Telephone number recognition model training method and device, storage medium and related equipment
By obtaining and processing the initial call data of phone numbers from the enterprise database, and using principal component analysis and support vector machine methods to train the phone number identification model, the problems of low efficiency and poor accuracy of car owner identity judgment in traditional marketing methods are solved, and a higher Marketing group positioning accuracy and identification accuracy.
Patent Information
- Application Number
- CN202411302981.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2025-01-03
AI Technical Summary
Traditional marketing methods rely on manual judgment of car owner identity, which is inefficient and poorly accurate. Especially in a society where mobile phone numbers are frequently changed, it is difficult to accurately judge the car owner's identity for a single phone number, resulting in insufficient accuracy in marketing group positioning.
By obtaining the initial call data of phone numbers from the enterprise database, after standardization, the principal component analysis and support vector machine method are used for dimensionality reduction and linear classification, sample call data is generated, and the phone number recognition model is trained to improve recognition accuracy and robustness sex.
It improves the generalization ability and robustness of the phone number recognition model, improves the accuracy of corporate marketing group positioning, and ensures the accuracy of phone recognition results.
Smart Images
Figure CN119249245A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a telephone number recognition model training method, device, storage medium and related equipment. Background Art
[0002] In today's society, with the continuous growth of car ownership and the rapid development of mobile Internet technology, car owners have become a target market that many companies and service providers pay close attention to. This group not only has clear consumer needs, such as vehicle maintenance, insurance, refueling, repairs, etc., but also shows high spending power and loyalty. Therefore, accurately identifying and targeting car owners has great commercial value for companies.
[0003] However, traditional marketing methods often require manual identification of the car owner of the call recipient, and then corresponding advertising marketing. This method is not only inefficient, but also has poor accuracy in identifying car owners. Especially in modern society where one person has multiple numbers and mobile phone numbers change frequently, it is difficult to fully portray the user portrait based on a single phone number, and it is even more difficult to accurately determine whether the person is the car owner, which leads to poor accuracy in the company's marketing group positioning. Summary of the invention
[0004] The purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defect that it is difficult to fully portray the user portrait based on a single phone number information in the prior art, and it is even more difficult to accurately determine whether the user is the car owner, which leads to the poor accuracy of the company's marketing group.
[0005] The present application provides a method for training a telephone number recognition model, the method comprising:
[0006] Acquire initial call data containing multiple phone numbers from an enterprise database, and standardize the initial call data to obtain standard call data; wherein the attributes of the phone numbers include vehicle owners and non-vehicle owners;
[0007] Performing dimensionality reduction analysis on the standard call data using a principal component analysis method to generate intermediate call data, and performing linear classification on the intermediate call data using a support vector machine to obtain sample call data;
[0008] The sample call data is used to train a preset initial number recognition model, and the trained initial number recognition model is used as a telephone number recognition model.
[0009] Optionally, obtaining the initial call data containing multiple telephone numbers from the enterprise database includes:
[0010] A plurality of vehicle owners' telephone numbers and a plurality of non-vehicle owners' telephone numbers are randomly selected from the enterprise database according to a preset ratio;
[0011] The call data of each phone number is obtained, and each call data is attributed to form initial call data.
[0012] Optionally, the standardizing the initial call data to obtain standard call data includes:
[0013] Determine the standard deviation of each call data in the initial call data; each call data is composed of multiple fields;
[0014] Each field in each call data is standardized based on the standard deviation of each call data to obtain standard call data.
[0015] Optionally, the adopting a principal component analysis method to perform dimensionality reduction analysis on the standard call data to generate intermediate call data includes:
[0016] Determine a covariance matrix of each call data in the standard call data, and determine a characteristic value of each field in each call data based on each covariance matrix;
[0017] For each call data, eigenvalues corresponding to a preset number are intercepted from each eigenvalue in order from large to small, and data dimension reduction is performed on the covariance matrix of the call data based on the intercepted eigenvalues;
[0018] The intermediate call data is generated according to the dimension reduction results of each covariance matrix.
[0019] Optionally, the adopting a support vector machine to perform linear classification on the intermediate call data to obtain sample call data includes:
[0020] Using a kernel function to perform high-dimensional mapping on the intermediate call data to obtain linearly separable data;
[0021] Optimizing the classification interval of the linearly separable data by Lagrange multiplier method, and constructing a classification decision function of the intermediate call data according to the optimization result;
[0022] The intermediate call data is classified using the classification decision function to obtain sample call data.
