Abnormal communication card number identification method and device, equipment and storage medium

Abstract

The invention discloses an abnormal communication card number identification method and device, equipment and a storage medium, and belongs to the technical field of communication, and the identification method comprises the steps: obtaining the communication characteristics of a to-be-identified communication card number in a preset time period; respectively inputting the communication features into a first base classifier, a second base classifier and a third base classifier of an abnormal communication card number identification model to obtain an abnormal probability value that the communication card number to be identified output by each base classifier is an abnormal card number and a normal probability value that the communication card number to be identified is a normal card number, and determining whether the communication card number to be identified is abnormal or not according to each abnormal probability value, each normal probability value and a weight corresponding to each base classifier, a loss function used when each base classifier is trained is obtained based on weight coefficients of positive and negative samples, the probability that the samples output by the base classifiers belong to abnormal communication card numbers and the similarity between the samples and sample center points. According to the invention, the identification accuracy of identifying the abnormal communication card number can be improved.

Description

Technical Field

[0001] The present application belongs to the field of communication technology, and in particular, relates to a method, device, equipment and storage medium for identifying abnormal communication card numbers. Background Art

[0002] An abnormal mobile phone card number means that the mobile phone card issued by the operator is not used normally by the user, but is used by unscrupulous users to conduct illegal activities and make profits.

[0003] In related technologies, whether a mobile phone card number is abnormal can be identified through a recognition model. The training of the recognition model requires sufficient abnormal label data identified by the mobile phone number. However, in reality, mobile phone cards are normally used by users, so the normal samples collected are far more than the abnormal samples, making the data distribution unbalanced. In an unbalanced situation, the directed graph model will be exposed to more categories with more samples during the training process, thereby learning more features and patterns related to the category, and is more inclined to use the normal category as the prediction result, resulting in a lower recognition accuracy. Summary of the invention

[0004] The embodiments of the present application provide a method, device, equipment and storage medium for identifying abnormal communication card numbers, which can improve the recognition accuracy of abnormal communication card numbers.

[0005] In a first aspect, an embodiment of the present application provides a method for identifying an abnormal communication card number, the method comprising:

[0006] Obtaining the communication characteristics of the communication card number to be identified within a preset time period;

[0007] Input the communication characteristics of the communication card number to be identified into the first base classifier of the abnormal communication card number identification model to obtain a first abnormal probability value of the communication card number to be identified as an abnormal card number and a first normal probability value of the communication card number to be identified as a normal card number; input the communication characteristics of the communication card number to be identified into the second base classifier of the abnormal communication card number identification model to obtain a second abnormal probability value of the communication card number to be identified as an abnormal card number and a second normal probability value of the communication card number to be identified as a normal card number; input the communication characteristics of the communication card number to be identified into the third base classifier of the abnormal communication card number identification model to obtain a third abnormal probability value of the communication card number to be identified as an abnormal card number and a third normal probability value of the communication card number to be identified as a normal card number; wherein each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, and the loss function used when training each base classifier is obtained based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to the abnormal communication card number, and the similarity between the sample and the sample center point, and the similarity between the sample and the sample center point is obtained based on a Gaussian kernel function;

[0008] Whether the communication card number to be identified is abnormal is determined based on the abnormal probability value and normal probability value output by each base classifier and the weight corresponding to each base classifier.

[0009] In some embodiments of the present application, before inputting the communication features of the communication card number to be identified into each base classifier of the abnormal communication card number identification model, the method further includes:

[0010] Acquire a first training sample set and a second training sample set, the first training sample set including communication features of a communication card number and a label of the communication card number, the label of the communication card number is used to identify an abnormal communication card number or a normal communication card number, and the second training sample set includes the communication features of the communication card number;

[0011] Screening out a first sub-training sample set, a second sub-training sample set, and a third sub-training sample set having differences from the first training sample set;

[0012] Performing first-stage training on the initial first base classifier, the second base classifier and the third base classifier according to the first sub-training sample set, the second sub-training sample set and the third sub-training sample set, respectively, to obtain the base classifiers after the first-stage training;

[0013] For any one of the three base classifiers, perform the following operations to obtain a trained base classifier, set corresponding weights for each trained base classifier, and obtain an abnormal communication card number recognition model: perform the second stage of training on the base classifier trained in the first stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier.

[0014] In some embodiments of the present application, obtaining a first training sample set and a second training sample set includes:

[0015] Collect communication data of a plurality of first sample communication card numbers and a plurality of second sample communication card numbers within a preset time period respectively, wherein the communication data of each first sample communication card number and each second sample communication card number includes at least one of the calling frequency, the calling and called ratio, the traffic usage data, the number of device changes corresponding to the communication card, the number of changes in the resident area, the number of changes in the commonly used base station, the number of changes in the place of attribution, the registration place, the registration time, and the network access time;

[0016] Preprocessing the communication data of the plurality of first sample communication card numbers and the plurality of second sample communication card numbers respectively to obtain the communication features corresponding to the plurality of first sample communication card numbers and the plurality of second sample communication card numbers;

[0017] Constructing a first training sample according to the communication feature of each first sample communication card number and its corresponding label to obtain a first training sample set;

[0018] A second training sample is constructed according to the communication characteristics of each second sample communication card number to obtain a second training sample set.

[0019] In some embodiments of the present application, the initial first base classifier, the second base classifier, and the third base classifier are respectively trained in the first stage according to the first sub-training sample set, the second sub-training sample set, and the third sub-training sample set to obtain each base classifier after the first stage training, including:

[0020] Input the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set into the initial first base classifier, the second base classifier and the third base classifier respectively to obtain a first prediction result, a second prediction result and a third prediction result;

[0021] Calculate the loss function value according to the first prediction result, the second prediction result, the third prediction result and their corresponding labels;

[0022] Update the parameters of each base classifier according to the gradient information corresponding to the loss function value, and update each base classifier according to the updated parameters of each base classifier;

[0023] Determine each updated base classifier as the initial base classifier;

[0024] Return and input the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set into the initial first base classifier, the second base classifier and the third base classifier respectively until the models of each base classifier converge, thereby obtaining each base classifier after the first stage of training.

[0025] In some embodiments of the present application, the first base classifier is a recurrent neural network, the second base classifier is a bidirectional long short-term memory network, and the third base classifier is a bidirectional gated recurrent network.

[0026] In some embodiments of the present application, the loss function is:

[0027]

[0028] Where y represents the sample label, p represents the probability value of the training sample output by the base classifier belonging to an abnormal communication card number, α represents the weight for balancing positive and negative samples, and r represents the degree of proximity between the training sample and the sample center;

[0029] Among them, r is obtained by the following formula:

[0030]

[0031] Among them, σ represents the Gaussian kernel function parameter, x represents the training sample, centi Represents class C i The sample center point, C i represents the i-th label data set, and n is the total number of samples.

