the implementation process of the mobile package recommendation method proposed in this application embodiment Figure 1 ;

[0063] Figure 4 This is a schematic diagram of pre-package categories for users with the same consumption level proposed in the embodiment of the present application;

[0064] Figure 5 A schematic diagram of the basic process of the Hopfield neural network based on the optimized particle swarm algorithm proposed in the embodiment of the present application;

[0065] Figure 6 This is a schematic diagram of the implementation process of building a target recommendation model proposed in an embodiment of the present application;

[0066] Figure 7 Schematic diagram of the implementation process of the mobile package recommendation method proposed in this application embodiment Figure 2 ;

[0067] Figure 8A schematic diagram of a package recommendation process performed by an exemplary target recommendation model proposed in an embodiment of the present application;

[0068] Figure 9 Schematic diagram of the composition structure of the mobile package recommendation device proposed in this application embodiment Figure 2 ;

[0069] Figure 10 Schematic diagram of the composition structure of the mobile package recommendation device proposed in this application embodiment Figure 3 . DETAILED DESCRIPTION

[0070] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to explain the related applications and are not intended to limit the applications. It should also be noted that for ease of description, only the portions relevant to the related applications are shown in the drawings.

[0071] Before further explaining the embodiments of the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0072] 1) K-Nearest Neighbor (KNN) algorithm: This is one of the simplest machine learning algorithms. The idea is that if a sample has a majority of its k most similar samples (i.e., its closest neighbors in the feature space) belonging to a certain category, then the sample also belongs to that category.

[0073] 2) Normalization: This is a method of simplifying calculations by transforming dimensioned expressions into dimensionless ones, becoming scalars. This is a dimensionless processing technique that transforms the absolute values ​​of physical system values ​​into relative values. It is an effective way to simplify calculations and reduce the size of quantities.

[0074] 3) Hopfield Feedback Neural Network: This is a recurrent neural network that combines a memory system with a binary system. It ensures convergence to a local minimum and provides a model for simulating human memory. A discrete Hopfield network is a single-layer network with n neurons, where the output of each neuron is connected to the input of every other neuron. Each node has no self-feedback. Each node can be in one of two possible states (1 or -1). When the stimulus received by the neuron exceeds its threshold, the neuron is in one state (e.g., 1); otherwise, the neuron remains in the other state (e.g., -1).

[0075] Figure 1This is a schematic diagram of the structure of the Hopfield feedback neural network composed of three neurons proposed in the embodiment of the present application, as shown in FIG. Figure 1 As shown, x j is the external input, w ij Represents the weight between neurons, w ij When i=j, it is equal to 0, indicating that the output of the neuron will not be fed back to its input. Since the discrete random Hopfield neural network, each neuron only takes a binary discrete value of -1 or 1. For binary neurons, the calculation formula is:

[0076]

[0077] In formula 1, u j is the output of the neuron at time j, x j is external input, w ij Represents the weight between neurons, y i is the output result of the neuron at the previous moment of time i, i.e. time j. If the output information of the neuron at time j is u j If it is greater than the threshold θ, then the neuron output at time j is y j Output 1, otherwise the output is -1, that is, formula 2:

[0078]

[0079] As shown in formula 2, if the neuron output information at time j is greater than the threshold, then the neuron output result y at time j is j If it is equal to 1, the neuron output information at time j is less than the threshold, then the neuron output result y at time j j Equal to -1. For a Hopfield feedback neural network with an output layer of n neurons, the output at time t is:

[0080] Y(t)=[y1(t),y2(t),...,y n (t)] T (3)

[0081] In formula 3, y i (t)(i=1,2,...,n) represents the i-th neuron, which takes the value of 1 or -1, so Y(t) has 2 n state.

[0082] The state of the node at time (t+1) is obtained using the following formula:

[0083] y j (t+1)=∫[u j (t)] (4)

[0084]

[0085] In Formula 4 and Formula 5, u j is the output of the neuron at time t, x j is external input, w ij Represents the weight between neurons, y i is the neuron output result at time t, θ j is the threshold at time t.

[0086] 4) Particle Swarm Optimization (PSO): Also known as Particle Swarm Optimization (PSO) or Particle Swarm Optimization (PSO), this algorithm is a type of evolutionary algorithm. Starting from a random solution, it iteratively searches for the optimal solution, evaluating the quality of the solution through fitness. In other words, it seeks the global optimum by following the currently found optimal value. The PSO is a parallel algorithm. Originating from the study of bird flocking behavior, the algorithm was initially inspired by the regularity of flocking bird movements and then developed a simplified model using swarm intelligence. Based on observations of animal flocking behavior, the PSO leverages information sharing among individuals within the swarm to evolve the swarm's movement from disorder to order within the problem-solving space, ultimately achieving the optimal solution. PSO's advantages lie in its simplicity and ease of implementation, as well as the lack of numerous parameters to adjust. It is currently widely used in function optimization, neural network training, fuzzy system control, and other genetic algorithm applications.

[0087] Due to the differences in consumption habits among different customers, the traditional extensive and wide-ranging marketing model does not adequately analyze customer needs. It not only consumes a lot of manpower and financial resources, resulting in high marketing costs and poor results, but also reduces users' consumption experience and satisfaction, fails to directly stimulate user consumption, and thus leads to customer churn. It cannot meet the development trend of mobile business diversity under the current situation of the mutual integration of mobile communications and the Internet.

[0088] In order to solve the problems existing in the traditional customer package recommendation model, the present application provides a mobile package recommendation method and device, and a storage medium. Specifically, the mobile package recommendation device builds different target recommendation models based on different business types, and then uses different target recommendation models to recommend packages to recommended objects of different business types. It can perform personalized package recommendations, realize efficient and intelligent package recommendations, and meet the development trend of mobile business diversity.

[0089] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application.

[0090] An embodiment of the present application provides a mobile package recommendation method, which is applied to Figure 2The mobile package recommendation device 300 shown, Figure 2 Schematic diagram of the composition structure of the mobile package recommendation device 300 proposed in this embodiment of the application Figure 1 ,like Figure 2 As shown, in an embodiment of the present application, the mobile package recommendation device 300 includes a data collection module 301 and a model building module 302, wherein the user data collection module 301 is mainly used to comprehensively analyze user data, define user attribute sets according to different attribute characteristics of user data, and mainly divide user data into attribute characteristic data, consumption characteristic data, and package usage characteristic data; the model building module 302 is mainly used to analyze user attribute characteristic data and consumption characteristic data, classify users, and build package recommendation models for different categories of users according to different user consumption characteristic data and package usage characteristic data under different categories, and then perform package recommendation processing on the recommended users through the package recommendation model.

[0091] Figure 3 Schematic diagram of the implementation process of the mobile package recommendation method proposed in this application embodiment Figure 1 ,like Figure 3 As shown, in an embodiment of the present application, the mobile package recommendation device executing the mobile package recommendation method may include the following steps:

[0092] Step 101: Acquire an initial data set; wherein the initial data set includes communication service data of a service object.

[0093] In an embodiment of the present application, the mobile package recommendation device may first obtain an initial data set; wherein the initial data set includes communication service data of multiple service objects.

[0094] It should be noted that in the embodiments of the present application, the mobile package recommendation device can be any device with computing and storage functions. Specifically, the mobile package recommendation method can be applied to physical servers, virtual servers (such as cloud servers), tablet computers, mobile phones, personal computers (PCs), laptop computers, wearable devices, and other devices with computing and storage functions. The embodiments of the present application do not specifically limit the devices to which the mobile package recommendation method is applied. Moreover, in the specific implementation process, the mobile package recommendation method in the embodiments of the present application can be executed by a single device alone or by multiple devices in cooperation.

