Number recognition model training method and device, number recognition method and device, storage medium and product
Through the number recognition model training method based on transfer learning, the model training is performed using CDR call list and supplementary features, and the problem of low accuracy in different regions is solved, and higher accuracy and generalization capabilities are achieved, abnormal numbers are intercepted in time and risks are reduced.
Patent Information
- Application Number
- CN202411483222.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-23
- Publication Date
- 2025-02-28
AI Technical Summary
The existing number identification model has poor performance in applications in different regions, resulting in low accuracy and the inability to intercept abnormal numbers in time, which brings risks.
The number recognition model training method based on transfer learning is adopted, and the first sample number set and the second sample number set are constructed, and the model is trained using CDR talk sheet and supplementary features to realize the transfer learning and secondary training of the model.
The accuracy and generalization ability of the number recognition model in specific scenarios is improved, and abnormal numbers can be intercepted in time and reduce risks to users.
Smart Images
Figure CN119538141A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of information security technology, and in particular to a number recognition model training method based on transfer learning, a number recognition method, an electronic device, a readable storage medium and a computer program product. Background Art
[0002] In the current work of identifying abnormal telecommunications numbers, data centers and computing resources in various places are mainly concentrated in higher-level departments in the region, and are uniformly built by higher-level departments. This centralized construction method helps to make full use of the data resources and computing power advantages of the entire region, thereby improving the overall performance of the number recognition model, including accuracy, recall rate, real-time performance, etc. However, most of the existing technical solutions start from the perspective of unified construction, and solve the identification of all abnormal or normal numbers by building a number recognition model. By improving the model architecture, optimizing rules, etc., the performance of the number recognition model in terms of accuracy and real-time performance in identifying abnormal numbers can be improved.
[0003] However, in actual applications, due to the differences in the abnormal situation and abnormal means of numbers in various regions, there are obvious differences in the performance of the unified number recognition model in different regions. It is unable to solve the problem of low accuracy of the number recognition model when it is actually applied in different regions, resulting in abnormal numbers being unable to be intercepted in time, bringing risks or unsafe factors to the called users. Summary of the invention
[0004] The purpose of the embodiments of the present application is to provide a number recognition model training, recognition method, device, storage medium and product to solve the problem of low accuracy of existing number recognition.
[0005] In order to solve the above technical problems, this specification is implemented as follows: In a first aspect, a method for training a number recognition model based on transfer learning is provided, comprising: Constructing a first sample number set based on the CDR call record of the user in the first scenario, wherein the first sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and features of samples in the first sample number set include a first number of call features corresponding to the CDR call record; Training a preset number recognition model based on positive samples and negative samples of the first sample number set to obtain a trained first number recognition model; Constructing a second sample number set based on the CDR bill of the user in the second scenario and the supplementary features of the second scenario, wherein the second sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the second sample number set include a first number of call features corresponding to the CDR bill and a second number of the supplementary features; The first number recognition model is trained twice based on the positive samples and negative samples of the second sample number set to obtain a trained second number recognition model as the number recognition model in the second scenario.
[0006] Optionally, the constructing a first sample number set based on a CDR of a user in the first scenario includes: Obtain user information and CDR of the target user in the first scenario, wherein the user information includes a number and a number identifier used to indicate whether the number is abnormal or normal, and the CDR includes a call date, a calling number, a called number, a call start time, a call duration, and a roaming location; Constructing a first number of call features corresponding to the CDR of the target user based on feature coverage, feature content and feature value; Taking features including a first number of call features corresponding to the target user as samples and the number identifier of the target user as a label, a sample number corresponding to the number of the target user in the first sample number set is constructed.
