A method and apparatus for pushing associated information of a communication exception

Abstract

Embodiments of the present application disclose a method and device for pushing associated information of communication exception, to improve the effectiveness of pushing associated information of communication exception. The embodiments of the present application include: obtaining user online information of a communication user authentication authorization charging system and information of an abnormal communication device of a communication alarm platform; performing comparison on the user online information and the information of the abnormal communication device to determine communication alarm information, the communication alarm information including abnormal communication data of the abnormal communication device and information of a communication user carried by the abnormal communication device; inputting the communication alarm information into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint, a training sample of the communication complaint prediction model including sample communication data of a sample communication device, and a training label of the communication complaint prediction model representing whether a sample communication user carried by the sample communication device initiates a communication complaint; and pushing associated information of an abnormal complaint to the abnormal communication device of a target communication user.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of communication, and in particular to a method and device for pushing associated information of communication exception. BACKGROUND

[0002] With the growth of home broadband services, communication services have higher requirements for active care of home broadband faults. A new complaint user analysis strategy is urgently needed to achieve accurate reach of associated information, improve accuracy, and reduce the amount of noise information for users.

[0003] In a communication exception scenario, one exception can affect multiple communication users. If the communication users affected by the exception are not accurately located, the pushing of associated information of the communication exception can cause unnecessary interference to users not affected by the exception. Too much associated information can cause information noise to users, resulting in a decrease in user satisfaction.

[0004] How to improve the effectiveness of the pushing of associated information of communication exception is a technical problem to be solved by the present application. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a method and device for pushing associated information of communication exception, so as to improve the effectiveness of the pushing of associated information of communication exception.

[0006] In a first aspect, a method for pushing associated information of communication exception is provided, comprising:

[0007] Obtaining user online information of a communication user authentication authorization charging system and information of an abnormal communication device of a communication alarm platform;

[0008] Comparing the user online information and the information of the abnormal communication device to determine communication alarm information, wherein the communication alarm information includes abnormal communication data of the abnormal communication device and information of a communication user carried by the abnormal communication device;

[0009] Inputting the communication alarm information into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint, wherein a training sample of the communication complaint prediction model includes sample communication data of a sample communication device, and a training label of the communication complaint prediction model represents whether a sample communication user carried by the sample communication device initiates a communication complaint;

[0010] Pushing associated information of an abnormal complaint to an abnormal communication device of a target communication user, wherein the target communication user is a communication user who initiates a communication complaint according to the prediction result of the communication complaint prediction model.

[0011] In a second aspect, a device for pushing associated information of communication exception is provided, comprising:

[0012] an acquisition module, configured to acquire user online information of a communication user authentication authorization charging system and information of an abnormal communication device of a communication alarm platform;

[0013] a determination module, configured to perform comparison between the user online information and the information of the abnormal communication device, and determine communication alarm information, the communication alarm information including abnormal communication data of the abnormal communication device and information of a communication user carried by the abnormal communication device;

[0014] a prediction module, configured to input the communication alarm information into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint, wherein a training sample of the communication complaint prediction model includes sample communication data of a sample communication device, and a training label of the communication complaint prediction model represents whether a sample communication user carried by the sample communication device initiates a communication complaint;

[0015] a pushing module, configured to push associated information of an abnormal complaint to an abnormal communication device of a target communication user, the target communication user being a communication user who initiates a communication complaint according to the prediction result of the communication complaint prediction model.

[0016] In a third aspect, an electronic device is provided, which includes a processor, a memory, and a computer program stored in the memory and executable in the processor, and when the computer program is executed by the processor, the steps of the method according to the first aspect are implemented.

[0017] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the steps of the method according to the first aspect are implemented.

[0018] In a fifth aspect, a computer program product is provided, which includes a non-transitory computer readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform some or all of the steps of the method according to the first aspect.

[0019] In the embodiment of the present application, the user online information of the communication user authentication authorization charging system and the information of the abnormal communication equipment of the communication alarm platform are acquired; the user online information and the information of the abnormal communication equipment are compared to determine communication alarm information, the communication alarm information including abnormal communication data of the abnormal communication equipment and information of a communication user borne by the abnormal communication equipment; the communication alarm information is input into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint, wherein the training sample of the communication complaint prediction model includes sample communication data of a sample communication equipment, and the training label of the communication complaint prediction model represents whether a sample communication user borne by the sample communication equipment initiates a communication complaint; and the associated information of the abnormal complaint is pushed to the abnormal communication equipment of a target communication user, the target communication user being a communication user predicted by the communication complaint prediction model to initiate a communication complaint. The user online information of the communication user authentication authorization charging system and the information of the abnormal communication equipment of the communication alarm platform can improve the accuracy of the determined communication alarm information from the dimensions of online users and equipment faults, determine which communication users are affected by the abnormal communication equipment, and effectively narrow the range of communication users affected by the abnormality. The pre-trained communication complaint prediction model can comprehensively predict the probability of the user initiating a communication complaint, thereby improving the prediction accuracy. Furthermore, the associated information of the abnormal complaint is pushed to the target communication user predicted to initiate a communication complaint, which can realize targeted active care for the target communication user, pacify the communication user affected by the communication abnormality, avoid disturbing normal communication users not affected by the abnormality, reduce the amount of noise information of the user, and effectively improve the effectiveness of the push of the associated information of the communication abnormality. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are included to provide a further understanding of the present application, constitute a part of the present application, illustrate the exemplary embodiments of the present application and their description serve to explain the present application, and do not constitute improper limitations on the present application. In the drawings:

