Communication service data audit method and device

Abstract

The present application provides a communication service data audit method and device, wherein the method includes: obtaining the user's communication service data and the service category to which the communication service data belongs; using an audit set to determine whether the service category is the first time to be audited, and the audit set includes all service categories that have not been audited; if the service category is the first time to be audited, then using a preset manual audit model to audit the communication service data to obtain an audit result; if the service category is not the first time to be audited, then using a machine learning audit model to audit the communication service data to obtain an audit result; wherein the machine learning audit model is trained using communication service data corresponding to multiple audited service categories. The method of the present application solves the problem that frequent changes in business rules bring huge pressure to the upgrade and maintenance of manual audit models.

Description

Technical Field

[0001] The present application relates to the technical field of communication service management, and in particular to a communication service data auditing method and device. Background Art

[0002] In the telecommunications industry, auditing of users' communication service data (such as billing auditing) is an important management and technical means to ensure the steady development of communication services.

[0003] The existing method for auditing billing is mainly a manual audit model method based on business rules. For example, one communication service corresponds to one business rule. The auditor classifies business rules with the same business rule attributes into a business category, and based on the business rules corresponding to the business category, writes an audit program (i.e., a manual audit model) for each business category. Based on the business category to which the user's communication service data belongs, the auditor uses the manual audit model corresponding to the business category to audit the user's communication service data, and obtains the audit results (such as normal billing or abnormal billing) for subsequent processing. When the business rules corresponding to the manual audit model change, such as the emergence of new business rules, it is necessary to upgrade the manual audit model or create a new manual audit model to meet the needs of accurate auditing of communication service data.

[0004] However, the frequent changes in user business demand scenarios have led to the frequent generation of new business rules, which in turn has led to frequent changes in business rules. The frequent changes in business rules have brought tremendous pressure on the upgrade and maintenance of manual audit models, which in turn has affected the timeliness of the audit of communication business data. Summary of the invention

[0005] The present application provides a communication service data auditing method and device to solve the problem that frequent changes in business rules bring huge pressure on the upgrade and maintenance of manual audit models, and ensure the timeliness of the audit of communication service data under frequent changes in business rules.

[0006] In a first aspect, the present application provides a communication service data audit method, comprising:

[0007] Acquire the user's communication service data and the service category to which the communication service data belongs;

[0008] Using an audit set, determining whether the business category is a business category being audited for the first time, the audit set including all business categories that have not been audited;

[0009] If the service category is a service category that is audited for the first time, a preset manual audit model is used to audit the communication service data to obtain an audit result;

[0010] If the service category is not the service category being audited for the first time, auditing the communication service data using a machine learning audit model to obtain an audit result;

[0011] The machine learning audit model is trained using communication service data corresponding to multiple audited service categories.

[0012] Optionally, the communication service data includes an actual outgoing value and a plurality of service parameter values; the audit result is any one of the following: outgoing normal, requiring reconfirmation, and outgoing abnormal;

[0013] The adopting of a machine learning audit model to audit the communication service data to obtain an audit result includes:

[0014] The communication service data is input and processed to obtain a multidimensional feature vector P consisting of an actual outgoing account feature value q and a plurality of service parameter feature values ​​Pi;

[0015] Input the multidimensional feature vector into the machine learning audit model to obtain an audit probability value γ of normal payment;

[0016] Based on the audit probability value γ, determining an audit result corresponding to the audit probability value γ;

[0017] Wherein, i is a natural number.

[0018] Optionally, determining the audit result corresponding to the audit probability value γ based on the audit probability value γ includes:

[0019] The audit probability value γ and the preset threshold δ 0 The comparison is performed and the audit results are determined as follows:

[0020] If δ 0 <γ, the audit result is determined to be normal;

[0021] If γ≤δ 0 , then the audit result is determined to be abnormal payment.

[0022] Optionally, determining the audit result corresponding to the audit probability value γ based on the audit probability value γ includes:

[0023] The audit probability value γ and the preset lower limit threshold δ L and the upper threshold δ H The comparison is performed and the audit results are determined as follows:

[0024] If δ H <γ, the audit result is determined to be normal;

[0025] If δ L≤γ≤δ H , then the audit result is determined to require reconfirmation;

[0026] If γ<δ L , then the audit result is determined to be abnormal payment.

[0027] Optionally, before the adopting of the preset manual audit model or the machine learning audit model to audit the communication service data to obtain the audit result, the method further includes:

[0028] The communication service data is cleansed to obtain corresponding pure data.

[0029] Optionally, the audit result includes an audit result that needs to be reconfirmed; after determining that the audit result needs to be reconfirmed, the method further includes:

[0030] Performing correction processing on the communication service data to obtain corrected service data;

[0031] Based on the business category corresponding to the modified business data, a preset manual audit model or the machine learning audit model is used to audit the modified business data to obtain a corresponding audit result.

[0032] Optionally, after obtaining the audit result, the method further includes:

[0033] Based on the communication business data corresponding to the audit result of the abnormal billing, an abnormal user work order for the user corresponding to the audit result is generated to notify the work order processing personnel in the user's location to handle it.

[0034] In a second aspect, the present application provides a communication service data auditing device, which is applied to a communication service data auditing system, wherein the system includes an auditing device and at least one data collection device; the auditing device includes:

[0035] Data transceiver unit, data audit unit and data correction unit; wherein,

[0036] The data transceiver unit is used to obtain the user's communication service data and the service category to which the communication service data belongs from at least one data collection device; using the audit set, judging whether the service category is the first time to be audited, and sending the judgment result of the communication service data and service category to the data audit unit; the audit set includes all service categories that have not been audited;

[0037] The data audit unit is used to audit the communication service data whose service category is the first audited service category using a preset manual audit model based on the judgment result of the communication service data service category to obtain an audit result; or, to audit the communication service data whose service category is not the first audited service category using a machine learning audit model to obtain an audit result; and send the audit result to the data transceiver unit; wherein the machine learning audit model is trained using communication service data corresponding to multiple audited service categories; and the audit result is any one of the following results: normal billing, reconfirmation required, and abnormal billing;

[0038] The data transceiver unit is also used to generate an abnormal user work order for the user corresponding to the audit result for the communication service data with abnormal billing results, so as to notify the work order processing personnel at the user's location to handle it.