[0007] Optionally, the constructing a second sample number set based on the CDR of the user in the second scenario and the supplementary features of the second scenario includes: Obtain user information and CDR of the target user in the second scenario, wherein the user information includes a number and a number identifier used to indicate whether the number is abnormal or normal, and the CDR includes a call date, a calling number, a called number, a call start time, a call duration, and a roaming location; Constructing a first number of call features corresponding to the CDR of the target user based on feature coverage, feature content and feature value; Acquire a second number of supplementary features corresponding to the target user in the second scenario; Using features including a first number of call features and a second number of supplementary features corresponding to the target user as samples and the number identifier of the target user as a label, construct a sample number corresponding to the number of the target user in the second sample number set.
[0008] Optionally, the preset number recognition model includes a deep neural network based on an Inception module, and the deep neural network includes an input layer, a hidden layer and an output layer. The input layer is used to input target samples; The hidden layer includes an Inception module and a fully connected layer connected in sequence, the Inception module is connected to the output of the input layer, the fully connected layer is connected to the output layer, and the hidden layer is used to extract features from the target sample; The output layer is used to output whether the target sample belongs to an abnormal number or a normal number.
[0009] Optionally, the target sample is a sample in the first sample number set, and the input layer corresponds to the preset number recognition model of the first number recognition model and is used to input features including a first number of call features of the target sample; The target sample is a sample in the second sample number set, and the input layer corresponds to the preset number recognition model of the second number recognition model and is used to input features including a first number of call features and a second number of supplementary features of the target sample.
[0010] Optionally, the supplementary features include: a combination of one or more of: user occupation, account opening status, package usage, account opening channel and place of residence.
[0011] In a second aspect, a number identification method is provided, comprising: Obtain a CDR of a target user in a target scenario and a second number of supplementary features of the target scenario; Constructing a first number of call features corresponding to the CDR; Input the features including the first number of call features and the second number of supplementary features into a pre-trained second number recognition model to identify whether the number of the target user in the target scenario is an abnormal number or a normal number, wherein the second number recognition model is trained based on the steps of the method described in the first aspect.
[0012] According to a third aspect, an electronic device is provided, comprising a processor and a memory, wherein the memory stores programs or instructions executable on the processor, and wherein the programs or instructions, when executed by the processor, implement the steps of the method described in the first aspect or the second aspect.
[0013] In a fourth aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect or the second aspect are implemented.
[0014] In a fifth aspect, a computer program product is provided, the computer program product comprising a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute the steps of the method described in the first aspect or the second aspect.
[0015] In an embodiment of the present application, a first sample number set is constructed based on the CDR call records of users in a first scenario, wherein the first sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the first sample number set include a first number of call features corresponding to the CDR call records; a preset number recognition model is trained based on the positive samples and negative samples of the first sample number set to obtain a trained first number recognition model; a second sample number set is constructed based on the CDR call records of users in a second scenario and supplementary features of the second scenario, wherein the second sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the second sample number set include a first number of call features and a second number corresponding to the CDR call records. the supplementary features; based on the positive samples and negative samples of the second sample number set, the first number recognition model is trained for the second number recognition model to obtain the trained second number recognition model and use it as the number recognition model for the second scenario. Based on the features extracted from the CDR call sheet with wide coverage, complete content and high integrity, the data basis of the number recognition model can be enriched, and for transfer learning of specific scenarios, only a small batch of new samples and supplementary features are used for secondary training to obtain a number recognition model that meets the needs of specific scenarios. Compared with the number recognition model trained from a unified perspective, the number recognition model for specific scenarios can improve the generalization ability and improve the accuracy of number recognition for specific scenarios in actual applications, so that abnormal numbers can be intercepted in time to avoid the risks brought to users by abnormal number communications. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 It is a flow chart of the number recognition model training method based on transfer learning in an embodiment of the present application.
[0017] Figure 2 It is a schematic diagram of the structure of the Inception module of an embodiment of the present application.
[0018] Figure 3 It is a flowchart of the number identification method of an embodiment of the present application.
[0019] Figure 4 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0020] 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 part of the embodiments of the present application, rather than 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. The numbering of the drawings in this application is only used to distinguish the various steps in the scheme, and is not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.
[0021] In order to solve the problems existing in the prior art, the present application embodiment provides a number recognition model training method based on transfer learning, such as Figure 1 As shown, the following steps 102 to 108 are included.