[0021] Figure 1a is one of flow diagrams of a communication abnormality associated information push method according to an embodiment of the present application;

[0022] Figure 1b is a scene flow diagram of a communication abnormality associated information push method according to an embodiment of the present application;

[0023] Figure 2 is one of flow diagrams of a communication abnormality associated information push method according to an embodiment of the present application;

[0024] Figure 3 is one of flow diagrams of a communication abnormality associated information push method according to an embodiment of the present application;

[0025] Figure 4 is a flowchart of a fourth embodiment of a method for pushing associated information of a communication exception;

[0026] Figure 5 is a flowchart of a fifth embodiment of a method for pushing associated information of a communication exception;

[0027] Figure 6 is a structural diagram of an apparatus for pushing associated information of a communication exception. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application. The figure numbers in the present application are only used to distinguish each step in the scheme, and are not used to limit the execution order of each step, and the specific execution order is subject to the description in the specification.

[0029] In the field of communication, the number of customers of home broadband services and the scale of networks develop rapidly. In some application scenarios, home broadband services cover a wide range, and once a home broadband service device fails, it may cause the home broadband services of thousands of home broadband users to be affected. Although the service device failure can be found and repaired according to failure alarms, user complaints and the like, this way only handles the failure at the service device side, and the user end only passively accepts the failure repair. In this way, the user cannot know the failure repair progress, which may cause the user to be dissatisfied with the home broadband service, and the user end cannot know the failure recovery information in the first time.

[0030] The scheme provided by the embodiments of the present application can be used to improve the effectiveness of pushing associated information of a communication exception, and is especially suitable for a communication service application scenario. After a communication service device fails, the scheme provided by the embodiments of the present application can efficiently realize effective pushing of associated information of the occurring communication exception, so that the users affected by the communication exception can receive the required pushed information, high-quality user care is realized, user satisfaction is improved, and information noise is avoided for normal users who are not affected.

[0031] In order to solve the problems in the prior art, the embodiments of the present application provide a method for pushing associated information of a communication exception, as shown in Figure 1a The method comprises the following steps:

[0032] S11: Obtain user online information of a communication user authentication authorization charging system and information of an abnormal communication device of a communication alarm platform.

[0033] The above-mentioned communication user authentication authorization accounting (AAA) system has authentication, authorization and accounting functions. The authentication function includes confirming the identity of the communication user accessing the network and judging whether the communication user is legal. The authorization function includes giving different permissions to different communication users and limiting the services that the communication user can use. The accounting function includes recording all operations of the communication user during the use of the service, and can realize the charging requirements for time and traffic. The user online information obtained through the AAA system can represent the information of the communication user in the online state, and can be used to determine which communication users are affected by the abnormality of the communication device in the subsequent steps.

[0034] The above-mentioned communication alarm platform can be used to monitor various communication parameters, and can be used to monitor the running state of the communication service device, and can also be used to monitor the fluctuation of the number of communication users, communication traffic and other parameters, so as to generate corresponding alarm information when an abnormality is monitored. The information of the abnormal communication device can include the identification of the communication device and the type of the abnormality. In actual application, the abnormality of the communication device can be various, which can include the abnormality of the running state of the communication device itself, and can also include the abnormality of the related parameters of the communication user carried by the communication device.

[0035] S12: Comparing the user online information and the information of the abnormal communication device, determining the communication alarm information, the communication alarm information including the abnormal communication data of the abnormal communication device and the information of the communication user carried by the abnormal communication device.

[0036] In this step, the user online information of the user AAA system and the information of the abnormal communication device of the alarm platform are obtained to determine the information of the communication user affected by the abnormality. Specifically, the influence range can be analyzed by analyzing the online user carried by the abnormal communication device and the user covered by the device in the asset management system.

[0037] Optionally, the AAA system online user number fluctuation is detected through the monitoring device, combined with the centralized alarm platform, to accurately locate and confirm the fault category. Optionally, the accuracy of fault range confirmation can be improved by referring to the data in the integrated resource device management system.

[0038] Wherein, the user account, user terminal MAC (Media Access Control), ONU (Optical Network Unit) information, PON (Passive Optical Network) port information, OLT (Optical line termina) IP (Internet Protocol) and other information carried by PPPOE+ can be obtained by analyzing the message through the AAA system using PPPoE (Point-to-Point Protocol over Ethernet) + and Radius technology, realizing real-time collection of the number of online users of the OLT, and determining the abnormal communication data of the abnormal communication equipment and the information of the communication user carried thereby.