[0095] It should be noted that, in the embodiment of the present application, the initial data set includes communication business data of multiple service objects, wherein the service objects may be users registered with different operators, and the communication business data of multiple service objects is the business usage information of multiple users of the operator during their network period. Specifically, the communication business data may be obtained from a local or other device, that is, it may be obtained by the mobile package recommendation device directly reading the operator log record locally, or it may be the communication business data held by other devices, which are sent to the mobile package recommendation device in the embodiment of the present application, and the communication business data of multiple users are comprehensively analyzed through the data acquisition module 301.

[0096] Specifically, in an embodiment of the present application, the data acquisition module 301 performs a comprehensive analysis on the communication service data of multiple users, defines a user attribute set according to different attribute characteristics of the user data, and mainly divides the user data into attribute parameters, consumption parameters and package parameters; wherein the attribute parameters are mainly used to divide the user's characteristics and group dimensions, including the user's gender, age, number home, habitual residence, card activation time, whether it is a smartphone, whether it is a fourth generation mobile communication technology (4G) terminal, the user's monthly online time, user identity information and other user identification information; consumption parameters mainly represent user consumption characteristics, including the user's monthly traffic usage, traffic Internet time, home broadband, number of SMS consumption, voice call duration, monthly consumption fee and other information since the card activation; package parameters mainly represent the user's current package consumption, that is, the user's package usage, such as: basic package fee, package traffic remaining amount, package SMS remaining amount, package call remaining amount and related information of some Internet value-added services, such as; video value-added services and game value-added services activated by the user.

[0097] Furthermore, in an embodiment of the present application, after acquiring the communication service data of a plurality of service objects, the service objects may be further classified based on the communication service data.

[0098] Step 102: Classify the initial data set according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types; wherein the preset classification strategy is used to distinguish service objects according to different business types; one business type corresponds to a group of sub-data sets.

[0099] In an embodiment of the present application, after the mobile package recommendation device obtains the communication service data of multiple service objects, it can classify and process the initial data set according to a preset classification strategy, thereby obtaining multiple groups of sub-data sets corresponding to multiple service types; wherein the preset classification strategy is used to distinguish service objects according to different service types; one service type corresponds to a group of sub-data sets.

[0100] It should be noted that in the embodiments of the present application, due to the differences in the consumption habits, that is, the consumption levels, of multiple users, the package requirements of the users are also different. Therefore, before building the package recommendation model, we first classify the multiple users according to the preset classification strategy. That is, we first classify the initial data set according to the preset classification strategy to obtain multiple groups of sub-data sets, each sub-data set corresponding to a business type, so that users can be distinguished according to different business types, that is, different business types corresponding to different users. Optionally, the business type can be the user's consumption level, that is, multiple users can be classified according to their consumption level and divided into different consumption levels. For example, users can be divided into four categories: high consumption level, medium-high consumption level, medium-low consumption level, and low consumption level.

[0101] It should be noted that in the embodiment of the present application, the mobile package recommendation device can use the improved K-nearest neighbor algorithm to classify users according to their consumption level, that is, if most of the K nearest neighbor data of an unknown data in the feature space belong to a certain category, then the unknown data is also divided into this category. Since the K-nearest neighbor algorithm belongs to lazy learning, that is, there is no explicit learning process (training process), the data set already has classifications and feature values ​​in advance, that is, the sample data set is already classified according to the feature values, and after receiving the new sample, it can be directly processed by the algorithm to determine the category. Specifically, classification can be performed by measuring the distance between different feature values. The distance can be one of Euclidean distance, cosine value, correlation, and Manhattan distance. The distance measurement method is not specifically limited in the embodiment of the present application; further, after calculating the distance between the new sample test data and each training data in the sample data set, it is sorted according to the increasing relationship of the distance, and the K points with the smallest distance are selected according to the specified K value. The frequency of occurrence of the category of the first K points is further determined, and the category with the highest frequency among the first K points is used as the predicted classification of the new sample test data.

[0102] Specifically, in an embodiment of the present application, the first consumption parameters of multiple users can be extracted from the communication service data, and the initial data set can be classified based on the first consumption parameter to obtain multiple groups of sub-data sets corresponding to multiple service types. In other words, the service type of the user is determined based on the first consumption parameter, wherein the service type can be the user's consumption level. Specifically, the first consumption parameter can be any consumption parameter in the user's communication service data. Optionally, the first consumption parameter can be the total monthly consumption cost or the consumption parameter of traffic usage. The embodiment of the present application does not specifically limit the first consumption parameter. For example, users are divided into four categories based on the parameter of total monthly consumption cost: those with a monthly consumption cost below 30 yuan are low-consumption groups, those with a monthly consumption cost between 30 and 60 yuan are low-middle consumption groups, those with a monthly consumption cost between 60 and 90 yuan are medium-high consumption groups, and those with a monthly consumption cost above 90 yuan are high-consumption groups.

[0103] It should be noted that in the embodiments of the present application, since some attribute parameters in the user's communication service data will also affect the user's service type, such as the user's age, whether the user is using a smartphone, or the user's identity information (student, employed, retired, etc.), the embodiment of the present application does not specifically limit the attribute parameters. Therefore, it is necessary to combine the attribute parameters on the basis of the first consumption parameter to jointly determine the consumption level, that is, it is necessary to extract the first consumption parameter and at least one attribute parameter of the service object from the communication service data, and jointly determine the user's service type based on the first consumption parameter and at least one attribute parameter.

[0104] For example, when a certain attribute parameter is age, determining the consumption level of older users and younger users is related to their respective age attribute parameters. Since the monthly total consumption expenses of younger users are generally higher, while the monthly total consumption expenses of older users are generally lower, therefore, when combining the attribute parameter of age and the monthly total consumption expenses to determine the consumption level, it is determined that the younger users with a monthly total consumption expense of 50 are in a medium-low consumption level, while the older users with a monthly total consumption expense of 50 are in a medium-high consumption level; or the monthly total consumption expenses of employed users are generally higher, while the monthly total consumption expenses of retired users are generally lower. Then, when combining the attribute parameter of identity information and the monthly consumption expenses to determine the consumption level, it is possible to determine that the employed users with a monthly consumption expense of 50 are in a medium-low consumption level, while the retired users with a monthly consumption expense of 50 are in a medium-high consumption level.

[0105] Furthermore, in an embodiment of the present application, after extracting the first consumption parameter and attribute parameter of the service object from the communication service data, since different attribute parameters have different degrees of influence on the service type, the corresponding relationship between the attribute parameter and the attribute weight value can be further obtained, so that multiple users can be classified according to the first consumption parameter, the attribute parameter, and the corresponding relationship between the attribute parameter and the attribute weight value to determine multiple service types. Specifically, the data is first normalized, and the attribute parameter values ​​in the initial data set are mapped to [0, 1]. The normalization formula for the attribute parameter value is:

[0106]

[0107] In formula 6, x 1n * For each sample with N attribute parameters in the classified sample set, min n and max n is the minimum / maximum value of one of the attribute parameters, and the normalized attribute parameter array X is obtained by formula 6. * , there are N attribute parameters of R groups of users in the initial data set. Calculate the attribute parameter influence coefficient of each attribute parameter of multiple users in the initial data set, that is, the attribute weight value:

[0108]

[0109] In Formula 7, M refers to the mean value of each attribute parameter in the initial data set. The attribute weight value corresponding to each attribute parameter is determined according to Formula 7, and the attribute influence coefficient array λ = [λ1,λ2,...λ n ], that is, the correspondence between attribute parameters and attribute weight values. Further, update the K-nearest neighbor distance formula:

[0110]

[0111] Among them, dist(X1,X2) is the number of different users with N attributes X1=(x 11 ,x 12 ,...x 1n ,) and X2=(x 21 ,x 22 ,...x 2n ,), that is, when the improved K-nearest neighbor algorithm processes the distance of multidimensional data, it repeatedly considers the impact of different attribute data on the results, which is beneficial to improving the classification accuracy of the K-nearest neighbor algorithm.