[0022] Step 102, constructing a first sample number set based on the CDR call record of the user in the first scenario, wherein the first sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the first sample number set include a first number of call features corresponding to the CDR call record.
[0023] The first scenario is based on the perspective of unified construction, including an overall scenario of multiple different scenarios, which is used to build a unified number recognition model. For the first scenario, the user's CDR, commonly known as the bill, is obtained, which includes the amount of services used by the user. A first number of call features are extracted based on the CDR to serve as features for constructing samples in the first sample number set.
[0024] Optionally, the constructing of a first sample number set based on the CDR call record of the user in the first scenario includes: obtaining user information and a CDR call record of the target user in the first scenario, the user information including a number and a number identifier used to indicate that the number is abnormal or normal, the CDR call record including a call date, calling number, called number, call start time, call duration and roaming location; constructing a first number of call features corresponding to the CDR call record of the target user based on feature coverage, feature content and feature value; constructing a sample number corresponding to the number of the target user in the first sample number set using the features including the first number of call features corresponding to the target user as a sample and the number identifier of the target user as a label.
[0025] User information and CDRs belong to the basic data of user calls. User information mainly includes three fields: user number, account opening time, and number identifier indicating whether the number is abnormal or normal. The number identifier in user information is used to construct positive and negative sample labels in number recognition model training. CDRs mainly include call date, calling and called numbers in call records, call start time, call duration, roaming location and other information. CDRs are used to construct features of positive and negative samples in number recognition model training. Tables 1 and 2 below give examples of corresponding data in user information and CDRs, respectively.
[0026] Table 1 User information
[0027] Table 2 CDR
[0028] Based on the CDR records of each user obtained in the first scenario, a feature set corresponding to each user is constructed.
[0029] In the embodiment of the present application, a feature set is first constructed from the perspective of unified construction, and is constructed according to the principle of making the feature coverage as wide as possible, the feature content as complete as possible, and the feature value as complete as possible, so as to ensure the accuracy of the subsequently unified number recognition model.
[0030] Specifically, in terms of feature coverage, the system analyzes 14 aspects, including the overall behavior of source calls, active users, short and long calls, etc., covering all aspects of the user's call process; in terms of feature content, it constructs 69 features, including the total number of contacts and the total number of calls, to ensure the number of valid features; in terms of feature value integrity, it is built entirely based on CDR records, which can guarantee 100% non-empty values, avoiding the impact of too many empty values on model training.
[0031] The following table 3 provides examples of 14 perspectives based on CDR call record analysis and 69 features constructed accordingly. Here, the 69 features are examples of the first number of call features corresponding to the user's CDR call record.
[0032] Table 3 Example of call features corresponding to CDR records
[0033] After obtaining the first number of call features corresponding to the CDR of each user in the first scenario, the sample number corresponding to each user's number is constructed using the first number of call features corresponding to each user as a sample and the number identifier of each user as a label, thereby obtaining a first sample number set based on multiple sample numbers.
[0034] Step 104: training a preset number recognition model based on the positive samples and negative samples of the first sample number set to obtain a trained first number recognition model.
[0035] The first number recognition model trained in this step is a unified number recognition model constructed from the perspective of unified construction. The training samples include positive samples and negative samples of the first sample number set, the labels of the positive samples correspond to the known abnormal numbers in the first scenario, and the labels of the negative samples correspond to the known normal numbers in the first scenario.
[0036] Optionally, the preset number recognition model includes a deep neural network based on the Inception module, and the deep neural network includes an input layer, a hidden layer and an output layer. The input layer is used to input the target sample; the hidden layer includes an Inception module and a fully connected layer connected in sequence, the Inception module is connected to the output of the input layer, the fully connected layer is connected to the output layer, the hidden layer is used to extract features from the target sample; the output layer is used to output whether the target sample belongs to an abnormal number or a normal number.