[0039] Optionally, the auditing device further includes a data correction unit;

[0040] The data transceiver unit is further used to send the communication service data that the audit result shows needs to be reconfirmed to the data correction unit;

[0041] The data correction unit is used to correct the received communication service data to obtain corrected service data, and send the corrected service data to the data transceiver unit;

[0042] The data transceiver unit is further used to send the modified business data to the data audit unit;

[0043] The data audit unit is also used to audit the modified business data based on the business category corresponding to the modified business data, using a preset manual audit model or the machine learning audit model to obtain a corresponding audit result.

[0044] Optionally, the data transceiver unit includes a data transceiver module, a data cleaning module, and a data input module; the data audit unit includes a manual audit module and a machine learning audit module; wherein,

[0045] The data transceiver module is used to obtain the user's communication service data and the service category to which the communication service data belongs from at least one data collection device; and use the audit set to determine whether the service category is the first time to be audited;

[0046] The data cleaning module is used to clean the communication service data in the following manner to obtain corresponding clean data:

[0047] Conflicting data is identified on the communication service data, and if no conflicting data is identified in the communication service data, the communication service data is determined to be primary cleansing data; the communication service data includes an actual outgoing value and service parameter values ​​corresponding to each of a plurality of service parameters; the conflicting data is a plurality of different service parameter values ​​corresponding to the same service parameter; the primary cleansing data is secondary cleansed to obtain pure data; the secondary cleansing includes data standardization conversion, missing data completion, and invalid data elimination;

[0048] The data input module is used to input the clean data whose business category is the first audited business category into the manual audit module to obtain the audit result based on the business category to which the clean data belongs; or to input the clean data whose business category is not the first audited business category to obtain the corresponding multidimensional feature vector, and input the multidimensional feature vector into the machine learning audit module to obtain the audit result; the multidimensional feature vector P is composed of the actual payment feature value q and multiple business parameter feature values ​​Pi; wherein i is a natural number;

[0049] The manual audit module is used to audit the input clean data using a preset manual audit model to obtain an audit result;

[0050] The machine learning audit module is used to audit the input clean data using a multi-layer feedforward neural network model to obtain an audit result; the multi-layer feedforward neural network model includes an input layer, at least one hidden layer, a normalization layer and an output layer; the node activation function of the hidden layer is a linear rectification function; the activation function of the normalization layer is a normalized exponential softmax function.

[0051] In a third aspect, the present application provides a communication service data auditing device, comprising:

[0052] Processor and memory;

[0053] The memory stores executable instructions executable by the processor;

[0054] The processor executes the executable instructions stored in the memory, so that the processor executes the method as described above.

[0055] The communication service data audit method and device provided in the present application judge whether the service category of the communication service data of each user obtained is the first time to be audited, and based on the judgment result, a preset manual audit model or a machine learning audit model is used to audit the communication service data of each user to obtain the audit result of each user. Among them, the machine learning audit model is suitable for the audit of the user communication service data corresponding to the communication service covered by the service category corresponding to the training sample. When the service category does not change but the communication service under the service category changes frequently, there is no need to upgrade the machine learning audit model to ensure the accurate audit of the user communication service data under the new communication service; and the manual audit model in the present application is only used for the audit of the communication service data of the newly added service category that is audited for the first time. The present application solves the problem that the frequent changes in business rules bring huge pressure to the upgrade and maintenance of the manual audit model, and ensures the timeliness of the audit of communication service data under the frequent changes in business rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0057] Figure 1 The communication service data audit system architecture diagram provided in the embodiment of the present application;

[0058] Figure 2 A flow chart of a communication service data audit method provided in an embodiment of the present application;

[0059] Figure 3 A structural diagram of a communication service data audit device provided in an embodiment of the present application;

[0060] Figure 4 A structural diagram of a BP neural network model provided in an embodiment of the present application;

[0061] Figure 5 This is a structural diagram of the communication service data auditing device provided in an embodiment of the present application.

[0062] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] With the technological innovation of the communications industry and the development of the Internet, user business demand scenarios are also changing frequently. In order to meet the frequently added business needs of users, operators are also frequently adding new communication services, that is, adding new business rules or changing the original business rules. If the preset manual audit model does not cover the new business rules, the manual audit model cannot accurately audit the user communication business data corresponding to the new business rules. In the prior art, in order to ensure that the manual audit model accurately audits the user communication business data corresponding to the new business rules, the following method is often used to maintain the manual audit model: if the new business rule does not belong to any existing business category, that is, a new business category is added, it is necessary to create a corresponding manual audit model for the new business category based on the new business rule; if the new business rule does not result in the addition of a new business category, it is necessary to upgrade the preset manual audit model corresponding to the business category to which the new business rule belongs based on the new business rule.

[0065] Whether it is adding new business rules or changing the original business rules, these changes in business rules inevitably bring about the need to create a new manual audit model or upgrade the preset manual audit model. The maintenance frequency of the manual audit model is usually comparable to the frequency of business rule changes. Therefore, even if the change in business rules does not involve changes in the original business categories (such as no new business categories), the frequent changes in business rules will also bring huge pressure to the upgrade and maintenance of the manual audit model, thereby affecting the timeliness of the audit of communication business data.