[0032] In some embodiments of the present application, for any of the three base classifiers, the following operations are performed respectively to obtain a trained base classifier, and corresponding weights are set for each trained base classifier to obtain an abnormal communication card number recognition model: the base classifier trained in the first stage is trained in the second stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier, including:

[0033] Input the samples in the second training sample set into each base classifier trained in the first stage respectively, and obtain the prediction results corresponding to each base classifier;

[0034] For each sample in the second training sample set, the following operations are performed cyclically until an expanded sample set for each base classifier is obtained:

[0035] When the prediction results of two base classifiers among the three base classifiers are abnormal communication card numbers and the prediction result of the other base classifier is a normal communication card number, the abnormal communication card number is used as the label of the sample to obtain the expanded sample data, and the expanded sample data is added to the expanded sample set of the other base classifier;

[0036] The following operations are performed for each base classifier until the preset stopping condition is reached, and the first base classifier, the second base classifier, and the third base classifier trained in the second stage are obtained:

[0037] The base classifier is trained according to the samples in the sub-training sample set and the samples in the expanded sample set corresponding to the base classifier to obtain a trained base classifier;

[0038] According to the prediction results of the base classifier for each sample and the label of each sample, the classification accuracy of the trained base classifier is calculated;

[0039] When the classification accuracy of this round is less than that of the previous round, some samples are randomly deleted from the expanded sample set corresponding to the base classifier to obtain an updated expanded sample set;

[0040] The base classifier is trained according to the samples in the sub-training sample set corresponding to the base classifier and the samples in the updated expanded sample set to obtain a trained base classifier;

[0041] The weights of each base classifier are set according to the classification accuracy of each trained base classifier to obtain an abnormal communication card number recognition model.

[0042] In some embodiments of the present application, determining whether the communication card number to be identified is abnormal according to the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier includes:

[0043] The sum of the product of the probability value of the communication card number to be identified output by each base classifier of the abnormal communication card number recognition model belonging to the abnormal card number and the weight corresponding to each base classifier is used as the target abnormal probability value;

[0044] The sum of the product of the probability value of the communication card number to be identified output by each base classifier of the abnormal communication card number identification model belonging to a normal card number and the weight corresponding to each base classifier is used as the target normal probability value;

[0045] Determine whether the communication card number to be identified is abnormal based on the target abnormal probability value and the target normal probability value.

[0046] In a second aspect, an embodiment of the present application provides a device for identifying an abnormal communication card number, the device comprising:

[0047] A first acquisition module is used to acquire the communication characteristics of the communication card number to be identified within a preset time period;

[0048] An input module is used to input the communication characteristics of the communication card number to be identified into the first base classifier of the abnormal communication card number identification model to obtain a first abnormal probability value that the communication card number to be identified is an abnormal card number and a first normal probability value that the communication card number to be identified is a normal card number; input the communication characteristics of the communication card number to be identified into the second base classifier of the abnormal communication card number identification model to obtain a second abnormal probability value that the communication card number to be identified is an abnormal card number and a second normal probability value that the communication card number to be identified is a normal card number; input the communication characteristics of the communication card number to be identified into the third base classifier of the abnormal communication card number identification model to obtain a third abnormal probability value that the communication card number to be identified is an abnormal card number and a third normal probability value that the communication card number to be identified is a normal card number; wherein each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, and the loss function used when training each base classifier is obtained based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to the abnormal communication card number, and the similarity between the sample and the sample center point, and the similarity between the sample and the sample center point is obtained based on a Gaussian kernel function;

[0049] The determination module is used to determine whether the communication card number to be identified is abnormal based on the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier.

[0050] In a third aspect, an embodiment of the present application provides a device for identifying an abnormal communication card number, the device comprising: a processor and a memory storing computer program instructions;

[0051] When the processor executes the computer program instructions, the method for identifying an abnormal communication card number of any of the above embodiments is implemented.

[0052] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the method for identifying an abnormal communication card number of any of the above embodiments is implemented.

[0053] According to the abnormal communication card number recognition method, device, equipment and storage medium provided in the embodiments of the present application, on the one hand, each base classifier of the abnormal communication card number recognition model is trained using a semi-supervised method, which can make full use of a large amount of unlabeled data for model training, solve the problem of uneven data distribution, enhance the model's robustness and generalization ability, and at the same time reduce manual intervention and reduce the cost of annotating data. The abnormal communication card number recognition model can also combine the advantages of multiple base classifiers, thereby improving the recognition accuracy of abnormal communication card numbers. On the other hand, the loss function used when training each base classifier is based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to an abnormal communication card number, and the similarity between the sample and the sample center point. Among them, the similarity between the sample and the sample center point is obtained based on the Gaussian kernel function. The weight of the abnormal class samples in the loss function can be increased by controlling the weight coefficients of the positive and negative samples, and the weight of the normal class samples in the loss function can be reduced, thereby reducing the influence of the normal class samples on the model, so that the model pays more attention to the abnormal class samples, and achieves the effect of balancing the distribution of the data set. The similarity between the sample and the sample center point is integrated into the loss function, so that the loss function value is related to the similarity between the sample and the sample center point. Therefore, the model can be controlled to pay more attention to difficult-to-distinguish samples, and the proportion of easy-to-distinguish samples in the loss function calculation can be gradually reduced. The model converges faster and has stronger generalization ability, and can also further improve the recognition accuracy of the model. Therefore, based on the abnormal communication card number recognition model, the communication characteristics of the communication card number to be identified within a preset time period are identified to determine whether the communication card number to be identified is abnormal, which can improve the recognition accuracy of abnormal communication card numbers. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1 A flow chart of a method for identifying an abnormal communication card number provided in an embodiment of the present application;

[0056] Figure 2 Another flowchart of the method for identifying an abnormal communication card number provided in an embodiment of the present application;

[0057] Figure 3 A flowchart of a method for training an abnormal communication card number recognition model provided in an embodiment of the present application;

[0058] Figure 4 A schematic diagram of the structure of a device for identifying abnormal communication card numbers provided in an embodiment of the present application;

[0059] Figure 5 A schematic diagram of the structure of a device for identifying abnormal communication card numbers provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0061] 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 statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0062] An abnormal mobile phone card number means that the mobile phone card issued by the operator is not used normally by the user, but is used by unscrupulous users to conduct illegal activities and make profits.

[0063] In related technologies, whether a mobile phone card number is abnormal can be identified through a recognition model. The training of the recognition model requires sufficient abnormal label data identified by the mobile phone number. However, in reality, mobile phone cards are normally used by users, so the normal samples collected are far more than the abnormal samples, making the data distribution unbalanced. In an unbalanced situation, the directed graph model will be exposed to more categories with more samples during the training process, thereby learning more features and patterns related to the category, and is more inclined to use the normal category as the prediction result, resulting in a lower recognition accuracy.