[0112] Furthermore, in an embodiment of the present application, after a plurality of users are classified according to the first consumption parameter, attribute parameter, and the correspondence between the attribute parameter and the attribute weight value, and the business types corresponding to the plurality of users are determined, the users under each classified business type can be used as a sample set, and the first consumption parameter and attribute parameter of the user under each business type can be used as feature data in the sample set. When it is necessary to determine the business type corresponding to the user to be recommended, the first consumption parameter and attribute parameter of the user to be recommended can be directly used in combination with the improved K-nearest neighbor algorithm to determine to which business type in the sample set the user to be recommended belongs.

[0113] Furthermore, in an embodiment of the present application, after the initial data set is classified according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types, training data and test data can be further extracted from the multiple groups of sub-data sets to construct a package recommendation model.

[0114] Step 103: extract training data and test data from one of the multiple sub-datasets.

[0115] In an embodiment of the present application, after the mobile package recommendation device classifies the initial data set according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types, training data and test data can be extracted from one group of sub-data sets in the multiple groups of sub-data sets.

[0116] It should be noted that in the embodiments of the present application, different sub-data sets correspond to different business types. Since the mobile package recommendation device builds a recommendation model based on different business types, before building the model, it is necessary to extract training data and test data from a sub-data set corresponding to each business type to further build the recommendation model.

[0117] Specifically, in an embodiment of the present application, the rules for extracting training data and test data from a set of sub-datasets may be as follows: first, extract a first package parameter from a set of sub-datasets, determine multiple pre-package recommendation categories corresponding to a service type based on the first package parameter, then extract a learning data set corresponding to multiple pre-package categories from a sub-dataset, and segment the learning data set according to a preset ratio, thereby obtaining training data and test data corresponding to a service type. In other words, first, based on the first package parameter in a sub-dataset, determine the pre-package category under the service type corresponding to the sub-dataset, that is, the package type recommended for unknown users under the same service type, then extract communication service data corresponding to the pre-package category from the sub-dataset as a learning data set, and then further segment the learning data set into training data and test data according to a certain ratio.

[0118] Optionally, the first package parameter may be a package name. By counting the package names used by users of the same business type, the top M packages may be taken, that is, the packages that are more frequently used by users of the same business type. These M package types may then be used as recommended package types for users of the same business type. Since these M package types are packages that are more frequently used by users of the same business type, they provide a better user package experience. Using these M package types as recommended package types can make the recommended package categories more intelligent and efficient, and provide a stronger user experience.

[0119] For example, Figure 4 Schematic diagram of pre-package categories for users with the same consumption level, as shown in Figure 4 As shown, ● indicates that the neuron state is 1, and ○ indicates that the neuron state is -1. Among the high-spending population, the first four package categories with the highest frequency of use are taken as pre-package categories. The first package is the family sharing package, which includes 500 minutes of voice, 500 text messages, 40G of traffic, 400M home broadband, Migu Video, Game Accelerator Service, and Home Edition Mobile Cloud. The basic package fee is 159. Figure 4 The second package is the Global Communication Package, which includes 400 minutes of voice calls, 400 text messages, 30G of data, 200M of home broadband, and no other value-added services. The basic package fee is 129. Figure 4 The third package is a large data package, with 200 minutes of voice, 200 text messages, 40G of data, no home broadband, video service, game accelerator service, and no mobile cloud. The basic package fee is 109. Figure 4 The fourth package is the 4G free travel package, which includes 100 minutes of voice calls, 100 text messages, 30G of traffic, 100M of home broadband, no value-added services, and a basic package fee of 99.

[0120] Specifically, after determining the pre-package categories, feature data corresponding to these multiple pre-package categories are extracted from the sub-datasets as a learning data set. For example, the feature data can be basic service data such as voice, SMS, traffic, broadband, etc., and Internet value-added service data such as video value-added services, game value-added services, and other value-added services. Then, the learning data set is divided into training data and test data according to a certain ratio. For example, the learning data set is divided into an 8:2 ratio, with 80% of the learning data set being training data and 20% being test data.

[0121] Furthermore, in an embodiment of the present application, after extracting training data and test data from one set of sub-datasets among the multiple sets of sub-datasets, a target recommendation model may be further constructed based on the training data and the test data.

[0122] Step 104: Construct a target recommendation model corresponding to a business type based on the training data and the test data.

[0123] In an embodiment of the present application, after extracting training data and test data from a set of sub-datasets among multiple sets of sub-datasets, the mobile package recommendation device can construct a target recommendation model for the service type corresponding to the sub-dataset based on the training data and test data in the set of sub-datasets.

[0124] It should be noted that, in the embodiments of the present application, the creation of a target recommendation model corresponding to the same business type is based on the training data and training data under the same business type, and the target recommendation model corresponding to the business type is further constructed in combination with a specific neural network structure. Specifically, a neural network structure determined based on a preset optimization algorithm can be used. Since the most important parameters affecting the training results of the neural network structure model are the structure of the neural network and the weights of the neural network, preferably, the present application can use a Hopfield feedback neural network as the specific neural network structure for model training, and then use a particle swarm algorithm as a weight optimization algorithm. The particle swarm algorithm is used to obtain the optimal initialization weights of the neural network, and then the parameters of the neural network are further optimized and adjusted through training data and test data until the target recommendation model with the highest recommendation accuracy is obtained. In the embodiments of the present application, there is no specific limitation on the neural network structure and the initial weight optimization algorithm.

[0125] Specifically, in the embodiments of the present application, for example Figure 1 The discrete Hopfield network shown here consists of n neurons and has an n*n weight matrix W = {w ij}, and there is an n-dimensional threshold vector θ=[θ1,θ2,...,θ n ] T When optimizing the neural network structure by particle swarm algorithm, according to the weight matrix, the D-dimensional search space is composed of n particles X=(X1,X2,...X n ), where X i =[x i1 ,x i2 ,...x id ] T represents the position of the i-th particle in D-dimensional space, and the velocity of the i-th particle is V i =[V i1 ,V i2 ,...V id ] T , Pbest is P i =[P i1 ,P i2 ,...P id ] T,Gbest is G i =[G i1 ,G i2 ,...G id ] T The particles update themselves through the individual extreme value Pbest and the group extreme value Gbest. Pbest represents the position, and Gbest represents the optimal solution searched by all particles in the group. In each iteration, the particles update their speed and position according to the following formula:

[0126]

[0127]

[0128] Wherein, Formula 9 is the iterative formula for updating particle velocity, Formula 10 is the iterative formula for updating particle position, ω is the inertia weight, c1 and c2 are acceleration factors, and r1 and r2 are random numbers distributed between [0, 1]. The inertia weight ω has a significant impact on the efficiency of the particle swarm algorithm. The embodiment of the present application adopts an inertia weight calculation method that changes linearly with the number of iterations:

[0129]

[0130] In formula 11, w min and w max is the minimum / maximum inertia weight, k max is the maximum number of iterations. This allows the algorithm to have a larger inertia weight in the early stages of the iteration, improving the algorithm's global search capability, and a smaller inertia weight in the later stages of the iteration, which is beneficial for local search.