[0039] S13: input the communication alarm information into the pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint, wherein the training sample of the communication complaint prediction model includes sample communication data of a sample communication equipment, and the training label of the communication complaint prediction model represents whether a sample communication user carried by the sample communication equipment initiates a communication complaint.

[0040] In this example, the communication complaint prediction model is pre-trained by a training sample including sample communication data and a corresponding training label representing whether a sample communication user initiates a communication complaint. In this step, the communication alarm information is input into the pre-trained communication complaint prediction model, and the model predicts the probability of the communication user initiating a communication complaint according to the abnormal communication data of the abnormal communication equipment and the information of the communication user carried thereby, thereby outputting the prediction result of whether the communication user initiates a communication complaint.

[0041] Wherein, the pre-trained communication complaint prediction model can comprehensively predict the communication abnormal features expressed by the abnormal communication data of the abnormal communication equipment and the features of the communication user carried thereby, and output the probability of the communication user initiating a communication complaint under the corresponding communication abnormal condition.

[0042] S14: push the associated information of the abnormal complaint to the abnormal communication equipment of the target communication user, wherein the target communication user is the communication user predicted to initiate a communication complaint by the communication complaint prediction model.

[0043] The associated information of the abnormal complaint can be information associated with the abnormal complaint and can include a user-side processing method for communication abnormality, communication abnormality repair progress, communication abnormality compensation, and the like. By pushing the associated information to the target communication user, active care can be provided to the communication user affected by the communication abnormality, the user's mood can be soothed, and the user's trust in the communication service can be improved.

[0044] In the scheme provided by the present application, the user online information of the communication user authentication authorization charging system and the information of the abnormal communication device of the communication alarm platform can improve the accuracy of the determined communication alarm information from the online user dimension and the device fault dimension, determine which communication users are affected by the abnormal communication device, and effectively narrow the range of communication users affected by the abnormality. The pre-trained communication complaint prediction model can comprehensively predict the probability of user initiating a communication complaint, thereby improving the prediction accuracy. Furthermore, pushing the associated information of the abnormal complaint to the predicted target communication user who will initiate a communication complaint can realize active care for the target communication user, sooth the communication user affected by the communication abnormality, and avoid disturbing normal communication users not affected by the abnormality, reduce the amount of noise information of the user, and effectively improve the effectiveness of pushing the associated information of the communication abnormality.

[0045] In the following, the present application will be described in detail in combination with Figure 1b The present application will be described in detail in combination with

[0046] The scheme provided by the present application can be applied to a communication system comprising a AAA system and a communication alarm platform, wherein the communication alarm platform can also be referred to as a centralized alarm platform. In actual application, device detection can be performed continuously or periodically based on a preset time period. On the one hand, user online information is obtained from the AAA system to obtain user online status, and on the other hand, information of abnormal communication devices is obtained from the centralized alarm platform to obtain information such as which OLT devices alarm and which device users fluctuate abnormally.

[0047] Then, the user online information and the information of the abnormal communication device are compared, i.e., the user online status of the AAA system and the device fault status are compared, so as to determine which communication users are affected by the abnormal communication device and determine the abnormal communication data of the abnormal communication device and the information of the communication users carried thereby.

[0048] Optionally, detailed information of the abnormal communication device and detailed information of the communication users carried thereby can also be obtained from a comprehensive resource device management system, further improving the accuracy of the communication alarm information.

[0049] Subsequently, the communication alarm information is input into a pre-trained communication complaint prediction model, also referred to as an alarm user complaint prediction model, which predicts the possibility of the communication user initiating a communication complaint according to the input communication alarm information, thereby determining the target communication user who will initiate a communication complaint.

[0050] Further, a short message containing the associated information of the abnormal complaint is pushed to the target communication user through a short message gateway or the like, so as to realize active care for the home broadband user affected by the communication abnormality.

[0051] The training sample of the alarm user complaint prediction model includes sample communication data of a sample communication device, and the training label represents whether the sample communication user carried by the sample communication device initiates a communication complaint. The training sample can be constructed based on multi-dimensional information. For example, the training sample can be constructed according to the communication alarm information of the centralized alarm platform and the communication device information and communication user information provided by the integrated resource device management system. The training sample can also be constructed according to the EOMS (Electric Operation Maintenance System) work order information, user basic attributes, and user consumption behavior characteristics. In addition, the training label corresponding to the training sample can be constructed according to the user complaint behavior characteristics.

[0052] Through the scheme provided in the embodiments of the present application, the accuracy of the abnormal influence range can be improved through the AAA system and the centralized alarm platform, and the prediction accuracy of whether the communication user complains can be effectively improved through the model.

[0053] Based on the scheme provided in the above embodiments, as shown in Figure 2 Before the step S13, that is, before the communication alarm information is input into the pre-trained communication complaint prediction model, the method further includes:

[0054] S21: Obtain original communication data of a sample communication device and complaint behavior information of whether a sample communication user carried by the sample communication device initiates a communication complaint.