[0064] In order to solve the above problems, the embodiments of the present application provide a method, device, equipment and storage medium for identifying abnormal communication card numbers, which can improve the recognition accuracy of abnormal communication card numbers.

[0065] Figure 1 A flow chart of a method for identifying abnormal communication card numbers provided in an embodiment of the present application. Figure 1 As shown, the method for identifying an abnormal communication card number may specifically include the following steps:

[0066] S110, obtaining the communication characteristics of the communication card number to be identified within a preset time period;

[0067] S120, input the communication characteristics of the communication card number to be identified into the first base classifier of the abnormal communication card number identification model, and obtain a first abnormal probability value of the communication card number to be identified as an abnormal card number and a first normal probability value of the communication card number to be identified as a normal card number; input the communication characteristics of the communication card number to be identified into the second base classifier of the abnormal communication card number identification model, and obtain a second abnormal probability value of the communication card number to be identified as an abnormal card number and a second normal probability value of the communication card number to be identified as a normal card number; input the communication characteristics of the communication card number to be identified into the third base classifier of the abnormal communication card number identification model, and obtain a third abnormal probability value of the communication card number to be identified as an abnormal card number and a third normal probability value of the communication card number to be identified as a normal card number; wherein each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, and the loss function used when training each base classifier is obtained based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to the abnormal communication card number, and the similarity between the sample and the sample center point, and the similarity between the sample and the sample center point is obtained based on a Gaussian kernel function;

[0068] S130, determining whether the communication card number to be identified is abnormal based on the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier.

[0069] According to the abnormal communication card number identification method provided in the embodiment of the present application, on the one hand, each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, which can make full use of a large amount of unlabeled data for model training, solve the problem of uneven data distribution, enhance the model's robustness and generalization ability, and at the same time reduce manual intervention and reduce the cost of annotating data. The abnormal communication card number identification model can also combine the advantages of multiple base classifiers, thereby improving the recognition accuracy of abnormal communication card numbers. On the other hand, the loss function used when training each base classifier is based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to an abnormal communication card number, and the similarity between the sample and the sample center point. Among them, the similarity between the sample and the sample center point is obtained based on the Gaussian kernel function. The weight of the abnormal class samples in the loss function can be increased by controlling the weight coefficients of the positive and negative samples, and the weight of the normal class samples in the loss function can be reduced, thereby reducing the influence of the normal class samples on the model, so that the model pays more attention to the abnormal class samples, and achieves the effect of balancing the distribution of the data set. The similarity between the sample and the sample center point is integrated into the loss function, so that the loss function value is related to the similarity between the sample and the sample center point. Therefore, the model can be controlled to pay more attention to difficult-to-distinguish samples, and the proportion of easy-to-distinguish samples in the loss function calculation can be gradually reduced. The model converges faster and has stronger generalization ability, and can also further improve the recognition accuracy of the model. Therefore, based on the abnormal communication card number recognition model, the communication characteristics of the communication card number to be identified within a preset time period are identified to determine whether the communication card number to be identified is abnormal, which can improve the recognition accuracy of abnormal communication card numbers.

[0070] The preset time period in the above S110 can be 1 month, and the communication data of the communication card number to be identified within 1 month can be collected, and the communication data can be cleaned, standardized and normalized to obtain communication characteristics, wherein the communication data can include calling frequency, calling and called ratio, traffic usage data, the number of device changes corresponding to the communication card, the number of changes in the permanent area, the number of changes in the commonly used base stations, the number of changes in the place of affiliation, the registration place, the registration time and at least one of the network access time.

[0071] Figure 2 Another flow chart of the method for identifying abnormal communication card numbers provided in the embodiment of the present application. Before the communication features of the communication card number to be identified are input into each base classifier of the abnormal communication card number identification model, such as Figure 2 As shown, the method for identifying abnormal communication card numbers may also include the following steps:

[0072] S210, obtaining a first training sample set and a second training sample set, wherein the first training sample set includes communication features of a communication card number and a label of the communication card number, the label of the communication card number is used to identify an abnormal communication card number or a normal communication card number, and the second training sample set includes the communication features of the communication card number;

[0073] S220, screening out a first sub-training sample set, a second sub-training sample set, and a third sub-training sample set having differences from the first training sample set;

[0074] S230, performing first-stage training on the initial first base classifier, the second base classifier, and the third base classifier according to the first sub-training sample set, the second sub-training sample set, and the third sub-training sample set, respectively, to obtain each base classifier after the first-stage training;

[0075] S240, for any one of the three base classifiers, perform the following operations respectively to obtain a trained base classifier, set corresponding weights for each trained base classifier, and obtain an abnormal communication card number recognition model: perform a second stage of training on the base classifier trained in the first stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier.

[0076] The above S210 to S240 are described in detail below, as shown below:

[0077] Regarding S210, the first training sample set is a labeled training sample set, and the second training sample set is an unlabeled training sample set. The communication card number may be a mobile phone card number, and the communication characteristics may be obtained by collecting and analyzing the communication data of a large number of mobile phone card numbers. For labeled mobile phone card numbers, the communication characteristics and labels of the mobile phone card numbers are used as labeled training data to obtain the first training sample set, and for unlabeled mobile phone card numbers, the communication characteristics of the mobile phone card numbers are used as unlabeled training data to obtain the second training sample set.

[0078] Regarding S220, the bootstrap sampling method can be used to screen out a first sub-training sample set, a second sub-training sample set, and a third sub-training sample set with differences from the first training sample set, and each sub-training sample set can be of a fixed size. Since the bootstrap sampling method is a sampling with replacement, the composition of samples in each sub-training sample set will be different. Although the size of each data set is fixed, the specific samples included and the number of times each sample appears will be different. This difference enables each data set to capture different features and patterns in the first training sample set, thereby training a differentiated base classifier.

[0079] Regarding S230, the first base classifier, the second base classifier and the third base classifier are three differentiated base classifiers, which can increase the coverage and information gain of the model, model and predict the data from different angles, thereby providing more diverse information and improving model performance.

[0080] The first stage is supervised training. Each base classifier is trained in the first stage using each sub-training sample set, and each base classifier after the first stage training can be obtained. The loss function used in training each base classifier is based on the weight coefficient of positive and negative samples, the probability that the sample output by the base classifier belongs to an abnormal communication card number, and the similarity between the sample and the sample center point. Among them, the similarity between the sample and the sample center point is obtained based on the Gaussian kernel function. The weight of the abnormal class samples in the loss function can be increased by controlling the weight coefficient of the positive and negative samples, and the weight of the normal class samples in the loss function can be reduced, thereby reducing the impact of the normal class samples on the model, making the model pay more attention to the abnormal class samples, and achieving the effect of balancing the distribution of the data set. The similarity between the sample and the sample center point is integrated into the loss function, so that the loss function value is related to the similarity between the sample and the sample center point. Therefore, the model can be controlled to pay more attention to difficult-to-distinguish samples, and the proportion of easy-to-distinguish samples in the loss function calculation can be gradually reduced. The model converges faster and has stronger generalization ability, and can further improve the recognition accuracy of the model.