[0023] Optionally, the using the sample call data to train a preset initial number recognition model, and using the trained initial number recognition model as a telephone number recognition model, includes:
[0024] Dividing the sample call data into a sample training set and a sample verification set;
[0025] Iteratively training a preset initial number recognition model using the sample training set to generate an intermediate number recognition model;
[0026] The sample verification set is used to perform prediction verification on the intermediate number recognition model, and the verified intermediate number recognition model is used as the telephone number recognition model.
[0027] Optionally, each call data in the sample training set is annotated with a corresponding real attribute result;
[0028] The iterative training of the preset initial number recognition model using the sample training set to generate an intermediate number recognition model includes:
[0029] Inputting the sample training set into a preset initial number recognition model to obtain a predicted attribute result output by the initial number recognition model;
[0030] The initial number recognition model is trained with the goal of making the predicted attribute result close to the real attribute result of the sample training set;
[0031] When the initial number recognition model meets the preset training end condition, the trained initial number recognition model is used as the telephone number recognition model.
[0032] The present application also provides a telephone number recognition model training device, comprising:
[0033] A data acquisition module, used to acquire initial call data containing multiple phone numbers from an enterprise database, and standardize the initial call data to obtain standard call data; wherein the attributes of the phone numbers include car owners and non-car owners;
[0034] A data processing module, configured to perform dimensionality reduction analysis on the standard call data using a principal component analysis method to generate intermediate call data, and perform linear classification on the intermediate call data using a support vector machine to obtain sample call data;
[0035] The model training module is used to train a preset initial number recognition model using the sample call data, and use the trained initial number recognition model as a telephone number recognition model.
[0036] The present application also provides a storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the telephone number identification method as described in any one of the above embodiments.
[0037] The present application also provides a computer device, comprising: one or more processors, and a memory;
[0038] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the telephone number recognition model training method described in any one of the above embodiments are performed.
[0039] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:
[0040] The telephone number recognition model training method, device, storage medium and related equipment provided by the present application, when it is necessary to train a telephone number recognition model for identifying telephone number attributes, the initial call data containing multiple telephones can be first obtained from the enterprise database, and the initial call data can be standardized to obtain standard call data, so as to avoid the inaccuracy of subsequent model recognition due to the different dimensions between the fields in the data. Then, the principal component analysis method can be used to perform dimensionality reduction analysis on the standard call data, and the intermediate call data generated in this way can reduce the data dimension, thereby improving the training efficiency and effect of the model. Then, the support vector machine can be used to perform linear classification on the intermediate call data to obtain sample call data, so as to improve the accuracy and robustness of data attribute classification. Therefore, the telephone number recognition model obtained by training the preset initial number recognition model with sample call data has high generalization ability and robustness, and its telephone recognition result has high accuracy, thereby improving the accuracy of the marketing group positioning of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0042] Figure 1 A flowchart of a telephone number recognition model training method provided in an embodiment of the present application;
[0043] Figure 2 A comparative schematic diagram of data dimension visualization before and after principal component analysis provided in an embodiment of the present application;
[0044] Figure 3 A schematic diagram of a flow chart of a linear classification process of intermediate call data provided in an embodiment of the present application;
[0045] Figure 4 A schematic diagram of the structure of a telephone number recognition model training device provided in an embodiment of the present application;
[0046] Figure 5 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0048] Traditional marketing methods often require manual identification of the car owner of the call recipient, and then corresponding advertising marketing. This method is not only inefficient, but also has poor accuracy in identifying car owners. Especially in modern society where one person has multiple numbers and mobile phone numbers change frequently, it is difficult to fully portray the user portrait based on a single phone number, and it is even more impossible to accurately determine whether the person is the car owner, which leads to poor accuracy in the company's marketing group positioning.
[0049] Based on this, this application proposes the following technical solutions, see below for details:
[0050] In one embodiment, Figure 1 As shown, Figure 1 A flowchart of a telephone number recognition model training method provided in an embodiment of the present application; the present application provides a telephone number recognition model training method, which specifically includes the following:
[0051] S110: Acquire initial call data containing multiple phone numbers from an enterprise database, and standardize the initial call data to obtain standard call data.