[0037] In the embodiment of the present application, a deep neural network (DNN) is selected to construct a number recognition model. Compared with traditional machine learning algorithms, deep neural networks have higher model complexity and are more suitable for the massive data scale of number recognition scenarios. The trained model has higher accuracy. In the network design part, the embodiment of the present application mainly uses the Inception module as the hidden layer of the recognition model. The Inception module can avoid the problem of reduced accuracy caused by the increase in the number of model layers. In the application scenario of number recognition, the Inception module can capture the features of different layers, thereby improving the model's ability to understand the features of abnormal numbers. The Inception module also reduces the computational complexity by reducing the number of channels. Especially for complex data sets containing high-dimensional features, the advantages of the Inception module are particularly obvious.
[0038] Figure 2 This is a schematic diagram of the structure of the Inception module of the embodiment of the present application. Figure 2 The single Inception module shown contains convolutional layers of different sizes (1x1, 3x3, 5x5) and maximum pooling layers.
[0039] The input of the Inception module of the hidden layer receives the data output by the input layer of the recognition model. The channel merging layer of the Inception module is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the output layer.
[0040] The hidden layer can be understood as a data processing factory module inside the network, which enables the neural network to learn the multi-level and abstract features of the input sample data from the input sample data. The Inception module of the hidden layer can capture features at different levels and improve the breadth of the model's feature extraction. The hidden layer can also include an attention mechanism (SE) block to increase the depth of the model's feature extraction. The hidden layer can also include a hole convolution layer, which captures more feature information by inserting zeros between the elements of the convolution kernel and improves the model's self-learning ability for features. The fully connected layer of the hidden layer plays the role of integrating and further processing the features processed by the previous layers, and is a one-dimensional vector. The output of the fully connected layer provides appropriate input for the output layer, and finally outputs the result of whether the user's number is a fraudulent number.
[0041] When the target sample is a sample in the first sample number set, the input layer corresponds to the preset number recognition model of the first number recognition model and is used to input features including a first number of call features of the target sample.
[0042] In this way, the feature set of the target sample of the first sample number set, for example, each user has a 69-dimensional feature vector, is input into the input layer of the recognition model, and then passes through the hidden layer and the output layer in sequence. The hidden layer includes two Inception modules and one fully connected layer, and the output layer outputs the predicted value of whether the target sample belongs to an abnormal number or a normal number. The loss between the predicted value and the label of the target sample is calculated, and the parameters of the preset number recognition model are adjusted based on the loss until the predicted value of the number recognition model and the loss of the corresponding label converge, so as to train and obtain the first number recognition model.
[0043] In the training phase, for example, a total of 67 million user sample data were used for training, where each user contained 69 dimensions of features as model input, and the entire data set contained 350,000 positive samples (abnormal numbers). Cross entropy was used as the loss function, and the Adaptive Moment Estimation (Adam) algorithm was used to optimize the training process. After testing, the accuracy of the first number recognition model was 96.9%, and the precision rate was 70%, which met the requirements for number recognition.
[0044] Step 106, construct a second sample number set based on the CDR call record of the user in the second scenario and the supplementary features of the second scenario, the second sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the second sample number set include a first number of call features corresponding to the CDR call record and a second number of the supplementary features.
[0045] The second scenario is based on a local perspective and includes a single specific scenario, which is used to construct a number recognition model for a specific scenario. For the second scenario, the user's CDR call record is obtained, a first number of call features are extracted based on the CDR call record, and supplementary features of the second scenario are obtained based on expert experience to serve as features for constructing samples in the second sample number set.
[0046] Optionally, the constructing of a second sample number set based on the CDR call record of the user in the second scenario and the supplementary features of the second scenario includes: obtaining user information and CDR call records of the target user in the second scenario, the user information including a number and a number identifier used to indicate that the number is abnormal or normal, the CDR call record including a call date, calling number, called number, call start time, call duration and roaming location; constructing a first number of call features corresponding to the CDR call record of the target user based on feature coverage, feature content and feature value; obtaining a second number of supplementary features corresponding to the target user in the second scenario; constructing a sample number corresponding to the number of the target user in the second sample number set using features including the first number of call features and the second number of supplementary features corresponding to the target user as samples and the number identifier of the target user as a label.