[0055] The original communication data can include communication data on the communication service device side and communication data on the communication user side. The original communication data can be time-based parameter information related to communication functions, which can be used to represent the communication state of the user side device and the communication state of the communication service device side.

[0056] The complaint behavior information can be generated according to the historical complaint records of the communication user initiating a communication complaint. The complaint behavior information can represent the information of the communication user initiating a complaint, and specifically can include the identifier of the complaint user, the behavior record of the complaint user, the historical complaint record of the complaint user, and the like.

[0057] S22: Extract a communication behavior feature of the sample communication device from the original communication data, the communication behavior feature comprising a user communication behavior feature and / or a device communication behavior feature.

[0058] For example, the user communication behavior feature extracted from the communication user side communication data can comprise: a user home broadband consumption behavior feature (a usage package level, a last month consumption amount, a last month consumption amount exceeding the package amount, a last month online traffic exceeding amount, etc.); a home broadband online behavior feature (an online duration, a concentrated use time period, etc.); an active seeking customer service willingness feature (contact behavior with an artificial customer service, a number of times of calling the customer service, a number of times of submitting a work order, a number of times of handling failure, etc.); a user basic attribute (gender, age, occupation, etc.).

[0059] Optionally, the user communication behavior feature comprises at least one of the following: a user consumption behavior feature, a user network behavior feature, and a user communication consulting behavior feature.

[0060] The device communication behavior feature extracted from the communication service device side can comprise: a home broadband fault type, a home broadband fault duration, a home broadband fault reason, a device type, a device fault level, a number of times of device fault in a certain time period, etc.

[0061] Optionally, the device communication behavior feature comprises at least one of the following: a communication abnormality type, a communication abnormality duration, a communication abnormality reason, a communication abnormality level, and a communication abnormality frequency.

[0062] S23: Training a prediction model based on a training sample constructed based on the communication behavior feature and a training label constructed based on the complaint behavior information, to obtain the communication complaint prediction model.

[0063] The type of the prediction model can be flexibly set according to actual needs, and in the embodiments of the present application, an XGBoost algorithm model is taken as an example for illustration. In the application scenario of the embodiments of the present application, the data contained in the complaint fault is relatively complex, and can specifically comprise data of the user end, the monitoring end, and the device end. The XGBoost algorithm is good at capturing the dependency relationship between complex data. In addition, the XGBoost can automatically process the default value. Moreover, the XGBoost uses the second derivative, has high precision and fast speed. The objective function of the XGBoost is composed of a loss function and a regular function, which can avoid underfitting and overfitting. Based on this, the scheme provided in the embodiments of the present application uses the XGBoost algorithm model, which can effectively improve the accuracy of the fault complaint prediction.

[0064] Specifically, XGBoost is a method based on additive model, the basic idea is that M base learners are added, where each learner is a regression tree, the number of features of the i-th sample is x i The predicted value of the i-th sample in M samples is is expressed as:

[0065]

[0066] It is optimized by forward distribution algorithm, t represents a certain tree, and w represents the value of each leaf node. The forward distribution algorithm can be expressed as:

[0067]

[0068]

[0069]

[0070] The XGBoost objective function is composed of a loss function and a regularization function, where the definition of the loss function is the accumulation of the difference between the actual result y and the predicted result to avoid underfitting, and the loss function is expressed as:

[0071]

[0072] The regularization function is the accumulation of the complexity of multiple trees to avoid overfitting, and the regularization function is expressed as:

[0073]

[0074] The objective function formula composed of the loss function and the regularization function is:

[0075]

[0076] The regularization function formula is:

[0077]

[0078] Where T is the number of leaf nodes, γ and λ are hyperparameters used to punish overly complex situations, T is the number of leaf nodes, and w is the node value.

[0079] Based on this, the forward distribution algorithm based on additive model can be expressed as:

[0080]

[0081] Where the previously optimized decision tree, the number of leaf nodes and the w node value are determined, so The value of is determined and recorded as a constant constant, which does not participate in the optimization of the objective function, based on which the optimization function can be expressed as:

[0082]

[0083] It can be seen that in the optimization function, the optimization effect of the regularization function is only related to the current optimization of the sub-node value w and the number of leaf nodes T.

[0084] The loss function is calculated by the square error loss, and the difference between the predicted value and the actual node value is squared. The loss function in the objective function can be expressed as:

[0085]

[0086] The sum of the losses of different sets I in all trees T can be expressed as:

[0087]

[0088] In addition, based on the forward distribution algorithm of the above addition model, the sum of the losses of different sets I in all trees T can be expressed as:

[0089]

[0090] Substitute the sum of the losses of different sets I in all trees T into the objective function, which can be expressed as:

[0091]

[0092] Using the second-order Taylor expansion, the node value w is separated out, and the formula is simplified to:

[0093]

[0094] Let G i＝ ∑ i∈I g i (All first-order gradient i nodes) H i＝ ∑ i∈I h i (All second-order gradient i nodes), the formula is simplified to:

[0095]

[0096] Since the loss function must have a minimum value to calculate the minimum loss, it is a convex function, and the second derivative of the convex function is greater than 0, so H i >0, G i >0, so Substitute it into the formula of the above objective function to represent it as:

[0097]

[0098] predicted value of the output of the XGBoost model It can be any value between (-∞, ∞). For the complaint prediction evaluation in the application scenario of the embodiment of the application, the output value can be introduced into a Logistic function to convert the output into the range of (0, 1) based on the above output value. Wherein is the probabilistic output, and the threshold value α = 0.5 is selected to obtain the final prediction result, which is used to represent whether the communication user will initiate a communication complaint.