[0066] In this regard, the present application provides a communication service data audit method, which determines whether the service category of the user's communication service data is the first service category to be audited by judging the service category to which the user's communication service data belongs. If the service category of the user's communication service data is the first service category to be audited, a preset manual audit model is used to audit the communication service data to obtain an audit result; conversely, if the service category of the user's communication service data is not the first service category to be audited, a machine learning audit model is used to audit the communication service data to obtain an audit result. Among them, the machine learning audit model is trained using communication service data corresponding to multiple audited service categories, and the machine learning audit model is applicable to all communication services corresponding to the service category corresponding to the communication service data used for training. In other words, changes in the communication services under the service category to which the machine learning audit model is applicable will not affect the accurate audit of all communication services under the service category to which the machine learning audit model is applicable. That is, if the communication services under the service category to which the machine learning audit model is applicable change, the machine learning audit model does not need to be upgraded with the same frequency, or even does not need to be upgraded. The communication service data auditing method provided in this application can solve the problem that frequent changes in business rules bring huge pressure on the upgrade and maintenance of manual auditing models when the changes in business rules do not involve changes in the original business categories, thereby ensuring the timeliness of the auditing of communication service data under frequent changes in business rules.

[0067] The communication service data auditing method provided by this application is described below in conjunction with some embodiments.

[0068] Figure 1 This is a diagram of the communication service data audit system architecture provided in the embodiment of the present application. Figure 1As shown, the system includes: an audit device 11, at least one data collection device 12, wherein the data collection device 12 can be a data collection device 12 in a central business system (also called a main system) for processing communication services at the operator headquarters, or a data collection device 12 in a regional business system (such as regional business system 1, regional business system 2, ..., regional business system n, where n is a natural number) for processing communication services at each regional operation department. When the auditor needs to audit the communication service data of each user of a certain communication service in a certain period of time in the central business system, the auditor sends an audit request to the audit device 11 through the central business system, such as the auditor sends the audit request to the audit device 11 through the data collection device 12 of the central business system; correspondingly, when the auditor needs to audit the communication service data of each user of a certain communication service in a certain period of time in a certain regional business system, the auditor sends the audit request to the audit device 11 through the regional business system, such as the auditor sends the audit request to the audit device 11 through the data collection device 12 of the regional business system. Among them, the audit request includes: the system identifier of the central business system or regional business system that sends the audit request, the device identifier that sends the audit request, the communication service identifier, and the time period (such as a certain year and a certain month). Usually, the data acquisition device 12 of the regional business system will regularly synchronize the communication service data with the data acquisition device 12 of the central business system. After receiving the audit request, the audit device 11 obtains the corresponding communication service data of each user and the business category to which the communication service data belongs from the data acquisition device 12 corresponding to the system identifier based on the audit request. Among them, the communication service data includes the actual billing value and the business parameter values ​​(abbreviated as: parameter values) corresponding to each of the multiple business parameters. The audit device 11 uses the audit set to determine whether the business category to which the communication service data belongs is the business category that is audited for the first time. The audit set includes all business categories that have not been audited. If the audit device 11 determines that the service category is the first audited service category, the preset manual audit model is used to audit the communication service data of each user to obtain the audit results of each user; if the audit device 11 determines that the service category is not the first audited service category, the machine learning audit model is used to audit the communication service data of each user to obtain the audit results of each user. The machine learning audit model is trained using the communication service data corresponding to multiple audited service categories.

[0069] Since the business category being audited for the first time is usually a newly added business category, the corresponding number of users is relatively small compared to the business categories that have been audited. It is accurate and efficient to use the preset manual audit model to audit the communication business data of users of the business category being audited for the first time. However, for the business categories that have been audited, the number of users is large, and there are frequent changes in the communication services under the business categories, which puts a lot of pressure on upgrading and maintaining the manual audit model. The machine learning audit model trained by using training samples composed of communication business data corresponding to multiple audited business categories is suitable for the audit of user communication business data covered by the business categories corresponding to the training samples. In other words, when the business category has not changed and only the communication services under the business category have changed frequently, there is no need to upgrade the machine learning audit model to ensure accurate auditing of the communication business data of users under the new communication services.

[0070] In the method provided in the present application, the manual audit model is only applicable to newly added business categories, that is, the manual audit model is newly created or upgraded for the newly added business category after the new business category is added. After the communication service data of the business category corresponding to the manual audit model is audited for the first time, if the communication service under the business category changes, it is necessary to audit the communication service data under the business category again. In this case, it is only necessary to update the training samples of the machine learning audit model and train the model based on the communication service data of the business category. The machine learning audit model after model training can be used to audit the communication service data of the business category, without the need to upgrade the manual audit model corresponding to the business category for changes in the communication service under the business category. Generally, the frequency of adding new business categories is low, and accordingly, the frequency of creating or upgrading manual audit models is also low, which will not cause pressure on the creation or upgrading of manual audit models.

[0071] The communication service data audit method provided by the present application determines whether the service category of the communication service data of each user obtained is the first time to be audited, and based on the judgment result, a preset manual audit model or a machine learning audit model is used to audit the communication service data of each user to obtain the audit result of each user. Among them, the machine learning audit model is suitable for the audit of user communication service data corresponding to the communication service covered by the service category corresponding to the training sample. When the service category does not change but the communication service under the service category changes frequently, there is no need to upgrade the machine learning audit model to ensure the accurate audit of the user communication service data under the new communication service; and the manual audit model is only used for the audit of the communication service data of the newly added service category that is audited for the first time. The method provided by the present application solves the problem that the frequent changes in business rules bring huge pressure to the upgrade and maintenance of the manual audit model, thereby ensuring the timeliness of the audit of communication service data under the frequent changes in business rules.

[0072] Combine the following Figure 2 The communication service data audit method provided in this application is described in detail. Figure 2 The specific process of auditing the communication service data of each user is as follows: Figure 2 shown. Figure 2 The execution subject of the embodiment shown is Figure 1 The audit device 11 in the embodiment shown. Figure 2 As shown, the method includes:

[0073] S201. Obtain the user's communication service data and the service category to which the communication service data belongs.

[0074] Specifically, the audit device 11 obtains the user's communication service data and the service category to which the communication service data belongs from the data collection device 12 of the central service system or the data collection device 12 of the regional service system.