[0081] Regarding S240, after the first stage of training is completed, the second stage of training is performed on each base classifier. The second stage is semi-supervised training. The trained base classifier is further trained according to the labeled sub-training sample set and the unlabeled second training sample set to obtain a trained base classifier. After setting corresponding weights for each base classifier, an abnormal communication card number recognition model is obtained. In this way, multiple base classifiers are trained by a semi-supervised training method, and the abnormal communication card number recognition model is obtained after weighting. This can solve the problem of data imbalance and reduce the labeling cost. The abnormal communication card number recognition model can also combine the advantages of multiple base classifiers, thereby improving the recognition accuracy of the abnormal communication card number recognition model.

[0082] In some embodiments of the present application, the above S210 includes:

[0083] Collect communication data of a plurality of first sample communication card numbers and a plurality of second sample communication card numbers within a preset time period respectively, wherein the communication data of each first sample communication card number and each second sample communication card number includes at least one of the calling frequency, the calling and called ratio, the traffic usage data, the number of device changes corresponding to the communication card, the number of changes in the resident area, the number of changes in the commonly used base station, the number of changes in the place of attribution, the registration place, the registration time, and the network access time;

[0084] Preprocessing the communication data of the plurality of first sample communication card numbers and the plurality of second sample communication card numbers respectively to obtain the communication features corresponding to the plurality of first sample communication card numbers and the plurality of second sample communication card numbers;

[0085] Constructing a first training sample according to the communication feature of each first sample communication card number and its corresponding label to obtain a first training sample set;

[0086] A second training sample is constructed according to the communication characteristics of each second sample communication card number to obtain a second training sample set.

[0087] The preset time may be one month, or other time periods, and this application does not impose any limitation thereto.

[0088] Preprocessing of communication data can include steps such as data cleaning, standardization and normalization.

[0089] The labeled first training sample set and the unlabeled second training sample set provide sufficient data accuracy for subsequent model training, which facilitates the subsequent first and second stage training of each base classifier. Moreover, the training sample set contains multiple communication features, so that the model can identify whether the communication card number is abnormal based on multiple communication features, thereby improving the recognition accuracy.

[0090] In some embodiments of the present application, the initial first base classifier, the second base classifier, and the third base classifier are respectively trained in the first stage according to the first sub-training sample set, the second sub-training sample set, and the third sub-training sample set to obtain each base classifier after the first stage training, including:

[0091] Input the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set into the initial first base classifier, the second base classifier and the third base classifier respectively to obtain a first prediction result, a second prediction result and a third prediction result;

[0092] Calculate the loss function value according to the first prediction result, the second prediction result, the third prediction result and their corresponding labels;

[0093] Update the parameters of each base classifier according to the gradient information corresponding to the loss function value, and update each base classifier according to the updated parameters of each base classifier;

[0094] Determine each updated base classifier as the initial base classifier;

[0095] Return and input the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set into the initial first base classifier, the second base classifier and the third base classifier respectively until the models of each base classifier converge, thereby obtaining each base classifier after the first stage of training.

[0096] The first stage of training is performed on each base classifier according to each sub-training sample set. The value of the loss function is calculated according to the prediction results of each base classifier on the sample and the true label of the sample. The corresponding gradient information is calculated according to the value of the loss function. The parameters of each base classifier are updated according to the gradient information. Specifically, the gradient descent algorithm can be used to optimize the parameters of each base classifier. The base classifiers are continuously iterated and trained. When the preset number of iterations is reached or the loss function value is less than the preset value, the training is terminated, and the base classifiers trained in the first stage can be obtained.

[0097] The supervised training in the first stage can lay a good foundation for the subsequent semi-supervised training, so that the subsequent semi-supervised training is carried out on the basis of the trained model, which can ensure the accuracy of the abnormal communication card number recognition model training.

[0098] In some embodiments of the present application, the first base classifier is a recurrent neural network, the second base classifier is a bidirectional long short-term memory network, and the third base classifier is a bidirectional gated recurrent network. Using base classifiers with different network structures can increase the coverage and information gain of the model, model and predict data from different angles, thereby providing more diverse information and improving model performance.

[0099] In some embodiments of the present application, the loss function is:

[0100]

[0101] Where y represents the sample label (y=1 means that the sample belongs to an abnormal communication card number, which is a positive sample, and y=0 means that the sample belongs to a normal communication card number, which is a negative sample), p represents the probability value of the training sample output by the base classifier belonging to an abnormal communication card number, α represents the weight for balancing positive and negative samples, and r represents the degree of proximity between the training sample and the sample center;

[0102] Among them, r is obtained by the following formula:

[0103]

[0104] Among them, σ represents the Gaussian kernel function parameter, x represents the training sample, cent i Represents class C i The sample center point, C i Represents the i-th label data set, n is the total number of samples, and i is 0 or 1.

[0105] By controlling the weight coefficients of positive and negative samples, the weight of abnormal samples in the loss function is increased, and the weight of normal samples in the loss function is reduced, thereby reducing the impact of normal samples on the model, making the model pay more attention to abnormal samples and achieving the effect of balancing the distribution of the data set.

[0106] r represents the training sample x and the sample center cent i The closer the sample x is to the center point cent, the smaller the value of r is. i The closer the distance is, the smaller the loss function value is, the easier it is for the sample to be correctly classified by the model, and the smaller the contribution of these easy-to-classify samples to the model parameter update is. As the number of iterations increases, the proportion of easy-to-classify samples increases, and the role of these samples in improving model performance decreases. i The similarity of is integrated into the loss function, making the model pay more attention to difficult-to-separate samples and gradually reducing the proportion of easy-to-separate samples in the loss function calculation. The model converges faster, has stronger generalization ability, and can further improve the recognition accuracy of the model.