[0052] In this step, when the enterprise needs to train a phone number recognition model for identifying the attributes of a phone number, the computer device can first obtain the initial call data between the enterprise and multiple phone numbers from the enterprise database as a data basis for analyzing user calling habits. The computer device can then standardize the initial call data to obtain standard call data, thereby avoiding the impact of different dimensions between various fields in the data on the training effect of subsequent models.
[0053] The enterprise database refers to a centralized system used within the enterprise to store and manage various data generated during the operation of the enterprise. In this application, when the enterprise makes a call with another user's phone number, the relevant data generated during the call will be directly stored in the enterprise database for summary management to form the initial call data corresponding to the phone number; the initial call data here may include data such as the activation time of the phone number, the number of active calls, the number of passive calls and the total call time, which are not limited here.
[0054] It should be noted that the telephone numbers of this application are divided by attributes, which can be divided into car owners and non-car owners. To elaborate, a car owner refers to a user who currently uses the phone number and has the right to use at least one car, while a non-car owner refers to a user who currently uses the phone number and does not have the right to use any car. Based on this, this application can classify the attributes of the phone numbers that have contacted the company, so that the company can conduct precision marketing for the car owner group and avoid telephone troubles for the non-car owner group.
[0055] It can be understood that since the initial call data is composed of multiple fields and the numerical dimensions of each field are different, for example, the number of active calls and the number of passive calls are measured in times, while the phone number activation time and the total call time can be measured in seconds, after obtaining the initial call data, this application needs to standardize the initial call data to unify the numerical dimensions of each field, thereby improving the efficiency and accuracy of subsequent model training.
[0056] S120: Use the principal component analysis method to perform dimensionality reduction analysis on the standard call data to generate intermediate call data, and use the support vector machine to perform linear classification on the intermediate call data to obtain sample call data.
[0057] In this step, after obtaining the standard call data through step S110, the computer device can first use the principal component analysis method to perform dimensionality reduction analysis on the standard call data to generate intermediate call data to simplify the data structure and reduce the data dimension, thereby improving the efficiency and effect of model training. Then the computer device can use a support vector machine to perform linear classification on the intermediate call data to obtain sample call data to improve the accuracy and robustness of data attribute classification, and further improve the efficiency and effect of model training.
[0058] It is understandable that principal component analysis (PCA) refers to a dimensionality reduction and feature extraction technology that can transform the original high-dimensional data into a new low-dimensional space through linear transformation while retaining the main features and information of the data as much as possible. Support vector machine (SVM) refers to a supervised learning model for classification and regression analysis, which can separate samples of different categories as much as possible by finding the best decision boundary. In this application, since the call data of car owners and non-car owners are similar in some fields, it is easy to underfit with low-dimensional curve fitting, and it is easy to overfit with high-dimensional polynomials, that is, linear inseparability. Therefore, before training the model, this application can first perform principal component analysis and support vector machine on the training data of the model, so as to improve the generalization ability and robustness of the model, and then improve the accuracy of model recognition.
[0059] S130: Using the sample call data to train a preset initial number recognition model, and using the trained initial number recognition model as a telephone number recognition model.
[0060] In this step, after obtaining the sample call data through step S120, the computer device can use the sample call data as a training sample, and then use the training sample to train the preset initial number recognition model, and use the trained initial number recognition model as the telephone number recognition model. The recognition effect of the model obtained by training has a high accuracy, so the present application can use this model to improve the accuracy of the company's marketing group positioning.
[0061] It is understandable that the computer device can use the sample call data as training samples, and the attributes annotated in each call data in the sample call data as sample labels, to train the preset initial number recognition model to generate a phone number recognition model. Therefore, when the enterprise makes a call with other users, it can obtain the call data of the user's phone number and input it into the phone number recognition model, thereby obtaining the recognition result corresponding to the phone number output by the phone number recognition model, so that the enterprise can conduct precision marketing based on the recognition result. Among them, the recognition result output by the phone number recognition model can be the owner of the car or the non-owner of the car.