[0075] Optionally, the audit device 11 obtains the communication service data of the same user and the service category to which the communication service data belongs from the data collection devices 12 of the central service system and the regional service system. In other words, the communication service data of the user may be composed of the communication service data from the data collection devices 12 of two different systems.

[0076] Since the data collection device 12 in the regional business system is a local customized device customized according to the characteristics of local business needs, the differences in the data transmission interfaces set on the local customized devices will result in the user's communication business data not being able to be transmitted to the audit device 11 through the data transmission interface. In this regard, with the authorization of the regional business system, the audit device 11 can obtain the complete user communication business data to be audited from the data collection device 12 in the regional business system through crawler technology.

[0077] To facilitate subsequent auditing, each communication service data may be pre-processed before auditing the communication service data.

[0078] Furthermore, before the audit device 11 uses a preset manual audit model or a machine learning audit model to audit the communication service data to obtain the audit results, the audit device 11 may first clean the communication service data to obtain corresponding pure data.

[0079] For example, if the user's communication service data only comes from the data collection device 12 of the central service system or the data collection device 12 of the regional service system, the user's communication service data is cleaned by a conventional cleaning method to obtain pure data. The conventional cleaning method includes data standardization conversion, missing data completion, and invalid data elimination.

[0080] An example of routine cleaning of a user's communication service data is shown in the following steps (1)-(3):

[0081] (1) Data standardization conversion: The communication service data obtained by the audit device 11 from the data acquisition device 12 usually contains multiple types of data such as integer, floating point, date, and string. However, whether it is a manual audit model or a machine learning audit model, the data type of the input information is required to be a digital type. Therefore, it is necessary to perform standardized conversion on the date type and string to uniformly convert all types of data into digital types, so as to facilitate the construction and input of subsequent model input features.

[0082] Exemplarily, the standardized conversion examples of date type data are as follows: ① Use regular matching to convert date type data expressed in strings into date arrays; ② Further simplify the date array to obtain simplified dates: For example, when the time period to be audited is a certain month, delete the year and month information in the date array, and only retain the day, hour, minute, and second information; ③ Calculate the offset of the simplified date to obtain a digital type date: For example, when the time period to be audited is a certain month, start from 0 seconds on the 1st of that month, and calculate the offset in seconds, such as "2021-10-01 01:00:00" is calculated and converted into a digital type date "3600".

[0083] Standardized conversion example of string type data: String type data is often used to represent information with practical meaning, such as communication location, SMS service type (such as "SMS", "MMS") and call type (such as "send", "receive", etc.). In this regard, string type data can be standardized and converted using digital coding conversion. For example, the communication location is converted using city code, such as converting the communication locations "Beijing" and "Guangzhou" into city codes "010" and "020" respectively; each service parameter (such as service type, call type) is enumerated and converted using numbers starting from 0.

[0084] (2) Missing data completion: Both manual audit models and machine learning audit models must be applicable to all business rules corresponding to the business categories to which they apply. A model usually needs to be applicable to multiple business rules at the same time, so the feature vector to be input into the model is the feature of all business parameter dimensions of multiple business rules to which the model applies. The user's communication service data only corresponds to one business rule. Therefore, in order to maintain the consistency of the input feature dimension of the input model, the user's communication service data needs to be completed based on the input features of the model. For example, the data of the fixed network and broadband number in the dimension of traffic usage is missing, so the missing data is directly completed by filling in zeros. That is, if user A's communication service is a fixed network, then based on user A's communication service data, a data completion operation is performed by adding a service parameter of "traffic usage" and a corresponding service parameter value of "0".

[0085] (3) Invalid data elimination: If the communication service data obtained by the audit device 11 from the data collection device 12 contains key data missing, such as the communication service identification code (i.e., product code) and the actual billing parameter value, the communication service data is determined to be invalid data and is directly eliminated. Optionally, after the communication service data is eliminated, a user abnormal work order can be generated to notify the user's local personnel to handle it.

[0086] Optionally, if the user's communication service data is composed of communication service data from two different systems, namely, the central service system and the regional service system, on the data collection device 12, the communication service data is cleaned in the following manner as shown in steps I-II to obtain corresponding pure data:

[0087] I. Clean the communication service data to obtain cleaned data.

[0088] One-time cleaning is a cleaning method that takes into account the business scenario settings of the communication business. In this solution, since the user's communication business data is composed of the user's data collected by the data collection equipment 12 of the central business system and the regional business system. Although the regional business system and the central business system will synchronize data regularly, it is inevitable that sudden failures will cause data synchronization failures and data loss will be caused by the update and upgrade of new and old services in the regional business system, which will lead to inconsistency of the user's communication business data in the central business system and the regional business system, and the user's accurate communication business data cannot be determined. In the case of inconsistency between the communication business data of the same user in the regional business system and the central business system, since it is impossible to determine the accurate communication business data of the user in the time period to be audited, it is not suitable to directly use a manual audit model or a machine learning audit model to audit the communication business data, and the communication business data needs to be removed from the communication business data to be audited, and the audit will be carried out after the accurate communication business data of the user is determined.

[0089] The audit device 11 performs a cleansing on the communication service data and obtains a cleansing data example as follows: the audit device 11 identifies conflicting data on the communication service data. If no conflicting data is identified in the communication service data, the communication service data is determined to be cleansing data. Among them, the conflicting data are multiple different service parameter values ​​corresponding to the same service parameter. The communication service data includes the actual outgoing value and the service parameter values ​​corresponding to each of the multiple service parameters. Generally speaking, the same service parameter of the same user corresponds to one service parameter value, and the same service parameter of the same user corresponds to multiple different service parameter values, which means that the communication service parameters of the user are from different data acquisition devices and there are conflicts on different data acquisition devices. If the communication service data of the same user in the regional service system and the central service system are consistent and there is no conflicting data, it means that the communication service data of the user is accurate and can be directly used for auditing.