[0131] Furthermore, in the embodiments of the present application, Figure 5 The basic flow chart of the Hopfield neural network based on the optimized particle swarm algorithm proposed in the embodiment of the present application is as follows: Figure 5As shown, first, the weight vector of the neural network is used as each particle individual of the particle swarm, and then the parameters and fitness function of the particle swarm are initialized: including the learning factor; the random number; the speed and position of the j-th dimension of the i particle at time t (step 1001), and then the inertia weight value of the particle is calculated (step 1002), and the mean square error function of the neural network is set as the fitness function of the particle, and the fitness of each particle is calculated (step 1003); then the particle swarm algorithm is used to update (step 1004), and the optimal particle individual after the update is completed is recorded, and the parameters and fitness function of the particle swarm are initialized N times, and the optimal particle individual after the new update is completed is recorded. Specifically, in the particle update process, it is necessary to determine whether the end condition is met (step 1005). The end condition can be the set fitness value size or the number of cycles. If the end condition is not met, then the steps 1002-1004 are continuously executed until the termination condition is reached, that is, the fitness value reaches a certain value or the number of cycles is a certain number. The optimal particle individual that meets the termination condition is the optimal initialization weight of the Hopfield neural network. Furthermore, after completing the initialization of the neural network structure (step 1006) and the first iteration of the network equation (step 1007), the optimal weights obtained based on the particle swarm algorithm are used as the initialization weights of the neural network to obtain the weight matrix of the network (step 1008), and the network state equation of the neural network is used to calculate the output results of the neurons in combination with the training data as external input data (step 1009). When calculating the output results, it is determined whether the end condition is currently met (step 1010). If the output result does not meet the end condition, the network state equation is iterated and the initialization weights are updated and adjusted until the accuracy of the output result of the neural network meets the threshold condition, that is, the end condition is met, and the output result is obtained, thereby obtaining a pre-recommendation model.

[0132] Specifically, in an embodiment of the present application, the training data under the same business type is first input into a specific neural network structure for machine learning to obtain a pre-recommendation model. In order to verify the recommendation accuracy of the pre-recommendation model, the test data is further input into the pre-recommendation model trained with the training data to obtain a recommendation test result, and then the accuracy of the pre-recommendation model is judged according to the recommendation test result, and the pre-recommendation model is further corrected according to the recommendation test result.

[0133] Optionally, in an embodiment of the present application, feature data for training can be first extracted from the training data. The feature data can be multiple consumption parameters and multiple package parameters, such as the basic service data such as voice, SMS, traffic, broadband, and Internet value-added service data such as video value-added services, game value-added services, and other value-added services. In the embodiment of the present application, there is no specific limitation on the feature data in the training data and test data. Then, these feature data are input into the neural network structure for machine learning to obtain a pre-recommendation model; then, feature data of the same category for testing the accuracy of the pre-recommendation model are further extracted from the test data, and these test feature data are input into the pre-recommendation model to obtain a recommendation test result. If the test result indicates that the pre-recommendation model meets the recommendation accuracy threshold, then the pre-recommendation model corresponding to the same business type is the target recommendation model corresponding to the same business type; if the recommendation test result shows that the pre-recommendation model does not meet the recommendation accuracy threshold, then the pre-recommendation model corresponding to the same business type needs to be corrected based on the recommendation test result, wherein the correction process can adjust the weights of the neural network structure to a certain extent until the pre-recommendation model meets the accuracy threshold, and then the target recommendation model corresponding to the same business type is obtained.

[0134] Furthermore, in an embodiment of the present application, after constructing a target recommendation model corresponding to a business type based on training data and test data, the mobile package recommendation device can traverse all sub-data sets in multiple groups of sub-data sets to further obtain multiple target recommendation models corresponding to multiple business types.

[0135] Step 105: traverse all sub-datasets in the multiple groups of sub-datasets until multiple target recommendation models corresponding to multiple business types are obtained.

[0136] In an embodiment of the present application, after constructing a target recommendation model corresponding to a business type based on training data and test data, the mobile package recommendation device can traverse all sub-data sets in multiple groups of sub-data sets, and further obtain multiple target recommendation models corresponding to multiple business types.

[0137] It should be noted that in the embodiments of the present application, since each sub-dataset corresponds to a business type, that is, different sub-datasets correspond to different business types, it is necessary to traverse each sub-dataset in multiple sub-datasets and complete the construction of the corresponding target recommendation model for each business type.

[0138] Specifically, in order to construct a target recommendation model corresponding to each business type, a first package parameter is extracted from the sub-data set corresponding to the business type, and then multiple pre-package recommendation categories corresponding to the business type are determined based on the first package parameter, and a learning data set including feature data corresponding to multiple pre-package categories under the business type is further extracted from the sub-data set. The learning data set is segmented according to a certain preset ratio, for example, 8:2, to obtain training data and test data corresponding to the business type, and then the training data is input into the neural network structure determined based on the initial weight optimization algorithm for machine learning to obtain a pre-recommendation model, and then the test data is further input into the pre-recommendation model for correction processing, so that the obtained target recommendation model meets the recommendation accuracy threshold.

[0139] For example, after dividing the communication service data of multiple users into four categories according to their consumption levels, namely high consumption, upper-middle consumption, lower-middle consumption, and low consumption, the package names at each consumption level can be counted, and the top M package categories with higher usage frequencies can be determined as pre-package categories for each consumption level. Multiple consumption parameters and package parameters corresponding to the pre-package categories at each consumption level are extracted from the communication service data of each consumption level as learning data, for example: basic service data such as voice, SMS, traffic, broadband, and Internet value-added service data such as video value-added services, game value-added services, and other value-added services. The learning data is then divided into training data and test data in a ratio of 8:2. The training data is input into a Hopfield feedback neural network determined based on a particle swarm optimization algorithm to obtain a pre-package recommendation model corresponding to each consumption level. The test data is then input into the pre-package recommendation model to obtain a target package recommendation model corresponding to each consumption level.

[0140] Furthermore, in an embodiment of the present application, the mobile package recommendation device traverses all sub-data sets in multiple groups of sub-data sets, obtains multiple target recommendation models corresponding to multiple business types, and can further use the multiple target recommendation models to perform package recommendation processing on the recommended users.

[0141] Step 106: Utilize multiple target recommendation models to perform recommendation processing on the recommended objects.

[0142] In an embodiment of the present application, the mobile package recommendation device traverses all sub-data sets in multiple groups of sub-data sets, obtains multiple target recommendation models corresponding to multiple business types, and can use the obtained multiple target recommendation models to perform recommendation processing on the recommended service objects.

[0143] It should be noted that in an embodiment of the present application, the mobile package recommendation device can perform package recommendation processing for users of different service types based on the determined multiple target recommendation models using the target recommendation model corresponding to their service type. Specifically, the mobile package recommendation device can first obtain the target communication service data of the user to be recommended, where the target communication service data includes the attribute parameters, consumption parameters, and package parameters of the user to be recommended, and then determine the service type of the user to be recommended. Thereafter, the feature data is input into the target recommendation model corresponding to the service type to obtain a package recommendation result for the user to be recommended, wherein the feature data input into the target recommendation model and the feature data in the training data and test data when constructing the target recommendation model are parameters of the same category.