[0062] In the above embodiment, when it is necessary to train a telephone number recognition model for identifying user attributes, the initial call data containing multiple telephone numbers can be first obtained from the enterprise database, and the initial call data can be standardized to obtain standard call data to avoid inaccurate subsequent model recognition due to different dimensions between various fields in the data. Then, the principal component analysis method can be used to perform dimensionality reduction analysis on the standard call data. The intermediate call data generated in this way can reduce the data dimension, thereby improving the training efficiency and effect of the model. Then, the support vector machine can be used to perform linear classification on the intermediate call data to obtain sample call data to improve the accuracy and robustness of data attribute classification. Therefore, the telephone number recognition model obtained by training the preset initial number recognition model using sample call data has high generalization ability and robustness, and its telephone recognition result has high accuracy, thereby improving the accuracy of enterprise marketing group positioning.
[0063] In one embodiment, obtaining the initial call data including multiple phone numbers from the enterprise database in step S110 may include:
[0064] S111: randomly extracting a plurality of vehicle owners' telephone numbers and a plurality of non-vehicle owners' telephone numbers from a corporate database according to a preset ratio.
[0065] S112: Acquire call data of each phone number, and annotate each call data with attributes to form initial call data.
[0066] In this embodiment, when the computer device obtains the initial call data, it can first randomly extract multiple car owners' phone numbers and multiple non-car owners' phone numbers from the enterprise database according to a preset ratio, and then obtain the communication data of each phone number, and annotate the call data with attributes to form initial call data as training data for the phone number recognition model.
[0067] It is understandable that the call data in the enterprise database is extremely unbalanced, where the number of call data of car owners may be hundreds or even thousands of times that of call data of non-car owners. If all the data in the enterprise database is directly used as the training data of the model, the trained model will be tilted towards the car owner during the recognition process, or even all phone numbers will be directly identified as the car owner. Therefore, in order to balance the call data of the two attributes in the training data, the computer device can extract data from the enterprise database according to a preset ratio, and then mix them to obtain balanced training data. In the present application, the computer device can extract data at a ratio of 1:1.
[0068] Specifically, when the computer device obtains the call data of each phone number, it can use the phone number as an index to index the call data corresponding to each phone number from the enterprise database, and then can also attribute the corresponding call data based on each phone number, that is, annotate whether each call data belongs to the car owner or non-car owner, and use this as a learning target for model training. After the attribute annotation is completed, the computer device can mix the call data of each car owner and each non-car owner to form the initial call data.
[0069] In one embodiment, the step S110 of normalizing the initial call data to obtain standard call data may include:
[0070] S113: Determine the standard deviation of each call data in the initial call data; each call data is composed of multiple fields.
[0071] S114: Standardize each field in each call data based on the standard deviation of each call data to obtain standard call data.
[0072] In this embodiment, when the computer device standardizes the initial call data, it can first determine the standard deviation of each call data in the initial data. For each call data, the computer device can standardize each field in the call data based on the standard deviation of the call data, and then obtain standard call data, so that each field can be converted into a unified scale, thereby improving the efficiency and effect of subsequent model training.
[0073] Specifically, the normalization equation used by the computer device to normalize the initial call data may be as follows:
[0074]
[0075] In the formula, Represents any field in standard call data; Indicates the initial call data The corresponding fields; Indicates the standard deviation corresponding to the field; Represents the mean of the initial call data.
[0076] It can be understood that, in the above formula, subtracting the standard deviation corresponding to the field in the initial call data from its call data can eliminate the discrete effect of the call data, and then dividing the difference by the mean of the call data can convert the corresponding data from the original scale to a new scale to unify the scale of each field data and improve the efficiency and accuracy of subsequent data processing.
[0077] In one embodiment, in step S120, the principal component analysis method is used to perform dimensionality reduction analysis on the standard call data to generate the intermediate call data, which may include:
[0078] S121: Determine a covariance matrix of each call data in the standard call data, and determine a characteristic value of each field in each call data based on each covariance matrix.
[0079] S122: For each call data, extract eigenvalues corresponding to a preset number from each eigenvalue in order from large to small, and perform data dimension reduction on the covariance matrix of the call data based on the extracted eigenvalues.
[0080] S123: Generate intermediate call data according to the dimension reduction results of each covariance matrix.