[0090] Optionally, if the user identification code of the cleaned data is a test number used by the operator to launch a new business, based on the special identification of the test number, a regular match is used to match the test number and the communication business data corresponding to the test number is removed without the need for auditing the subsequent steps S202-S204.

[0091] II. Perform secondary cleaning on the once cleaned data to obtain pure data.

[0092] The auditing device 11 performs a primary cleaning on the communication service data and determines the primary cleaning data; the primary cleaning data is secondary cleansed, that is, the primary cleaning data is cleansed using the above conventional cleaning method to obtain pure data.

[0093] S202: Determine whether the business category is the first business category to be audited.

[0094] Specifically, the audit device 11 uses the audit set to determine whether the business category is the business category being audited for the first time. If the determination result is yes, step S203 is performed; if the determination result is no, step S204 is performed.

[0095] Among them, the audit focus includes all business categories that have not been audited.

[0096] S203: Use a preset manual audit model to audit the communication service data to obtain an audit result.

[0097] Specifically, after the audit device 11 completes step S202, if the service category is the first service category to be audited, the preset manual audit model is used to audit the communication service data to obtain the audit result. That is, the audit device 11 inputs the clean data corresponding to the communication service data into the preset manual audit model to audit and obtain the audit result.

[0098] Exemplarily, the audit device 11 uses a preset manual audit model to audit the clean data corresponding to the communication service data, and obtains the audit output value Q corresponding to the clean data. j Audit equipment 11 according to the formula The audit probability value γ of the clean data being issued normally is calculated. The audit device 11 compares the audit probability value γ with the preset lower limit threshold δ L and the upper threshold δ H The following comparisons are made in the manual audit model to determine the audit results:

[0099] If δ H <γ, the audit result is determined to be normal;

[0100] If δ L ≤γ≤δ H , then the audit result is determined to require reconfirmation;

[0101] If γ<δ L , then the audit result is determined to be abnormal payment.

[0102] For example, the lower threshold δ L can be 50%, the upper threshold δ H It can be 80%.

[0103] Wherein, Q is the actual outgoing value in the communication service data or the pure data.

[0104] Since the services of the service category being audited for the first time are usually new communication services, and the number of users of new communication services is small in the initial application, there are fewer samples that can be used to train the machine learning audit model, which is not conducive to the training of the machine learning audit model. Therefore, using a manual audit model to audit the communication service data of this service category is the preferred and efficient method. Usually, after an operator develops a new service category, it will design a manual audit model for the new service category and add the new service category to the audit set. After the number of users of the new service category accumulates to the point where its corresponding communication service data can be used for machine learning audit model training (such as after changes in the communication services under the service category), the machine learning audit model training samples can be updated and model trained based on the communication service data corresponding to the users of the service category, and the service category can be removed from the audit set. The machine learning audit model trained with the updated training samples can be used to audit the communication service data of this service category.

[0105] Optionally, the audit result is any one of the following: normal payment, reconfirmation required, and abnormal payment.

[0106] S204. Use a machine learning audit model to audit communication business data to obtain audit results.

[0107] Specifically, after the audit device 11 completes step S202, if the service category is not the service category being audited for the first time, a machine learning audit model is used to audit the communication service data to obtain an audit result.

[0108] The machine learning audit model is trained using communication service data corresponding to multiple audited service categories. The communication service data includes actual outgoing bill values ​​and multiple service parameter values.

[0109] Optionally, the audit result may be any one of the following: normal payment, reconfirmation required, or abnormal payment.

[0110] The communication service data is cleaned in step S201 and becomes pure data, and its data reliability, consistency and format are effectively guaranteed. The number of digits of the parameter values ​​of each service parameter in the pure data is quite different. For example, in the communication service data, the product code parameter value is 88888888, and the discount rate parameter value is 0 to 1 with two decimal places. The number of digits of the product code parameter value and the discount rate parameter value is 8 and 3 respectively, which is quite different. The difference in the number of digits of the parameter values ​​of each service parameter in the pure data will have a certain impact on the speed and fitting effect of the subsequent machine learning audit model training. In this regard, the pure data can be input processed, and then the machine learning audit model can be used to audit the input processed pure data.

[0111] Optionally, the process of the audit device 11 using the machine learning audit model to audit the communication service data to obtain the audit result is shown in S2041-S2043:

[0112] S2041. Process the communication service data input to obtain a multi-dimensional feature vector P consisting of an actual outgoing payment feature value q and multiple service parameter feature values ​​Pi.

[0113] Specifically, the audit device 11 processes the input of the communication service data to obtain a multi-dimensional feature vector P composed of an actual payment feature value q and multiple service parameter feature values ​​Pi.

[0114] Among them, the service parameter characteristic value is referred to as: characteristic value in this application.

[0115] Exemplarily, the input processing includes standardizing each service parameter value in the communication service data to obtain a standardized parameter value, and constructing a multi-dimensional feature vector based on the standardized parameter value.

[0116] Among them, the standardized processing of each service parameter value in the communication service data includes: reducing the number of bits of parameter values ​​with a large number of digits and unifying the standard processing of parameter value units, as shown in the following examples.

[0117] Reduce the number of digits of parameter values ​​with a large number of digits: for example, establish an identification mapping relationship from communication service type to audit service type, and map codes with a large number of digits such as product codes to audit service type codes with a reduced number of digits.

[0118] Standardized processing of parameter value units: For example, the parameter value of the parameter "traffic usage" in user communication service data is different in the communication service data of different users. Some users use KB as the traffic unit, while others use MB or GB as the traffic unit. Since the traffic charges in communication services are based on MB as the basic unit, for service parameters involving traffic, the traffic is expressed in scientific notation based on the unit KB, where the basic value retains two decimal places and the power is a multiple of 3. For example, the parameter value "1,000KB" is expressed as "(1.0,3)", and the parameter value "50,101,000KB" is expressed as "(50.10,6)".

[0119] By processing the input of communication business data, the parameter values ​​in the communication business data can be standardized to a preset value range, which is convenient for the subsequent model training and audit processing of the machine learning audit model.