[0047] As mentioned above, user information and CDR bills belong to the user's basic call data, where user information mainly includes three fields: user number, account opening time, and number identifier indicating whether the number is abnormal or normal. The number identifier in the user information is used to construct positive and negative sample labels in the number recognition model training. CDR bills mainly include call date, calling and called numbers in the call record, call start time, call duration, roaming location and other information. CDR bills are used to construct positive and negative sample features in the number recognition model training. For examples of user information and corresponding data in the second scenario, please refer to Tables 1 and 2 above.
[0048] Taking the telephone anomaly scenario in the target area as an example, the CDR bills of the corresponding users are extracted from the abnormal numbers provided by the relevant departments. Assuming that there are 8 abnormal numbers in total, the 69-dimensional feature set corresponding to the abnormal numbers in the second scenario can be extracted in the same way as the feature set corresponding to the user is constructed based on the CDR bills of the users in the first scenario, as shown in Table 4 below: Table 4 Example of feature set of abnormal numbers in the target area
[0049] If the feature set extracted in the second scenario of Table 4 is input into the first number recognition model obtained in step 104, the corresponding test results are shown in Table 5 below: Table 5
[0050] Through testing, it was found that the generalization ability of the first number recognition model in the abnormal telephone scenarios in the target area was low, and only the user number 101 was recognized as an abnormal number. Therefore, the embodiment of the present application proposes to improve the generalization ability of the first number recognition model in the abnormal telephone scenarios in the target area through transfer learning.
[0051] Therefore, in the embodiment of the present application, feature supplementation is first performed for the second scenario, and the supplementary features are generally provided by business experts in related fields for the specific second scenario. Since the number of abnormal numbers in a specific scenario is often small, the number of supplementary features is small during the transfer learning process.
[0052] Optionally, the supplementary features include: a combination of one or more of: user occupation, account opening status, package usage, account opening channel and place of residence.
[0053] Table 6 gives examples of supplementary features of abnormal phone calls in the target area corresponding to Tables 4 and 5.
[0054] Table 6
[0055] In the above embodiment, 3 features are added for the second scenario based on the 69 features. When training the second number recognition model applicable to the second scenario, the entire transfer learning includes 8 samples, each of which has 72 features.
[0056] Step 108: Perform secondary training on the first number recognition model based on the positive samples and negative samples of the second sample number set to obtain a trained second number recognition model as the number recognition model in the second scenario.
[0057] The second number recognition model obtained by training in this step is a number recognition model for a specific scenario constructed from a local perspective.
[0058] In this step, the initial number recognition model is a first number recognition model trained based on a first number of call features of a first scenario. The training samples include positive samples and negative samples of a second sample number set. The labels of the positive samples correspond to known abnormal numbers in the second scenario, and the labels of the negative samples correspond to known normal numbers in the second scenario.
[0059] Therefore, the first number recognition model includes a deep neural network based on the Inception module, and the deep neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to input features corresponding to the target sample; the hidden layer includes an Inception module and a fully connected layer connected in sequence, the Inception module is connected to the output of the input layer, and the fully connected layer is connected to the output layer; the output layer is used to output whether the target sample belongs to an abnormal number or a normal number.
[0060] When the target sample is a sample in the second sample number set, the input layer corresponds to the preset number recognition model of the second number recognition model and is used to input features including a first number of call features and a second number of supplementary features of the target sample.
[0061] That is, in the transfer learning stage, the model parameters of the first number recognition model trained in step 106 are directly used, except that the input layer of the model is expanded from a 69-dimensional feature vector to a 72-dimensional one, and the parameters of the newly added 3-dimensional input vectors are randomly initialized.