[0144] Specifically, Figure 6 This is a schematic diagram of the implementation process of building a target recommendation model proposed in the embodiment of this application. First, the user's communication service data is obtained, such as Figure 6 As shown, an improved K-nearest neighbor algorithm is used to classify users, and an improved neural network algorithm is used to construct a target package recommendation model for users in the same category. Based on different target recommendation models corresponding to different user categories, package recommendations are then implemented for the recommended users. For example, if communication service data of the user to be recommended is obtained and it is determined that the user category is high-consumption, feature data is extracted from the user's communication service data and input into the high-consumption target recommendation model to obtain a recommendation result. If it is determined that the user to be recommended is low-consumption, feature data is extracted from the user's communication service data and input into the low-consumption target recommendation model to obtain a recommendation result.

[0145] The present application provides a mobile package recommendation method and device, and a storage medium, wherein the mobile package recommendation device obtains an initial data set; wherein the initial data set includes communication service data of a service object; the initial data set is classified and processed according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple service types; wherein the preset classification strategy is used to distinguish service objects according to different service types; one service type corresponds to one group of sub-data sets; training data and test data are extracted from one group of sub-data sets in the multiple groups of sub-data sets; a target recommendation model corresponding to one service type is constructed based on the training data and the test data; all sub-data sets in the multiple groups of sub-data sets are traversed until multiple target recommendation models corresponding to multiple service types are obtained; and multiple target recommendation models are used to perform recommendation processing on the recommended objects. That is, in the present application, the mobile package recommendation device builds different target recommendation models based on different service types, and then uses different target recommendation models to perform package recommendation processing on the recommended objects of different service types, which can perform personalized package recommendation, realize the efficiency and intelligence of package recommendation, and meet the development trend of mobile service diversity.

[0146] Based on the above embodiment, in another embodiment of the present application, Figure 7 Schematic diagram of the implementation process of the mobile package recommendation method proposed in this application embodiment Figure 2 ,like Figure 7 As shown, the method for the mobile package recommendation device to perform recommendation processing on the recommendation object using multiple target recommendation models includes:

[0147] Step 201: Obtain target communication service data of the object to be recommended.

[0148] In an embodiment of the present application, after the mobile package recommendation device successfully constructs multiple target recommendation models corresponding to multiple service types based on the training data and test data in each sub-data set, it can obtain the communication service data of the user to be recommended and further perform package recommendation processing on the recommended user based on the communication service data.

[0149] It should be noted that in the embodiments of the present application, the target communication service data of the recommended object is the communication service usage information of users registered with different operators during their network usage. The service information of the recommended object can be obtained by the mobile package recommendation device directly reading the operator's log records locally, or it can be obtained by storing the communication service data in another device and sending it to the mobile package recommendation device in the embodiments of the present application. Specifically, the target communication service data includes the target attribute parameters, target consumption parameters, and target package parameters of the recommended object.

[0150] Furthermore, in an embodiment of the present application, after obtaining the target communication service data of the object to be recommended, the service type of the object to be recommended may be further determined.

[0151] Step 202: Determine the target business type corresponding to the object to be recommended based on a preset classification strategy.

[0152] In an embodiment of the present application, after the target communication service data of the object to be recommended is acquired, the target service type corresponding to the object to be recommended may be determined based on a preset classification strategy.

[0153] It should be noted that in an embodiment of the present application, the mobile package recommendation device can extract the second consumption parameter and target attribute parameter from the target communication service data, and determine the target service type corresponding to the object to be recommended based on the second consumption parameter, the target attribute parameter, and the correspondence between the attribute parameter and the attribute weight value.

[0154] Specifically, in an embodiment of the present application, the mobile package recommendation device can classify users according to their consumption levels using an improved K-nearest neighbor algorithm as shown in Formula 1. As can be seen from step 102, the mobile package recommendation device classifies multiple users based on the first consumption parameter, the attribute parameter, and the corresponding relationship between the attribute parameter and the attribute weight, further obtaining an improved K-nearest neighbor algorithm, and using the classified users of different service types as a sample set of users to be recommended, and the first consumption parameter and the attribute parameter as sample feature data. When determining the service type of the user to be recommended, feature data of the object to be recommended and the sample set are obtained, namely, the second consumption parameter and the target attribute parameter, which are of the same category but have different data values. The second consumption parameter, the target attribute parameter, and the attribute weight corresponding to the target attribute parameter are substituted into the improved K-nearest neighbor algorithm shown in Formula 1. The distance between the object to be recommended and each classified sample in the sample set is calculated according to Formula 1, the top K classified samples with the smallest distance are extracted, and the service type of the classified sample with the highest frequency among the top K points is used as the target service type of the object to be recommended.

[0155] Exemplarily, when determining the consumption level of the user to be recommended, the total monthly consumption cost of the user to be recommended, and attribute characteristic parameters such as age, identity information, and number location are obtained, and the attribute weight values ​​corresponding to the attribute characteristic parameters such as age, identity information, and number location are determined. According to Formula 1, among the top 5 samples with the smallest distance, there are 3 medium-low consumption levels, one medium-high consumption level, and one high consumption level. It can be seen that the medium-low consumption level appears most frequently among the top 5 classified samples, so the consumption level of the user to be recommended is the medium-low consumption level.

[0156] Furthermore, in an embodiment of the present application, after determining the service type corresponding to the object to be recommended, in order to further perform package recommendation processing on the object to be recommended, it is necessary to first extract input feature data of the target recommendation model from the target communication service data.

[0157] Step 203: Extract target consumption parameters and target package parameters from the target communication service data.

[0158] In an embodiment of the present application, after determining the target service type corresponding to the object to be recommended based on a preset classification strategy, the input feature data of the target recommendation model, that is, the target consumption parameters and target package parameters, can be extracted from the target communication service data.

[0159] It should be noted that, in the embodiment of the present application, the target recommendation model input feature data extracted from the target communication service data, i.e., the target consumption parameters and target package parameters, are feature data of the same category as the feature data in the training data and test data input when constructing the target recommendation model in steps 103 and 104. In other words, if the feature data in the training data and test data are basic service data such as voice, SMS, traffic, broadband, and Internet value-added service data such as video value-added services, game value-added services, and other value-added services, then the target consumption parameters and target package parameters extracted from the communication service data of the object to be recommended are also basic service data such as voice, SMS, traffic, broadband, and Internet value-added service data such as video value-added services, game value-added services, and other value-added services, and are further input into the target recommendation model corresponding to the user to be recommended.

[0160] Furthermore, in an embodiment of the present application, after extracting the target consumption parameters and the target package parameters, these feature data can be further input into a recommendation model for package recommendation processing.

[0161] Step 204: Input the target consumption parameters and the target package parameters into a target recommendation model corresponding to the target service type for recommendation processing.

[0162] In an embodiment of the present application, after extracting target consumption parameters and target package parameters from the communication service data of the object to be recommended, the mobile package recommendation device inputs them into a target recommendation model corresponding to the target service type of the object to be recommended for recommendation processing.

[0163] It should be noted that, in the embodiment of the present application, different target recommendation models corresponding to different business types are constructed based on steps 101-105. That is to say, after determining the business type of the object to be recommended, its target consumption parameters and target package parameters need to be input into the target recommendation model corresponding to the target business type in order to accurately perform package recommendation processing on the object to be recommended. For example, when the consumption level of the object to be recommended, that is, the business type is a high consumption level, the target consumption parameters and target package parameters are input into the high consumption level target recommendation model; when the consumption level of the object to be recommended, that is, the business type is a low consumption level, the target consumption parameters and target package parameters are input into the low consumption level target recommendation model.