[0081] In this embodiment, when the computer device performs dimensionality reduction analysis on the standard call data, it can first determine the covariance matrix of each call data in the standard call data, and determine the eigenvalues of each field in each call data based on each covariance matrix. For each call data, the computer device can intercept the eigenvalues corresponding to the preset number from each eigenvalue in order from large to small, and perform data dimensionality reduction on the covariance matrix of the call data based on the intercepted eigenvalues to simplify the data structure and reduce the dimension; finally, the computer device can generate intermediate call data with lower dimensions based on the dimensionality reduction results of each covariance matrix to further improve the efficiency and effect of subsequent model training.
[0082] Specifically, after obtaining the eigenvalues of each field in the call data, the computer device can calculate the eigenvector of each eigenvalue, and then sort the eigenvalues from large to small, and select the eigenvalues corresponding to the preset number and their corresponding eigenvectors in turn. The percentage of the selected eigenvalues in all eigenvalues is the percentage of all fields that can be explained by the corresponding eigenvector. Therefore, the computer device can construct a new matrix based on the selected eigenvalues and their corresponding eigenvectors, and reduce the dimension of the covariance matrix based on the matrix to form a data matrix with lower dimension as the intermediate call data.
[0083] For example, in a covariance matrix consisting of 30 fields, let the first field be x1, the second field be x2, ..., and the 30th field be x30. The covariance matrix of these 30 fields is:
[0084]
[0085] Among them, cov(xi,xj) is the covariance between the i-th field and the j-th field.
[0086] Furthermore, the computer device can calculate the eigenvalues of the covariance matrix cov(X) and their corresponding eigenvectors, assuming that the eigenvalues are r1, r2...r30, and the corresponding eigenvectors are a1, a2...a30; then the eigenvectors corresponding to the first k largest eigenvalues can be selected, and the matrix composed of the k eigenvalues and their corresponding eigenvectors is w(30 k), the matrix composed of the covariance matrix in the standard call data is X(454387 30), then the final projection matrix of the intermediate call data is = X(454387 30) w(30 k)= (454387 k).
[0087] Indicatively, if Figure 2 As shown, Figure 2 A comparative schematic diagram of data dimension visualization before and after principal component analysis provided in an embodiment of the present application; Figure 2 The two figures in the figure are obtained by using MATLAB (Matrix Laboratory) to visualize the dimensions of the data matrix before and after principal component analysis. It can be seen that when k is 10, the 30-dimensional field can be reduced to a 10-dimensional field, and its principal component can be retained to 98%.
[0088] In one embodiment, Figure 3 As shown, Figure 3 A schematic diagram of a flow chart of a linear classification process of intermediate call data provided in an embodiment of the present application; Figure 3 In step S120, the support vector machine is used to perform linear classification on the intermediate call data to obtain sample call data, which may include:
[0089] S124: Use a kernel function to perform high-dimensional mapping on the intermediate call data to obtain linearly separable data.
[0090] S125: Optimize the classification interval of linearly separable data by Lagrange multiplier method, and construct a classification decision function for the intermediate call data according to the optimization result.
[0091] S126: Use the classification decision function to classify the intermediate call data to obtain sample call data.
[0092] In this embodiment, when the computer device performs linear classification on the intermediate call data, it can first use the kernel function to perform high-dimensional mapping on the intermediate call data to obtain linearly separable data, and then optimize the classification interval of the linearly separable data through the Lagrange multiplier method, and construct a classification decision function for the intermediate call data based on the optimization result. Finally, the computer device can use the classification decision function to classify the intermediate call data to obtain sample call data, so as to improve the accuracy and robustness of data attribute classification, thereby improving the efficiency and effectiveness of subsequent model training.
[0093] It can be understood that the kernel function is a key component of the support vector machine, which is used to map the data in the low-dimensional space to the high-dimensional space to make it linearly separable. Commonly used kernel functions include linear kernel, radial basis kernel (RBF), polynomial kernel, etc. The Lagrange multiplier rule is a mathematical optimization method used to solve the extreme value problem of the function under given constraints. The solution process of the support vector machine is actually an optimization problem with constraints. Therefore, by introducing the Lagrange multiplier, this application can transform the original constrained optimization problem into an unconstrained Lagrangian function extreme value problem, and then find the parameters of the optimal classification hyperplane for the intermediate call data by solving the extreme value of the Lagrangian function.