[0099] In this way, the output result of the XGBoost model can be converted into two categories. When close to 1, the model has a higher degree of certainty in classifying the sample as 1, and when close to 0, the model has a lower degree of certainty in classifying the sample as 0.

[0100] Through the scheme provided by the embodiment of the application, the communication complaint prediction model is trained based on XGBoost, the abnormal communication data containing the abnormal communication equipment and the information of the communication user carrying the abnormal communication data are input into the pre-trained communication complaint prediction model, the user with a high complaint probability is output, and then the fault care information is sent to the home broadband user. The present scheme adopts a regression decision tree and selects XGBoost improved based on GBDT as an algorithm model. It can be used for a complex task to appropriately integrate the judgments of multiple experts, thereby improving the accuracy of the judgment result.

[0101] Based on the scheme provided in the above embodiment, optionally, as shown in Figure 3 Before the above step S22, that is, before the communication behavior features of the sample communication equipment are extracted from the original communication data, the method further includes:

[0102] S31: performing normalization processing on the original communication data and the complaint behavior information to obtain normalized original communication data and normalized complaint behavior information.

[0103] This step performs preprocessing on the original communication data and the complaint behavior information by normalization. Specifically, the original communication data and the complaint behavior information are expressed in numerical form, and the numerical values are converted into values between 0 and 1. In this way, various data can have a unified measurement scale.

[0104] Optionally, based on the complaint behavior information, the user initiating a complaint can be represented as 1, and the user not initiating a complaint can be represented as 0. Further, the content of the user's communication with the customer service can be used to assist in optimizing the corresponding feature value. For example, the user does not initiate a complaint, but expresses negative emotions in communication with the customer service, and the corresponding feature value can be 0.5.

[0105] In this example, the data sent into the model defines the following examples: (user overdue amount), a (home broadband user offline duration), β (recently sent similar SMS time duration), γ (unsubscribe time duration), δ (user blacklist existence duration).

[0106] The normalized processing value is as follows (taking the user overdue data as an example):

[0107]

[0108] The advantage of normalization processing is that the processed input data can be used as parameters in a controllable range. The prediction model needs to accurately judge whether each data will lead to complaints. The normalized model can clearly express the correlation between different data indicators and output results, that is, the influence of data on whether the user initiates a complaint, which is beneficial to the model learning the correlation between each data and whether the user complains.

[0109] S32: Based on the data missing rate, the normalized original communication data and the normalized complaint behavior information are subjected to data screening to obtain screened original communication data and screened complaint behavior information.

[0110] In this step, data with a missing rate greater than a preset missing rate is screened out to exclude data that has lost reference value. For example, data features with a missing rate greater than 90% are screened out. After removing data with a large missing rate, for original data with continuity, the XGBoost algorithm can be used to fill in the missing values, and for non-continuous data features (such as age, gender, etc. Categorical data), the missing values can be filled in by using the mode. This step can optimize the overall effectiveness of the data by data removal and filling.

[0111] In the above step S22, the communication behavior features of the sample communication device are extracted from the original communication data, including:

[0112] S33: The communication behavior features of the sample communication device are extracted from the screened original communication data and the screened complaint behavior information original communication data.

[0113] The scheme provided by the embodiments of the present application can effectively optimize the quality of the training data through normalization and missing data removal processing, simplify the data features, delete redundant information, reduce the dimension of the data, facilitate the import of the data into the algorithm model, improve the model learning efficiency, and optimize the communication complaint prediction accuracy of the trained model.

[0114] Based on the scheme provided in the above embodiments, optionally, Figure 4As shown, in the step S33, the prediction model is trained based on the training sample constructed based on the communication behavior feature and the training label constructed based on the complaint behavior information, to obtain the communication complaint prediction model, including:

[0115] S41: constructing a training sample based on at least one segmentation point and the communication behavior feature.

[0116] In this step, the normalized data feature values are segmented and positioned multiple times. For example, for the communication behavior feature value, the 0.1, 0.3, and 0.7 in each group of data are set as segmentation points, and the communication behavior feature value is segmented into three groups: less than or equal to 0.3, greater than 0.3 and less than or equal to 0.7, and greater than 0.7. It is assumed that 0.1 to 0.3 is considered as no complaint by the user, 0.3 to 0.7 needs to be judged according to other parameter conditions whether the user will complain, and greater than 0.7 will complain. Through this step, 0.1, 0.3, and 0.7 in each group of data are set, and finally each parameter is transmitted into the model prediction in a similar manner.