[0164] Furthermore, in an embodiment of the present application, after the target consumption parameters and the target package parameters are input into a target recommendation model corresponding to the target service type for recommendation processing, a recommendation result can be further obtained.

[0165] Step 205: Obtain recommendation results and send the recommendation results to the recommended object.

[0166] In an embodiment of the present application, after inputting the target consumption parameters and the target package parameters into the target recommendation model corresponding to the target business type for recommendation processing, a recommendation result is obtained, and the recommendation result can be sent to the object to be recommended.

[0167] It should be noted that in the embodiments of the present application, the recommendation results can be sent to the user by the operator in various prompt forms such as text messages, phone calls, or operator APP software push messages. This application does not specifically limit the method of sending the recommendation results.

[0168] Furthermore, after the user obtains the recommendation results, he can compare them with the currently used package. If they are similar, the currently used package is the optimal package category. If there are differences, the recommended user can change the package and use the package recommendation results pushed by the operator.

[0169] For example, Figure 8 A schematic diagram of the target recommendation model provided in the embodiment of the present application for performing package recommendation processing, as shown in FIG. Figure 8 As shown in the figure, the first row of the matrix represents the recognition results of the target recommendation model for the feature data of the recommended users with the same consumption level. When the index value is greater than or equal to a certain classification, the corresponding neuron state output is 1, otherwise it is -1, where ● indicates that the neuron state is 1, and ○ indicates that the neuron state is -1. The second row of the matrix represents the package category predicted by the target recommendation model for users with the same consumption level. It can be seen that the target recommendation model determines that the recommendation result for user 1 is pre-package category 4 based on the feature data of user 1 with the same consumption level, including phone calls, text messages, data traffic, home broadband, video value-added services, game value-added services, other services and package prices; based on the feature data of user 2 with the same consumption level, the recommendation result for user 2 is determined to be pre-package category 3; the same applies to users 3 and 4.

[0170] The present application provides a mobile package recommendation method and device, and a storage medium, wherein the mobile package recommendation device obtains an initial data set; wherein the initial data set includes communication service data of a service object; the initial data set is classified and processed according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple service types; wherein the preset classification strategy is used to distinguish service objects according to different service types; one service type corresponds to one group of sub-data sets; training data and test data are extracted from one group of sub-data sets in the multiple groups of sub-data sets; a target recommendation model corresponding to one service type is constructed based on the training data and the test data; all sub-data sets in the multiple groups of sub-data sets are traversed until multiple target recommendation models corresponding to multiple service types are obtained; and multiple target recommendation models are used to perform recommendation processing on the recommended objects. That is, in the present application, the mobile package recommendation device builds different target recommendation models based on different service types, and then uses different target recommendation models to perform package recommendation processing on the recommended objects of different service types, which can perform personalized package recommendation, realize the efficiency and intelligence of package recommendation, and meet the development trend of mobile service diversity.

[0171] Based on the above embodiment, in another embodiment of the present application, Figure 9 Schematic diagram of the composition structure of the mobile package recommendation device proposed in this application Figure 2 ,like Figure 9 As shown, the mobile package recommendation device 300 proposed in the embodiment of the present application may include an acquisition unit 303, a classification unit 304, an extraction unit 305, a creation unit 306 and a recommendation unit 307.

[0172] The acquisition unit 303 is configured to acquire an initial data set, wherein the initial data set includes communication service data of the service object;

[0173] The classification unit 304 is configured to classify the initial data set according to a preset classification strategy to obtain multiple sub-data sets corresponding to multiple business types; wherein the preset classification strategy is used to distinguish the service objects according to different business types; and each business type corresponds to a group of sub-data sets;

[0174] The extraction unit 305 is configured to extract training data and test data from the set of sub-datasets in the plurality of sub-datasets;

[0175] The establishing unit 306 is configured to construct a target recommendation model corresponding to the business type based on the training data and the test data;

[0176] The acquisition unit 303 is further configured to traverse all sub-data sets in the multiple sub-data sets until a plurality of target recommendation models corresponding to the plurality of business types are obtained;

[0177] The recommendation unit 307 is configured to perform recommendation processing on the to-be-recommended objects using the multiple target recommendation models.

[0178] Furthermore, in an embodiment of the present application, the classification unit 304 is specifically used to extract the first consumption parameter of the service object from the communication service data; and classify the initial data set based on the first consumption parameter to obtain the multiple groups of sub-data sets corresponding to the multiple service types.

[0179] Furthermore, in an embodiment of the present application, the classification unit 304 is also specifically used to extract the first consumption parameter and attribute parameter of the service object from the communication service data; and based on the first consumption parameter and the attribute parameter, classify the initial data set to obtain the multiple groups of sub-data sets corresponding to the multiple service types.

[0180] Furthermore, in an embodiment of the present application, the classification unit 304 is also specifically used to obtain the correspondence between attribute parameters and attribute weight values; and based on the first consumption parameter, the attribute parameter and the correspondence between the attribute parameter and the attribute weight value, the initial data set is classified and processed to obtain the multiple groups of sub-data sets.

[0181] Furthermore, in an embodiment of the present application, the extraction unit 305 is specifically used to extract a first package parameter from the group of sub-data sets; and determine multiple pre-package categories corresponding to the one business type based on the first package parameter; and extract a learning data set corresponding to the multiple pre-package categories from the one sub-data set; and segment the learning data set according to a preset ratio to obtain the training data and the test data corresponding to the one business type.

[0182] Furthermore, in an embodiment of the present application, the establishment unit 306 is specifically used to determine a neural network structure according to a preset optimization algorithm; wherein the preset optimization algorithm is used to determine the initialization weights of the neural network structure; and input the training data into the neural network structure for machine learning to obtain a pre-recommendation model; and correct the pre-recommendation model through the test data to obtain the target recommendation model.

[0183] Furthermore, in an embodiment of the present application, the establishment unit 306 is also specifically used to extract training consumption parameters and training package parameters from the training data; and input the training consumption parameters and the training package parameters into the neural network structure for machine learning to obtain the pre-recommendation model.

[0184] Furthermore, in an embodiment of the present application, the establishment unit 306 is also specifically used to extract test consumption parameters and test package parameters from the test data; and input the test consumption parameters and the test package parameters into the pre-recommendation model to obtain a recommendation test result; and perform the correction processing on the pre-recommendation model according to the recommendation test result to obtain the target recommendation model.

[0185] Furthermore, in an embodiment of the present application, the establishment unit 306 is also specifically used to determine whether the recommendation test result meets a preset recommendation threshold; wherein the preset recommendation threshold is used to reflect the recommendation accuracy of the pre-recommendation model; and if it is determined that the recommendation test result does not meet the preset recommendation threshold, the pre-recommendation model is corrected based on the recommendation test result to obtain the target recommendation model.

[0186] Furthermore, in an embodiment of the present application, the recommendation unit 307 is specifically used to obtain target communication service data of the object to be recommended; and determine the target service type corresponding to the object to be recommended based on the preset classification strategy; and extract target consumption parameters and target package parameters from the target communication service data; and input the target consumption parameters and target package parameters into the target recommendation model corresponding to the target service type for recommendation processing; and obtain a recommendation result, and send the recommendation result to the object to be recommended.

[0187] Furthermore, in an embodiment of the present application, the recommendation unit 307 is also specifically used to extract a second consumption parameter and a target attribute parameter from the target communication service data; and determine the target service type corresponding to the object to be recommended based on the second consumption parameter, the target attribute parameter, and the correspondence between the attribute parameter and the attribute weight value.