[0107] In some embodiments of the present application, for any of the three base classifiers, the following operations are performed respectively to obtain a trained base classifier, and corresponding weights are set for each trained base classifier to obtain an abnormal communication card number recognition model: the base classifier trained in the first stage is trained in the second stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier, including:

[0108] Input the samples in the second training sample set into each base classifier trained in the first stage respectively, and obtain the prediction results corresponding to each base classifier;

[0109] For each sample in the second training sample set, the following operations are performed cyclically until an expanded sample set for each base classifier is obtained:

[0110] When the prediction results of two base classifiers among the three base classifiers are abnormal communication card numbers and the prediction result of the other base classifier is a normal communication card number, the abnormal communication card number is used as the label of the sample to obtain the expanded sample data, and the expanded sample data is added to the expanded sample set of the other base classifier;

[0111] The following operations are performed for each base classifier until the preset stopping condition is reached, and the first base classifier, the second base classifier, and the third base classifier trained in the second stage are obtained:

[0112] The base classifier is trained according to the samples in the sub-training sample set and the samples in the expanded sample set corresponding to the base classifier to obtain a trained base classifier;

[0113] According to the prediction results of the base classifier for each sample and the label of each sample, the classification accuracy of the trained base classifier is calculated;

[0114] When the classification accuracy of this round is less than that of the previous round, some samples are randomly deleted from the expanded sample set corresponding to the base classifier to obtain an updated expanded sample set;

[0115] The base classifier is trained according to the samples in the sub-training sample set corresponding to the base classifier and the samples in the updated expanded sample set to obtain a trained base classifier;

[0116] The weights of each base classifier are set according to the classification accuracy of each trained base classifier to obtain an abnormal communication card number recognition model.

[0117] It can make full use of unlabeled data to generate pseudo-labeled samples, reduce the dependence on manually labeled data, and better deal with the problem of uneven data distribution. Moreover, by adopting the method of "the minority obeys the majority", the expanded sample set corresponding to each base classifier is provided by the collaboration of the other two base classifiers, which can improve the accuracy of sample labeling.

[0118] The base classifier can be trained in the second stage according to the samples in the sub-training sample set and the samples in the expanded sample set. If the classification accuracy of the base classifier calculated in the current training round is less than the classification accuracy of the previous training round, it means that the classification effect of the base classifier in the current round is not as good as that of the base classifier in the previous round. At this time, several samples can be randomly deleted from the expanded sample set to update the expanded sample set. Because the expanded sample set contains pseudo-label data and there is a certain probability of mislabeling, deletion can appropriately reduce the impact on model training.

[0119] The trained classifiers are obtained by continuously iteratively training each base classifier until no samples are available in the second training sample set or a preset stopping condition (such as reaching a maximum number of iterations) is reached.

[0120] The classification accuracy of each base classifier can be obtained by the following formula:

[0121]

[0122] Among them, ω i is the weight of the base classifier, P i is the classification accuracy of the i-th base classifier, and j is the label of the abnormal communication card number.

[0123] Finally, the weights of each base classifier are set according to the calculated classification accuracy of each trained base classifier to obtain the abnormal communication card number recognition model.

[0124] Figure 3 A flow chart of a training method for an abnormal communication card number recognition model provided in an embodiment of the present application is shown below in combination with Figure 3 Briefly summarize the entire process of the training method for the abnormal communication card number recognition model provided in the embodiment of the present application.

[0125] S310, first, obtain a labeled data set;

[0126] S320, screening the difference training subsets to obtain three training subsets;

[0127] S330, training a base classifier according to the difference training subset;

[0128] S340, then, obtaining an unlabeled data set;

[0129] S350, obtaining a pseudo-labeled dataset according to the unlabeled dataset and the trained base classifier;

[0130] S360, next, merge the original training subset and the pseudo-label dataset, retrain the base classifier, and randomly delete some samples according to the model accuracy;

[0131] S370, determining whether the model training meets the preset conditions, if so, executing S380, if not, returning to executing S350;

[0132] S380, calculating the weight of the base classifier;

[0133] S390, finally, after weighted voting on each base classifier, an abnormal communication card number recognition model is obtained.

[0134] In some embodiments of the present application, determining whether the communication card number to be identified is abnormal according to the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier includes:

[0135] The sum of the product of the probability value of the communication card number to be identified output by each base classifier of the abnormal communication card number recognition model belonging to the abnormal card number and the weight corresponding to each base classifier is used as the target abnormal probability value;

[0136] The sum of the product of the probability value of the communication card number to be identified output by each base classifier of the abnormal communication card number identification model belonging to a normal card number and the weight corresponding to each base classifier is used as the target normal probability value;

[0137] Determine whether the communication card number to be identified is abnormal based on the target abnormal probability value and the target normal probability value.

[0138] After obtaining the abnormal probability value of the communication card number to be identified belonging to an abnormal card number and the normal probability value of the communication card number to be identified output by each base classifier, each abnormal probability value and normal probability value can be weighted by the weights corresponding to each base classifier, and finally a target abnormal probability value and a target normal probability value are obtained. Whether the communication card number to be identified is abnormal is determined according to the weighted target abnormal probability value and target normal probability value, which can improve the accuracy of judging whether the communication card number is abnormal.

[0139] In some embodiments of the present application, before taking the sum of the products of the probability value that the communication card number to be identified output by each base classifier of the abnormal communication card number recognition model belongs to an abnormal card number and the weight corresponding to each base classifier as the target abnormal probability value, the method further includes:

[0140] The weight of each base classifier is obtained, and the weight of each base classifier is calculated according to the classification accuracy of each base classifier.

[0141] The calculation method of the weight of each base classifier can refer to the calculation formula provided in the embodiment of the abnormal communication card number recognition model training method. The corresponding weight is calculated according to the classification accuracy of each base classifier, and a high weight can be given to the base classifier with high classification accuracy, thereby improving the accuracy of recognizing the communication card number to be recognized.

[0142] In some embodiments of the present application, determining whether the communication card number to be identified is abnormal according to the target abnormal probability value and the target normal probability value includes:

[0143] When the target abnormal probability value is greater than the target normal probability value, the communication card number to be identified is determined to be an abnormal card number;

[0144] When the target normal probability value is greater than the target abnormal probability value, the communication card number to be identified is determined to be a normal card number.

[0145] Specifically, the following formula can be used to calculate the target abnormal probability value and the target normal probability value and determine the maximum value therebetween:

[0146]

[0147] Among them, y represents the final recognition result, which includes two recognition results: normal card number and abnormal card number. x is the communication feature vector corresponding to the communication card number to be recognized. i represents the i-th base classifier. j represents the j-th communication card number label, which takes the value of 0 or 1, representing the normal card number and the abnormal card number, respectively. ω i is the weight corresponding to the i-th base classifier, is the recognition probability of the communication card number to be recognized.

[0148] For example, the communication features of the communication card number to be identified are respectively sent to the three base classifiers. Assume that the weight w1 of the first base classifier C1 is 0.2, the weight w2 of the second base classifier C2 is 0.3, and the weight w3 of the third base classifier C3 is 0.5. If the output of C1 is that the probability that the communication card number to be identified belongs to a normal card number is p10=0.3, and the probability that it belongs to an abnormal card number is p11=0.7; the output of C2 is that the probability that the communication card number to be identified belongs to a normal card number is p20=0.1, and the probability that it belongs to an abnormal card number is p21=0.9; the output of C3 is that the probability that the communication card number to be identified belongs to a normal card number is p30=0.6, and the probability that it belongs to an abnormal card number is p31=0.4. Then calculate the probability that the communication card number to be identified is a normal card number = w1*p10+w2*p20+w3*p30=0.2*0.3+0.3*0.1+0.5*0.6=0.39; the probability that it is an abnormal card number = w1*p11+w2*p21+w3*p31=0.2*0.7+0.3*0.9+0.5*0.4=0.61.