[0120] After the audit device 11 obtains the standardized parameter values ​​by standardizing the service parameter values ​​in the communication service data, it constructs a multidimensional feature vector as shown in Table 1 below based on the standard parameter values. The multidimensional vector feature is composed of an actual outgoing payment feature value q and a plurality of service parameter feature values ​​Pi. The actual outgoing payment feature value q is equal to the actual outgoing payment value Q in the corresponding communication service data or pure data.

[0121] Table 1 Multidimensional feature vector of user A

[0122]

[0123] That is to say, the multi-dimensional feature vector of user A is P = [0, 0, 6, 0, 1, 6, 1, …, 112, 3, 0, 0, 0, …, 0, 0, 0, …, 20, 0.3, 0, 0, 0, 0, 0, …, 100].

[0124] The feature dimension in the multidimensional feature vector, i.e., the number of feature values, is determined by the business rules applicable to the machine learning audit model. The feature dimensions in the multidimensional feature vector corresponding to the communication service data of each user are the same. Each multidimensional feature vector contains all the business parameters of the business rules applicable to the machine learning audit model. Generally speaking, the business rules of new communication services are often formed by changing the parameter values ​​of the original business parameters. For example, the parameter value (or feature value P64) of the business parameter "package discount rate" shown in Table 1 is increased from 0.3 to 0.6, and the parameter value (or feature value P63) of the business parameter "grant" is increased from 20 to 60. Although the business rules have changed and a new communication service has been formed, its business parameters have not changed, and the business category has not changed. In this regard, if the business parameters remain unchanged, the new communication service formed by changing the business parameter value will not change the scope of application and audit accuracy of the machine learning audit model.

[0125] S2042. Input the multi-dimensional feature vector into the machine learning audit model to obtain the audit probability value γ of normal account issuance.

[0126] Specifically, the audit device 11 inputs the multi-dimensional feature vector into the machine learning audit model to obtain the audit probability value γ of normal payment.

[0127] S2043. Based on the audit probability value γ, determine the audit result corresponding to the audit probability value γ.

[0128] Specifically, the audit device 11 determines the audit result corresponding to the audit probability value γ based on the audit probability value γ.

[0129] Optionally, the audit device 11 compares the audit probability value γ obtained in step S2042 with a preset threshold value δ 0 The comparison is performed and the audit results are determined as follows:

[0130] If δ 0 <γ, the audit result is determined to be normal;

[0131] If γ≤δ 0 , then the audit result is determined to be abnormal payment.

[0132] For example, the audit device 11 compares the audit probability value γ with the preset threshold value δ 0 The following comparisons are made in the machine learning audit model to determine the audit results:

[0133] If δ 0 <γ, the audit result is determined to be normal;

[0134] If γ≤δ 0 , then the audit result is determined to be abnormal payment.

[0135] Exemplarily, the preset threshold δ 0 It can be 80%.

[0136] In the model training of the machine learning audit model, it is preferred to compare the audit probability value γ with the preset threshold δ 0 Determining the audit results by comparison can improve the efficiency of model training.

[0137] Optionally, the audit device 11 compares the audit probability value γ obtained in step S2042 with a preset lower limit threshold δ L and the upper threshold δ H The comparison is performed and the audit results are determined as follows:

[0138] If δ H <γ, the audit result is determined to be normal;

[0139] If δ L ≤γ≤δ H , then the audit result is determined to require reconfirmation;

[0140] If γ<δ L , then the audit result is determined to be abnormal payment.

[0141] For example, the audit device 11 compares the audit probability value γ with the preset lower limit threshold δ L and the upper threshold δ H The following comparisons are made in the machine learning audit model to determine the audit results:

[0142] If δ H <γ, the audit result is determined to be normal;

[0143] If δ L ≤γ≤δ H, then the audit result is determined to require reconfirmation;

[0144] If γ<δ L , then the audit result is determined to be abnormal payment.

[0145] For example, the lower threshold δ L can be 50%, the upper threshold δ H It can be 80%.

[0146] Through the preset upper and lower thresholds, the audit results that need to be reconfirmed are determined to select the communication business data of users with smaller billing anomalies, so as to audit their automatically corrected business data. If the corresponding audit result is that the billing is normal, the workload of user abnormality work order processing personnel will be greatly reduced.

[0147] In the model application stage of the machine learning audit model, a preset upper threshold δ H and the lower threshold δ L The preferred method is to determine the audit results.

[0148] After the audit result is determined through step S2043, if the audit result is abnormal billing, the audit device 11 generates an abnormal user work order for the user corresponding to the audit result based on the communication service data corresponding to the audit result of abnormal billing, so as to notify the work order processing personnel at the user's place of origin to handle it; if the audit result is normal billing, the audit device 11 sends the user's audit result to the business system (such as the central business system or the regional business system n) requesting the audit. For example, the audit device 11 generates an audit work order for the user corresponding to the audit result based on the communication service data corresponding to the audit result of normal billing, and sends the audit work order to the business system (such as the central business system or the regional business system n) requesting the audit.

[0149] The audit result includes an audit result that needs to be reconfirmed. After determining in step S2043 that the audit result needs to be reconfirmed, the audit device 11 corrects the communication service data corresponding to the audit result to obtain corrected service data. The audit device 11 uses a preset manual audit model or a machine learning audit model to audit the corrected service data based on the service category corresponding to the corrected service data to obtain the corresponding audit result. If the audit result after auditing the revised business data is that the billing is abnormal or needs to be confirmed again, the audit device 11 generates an abnormal user work order for the user corresponding to the audit result based on the communication business data corresponding to the audit result, so as to notify the work order processing personnel at the user's location to handle it; if the audit result after auditing the revised business data is that the billing is normal, the user's audit result, revised business data, audit times and other information are sent to the business system requesting the audit (such as the central business system or the regional business system). For example, the audit device 11 generates an audit work order based on the communication business data whose audit result after auditing the revised business data is that the billing is normal, and sends it to the business system requesting the audit (such as the central business system or the regional business system n). The audit work order covers the user's audit result, revised business data, audit times and other information.