[0094] Specifically, the computer device can use the Lagrange multiplier method to optimize the classification interval of linearly separable data to obtain the optimal normal vector and bias term of the classification hyperplane, and then construct a classification decision function based on the normal vector and bias term, which can be specifically expressed as follows:
[0095]
[0096] In the formula, represents the normal vector, Represents the bias term. For the input call data , can be The sign of determines its category. , then it is classified as the owner; if , then it is classified as a non-car owner.
[0097] In one embodiment, in step S130, the preset initial number recognition model is trained using the sample call data, and the trained initial number recognition model is used as the telephone number recognition model, which may include:
[0098] S131: Divide the sample call data into a sample training set and a sample verification set.
[0099] S132: Iteratively train the preset initial number recognition model using the sample training set to generate an intermediate number recognition model.
[0100] S133: Use the sample verification set to perform prediction verification on the intermediate number recognition model, and use the verified intermediate number recognition model as the telephone number recognition model.
[0101] In this embodiment, before performing model training, the computer device can first divide the sample call data into a sample training set and a sample verification set, and then use the sample training set to iteratively train the preset initial number recognition model to generate an intermediate number recognition model. After that, the computer device can also use the sample verification set to predict and verify the intermediate number recognition model, and use the verified intermediate number recognition model as the final telephone number recognition model, thereby ensuring the recognition accuracy of the model.
[0102] It is understandable that through the sample training set, the model can learn the field features in the call data and continuously adjust its internal parameters to minimize the prediction error on the sample training set until the prediction error is less than the target value, and the training is considered complete. The sample validation set can be used to evaluate the performance of the trained model, and it does not participate in the feature learning of the model training process; if the model is verified, the computer device can use the intermediate number recognition model obtained through training as the final phone number recognition model; if the model is not verified, the computer device can continue to use the sample training set to iterate the model.
[0103] It should be noted that the present application can divide the sample call data into a sample training set and a sample verification set in a ratio of 3:1.
[0104] In one embodiment, each call data in the sample training set in step S132 is annotated with a corresponding real attribute result; wherein the step of iteratively training the preset initial number recognition model using the sample training set to generate the intermediate number recognition model may include:
[0105] S1321: Input the sample training set into a preset initial number recognition model to obtain the predicted attribute result output by the initial number recognition model.
[0106] S1322: training the initial number recognition model with the goal of making the predicted attribute result close to the actual attribute result of the sample training set;
[0107] S1323: When the initial number recognition model meets the preset training end condition, the trained initial number recognition model is used as the telephone number recognition model.
[0108] In this embodiment, when the computer device iteratively trains the preset initial number recognition model, it can first input the sample training set into the preset initial number recognition model to obtain the predicted attribute result output by the initial number recognition model, and then the computer device can train the initial number recognition model with the goal of making the predicted attribute result approach the real attribute result of the sample training set. When the initial number recognition model meets the preset training end condition, the training can be regarded as completed, and the computer device can use the trained initial number recognition model as the telephone number recognition model.
[0109] Specifically, during model training, the computer device may first select a corresponding semantic classification model as a preset initial number recognition model for improvement and training. During the training process, the computer device may first obtain a sample training set, which is annotated with a corresponding sample label, and the sample label is the real attribute result of each call data in the sample training set, such as owner or non-owner. In this way, after the computer device inputs the sample training set into the preset initial number recognition model, it can obtain the predicted attribute result output by the initial number recognition model. Then, the computer device can train the initial number recognition model with the goal of making the predicted attribute result approach the real attribute result of the sample training set. When the initial number recognition model meets certain training conditions or parameter convergence conditions, such as the number of iterations reaching the set value, the model training can be considered to be completed, and a telephone number recognition model can be obtained.
[0110] The following is a description of a telephone number recognition model training device provided in an embodiment of the present application. The telephone number recognition model training device described below and the telephone number recognition model training method described above can be referenced to each other.
[0111] In one embodiment, Figure 4 As shown, Figure 4 A schematic diagram of the structure of a telephone number recognition model training device provided in an embodiment of the present application; the present application also provides a telephone number recognition model training device, including a data acquisition module 210, a data processing module 220 and a model training module 230, specifically including the following:
[0112] The data acquisition module 210 is used to acquire initial call data including multiple phone numbers from the enterprise database, and standardize the initial call data to obtain standard call data; wherein the attributes of the phone numbers include car owners and non-car owners.