[0117] S42: constructing a training label based on the at least one segmentation point and the complaint behavior information.

[0118] In this step, the training label can be set according to actual needs, for example, the complaint behavior information with a feature value greater than 0.5 is determined as a complaint behavior, and the complaint behavior information with a feature value less than or equal to 0.5 is determined as a non-complaint behavior, thereby constructing a binary classification model. In actual application scenarios, a classification between complaint and non-complaint can also be set to represent that the communication user has a complaint intention but does not execute the complaint.

[0119] S43: training a prediction model based on the training sample and the training label.

[0120] Through the scheme provided in the embodiments of the present application, in order to make each decision model reliable and the number of segmentations controllable, the number of segmentations can be positioned multiple times, and the data fluctuation after the basic model experiences decision is controlled to be not too large, so as to optimize the model training effect.

[0121] Based on the scheme provided in the above embodiments, optionally, as shown in Figure 5 As shown, before the step S41, that is, before the training sample is constructed based on the at least one segmentation point and the communication behavior feature, the method further includes:

[0122] S51: performing numerical segmentation on the communication behavior feature based on at least one segmentation point, to obtain a communication behavior feature segmentation group corresponding to the segmentation point.

[0123] For example, it is assumed that the feature value accuracy is 0.1, and the 0.3 and 0.7 in each group of data are set as segmentation points, thereby delimiting the communication behavior feature segmentation group as less than or equal to 0.3, greater than 0.3 and less than or equal to 0.7, and greater than 0.7.

[0124] S52: determine a corresponding first entropy value for the value of the communication behavior feature, and determine a corresponding second entropy value for each of the plurality of communication behavior feature segmentation groups.

[0125] In this step, the entropy values of the data before and after segmentation are calculated respectively, and the entropy values are determined according to the following formula:

[0126]

[0127] Where the entropy of the initial data is recorded as the first entropy value, and the second entropy value is calculated for each communication behavior feature segmentation group after segmentation.

[0128] S53: Adjust the at least one cut point based on the similarity between the weighted average of the plurality of second entropy values and the first entropy value.

[0129] In this step, the weighted average of the second entropy values corresponding to the plurality of communication behavior feature segmentation groups is calculated, and the similarity with the first entropy value is compared. The greater the similarity, the smaller the data fluctuation after segmentation.

[0130] In this step, if the similarity is less than the preset similarity, the cut point position or the cut point number is adjusted, and then the entropy value is calculated using the adjusted cut point to determine the data fluctuation degree, until the similarity between the adjusted first entropy value and the weighted average of the plurality of second entropy values is greater than or equal to the preset similarity.

[0131] In practical applications, the adjustment can be based on a preset step size, for example, the segmentation process initially increases by 0.1 every 0.2, and only a few times of entropy calculation are needed to obtain the minimum data fluctuation after segmentation.

[0132] Optionally, the number of compressed features and the number of candidate cut points can also be optimized. In terms of feature number, a random layer-by-layer approach is used to calculate and learn all trees to improve prediction accuracy. In terms of candidate cut points, the weighted quantile method is used to optimize the cut points and compress the number of candidate cut points. A local strategy is used to maximize the learning time and ensure reasonable selection of cut points.

[0133] According to the optimization result, the parameters are adjusted, for example, the final result sets the learning eta in the XGBoost model parameters to 0.001, the tree depth to 12, the nround to 160, the regular term weight reg_alpha to 0.8, and the rest of the parameters to default parameters. 70% of the samples are used as the training set for training, and the other 30% of the samples are used as the test set. The final output of the data features with high relative value is as follows:

[0134]

[0135] Based on the above parameter adjustment results, the data of the test set is placed in the 160 CART trees generated by the training set to make a prediction, the sum of the predicted values on all leaf nodes is calculated, the predicted result is calculated according to the formula, and the transformation is performed through the logistic function:

[0136]

[0137] If the probability is greater than 0.5, the predicted score is 1, that is, the complaint risk is high, the target communication user is determined, and the associated information of the abnormal complaint is pushed to the target communication user; otherwise, it is 0, and the associated information of the abnormal complaint does not need to be sent.

[0138]

[0139] For the complaint prediction result judgment of this binary classification problem, the present application selects four evaluation indexes of accuracy (Accuracy), recall (Recall), receiver operating characteristic curve (ROC) and area under the ROC curve (AUC) to evaluate the effect of the model. In the test phase, the test data set is used, and the complaint probability of the home broadband user is predicted according to the above model, and the accuracy of the predicted result is compared and judged. In the test set, the confusion matrix can obtain the accuracy of the model 91.05%, the recall of the complaint user is 82.01%, and the accuracy of the model meets the requirement that the accuracy is more than 85%. It can be seen from the prediction result of the test set that the generalization ability of the model is good.

[0140] In the validation set, according to the confusion matrix of the validation set, the accuracy of the model on the validation set is calculated as 92.16%, the recall of the complaint user is 79.7%, the accuracy of the model is high, and can meet the requirement that the accuracy is more than 85%, and the generalization ability of the model is good.