[0188] Furthermore, in an embodiment of the present application, the communication service data includes attribute parameters, consumption parameters and package parameters; wherein the attribute parameters are identification information of the service object.

[0189] Furthermore, in an embodiment of the present application, the consumption parameters include monthly consumption fee, voice call duration, number of SMS consumption, traffic usage, and home broadband.

[0190] Furthermore, in an embodiment of the present application, the package parameters include package name, package price and Internet value-added services.

[0191] In the embodiments of the present application, further, Figure 10 Schematic diagram of the composition structure of the mobile package recommendation device proposed in this application Figure 3 ,like Figure 10As shown, the mobile package recommendation device 300 proposed in the embodiment of the present application may also include a processor 308, a memory 309 storing executable instructions of the processor 308, and further, the mobile package recommendation device 300 may also include a communication interface 310, and a bus 311 for connecting the processor 308, the memory 309 and the communication interface 310.

[0192] In an embodiment of the present application, the processor 308 may be at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, and a microprocessor. It is understandable that for different devices, the electronic device used to implement the above-mentioned processor function may also be other, and the embodiment of the present application does not specifically limit this. The mobile package recommendation device 300 may also include a memory 309, which may be connected to the processor 308, wherein the memory 309 is used to store executable program code, which includes computer operating instructions. The memory 309 may include a high-speed RAM memory, and may also include a non-volatile memory, for example, at least two disk memories.

[0193] In the embodiment of the present application, the bus 311 is used to connect the communication interface 310, the processor 308 and the memory 309, and to facilitate mutual communication between these devices.

[0194] In the embodiment of the present application, the memory 309 is used to store instructions and data.

[0195] Further, in an embodiment of the present application, the above-mentioned processor 308 is used to obtain an initial data set; wherein, the initial data set includes communication business data of the service object; the initial data set is classified and processed according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types; wherein, the preset classification strategy is used to distinguish the service objects according to different business types; one business type corresponds to a group of sub-data sets; training data and test data are extracted from the group of sub-data sets in the multiple groups of sub-data sets; a target recommendation model corresponding to the one business type is constructed based on the training data and the test data; all sub-data sets in the multiple groups of sub-data sets are traversed until multiple target recommendation models corresponding to the multiple business types are obtained; and the recommendation objects are recommended using the multiple target recommendation models.

[0196] In practical applications, the memory 309 may be a volatile memory, such as a random-access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 308.

[0197] In addition, the functional modules in this embodiment can be integrated into a recommended unit, or each unit can exist physically separately, or two or more units can be integrated into a single unit. The above-mentioned integrated unit can be implemented in the form of hardware or software functional modules.

[0198] If the integrated unit is implemented in the form of a software functional module and is not 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 this embodiment, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method of this embodiment. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0199] The present application provides a mobile package recommendation device, which obtains an initial data set; wherein the initial data set includes communication service data of a service object; the initial data set is classified and processed according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple service types; wherein the preset classification strategy is used to distinguish service objects according to different service types; one service type corresponds to one group of sub-data sets; training data and test data are extracted from one group of sub-data sets in the multiple groups of sub-data sets; a target recommendation model corresponding to one service type is constructed based on the training data and the test data; all sub-data sets in the multiple groups of sub-data sets are traversed until multiple target recommendation models corresponding to multiple service types are obtained; and multiple target recommendation models are used to perform recommendation processing on the recommended objects. That is, in the present application, the mobile package recommendation device builds different target recommendation models based on different service types, and then uses different target recommendation models to perform package recommendation processing on the recommended objects of different service types, which can perform personalized package recommendation, realize the efficiency and intelligence of package recommendation, and meet the development trend of mobile service diversity.

[0200] An embodiment of the present application provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the mobile package recommendation method described above.

[0201] Specifically, the program instructions corresponding to the mobile package recommendation method in this embodiment can be stored on a storage medium such as a CD, a hard disk, or a USB flash drive. When the program instructions corresponding to the mobile package recommendation method in the storage medium are read or executed by an electronic device, the following steps are included:

[0202] Acquire an initial data set; wherein the initial data set includes communication service data of the service object;

[0203] Classifying the initial data set according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types; wherein the preset classification strategy is used to distinguish the service objects according to different business types; and one business type corresponds to one group of sub-data sets;

[0204] extracting training data and test data from the set of sub-datasets among the plurality of sets of sub-datasets;

[0205] Constructing a target recommendation model corresponding to the business type according to the training data and the test data;

[0206] Traversing all sub-data sets in the multiple groups of sub-data sets until a plurality of target recommendation models corresponding to the multiple business types are obtained;

[0207] The multiple target recommendation models are used to perform recommendation processing on the recommended objects.

[0208] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0209] The present application is described with reference to the implementation flow charts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flow charts and / or block diagrams, as well as the combination of processes and / or boxes in the flow charts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the implementation flow charts. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0210] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which is implemented in the implementation flow diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0211] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process described in the flowchart. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0212] The above description is merely a preferred embodiment of the present application and is not intended to limit the scope of protection of the present application.

Claims

1. A mobile package recommendation method, characterized in that: The method comprises: Acquire an initial data set; wherein the initial data set includes communication service data of the service object; Classifying the initial data set according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types; wherein the preset classification strategy is used to distinguish the service objects according to different business types; and one business type corresponds to one group of sub-data sets; extracting training data and test data from the set of sub-datasets among the plurality of sets of sub-datasets; Constructing a target recommendation model corresponding to the business type according to the training data and the test data; Traversing all sub-data sets in the multiple groups of sub-data sets until a plurality of target recommendation models corresponding to the multiple business types are obtained; Performing recommendation processing on the recommended objects using the multiple target recommendation models; Extracting training data and test data from the set of sub-datasets in the plurality of sub-datasets comprises: extracting first package parameters from the set of sub-data sets; Determining, based on the first package parameter, a plurality of pre-package categories corresponding to the one service type; the pre-package categories being the top N package categories with the highest usage frequency in the set of sub-data sets; Extracting learning data sets corresponding to the plurality of pre-package categories from the set of sub-data sets; Segmenting the learning data set according to a preset ratio to obtain the training data and the test data corresponding to the one business type; The constructing a target recommendation model corresponding to the business type according to the training data and the test data includes: Determining a neural network structure according to a preset optimization algorithm; wherein the preset optimization algorithm is used to determine the initialization weights of the neural network structure; Inputting the training data into the neural network structure for machine learning to obtain a pre-recommendation model; The pre-recommendation model is corrected using the test data to obtain a target recommendation model.

2. The method according to claim 1, characterized in that The initial data set is classified according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types, including: extracting a first consumption parameter of the service object from the communication service data; The initial data set is classified based on the first consumption parameter to obtain the multiple groups of sub-data sets corresponding to the multiple business types.

3. The method according to claim 1, characterized in that The initial data set is classified according to a preset classification strategy to obtain multiple groups of sub-data sets corresponding to multiple business types, including: Extracting the first consumption parameter and attribute parameter of the service object from the communication service data; Based on the first consumption parameter and the attribute parameter, the initial data set is classified to obtain the multiple groups of sub-data sets corresponding to the multiple business types.

4. The method according to claim 3, characterized in that The classifying the initial data set based on the first consumption parameter and the attribute parameter to obtain the multiple sub-data sets corresponding to the multiple business types includes: Obtain the corresponding relationship between attribute parameters and attribute weight values; Based on the first consumption parameter, the attribute parameter, and the corresponding relationship between the attribute parameter and the attribute weight value, the initial data set is classified to obtain the multiple groups of sub-data sets.