[0150] The technical effects of the communication service data audit method provided in this application are explained below with reference to specific examples.

[0151] The audit device 11 obtains the communication service data of three users (assuming they are users A, B, and C). The audit probability value γ of normal payment is obtained. The audit device 11 presets the lower limit threshold δ L =50%, upper threshold δ H =80%. The audit device 11 cleans the communication service data of users A, B, and C respectively, and obtains the clean data shown in Table 2-4.

[0152] Table 2: User A’s clean data

[0153]

[0154] Table 3: User B’s clean data

[0155]

[0156] Table 4: Clean data of user C

[0157]

[0158] Among them, the business rules to which user A's communication business data belongs are: Business rules for the 399 yuan package of the refreshing ice cream 5G package;

[0159] The business rules to which user B's communication business data belongs are: Guangdong 500M broadband monthly package 5G integrated dedicated product business rules;

[0160] The business rules to which user C's communication business data belongs are: Guangdong Woqi Cloud Call Product Business Rules.

[0161] The audit device 11 determines that the service category to which the communication service data of user C belongs is the service category being audited for the first time, and determines that the service category to which the communication service data of user A and user B belongs is not the service category being audited for the first time.

[0162] Therefore, the audit device 11 uses a preset manual audit model to audit the clean data of user C, and obtains the audited account value of user C as 67.86, as well as the audit result that the account is normal.

[0163] After the audit device 11 processes the clean data of user A and user B, it obtains the multi-dimensional feature vectors of user A and user B. The audit device 11 uses the machine learning audit model to audit the multi-dimensional feature vectors of user A and user B, and obtains the audit probability value of user A of 92.18% and the audit result that user A's payment is normal; the audit probability value of user B of 15.64% and the audit result that user B's payment is abnormal.

[0164] The communication service data audit method provided by the present application obtains pure data by cleaning the acquired user communication service data, and then determines whether to use a manual audit model or a machine learning audit model to perform an audit based on the service category corresponding to the pure data to obtain an audit result. Among them, before the pure data is input into the machine learning audit model, the pure data is input processed to obtain a multi-dimensional feature vector, which ensures the efficiency and accuracy of the machine learning audit model in auditing user communication service data. The multi-dimensional feature vector input into the machine learning audit model covers the business parameters of all business rules applicable to the model. When the business parameters remain unchanged, new communication services formed by changes in business parameter values ​​will not change the scope of application and audit accuracy of the machine learning audit model. The present application solves the problem that when there are no new business categories, the frequent changes in business rules bring huge pressure to the upgrade and maintenance of the manual audit model, thereby ensuring the efficiency of the audit of communication service data.

[0165] The present application also provides a communication service data audit device. The audit device is used for Figure 1 The communication service data auditing system shown includes an auditing device and at least one data collection device. Figure 3 This is a structural diagram of the communication service data auditing equipment provided in an embodiment of the present application. Figure 4This is a BP neural network model structure diagram provided in the embodiment of the present application. Figure 1 , Figure 3 and Figure 4 The communication service data auditing device provided by this application is described. Figure 3 As shown, the audit device includes: a data transceiver unit 31, a data audit unit 32 and a data correction unit 33. Among them,

[0166] The data transceiver unit 31 is used to receive data from at least one data acquisition device 12 (such as Figure 1 The communication service data of the user and the service category to which the communication service data belongs are obtained; the audit set is used to determine whether the service category is the first service category to be audited, and the communication service data and the judgment result of the service category are sent to the data audit unit 32. The audit set includes all service categories that have not been audited.

[0167] The data audit unit 32 is used to audit the communication service data whose service category is the first audited service category based on the judgment result of the communication service data service category, using a preset manual audit model to obtain an audit result; or, using a machine learning audit model to audit the communication service data whose service category is not the first audited service category to obtain an audit result; and send the audit result to the data transceiver unit 31. Among them, the machine learning audit model is trained using communication service data corresponding to multiple audited service categories. The audit result is any one of the following results: normal billing, reconfirmation required, and abnormal billing.

[0168] The data transceiver unit 31 is also used to generate an abnormal user work order for the user corresponding to the audit result for the communication service data with abnormal billing results, so as to notify the work order processing personnel at the user's location to handle it.

[0169] Optionally, the data transceiver unit 31 is further used to generate and send an audit work order to a user corresponding to the audit result for the communication service data whose audit result shows that the billing is normal.

[0170] Optionally, the auditing device further includes a data correction unit 33 .

[0171] The data transceiver unit 31 is also used to send the communication service data that the audit result shows needs to be reconfirmed to the data correction unit 33.

[0172] The data correction unit 33 is used to correct the received communication service data, obtain corrected service data, and send the corrected service data to the data transceiver unit 31.

[0173] The data transceiver unit 31 is also used to send the modified business data to the data audit unit 32 .

[0174] The data audit unit 32 is also used to audit the modified business data based on the business category corresponding to the modified business data, using a preset manual audit model or a machine learning audit model to obtain a corresponding audit result.

[0175] Optionally, the data transceiver unit 31 includes a data transceiver module 311, a data cleaning module 312, and a data input module 313; the data audit unit 32 includes a manual audit module 321 and a machine learning audit module 322.

[0176] The data transceiver module 311 is used to receive data from at least one data acquisition device 12 (such as Figure 1 As shown) obtain the user's communication service data and the service category to which the communication service data belongs; and use the audit set to determine whether the service category is the first time to be audited.

[0177] The data cleaning module 312 is used to clean the communication service data in the following manner to obtain corresponding clean data:

[0178] Conflicting data is identified for communication service data. If no conflicting data is identified in the communication service data, the communication service data is determined to be primary cleansing data. The communication service data includes the actual outgoing value and the service parameter values ​​corresponding to each of the multiple service parameters; the conflicting data is multiple different service parameter values ​​corresponding to the same service parameter. The primary cleansing data is secondary cleansed to obtain pure data; secondary cleaning includes data standardization conversion, missing data completion, and invalid data removal.