[0113] The data processing module 220 is used to perform dimensionality reduction analysis on the standard call data using the principal component analysis method to generate intermediate call data, and to perform linear classification on the intermediate call data using a support vector machine to obtain sample call data.
[0114] The model training module 230 is used to train a preset initial number recognition model using sample call data, and use the trained initial number recognition model as a telephone number recognition model.
[0115] In the above embodiment, when it is necessary to train a telephone number recognition model for identifying telephone number attributes, initial call data containing multiple telephone numbers can be obtained from the enterprise database, and the initial call data can be standardized to obtain standard call data to avoid inaccurate subsequent model recognition due to different dimensions between various fields in the data. Then, the principal component analysis method can be used to perform dimensionality reduction analysis on the standard call data. The intermediate call data generated in this way can reduce the data dimension, thereby improving the training efficiency and effect of the model. Then, the support vector machine can be used to perform linear classification on the intermediate call data to obtain sample call data to improve the accuracy and robustness of data attribute classification. Therefore, the telephone number recognition model obtained by training the preset initial number recognition model using sample call data has high generalization ability and robustness, and its telephone recognition result has high accuracy, thereby improving the accuracy of the marketing group positioning of the enterprise.
[0116] In one embodiment, the data acquisition module 210 may include:
[0117] A data extraction submodule, used to randomly extract a plurality of vehicle owners' telephone numbers and a plurality of non-vehicle owners' telephone numbers from an enterprise database according to a preset ratio;
[0118] The attribute labeling submodule is used to obtain the call data of each phone number and label the attributes of each call data to form initial call data.
[0119] In one embodiment, the data acquisition module 210 may further include:
[0120] The standard deviation determination submodule is used to determine the standard deviation of each call data in the initial call data; each call data is composed of multiple fields.
[0121] The data standardization submodule is used to standardize each field in each call data based on the standard deviation of each call data to obtain standard call data.
[0122] In one embodiment, the data processing module 220 may include:
[0123] The matrix determination submodule is used to determine the covariance matrix of each call data in the standard call data, and determine the characteristic value of each field in each call data based on each covariance matrix.
[0124] The data dimension reduction submodule is used to intercept the eigenvalues corresponding to the preset number from each eigenvalue in order from large to small for each call data, and perform data dimension reduction on the covariance matrix of the call data based on the intercepted eigenvalues.
[0125] The data generation submodule is used to generate intermediate call data according to the dimension reduction results of each covariance matrix.
[0126] In one embodiment, the data processing module 220 may further include:
[0127] The data mapping submodule is used to perform high-dimensional mapping on the intermediate call data using a kernel function to obtain linearly separable data.
[0128] The function construction submodule is used to optimize the classification interval of linearly separable data through the Lagrange multiplier method, and to construct the classification decision function of the intermediate call data based on the optimization results.
[0129] The data classification submodule is used to classify the intermediate call data using the classification decision function to obtain sample call data.
[0130] In one embodiment, the model training module 230 may include:
[0131] The data partitioning submodule is used to divide the sample call data into a sample training set and a sample verification set.
[0132] The model training submodule is used to iteratively train the preset initial number recognition model using the sample training set to generate an intermediate number recognition model.
[0133] The model verification submodule is used to use a sample verification set to predict and verify the intermediate number recognition model, and use the verified intermediate number recognition model as the telephone number recognition model.
[0134] In one embodiment, each call data in the sample training set in step S132 is annotated with a corresponding real attribute result; wherein the step of iteratively training the preset initial number recognition model using the sample training set to generate the intermediate number recognition model may include:
[0135] The model prediction unit is used to input the sample training set into the preset initial number recognition model to obtain the predicted attribute results output by the initial number recognition model.
[0136] A model optimization unit is used to train the initial number recognition model with the goal of predicting attribute results close to the actual attribute results of the sample training set;
[0137] The model generation unit is used to use the trained initial number recognition model as a telephone number recognition model when the initial number recognition model meets a preset training end condition.
[0138] In one embodiment, the present application further provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the telephone number recognition model training method as described in any of the above embodiments.
[0139] In one embodiment, the present application further provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the telephone number recognition model training method as described in any one of the above embodiments.