5. The method according to claim 1, wherein The step of inputting the training data into the neural network structure for machine learning to obtain a pre-recommendation model includes: Extracting training consumption parameters and training package parameters from the training data; The training consumption parameters and the training package parameters are input into the neural network structure to perform the machine learning to obtain the pre-recommendation model.

6. The method according to claim 2, characterized in that The correcting the pre-recommendation model using the test data to obtain a target recommendation model includes: Extracting test consumption parameters and test package parameters from the test data; Inputting the test consumption parameters and the test package parameters into the pre-recommendation model to obtain a recommended test result; The pre-recommendation model is corrected according to the recommendation test result to obtain the target recommendation model.

7. The method according to claim 6, characterized in that The correcting the pre-recommendation model according to the recommendation test result to obtain the target recommendation model includes: Determining whether the recommendation test result meets a preset recommendation threshold; wherein the preset recommendation threshold is used to reflect the recommendation accuracy of the pre-recommendation model; If it is determined that the recommendation test result does not meet the preset recommendation threshold, the correction process is performed on the pre-recommendation model based on the recommendation test result to obtain the target recommendation model.

8. The method according to claim 4, characterized in that The using the multiple target recommendation models to perform recommendation processing on the recommended objects includes: Obtain target communication service data of the recommended object; Determining the target business type corresponding to the object to be recommended based on the preset classification strategy; Extracting target consumption parameters and target package parameters from the target communication service data; Inputting the target consumption parameter and the target package parameter into the target recommendation model corresponding to the target service type for recommendation processing; Obtain recommendation results, and send the recommendation results to the object to be recommended.

9. The method according to claim 8, characterized in that The determining the target business type corresponding to the object to be recommended based on the preset classification strategy includes: extracting a second consumption parameter and a target attribute parameter from the target communication service data; The target business type corresponding to the object to be recommended is determined according to the second consumption parameter, the target attribute parameter, and the corresponding relationship between the attribute parameter and the attribute weight value.

10. The method according to claim 1, characterized in that The communication service data includes attribute parameters, consumption parameters and package parameters; wherein the attribute parameters are identification information of the service object.

11. The method according to claim 10, characterized in that The consumption parameters include monthly consumption fees, voice call duration, number of SMS messages, data usage, and home broadband.

12. The method according to claim 10, characterized in that The package parameters include package name, package price and Internet value-added services.

13. A mobile package recommendation device, characterized in that: The mobile package recommendation device includes: an acquisition unit, a classification unit, an extraction unit, a creation unit and a recommendation unit. The acquisition unit is configured to acquire an initial data set, wherein the initial data set includes communication service data of the service object; The classification unit is configured to classify the initial data set according to a preset classification strategy to obtain a plurality of sub-data sets corresponding to a plurality of business types; wherein the preset classification strategy is configured to distinguish the service objects according to different business types; and each business type corresponds to a group of sub-data sets; The extraction unit is configured to extract training data and test data from the set of sub-datasets in the plurality of sets of sub-datasets; The establishing unit is configured to construct a target recommendation model corresponding to the business type based on the training data and the test data; The acquisition unit is further configured to traverse all sub-data sets in the multiple groups of sub-data sets until a plurality of target recommendation models corresponding to the multiple business types are obtained; The recommendation unit is configured to perform recommendation processing on the recommended objects using the multiple target recommendation models; The extraction unit is specifically configured to extract a first package parameter from the set of sub-datasets; determine a plurality of pre-package categories corresponding to the one service type based on the first package parameter; the pre-package categories being the top N package categories with the highest frequency of use in the set of sub-datasets; extract a learning data set corresponding to the plurality of pre-package categories from the set of sub-datasets; and segment the learning data set according to a preset ratio to obtain the training data and the test data corresponding to the one service type; The establishment unit is specifically used to determine the neural network structure according to a preset optimization algorithm; wherein the preset optimization algorithm is used to determine the initialization weights of the neural network structure; and input the training data into the neural network structure for machine learning to obtain a pre-recommendation model; and correct the pre-recommendation model through the test data to obtain a target recommendation model.

14. The mobile package recommendation device according to claim 13, wherein: The classification unit is specifically configured to extract the first consumption parameter of the service object from the communication service data; and classify the initial data set based on the first consumption parameter to obtain the multiple groups of sub-data sets corresponding to the multiple service types.

15. The mobile package recommendation device according to claim 13, wherein: The classification unit is further specifically used to extract the first consumption parameter and attribute parameter of the service object from the communication service data; and based on the first consumption parameter and the attribute parameter, classify the initial data set to obtain the multiple groups of sub-data sets corresponding to the multiple service types.

16. The mobile package recommendation device according to claim 15, characterized in that: The classification unit is further specifically used to obtain the correspondence between attribute parameters and attribute weight values; and based on the first consumption parameter, the attribute parameter and the correspondence between the attribute parameter and the attribute weight value, classify the initial data set to obtain the multiple groups of sub-data sets.

17. The mobile package recommendation device according to claim 13, wherein: The establishment unit is further specifically used to extract training consumption parameters and training package parameters from the training data; and input the training consumption parameters and the training package parameters into the neural network structure for machine learning to obtain the pre-recommendation model.

18. The mobile package recommendation device according to claim 17, wherein: The establishment unit is also specifically used to extract test consumption parameters and test package parameters from the test data; and input the test consumption parameters and the test package parameters into the pre-recommendation model to obtain the recommended test results; and perform the correction processing on the pre-recommendation model according to the recommended test results to obtain the target recommendation model.

19. The mobile package recommendation device according to claim 18, wherein: The establishment unit is also specifically used to determine whether the recommendation test result meets a preset recommendation threshold; wherein the preset recommendation threshold is used to reflect the recommendation accuracy of the pre-recommendation model; and if it is determined that the recommendation test result does not meet the preset recommendation threshold, the pre-recommendation model is corrected based on the recommendation test result to obtain the target recommendation model.

20. The mobile package recommendation device according to claim 16, wherein: The recommendation unit is specifically configured to obtain target communication service data of the object to be recommended; determine the target service type corresponding to the object to be recommended based on the preset classification strategy; and extract target consumption parameters and target package parameters from the target communication service data; and inputting the target consumption parameter and the target package parameter into the target recommendation model corresponding to the target service type for recommendation processing; And obtain the recommendation result, and send the recommendation result to the object to be recommended.

21. The mobile package recommendation device according to claim 20, characterized in that: The recommendation unit is further specifically used to extract a second consumption parameter and a target attribute parameter from the target communication service data; and determine the target service type corresponding to the object to be recommended based on the second consumption parameter, the target attribute parameter, and the correspondence between the attribute parameter and the attribute weight value.

22. The mobile package recommendation device according to claim 13, wherein: The communication service data includes attribute parameters, consumption parameters and package parameters; wherein the attribute parameters are identification information of the service object.

23. The mobile package recommendation device according to claim 22, wherein: The consumption parameters include monthly consumption fees, voice call duration, number of SMS messages, data usage, and home broadband.

24. The mobile package recommendation device according to claim 22, wherein: The package parameters include package name, package price and Internet value-added services.

25. A mobile package recommendation device, characterized in that: The mobile package recommendation device includes a processor and a memory storing instructions executable by the processor. When the instructions are executed by the processor, the method according to any one of claims 1 to 12 is implemented.

26. A computer-readable storage medium having a program stored thereon, applied to a mobile package recommendation device, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

Images