[0062] In this way, the 72-dimensional feature vector of the target sample of the second sample number set is input into the input layer of the trained first number recognition model, and then passes through the hidden layer and the output layer in sequence, and the output layer outputs the predicted value of whether the target sample belongs to an abnormal number or a normal number. The loss between the predicted value and the label of the target sample is calculated, and the parameters of the preset number recognition model are adjusted based on the loss until the predicted value of the number recognition model and the loss of the corresponding label converge, so as to train and obtain the second number recognition model.
[0063] For example, the first 5 of the 8 samples in Table 6 are used as positive samples for transfer learning, and the obtained second number recognition model is the new model after transfer learning for the telephone abnormality scenarios in the target areas of Tables 4 and 5. By testing the effect of the new model on the remaining samples in Table 6, the effect test shown in the following Table 7 is obtained.
[0064] Table 7
[0065] It can be found that the performance of the second number recognition model obtained based on transfer learning in the second scenario is significantly better than the first number recognition model.
[0066] In an embodiment of the present application, a first sample number set is constructed based on the CDR call records of users in a first scenario, wherein the first sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the first sample number set include a first number of call features corresponding to the CDR call records; a preset number recognition model is trained based on the positive samples and negative samples of the first sample number set to obtain a trained first number recognition model; a second sample number set is constructed based on the CDR call records of users in a second scenario and supplementary features of the second scenario, wherein the second sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the second sample number set include a first number of call features and a second number corresponding to the CDR call records. the supplementary features; based on the positive samples and negative samples of the second sample number set, the first number recognition model is trained for the second number recognition model to obtain the trained second number recognition model and use it as the number recognition model for the second scenario. Based on the features extracted from the CDR call sheet with wide coverage, complete content and high integrity, the data basis of the number recognition model can be enriched, and for transfer learning of specific scenarios, only a small batch of new samples and supplementary features are used for secondary training to obtain a number recognition model that meets the needs of specific scenarios. Compared with the number recognition model trained from a unified perspective, the number recognition model for specific scenarios can improve the generalization ability and improve the accuracy of number recognition for specific scenarios in actual applications, so that abnormal numbers can be intercepted in time to avoid the risks brought to users by abnormal number communications.
[0067] Alternatively, if Figure 3 As shown, the embodiment of the present application also provides a number identification method, comprising the following steps: Step 202, obtaining a CDR of a target user in a target scenario and a second number of supplementary features of the target scenario; Step 204, constructing a first number of call features corresponding to the CDR; Step 206: Input the features including the first number of call features and the second number of supplementary features into a pre-trained second number recognition model to recognize whether the number of the target user in the target scenario is an abnormal number or a normal number, wherein the second number recognition model is based on Figure 1 to Figure 2 The number recognition model training method based on transfer learning is obtained by training according to the steps of the method described in the embodiment.
[0068] In this embodiment, based on the second number recognition model of the target scenario obtained by transfer learning, the user of the target scenario can be analyzed by reasoning, and potential abnormal numbers in abnormal phone scenarios can be found to improve the accuracy of number recognition. The second number recognition model obtained by training different scenarios is used for number recognition in corresponding scenarios.
[0069] The number recognition model training method based on transfer learning can be referred to in the embodiment Figure 1 to Figure 2 , in order to avoid repetition, it will not be repeated here.
[0070] Alternatively, if Figure 4 As shown, an embodiment of the present application also provides an electronic device 2000, including a processor 2400 and a memory 2200, wherein the memory 2200 stores programs or instructions that can be executed on the processor 2400, and when the program or instructions are executed by the processor 2400, the various steps of the above-mentioned number recognition model training method based on transfer learning or the number recognition method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.
[0071] The embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by the processor, each process of any of the above-mentioned number recognition model training methods based on transfer learning or number recognition method embodiments is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the readable storage medium includes a computer-readable storage medium, such as a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.