[0141] Based on the pre-trained communication complaint prediction model, in the model application stage, the communication alarm information is also normalized and preprocessed according to the above example, and then the preprocessed communication alarm information is input into the pre-trained communication complaint prediction model to obtain the user complaint probability output by the model. If the probability of belonging to the positive class is greater than 0.5 and closer to 1, the predicted score is 1, then the complaint probability is large, and the associated information of the abnormal complaint needs to be sent to the user. In addition, in the process of actual application, a timing data processing process needs to be set up, and the prediction result of the model is recorded in the database, which can facilitate the developers to retrieve the prediction result, and use the result to send the associated information of the abnormal complaint to the user with high complaint risk.

[0142] Through the scheme provided by the embodiment of the application, by comparing the online user fluctuation data of the AAA with the information of the centralized alarm platform, and in combination with the data of the resource device management system, the influence range of the faulty user can be accurately located. Moreover, through the pre-trained model, the user with high complaint probability can be efficiently and accurately predicted. The model is trained based on training samples generated based on multiple data sources, which not only includes device data on the user side, but also includes data on the communication service device side. The scheme uses an improved XGBoost algorithm, and the effect of machine learning is better, and the prediction result is more accurate.

[0143] In order to solve the problems in the prior art, the embodiment of the application also provides a device 60 for pushing associated information of communication exception, as shown in Figure 6 The device 60 comprises:

[0144] An acquisition module 61 acquires user online information of a communication user authentication authorization charging system and information of an abnormal communication device of a communication alarm platform;

[0145] A determination module 62 performs comparison between the user online information and the information of the abnormal communication device, and determines information of a communication user borne by the abnormal communication device. The communication alarm information comprises abnormal communication data of the abnormal communication device and the information of the communication user borne by the abnormal communication device;

[0146] A prediction module 63 inputs the communication alarm information into a pre-trained communication complaint prediction model, and obtains a prediction result of whether the communication user initiates a communication complaint. The training sample of the communication complaint prediction model comprises sample communication data of a sample communication device, and the training label of the communication complaint prediction model represents whether a sample communication user borne by the sample communication device initiates a communication complaint;

[0147] A pushing module 64 pushes associated information of an abnormal complaint to an abnormal communication device of a target communication user. The target communication user is a communication user who initiates a communication complaint according to the prediction of the communication complaint prediction model.

[0148] The device provided by the embodiment of the application can obtain user online information of a communication user authentication authorization charging system and information of an abnormal communication device of a communication alarm platform, perform comparison on the user online information and the information of the abnormal communication device, determine communication alarm information, the communication alarm information including abnormal communication data of the abnormal communication device and information of a communication user borne by the abnormal communication device, input the communication alarm information into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint, wherein a training sample of the communication complaint prediction model includes sample communication data of a sample communication device, and a training label of the communication complaint prediction model represents whether a sample communication user borne by the sample communication device initiates a communication complaint, and push associated information of an abnormal complaint to an abnormal communication device of a target communication user, the target communication user being a communication user who initiates a communication complaint according to the prediction result output by the communication complaint prediction model. The user online information of the communication user authentication authorization charging system and the information of the abnormal communication device of the communication alarm platform can improve the accuracy of the determined communication alarm information from the dimensions of online users and device faults, determine which communication users are affected by the abnormal communication device, and effectively narrow the range of the communication users affected by the abnormality. The pre-trained communication complaint prediction model can comprehensively predict the probability of the user initiating the communication complaint, thereby improving the prediction accuracy. Furthermore, the associated information of the abnormal complaint is pushed to the target communication user who is predicted to initiate the communication complaint, which can realize targeted active care for the target communication user, pacify the communication user affected by the communication abnormality, avoid disturbing normal communication users who are not affected by the abnormality, reduce the amount of noise information of the user, and effectively improve the effectiveness of pushing the associated information of the communication abnormality.

[0149] In terms of fault influence range positioning, the AAA online user fluctuation data and the information of the centralized alarm platform are compared, and the resource device management system data is combined, so that the user influence range of the fault can be accurately positioned. In terms of user complaint possibility analysis, a regression decision tree commonly used for prediction is adopted. Specifically, an XGBoost algorithm improved on the basis of GBDT is selected, and accurate prediction analysis is realized through machine learning.

[0150] The modules in the device provided by the embodiment of the application can also implement the method steps provided by the method embodiment. Alternatively, the device provided by the embodiment of the application can include other modules in addition to the above modules to implement the method steps provided by the method embodiment. The device provided by the embodiment of the application can achieve the technical effects achieved by the method embodiment.

[0151] Preferably, the embodiment of the present application further provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable in the processor, wherein the computer program is executed by the processor to implement each process of the method for pushing the associated information of communication exception and achieve the same technical effects. To avoid repetition, details are not described herein.

[0152] The embodiment of the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement each process of the method for pushing the associated information of communication exception and achieve the same technical effects. To avoid repetition, details are not described herein. The computer readable storage medium includes a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.