[0149] That is, the three base classifiers determine that the probability that the communication card number to be identified belongs to a normal card number is 0.39, and the probability that it is an abnormal card number is 0.61. The probability of belonging to an abnormal card number obtained through argmax calculation is the largest, and the final recognition result is an abnormal card number.

[0150] The maximum value of the target abnormal probability value and the target normal probability value is determined as the final recognition result of the communication card number to be recognized, so that a more accurate recognition result can be obtained.

[0151] Figure 4 A schematic diagram of the structure of the device for identifying abnormal communication card numbers provided in the embodiment of the present application is shown below in combination with Figure 4 The present invention introduces an apparatus for identifying abnormal communication card numbers in an embodiment of the present invention, the apparatus comprising:

[0152] The first acquisition module 401 is used to acquire the communication characteristics of the communication card number to be identified within a preset time period;

[0153] Input module 402, used for inputting the communication characteristics of the communication card number to be identified into the first base classifier of the abnormal communication card number identification model, and obtaining a first abnormal probability value of the communication card number to be identified as an abnormal card number and a first normal probability value of the communication card number to be identified as a normal card number; inputting the communication characteristics of the communication card number to be identified into the second base classifier of the abnormal communication card number identification model, and obtaining a second abnormal probability value of the communication card number to be identified as an abnormal card number and a second normal probability value of the communication card number to be identified as a normal card number; inputting the communication characteristics of the communication card number to be identified into the third base classifier of the abnormal communication card number identification model, and obtaining a third abnormal probability value of the communication card number to be identified as an abnormal card number and a third normal probability value of the communication card number to be identified as a normal card number; each base classifier of the abnormal communication card number identification model is trained by a semi-supervised method, and the loss function used when training each base classifier is obtained based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to the abnormal communication card number, and the similarity between the sample and the sample center point, and the similarity between the sample and the sample center point is obtained based on the Gaussian kernel function;

[0154] The determination module 403 is used to determine whether the communication card number to be identified is abnormal according to the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier.

[0155] In some embodiments of the present application, the identification device further includes:

[0156] A second acquisition module is used to acquire a first training sample set and a second training sample set before inputting the communication features of the communication card number to be identified into each base classifier of the abnormal communication card number identification model, wherein the first training sample set includes the communication features of the communication card number and the label of the communication card number, the label of the communication card number is used to identify an abnormal communication card number or a normal communication card number, and the second training sample set includes the communication features of the communication card number;

[0157] A screening module, used for screening out a first sub-training sample set, a second sub-training sample set and a third sub-training sample set having differences from the first training sample set;

[0158] A first training module is used to perform first-stage training on the initial first base classifier, the second base classifier and the third base classifier according to the first sub-training sample set, the second sub-training sample set and the third sub-training sample set, respectively, to obtain each base classifier after the first-stage training;

[0159] The second training module is used to perform the following operations on any one of the three base classifiers to obtain a trained base classifier, set corresponding weights for each trained base classifier, and obtain an abnormal communication card number recognition model: the base classifier trained in the first stage is trained in the second stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier.

[0160] In some embodiments of the present application, the second acquisition module includes:

[0161] A collection unit, used to collect communication data of a plurality of first sample communication card numbers and a plurality of second sample communication card numbers within a preset time period, respectively, wherein the communication data of each first sample communication card number and each second sample communication card number includes at least one of the calling frequency, the calling and called ratio, the traffic usage data, the number of device changes corresponding to the communication card, the number of changes in the resident area, the number of changes in the commonly used base station, the number of changes in the place of attribution, the registration place, the registration time, and the network access time;

[0162] A preprocessing unit, used to preprocess the communication data of the plurality of first sample communication card numbers and the plurality of second sample communication card numbers respectively, to obtain the communication features corresponding to the plurality of first sample communication card numbers and the plurality of second sample communication card numbers respectively;

[0163] A construction unit, configured to construct a first training sample according to the communication feature of each first sample communication card number and its corresponding label, so as to obtain a first training sample set;

[0164] The construction unit is also used to construct a second training sample according to the communication characteristics of each second sample communication card number to obtain a second training sample set.

[0165] In some embodiments of the present application, the first training module includes:

[0166] A first input unit is used to input the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set into the initial first base classifier, the second base classifier and the third base classifier respectively to obtain a first prediction result, a second prediction result and a third prediction result;

[0167] A first calculation unit, used to calculate a loss function value according to the first prediction result, the second prediction result, and the third prediction result and their corresponding labels;

[0168] A first updating unit, used for updating the parameters of each base classifier according to the gradient information corresponding to the loss function value, and updating each base classifier according to the updated parameters of each base classifier;

[0169] A first determining unit, used for determining each updated base classifier as an initial base classifier;

[0170] The return unit is used to return the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set to the initial first base classifier, the second base classifier and the third base classifier respectively until the models of each base classifier converge to obtain each base classifier after the first stage of training.

[0171] In some embodiments of the present application, the first base classifier is a recurrent neural network, the second base classifier is a bidirectional long short-term memory network, and the third base classifier is a bidirectional gated recurrent network.

[0172] In some embodiments of the present application, the loss function is:

[0173]

[0174] Where y represents the sample label, p represents the probability value of the training sample output by the base classifier belonging to an abnormal communication card number, α represents the weight for balancing positive and negative samples, and r represents the degree of proximity between the training sample and the sample center;

[0175] Among them, r is obtained by the following formula:

[0176]

[0177] Among them, σ represents the Gaussian kernel function parameter, x represents the training sample, cent i Represents class C i The sample center point, C i represents the i-th label data set, and n is the total number of samples.

[0178] In some embodiments of the present application, the second training module includes:

[0179] The second input unit is used to input the samples in the second training sample set into each base classifier trained in the first stage to obtain the prediction results corresponding to each base classifier;

[0180] For each sample in the second training sample set, the following operations are performed cyclically until an expanded sample set for each base classifier is obtained:

[0181] An adding unit is used for obtaining extended sample data by taking the abnormal communication card number as a label of the sample when the prediction results of two base classifiers among the three base classifiers are abnormal communication card numbers and the prediction result of the other base classifier is a normal communication card number, and adding the extended sample data to the extended sample set of the other base classifier;

[0182] The following operations are performed for each base classifier until the preset stopping condition is reached, and the first base classifier, the second base classifier, and the third base classifier trained in the second stage are obtained:

[0183] A training unit, used for training the base classifier according to the samples in the sub-training sample set and the samples in the expanded sample set corresponding to the base classifier, to obtain a trained base classifier;

[0184] A second calculation unit is used to calculate the classification accuracy of the trained base classifier according to the prediction results of the base classifier for each sample and the label of each sample;

[0185] The second updating unit is used to randomly delete some samples from the expanded sample set corresponding to the base classifier to obtain an updated expanded sample set when the classification accuracy of the current round is less than the classification accuracy of the previous round;

[0186] The training unit is also used to train the base classifier according to the samples in the sub-training sample set corresponding to the base classifier and the samples in the updated expanded sample set to obtain a trained base classifier;

[0187] The setting unit is used to set the weights of each base classifier according to the classification accuracy of each trained base classifier to obtain an abnormal communication card number recognition model.