[0072] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to enable a computer to implement the various processes of any one of the above-mentioned vehicle control method embodiments when executed, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0073] It should be noted that, in this article, the terms "include", "comprises" or any other variations 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, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0074] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0075] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A number recognition model training method based on transfer learning, characterized in that: include: Constructing a first sample number set based on the CDR call record of the user in the first scenario, wherein the first sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and features of samples in the first sample number set include a first number of call features corresponding to the CDR call record; Training a preset number recognition model based on positive samples and negative samples of the first sample number set to obtain a trained first number recognition model; Constructing a second sample number set based on the CDR bill of the user in the second scenario and the supplementary features of the second scenario, wherein the second sample number set includes positive samples belonging to abnormal numbers and negative samples belonging to normal numbers, and the features of the samples in the second sample number set include a first number of call features corresponding to the CDR bill and a second number of the supplementary features; The first number recognition model is trained twice based on the positive samples and negative samples of the second sample number set to obtain a trained second number recognition model as the number recognition model in the second scenario.
2. The method according to claim 1, characterized in that The step of constructing a first sample number set based on a CDR of a user in the first scenario includes: Obtain user information and CDR of the target user in the first scenario, wherein the user information includes a number and a number identifier used to indicate whether the number is abnormal or normal, and the CDR includes a call date, a calling number, a called number, a call start time, a call duration, and a roaming location; Constructing a first number of call features corresponding to the CDR of the target user based on feature coverage, feature content and feature value; Taking features including a first number of call features corresponding to the target user as samples and the number identifier of the target user as a label, a sample number corresponding to the number of the target user in the first sample number set is constructed.
3. The method according to claim 1, characterized in that The constructing a second sample number set based on the CDR of the user in the second scenario and the supplementary features of the second scenario includes: Obtain user information and CDR of the target user in the second scenario, wherein the user information includes a number and a number identifier used to indicate whether the number is abnormal or normal, and the CDR includes a call date, a calling number, a called number, a call start time, a call duration, and a roaming location; Constructing a first number of call features corresponding to the CDR of the target user based on feature coverage, feature content and feature value; Acquire a second number of supplementary features corresponding to the target user in the second scenario; Using features including a first number of call features and a second number of supplementary features corresponding to the target user as samples and the number identifier of the target user as a label, construct a sample number corresponding to the number of the target user in the second sample number set.
4. The method according to claim 1, characterized in that: The preset number recognition model includes a deep neural network based on the Inception module, and the deep neural network includes an input layer, a hidden layer and an output layer. The input layer is used to input target samples; The hidden layer includes an Inception module and a fully connected layer connected in sequence, the Inception module is connected to the output of the input layer, the fully connected layer is connected to the output layer, and the hidden layer is used to extract features from the target sample; The output layer is used to output whether the target sample belongs to an abnormal number or a normal number.
5. The method according to claim 4, characterized in that The target sample is a sample in the first sample number set, the input layer corresponds to the preset number recognition model of the first number recognition model and is used to input features including a first number of call features of the target sample; The target sample is a sample in the second sample number set, and the input layer corresponds to the preset number recognition model of the second number recognition model and is used to input features including a first number of call features and a second number of supplementary features of the target sample.
6. The method according to claim 1, characterized in that The supplementary features include: one or more combinations of user occupation, account opening status, package usage, account opening channel and place of residence.
7. A number identification method, characterized in that: include: Obtain a CDR of a target user in a target scenario and a second number of supplementary features of the target scenario; Constructing a first number of call features corresponding to the CDR; Input the features including the first number of call features and the second number of supplementary features into a pre-trained second number recognition model to identify whether the number of the target user in the target scenario is an abnormal number or a normal number, wherein the second number recognition model is trained based on the steps of the method described in any one of claims 1-6.
8. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method according to any one of claims 1 to 6 or the steps of the method according to claim 7 are implemented.
9. A readable storage medium, characterized in that: The readable storage medium stores a program or an instruction, and when the program or the instruction is executed by the processor, the steps of the method according to any one of claims 1 to 6 or the steps of the method according to claim 7 are implemented.
10. A computer program product, characterized in that The computer program product comprises a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute the steps of the method according to any one of claims 1 to 6 or to implement the steps of the method according to claim 7.