[0140] Indicatively, if Figure 5 As shown, Figure 5 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 may be provided as a server. Figure 5 The computer device 300 includes a processing component 302, which further includes one or more processors, and a memory resource represented by a memory 301 for storing instructions executable by the processing component 302, such as an application. The application stored in the memory 301 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 302 is configured to execute instructions to perform the telephone number recognition model training method of any of the above embodiments.
[0141] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.
[0142] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0143] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0144] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.
[0145] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A telephone number recognition model training method, characterized in that: The method comprises: Acquire initial call data containing multiple phone numbers from an enterprise database, and standardize the initial call data to obtain standard call data; wherein the attributes of the phone numbers include vehicle owners and non-vehicle owners; Performing dimensionality reduction analysis on the standard call data using a principal component analysis method to generate intermediate call data, and performing linear classification on the intermediate call data using a support vector machine to obtain sample call data; The sample call data is used to train a preset initial number recognition model, and the trained initial number recognition model is used as a telephone number recognition model.
2. The telephone number recognition model training method according to claim 1, characterized in that: The obtaining of initial call data containing multiple phone numbers from the enterprise database includes: A plurality of vehicle owners' telephone numbers and a plurality of non-vehicle owners' telephone numbers are randomly selected from the enterprise database according to a preset ratio; The call data of each phone number is obtained, and each call data is attributed to form initial call data.
3. The telephone number recognition model training method according to claim 1, characterized in that: The step of normalizing the initial call data to obtain standard call data includes: Determine the standard deviation of each call data in the initial call data; each call data is composed of multiple fields; Each field in each call data is standardized based on the standard deviation of each call data to obtain standard call data.
4. The telephone number recognition model training method according to claim 1, characterized in that: The step of performing dimensionality reduction analysis on the standard call data using the principal component analysis method to generate intermediate call data includes: Determine a covariance matrix of each call data in the standard call data, and determine a characteristic value of each field in each call data based on each covariance matrix; For each call data, eigenvalues corresponding to a preset number are intercepted from each eigenvalue in order from large to small, and data dimension reduction is performed on the covariance matrix of the call data based on the intercepted eigenvalues; The intermediate call data is generated according to the dimension reduction results of each covariance matrix.
5. The telephone number recognition model training method according to claim 1, characterized in that: The use of a support vector machine to perform linear classification on the intermediate call data to obtain sample call data includes: Using a kernel function to perform high-dimensional mapping on the intermediate call data to obtain linearly separable data; Optimizing the classification interval of the linearly separable data by Lagrange multiplier method, and constructing a classification decision function of the intermediate call data according to the optimization result; The intermediate call data is classified using the classification decision function to obtain sample call data.
6. The telephone number recognition model training method according to claim 1, characterized in that: The method of training a preset initial number recognition model by using the sample call data and using the trained initial number recognition model as a telephone number recognition model includes: Dividing the sample call data into a sample training set and a sample verification set; Iteratively training a preset initial number recognition model using the sample training set to generate an intermediate number recognition model; The sample verification set is used to perform prediction verification on the intermediate number recognition model, and the verified intermediate number recognition model is used as the telephone number recognition model.
7. The telephone number recognition model training method according to claim 6, characterized in that: Each call data in the sample training set is annotated with corresponding real attribute results; The iterative training of the preset initial number recognition model using the sample training set to generate an intermediate number recognition model includes: Inputting the sample training set into a preset initial number recognition model to obtain a predicted attribute result output by the initial number recognition model; The initial number recognition model is trained with the goal of making the predicted attribute result close to the real attribute result of the sample training set; When the initial number recognition model meets the preset training end condition, the trained initial number recognition model is used as the telephone number recognition model.
8. A telephone number recognition model training device, characterized in that: include: A data acquisition module, used to acquire initial call data containing multiple phone numbers from an enterprise database, and standardize the initial call data to obtain standard call data; wherein the attributes of the phone numbers include car owners and non-car owners; A data processing module, configured to perform dimensionality reduction analysis on the standard call data using a principal component analysis method to generate intermediate call data, and perform linear classification on the intermediate call data using a support vector machine to obtain sample call data; The model training module is used to train a preset initial number recognition model using the sample call data, and use the trained initial number recognition model as a telephone number recognition model.
9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the telephone number identification method according to any one of claims 1 to 7.
10. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the telephone number recognition model training method as described in any one of claims 1 to 7 are performed.