[0188] In some embodiments of the present application, the determination module 403 is specifically configured to:

[0189] The sum of the product of the probability value of the communication card number to be identified output by each base classifier of the abnormal communication card number recognition model belonging to the abnormal card number and the weight corresponding to each base classifier is used as the target abnormal probability value;

[0190] The sum of the product of the probability value of the communication card number to be identified output by each base classifier of the abnormal communication card number identification model belonging to a normal card number and the weight corresponding to each base classifier is used as the target normal probability value;

[0191] Determine whether the communication card number to be identified is abnormal based on the target abnormal probability value and the target normal probability value.

[0192] In some embodiments of the present application, the first acquisition module 401 is further used to:

[0193] Before using the sum of the products of the probability value that the communication card number to be identified output by each base classifier of the abnormal communication card number recognition model belongs to an abnormal card number and the weight corresponding to each base classifier as the target abnormal probability value, the weights of each base classifier are obtained. The weights of each base classifier are calculated based on the classification accuracy of each base classifier.

[0194] In some embodiments of the present application, the determination module 403 is specifically used to:

[0195] When the target abnormal probability value is greater than the target normal probability value, the communication card number to be identified is determined to be an abnormal card number;

[0196] When the target normal probability value is greater than the target abnormal probability value, the communication card number to be identified is determined to be a normal card number.

[0197] According to the abnormal communication card number identification device provided in the embodiment of the present application, on the one hand, each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, which can make full use of a large amount of unlabeled data for model training, solve the problem of uneven data distribution, enhance the model's robustness and generalization ability, and at the same time reduce manual intervention and reduce the cost of annotating data. The abnormal communication card number identification model can also combine the advantages of multiple base classifiers, thereby improving the recognition accuracy of abnormal communication card numbers. On the other hand, the loss function used when training each base classifier is based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to an abnormal communication card number, and the similarity between the sample and the sample center point. Among them, the similarity between the sample and the sample center point is obtained based on the Gaussian kernel function. The weight of the abnormal class samples in the loss function can be increased by controlling the weight coefficients of the positive and negative samples, and the weight of the normal class samples in the loss function can be reduced, thereby reducing the influence of the normal class samples on the model, so that the model pays more attention to the abnormal class samples, and achieves the effect of balancing the distribution of the data set. The similarity between the sample and the sample center point is integrated into the loss function, so that the loss function value is related to the similarity between the sample and the sample center point. Therefore, the model can be controlled to pay more attention to difficult-to-distinguish samples, and the proportion of easy-to-distinguish samples in the loss function calculation can be gradually reduced. The model converges faster and has stronger generalization ability, and can also further improve the recognition accuracy of the model. Therefore, based on the abnormal communication card number recognition model, the communication characteristics of the communication card number to be identified within a preset time period are identified to determine whether the communication card number to be identified is abnormal, which can improve the recognition accuracy of abnormal communication card numbers.

[0198] Figure 5 A schematic diagram of the structure of a device for identifying abnormal communication card numbers provided in an embodiment of the present application;

[0199] The device for identifying abnormal communication card numbers may include a processor 501 and a memory 502 storing computer program instructions.

[0200] Specifically, the processor 501 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0201] The memory 502 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 502 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 502 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 502 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 502 is a non-volatile solid-state memory.

[0202] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical or other physical / tangible memory storage devices. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the various method embodiments.

[0203] The processor 501 implements the method for identifying abnormal communication card numbers in the above embodiment by reading and executing computer program instructions stored in the memory 502 .

[0204] In one example, the device for identifying abnormal communication card numbers may further include a communication interface 503 and a bus 510. Figure 5 As shown, the processor 501, the memory 502, and the communication interface 503 are connected via a bus 510 and communicate with each other.

[0205] The communication interface 503 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0206] Bus 510 includes hardware, software or both, and the components of the determination device of the base station configuration parameters are coupled to each other. For example, but not limitation, the bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-end bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 510 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the present application considers any suitable bus or interconnection.

[0207] The abnormal communication card number identification device executes the abnormal communication card number identification method in the embodiment of the present application, thereby achieving Figure 1 , Figure 2 A method for identifying abnormal communication card numbers.

[0208] In addition, in combination with the method for identifying abnormal communication card numbers in the above embodiments, the present application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the methods for identifying abnormal communication card numbers in the above embodiments is implemented.

[0209] In combination with the method for identifying abnormal communication card numbers in the above embodiments, an embodiment of the present application also provides a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for identifying abnormal communication card numbers in any of the above embodiments.

[0210] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0211] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0212] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0213] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0214] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A method for identifying abnormal communication card numbers, characterized in that: The method comprises: Obtaining the communication characteristics of the communication card number to be identified within a preset time period; Input the communication characteristics of the communication card number to be identified into the first base classifier of the abnormal communication card number identification model to obtain a first abnormal probability value that the communication card number to be identified is an abnormal card number and a first normal probability value that it is a normal card number; input the communication characteristics of the communication card number to be identified into the second base classifier of the abnormal communication card number identification model to obtain a second abnormal probability value that the communication card number to be identified is an abnormal card number and a second normal probability value that it is a normal card number; input the communication characteristics of the communication card number to be identified into the third base classifier of the abnormal communication card number identification model to obtain a third abnormal probability value that the communication card number to be identified is an abnormal card number and a third normal probability value that it is a normal card number; wherein each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, and the loss function used when training each base classifier is obtained based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to the abnormal communication card number, and the similarity between the sample and the sample center point, and the similarity between the sample and the sample center point is obtained based on a Gaussian kernel function; Whether the communication card number to be identified is abnormal is determined based on the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier.