[0153] The embodiment of the present application further provides a computer program product, which comprises a non-transitory computer readable storage medium storing a computer program. The computer program is operable to cause a computer to perform some or all of the steps of the method for pushing the associated information of communication exception and achieve the same technical effects. To avoid repetition, details are not described herein.

[0154] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. In addition, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) containing computer-usable program code.

[0155] The present application is described with reference to the flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a device that implements the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0156] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart or flowsheets and / or block or blocks of the block diagrams.

[0157] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flowsheets and / or block or blocks of the block diagrams.

[0158] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0159] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer readable media.

[0160] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to computing devices. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0161] It is also to be noted that the terms "comprising", "including", and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0162] Those skilled in the art will appreciate that embodiments of the present application can be devised for a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0163] The embodiments of the present application described above are only used to explain the technical solutions of the present application and not to limit the present application. Although the present application has been described in detail, those skilled in the art will understand that the present application can make various modifications and changes without departing from the spirit and scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.

Claims

1. A method of pushing associated information of a communication anomaly, characterized by, include: Obtain online user information from the communication user authentication, authorization, and billing system, and information on abnormal communication devices from the communication alarm platform; The user's online information is compared with the information of the abnormal communication device to determine communication alarm information. The communication alarm information includes abnormal communication data of the abnormal communication device and information of the communication user carried by the abnormal communication device. The communication alarm information is input into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint. The training samples of the communication complaint prediction model include sample communication data of sample communication devices, and the training labels of the communication complaint prediction model characterize whether the sample communication user carried by the sample communication device initiates a communication complaint. The system pushes relevant information about abnormal complaints to the abnormal communication devices of the target communication user, where the target communication user is the communication user predicted to initiate the communication complaint by the communication complaint prediction model.

2. The method as described in claim 1, characterized in that, Before inputting the communication alarm information into the pre-trained communication complaint prediction model, the method further includes: Obtain the original communication data of the sample communication device and the complaint behavior information of whether the sample communication user carried by the sample communication device initiated a communication complaint; The communication behavior characteristics of the sample communication devices are extracted from the original communication data, and the communication behavior characteristics include user communication behavior characteristics and / or device communication behavior characteristics. The prediction model is trained using training samples constructed based on the communication behavior features and training labels constructed based on the complaint behavior information, thus obtaining the communication complaint prediction model.

3. The method as described in claim 2, characterized in that, Before extracting the communication behavior features of the sample communication device from the original communication data, the method further includes: The original communication data and the complaint behavior information are normalized to obtain normalized original communication data and normalized complaint behavior information; Based on the data missing rate, data filtering is performed on the normalized original communication data and the normalized complaint behavior information to obtain the filtered original communication data and the filtered complaint behavior information. The extraction of communication behavior features of the sample communication device from the original communication data includes: The communication behavior features of the sample communication devices are extracted from the filtered raw communication data.

4. The method as described in claim 2, characterized in that, The communication complaint prediction model is obtained by training a prediction model based on training samples constructed from the communication behavior features and training labels constructed from the complaint behavior information, including: Training samples are constructed based on at least one segmentation point and the aforementioned communication behavior features; Training labels are constructed based on the at least one segmentation point and the complaint behavior information; The prediction model is trained based on the training samples and the training labels.

5. The method as described in claim 4, characterized in that, Before constructing training samples based on at least one segmentation point and the aforementioned communication behavior features, the following steps are also included: Numerical segmentation of the communication behavior features is performed based on at least one segmentation point to obtain communication behavior feature segmentation groups corresponding to the segmentation points; A first entropy value is determined for the value of the communication behavior feature, and a second entropy value is determined for each of the multiple communication behavior feature segmentation groups; The at least one segmentation point is adjusted based on the similarity between the weighted average of multiple second entropy values ​​and the first entropy value.

6. The method according to any one of claims 2 to 5, characterized in that, The user communication behavior characteristics include at least one of the following: user consumption behavior characteristics, user network behavior characteristics, and user communication consultation behavior characteristics; The device communication behavior characteristics include at least one of the following: communication anomaly type, communication anomaly duration, communication anomaly cause, communication anomaly level, and communication anomaly frequency.

7. A device for pushing information related to communication anomalies, characterized in that, include: The acquisition module acquires online user information from the communication user authentication, authorization, and billing system and information on abnormal communication devices from the communication alarm platform. The determination module compares the user's online information with the information of the abnormal communication device to determine communication alarm information. The communication alarm information includes abnormal communication data of the abnormal communication device and information of the communication user carried by the abnormal communication device. The prediction module inputs the communication alarm information into a pre-trained communication complaint prediction model to obtain a prediction result of whether the communication user initiates a communication complaint. The training samples of the communication complaint prediction model include sample communication data of sample communication devices, and the training labels of the communication complaint prediction model characterize whether the sample communication user carried by the sample communication device initiates a communication complaint. The push module pushes relevant information about abnormal complaints to the abnormal communication devices of the target communication user, where the target communication user is the communication user predicted to initiate the communication complaint by the communication complaint prediction model.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of the method as described in any one of claims 1 to 6.

Images