2. The method according to claim 1, characterized in that Before inputting the communication features of the communication card number to be identified into each base classifier of the abnormal communication card number identification model, the method further includes: Acquire a first training sample set and a second training sample set, wherein the first training sample set includes communication features of a communication card number and a label of the communication card number, wherein the label of the communication card number is used to identify an abnormal communication card number or a normal communication card number, and the second training sample set includes the communication features of the communication card number; Screening out a first sub-training sample set, a second sub-training sample set, and a third sub-training sample set having differences from the first training sample set; Performing first-stage training on the initial first base classifier, the second base classifier and the third base classifier according to the first sub-training sample set, the second sub-training sample set and the third sub-training sample set, respectively, to obtain the base classifiers after the first-stage training; For any one of the three base classifiers, perform the following operations to obtain a trained base classifier, set corresponding weights for each trained base classifier, and obtain an abnormal communication card number recognition model: perform a second stage of training on the base classifier trained in the first stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier.

3. The method according to claim 2, characterized in that The obtaining of the first training sample set and the second training sample set comprises: Collect communication data of a plurality of first sample communication card numbers and a plurality of second sample communication card numbers within a preset time period respectively, wherein the communication data of each of the first sample communication card numbers and each of the second sample communication card numbers includes at least one of the calling frequency, the calling and called ratio, the traffic usage data, the number of device changes corresponding to the communication card, the number of changes in the resident area, the number of changes in the commonly used base stations, the number of changes in the place of attribution, the registration place, the registration time, and the network access time; Preprocessing the communication data of the plurality of first sample communication card numbers and the plurality of second sample communication card numbers respectively to obtain communication features corresponding to the plurality of first sample communication card numbers and the plurality of second sample communication card numbers; Constructing a first training sample according to the communication feature of each first sample communication card number and its corresponding label to obtain the first training sample set; A second training sample is constructed according to the communication characteristics of each of the second sample communication card numbers to obtain the second training sample set.

4. The method according to claim 3, characterized in that The first stage of training the initial first base classifier, the second base classifier and the third base classifier according to the first sub-training sample set, the second sub-training sample set and the third sub-training sample set to obtain the base classifiers after the first stage of training includes: Input samples in the first sub-training sample set, the second sub-training sample set, and the third sub-training sample set into an initial first base classifier, a second base classifier, and a third base classifier, respectively, to obtain a first prediction result, a second prediction result, and a third prediction result; Calculate the loss function value according to the first prediction result, the second prediction result, the third prediction result and their corresponding labels; Update the parameters of each base classifier according to the gradient information corresponding to the loss function value, and update each base classifier according to the updated parameters of each base classifier; Determine each updated base classifier as the initial base classifier; Return and input the samples in the first sub-training sample set, the second sub-training sample set and the third sub-training sample set into the initial first base classifier, the second base classifier and the third base classifier respectively until the models of each base classifier converge, thereby obtaining each base classifier after the first stage of training.

5. The method according to claim 4, characterized in that The loss function is: Where y represents the sample label, p represents the probability value of the training sample output by the base classifier belonging to an abnormal communication card number, α represents the weight for balancing positive and negative samples, and r represents the degree of proximity between the training sample and the sample center; Among them, r is obtained by the following formula: Among them, σ represents the Gaussian kernel function parameter, x represents the training sample, cent i Represents class C i The sample center point, C i represents the i-th label data set, and n is the total number of samples.

6. The method according to claim 2, characterized in that For any one of the three base classifiers, the following operations are performed respectively to obtain a trained base classifier, and corresponding weights are set for each trained base classifier to obtain an abnormal communication card number recognition model: the base classifier trained in the first stage is trained in the second stage according to the corresponding sub-training sample set and the second training sample set to obtain a trained base classifier, including: Inputting samples in the second training sample set into each base classifier trained in the first stage respectively to obtain prediction results corresponding to each base classifier; For each sample in the second training sample set, the following operations are performed cyclically until an expanded sample set of each base classifier is obtained: In the case where the prediction results of two base classifiers among the three base classifiers are abnormal communication card numbers and the prediction result of the other base classifier is a normal communication card number, the abnormal communication card number is used as the label of the sample to obtain expanded sample data, and the expanded sample data is added to the expanded sample set of the other base classifier; The following operations are performed for each base classifier respectively until a preset stop condition is reached, thereby obtaining the first base classifier, the second base classifier, and the third base classifier trained in the second stage: The base classifier is trained according to the samples in the sub-training sample set and the samples in the expanded sample set corresponding to the base classifier to obtain a trained base classifier; According to the prediction results of the base classifier for each sample and the label of each sample, the classification accuracy of the trained base classifier is calculated; When the classification accuracy of this round is less than that of the previous round, some samples are randomly deleted from the expanded sample set corresponding to the base classifier to obtain an updated expanded sample set; The base classifier is trained according to the samples in the sub-training sample set corresponding to the base classifier and the samples in the updated expanded sample set to obtain a trained base classifier; The weights of the base classifiers are set according to the classification accuracy of the trained base classifiers to obtain the abnormal communication card number recognition model.

7. The method according to any one of claims 1 to 6, characterized in that: The determining whether the communication card number to be identified is abnormal according to the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier includes: The sum of the products of the probability value of the communication card number to be identified belonging to an abnormal card number output by each base classifier of the abnormal communication card number identification model and the weight corresponding to each base classifier is used as the target abnormal probability value; The sum of the product of the probability value of the communication card number to be identified belonging to a normal card number output by each base classifier of the abnormal communication card number identification model and the weight corresponding to each base classifier is used as the target normal probability value; Whether the communication card number to be identified is abnormal is determined based on the target abnormal probability value and the target normal probability value.

8. A device for identifying abnormal communication card numbers, characterized in that: The device comprises: A first acquisition module is used to acquire the communication characteristics of the communication card number to be identified within a preset time period; An input module is used to input the communication characteristics of the communication card number to be identified into the first base classifier of the abnormal communication card number identification model to obtain a first abnormal probability value that the communication card number to be identified is an abnormal card number and a first normal probability value that the communication card number to be identified is a normal card number; input the communication characteristics of the communication card number to be identified into the second base classifier of the abnormal communication card number identification model to obtain a second abnormal probability value that the communication card number to be identified is an abnormal card number and a second normal probability value that the communication card number to be identified is a normal card number; input the communication characteristics of the communication card number to be identified into the third base classifier of the abnormal communication card number identification model to obtain a third abnormal probability value that the communication card number to be identified is an abnormal card number and a third normal probability value that the communication card number to be identified is a normal card number; wherein each base classifier of the abnormal communication card number identification model is trained using a semi-supervised method, and the loss function used when training each base classifier is obtained based on the weight coefficients of positive and negative samples, the probability that the sample output by the base classifier belongs to the abnormal communication card number, and the similarity between the sample and the sample center point, and the similarity between the sample and the sample center point is obtained based on a Gaussian kernel function; The determination module is used to determine whether the communication card number to be identified is abnormal according to the abnormal probability value and the normal probability value output by each base classifier and the weight corresponding to each base classifier.

9. A device for identifying abnormal communication card numbers, characterized in that: The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for identifying an abnormal communication card number as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for identifying an abnormal communication card number as described in any one of claims 1 to 7 is implemented.

Images