[0179] The data input module 313 is used to input the clean data whose business category is the first audited business category into the manual audit module 321 to obtain the audit result based on the business category to which the clean data belongs; or, to input the clean data whose business category is not the first audited business category, obtain the corresponding multidimensional feature vector, and input the multidimensional feature vector into the machine learning audit module 322 to obtain the audit result. Among them, the multidimensional feature vector P is composed of the actual account feature value P0 and multiple business parameter feature values ​​Pi; i is a natural number.

[0180] The manual audit module 321 is used to audit the input clean data using a preset manual audit model to obtain an audit result.

[0181] Machine learning audit module 322 is used to adopt Figure 4The multi-layer feedforward neural network (Back Propagation Neural Network, referred to as: BP neural network) model shown audits the input clean data to obtain the audit result. The multi-layer feedforward neural network model includes an input layer, at least one hidden layer, a normalization layer and an output layer. The node activation function of the hidden layer is a linear rectification function (i.e., ReLU function). The activation function of the normalization layer is a normalized exponential (softmax) function, and the normalization layer is also called a softmax layer. Both the input layer and the hidden layer are provided with a plurality of neurons or nodes.

[0182] The node activation function of the hidden layer is the ReLU function, and the ReLU function is:

[0183]

[0184] In the communication service, when x>0, ReLU(x) is in a linear relationship with x, and has a high degree of match with the linear relationship between the actual billing of the communication service and the service usage; and when x<0, ReLU(x)=0, which can be represented as the switch of free resource volume or step-type deduction in the communication service. The machine learning audit module 322 uses the ReLU function as the node activation function of the hidden layer, which can ensure the accuracy of the machine learning audit model in auditing the communication service data. The softmax layer is used to map the neuron output of the previous hidden layer into a normal billing probability value and an abnormal billing probability value. Among them, the normal probability value is the audit probability value γ of normal billing; the normal billing probability value + the abnormal billing probability value = 1. The output layer is used to calculate the normal billing probability value γ based on the normal billing audit probability value. Figure 2 Step S2043 in the illustrated embodiment determines the audit result and outputs the audit result.

[0185] Optionally, the machine learning audit module 322 can also be used to audit the input clean data using other machine learning algorithm models with equivalent functions to the BP neural network model to obtain audit results.

[0186] The communication service data auditing device provided by this application has the following specific implementation principles and technical effects: Figure 2 The implementation principle and technical effect of the illustrated embodiment are similar, and this embodiment will not be repeated here. In addition, the communication service data auditing device provided by this application adopts a machine learning auditing model as a BP neural network model. The algorithm principle of the BP neural network model has a high degree of matching with the billing algorithm of the communication service, which further ensures the higher accuracy of the machine learning auditing model for the communication service data audit.

[0187] The embodiment of the present application also provides a communication service data auditing device. Figure 5 This is a structural diagram of the communication service data auditing device provided in the embodiment of the present application. Figure 5As shown, the audit device includes a processor 51 and a memory 52, and the memory 52 stores executable instructions of the processor 51, so that the processor 51 can be used to execute the technical solution of the above method embodiment, and its implementation principle and technical effect are similar, and this embodiment will not be repeated here. It should be understood that the above processor 51 can be a central processing unit (English: Central Processing Unit, referred to as: CPU), or other general-purpose processors, digital signal processors (English: Digital Signal Processor, referred to as: DSP), application-specific integrated circuits (English: Application Specific Integrated Circuit, referred to as: ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the invention can be directly embodied as being executed by a hardware processor, or by a combination of hardware and software modules in the processor. The memory 52 may include a high-speed random access memory (English: Random Access Memory, referred to as: RAM), and may also include a non-volatile memory (English: Non-volatile memory, referred to as: NVM), such as at least one disk storage, and can also be a U disk, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0188] The embodiment of the present application also provides a storage medium, in which computer-executable instructions are stored, and when these computer-executable instructions are executed by a processor, the above-mentioned communication service data audit method is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (English: Static Random-Access Memory, referred to as: SRAM), electrically erasable programmable read-only memory (English: Electrically-Erasable Programmable Read-Only Memory, referred to as: EEPROM), erasable programmable read-only memory (English: Erasable Programmable Read-Only Memory, referred to as: EPROM), programmable read-only memory (English: Programmable Read-Only Memory, referred to as: PROM), read-only memory (English: Read-Only Memory, referred to as: ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0189] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (English: Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0190] An embodiment of the present application also provides a program product, such as a computer program, which, when executed by a processor, implements the communication service data audit method covered by the present application.

[0191] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0192] Finally, it should be noted that the above implementation modes are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned implementation modes, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned implementation modes, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the implementation modes of the present invention.

Claims

1. A communication service data auditing device, characterized in that: Applied to a communication service data audit system, the system includes an audit device and at least one data acquisition device; the audit device includes: a data transceiver unit, a data audit unit and a data correction unit; wherein, The data transceiver unit is used to obtain the user's communication service data and the service category to which the communication service data belongs from at least one data collection device; using the audit set, judging whether the service category is the first time to be audited, and sending the judgment result of the communication service data and service category to the data audit unit; the audit set includes all service categories that have not been audited; The data audit unit is used to audit the communication service data whose service category is the first audited service category using a preset manual audit model based on the judgment result of the communication service data service category to obtain an audit result; or, to audit the communication service data whose service category is not the first audited service category using a machine learning audit model to obtain an audit result; and send the audit result to the data transceiver unit; wherein the machine learning audit model is trained using communication service data corresponding to multiple audited service categories; and the audit result is any one of the following results: normal billing, reconfirmation required, and abnormal billing; The data transceiver unit is further used to generate an abnormal user work order for the user corresponding to the audit result for the communication service data with an abnormal billing result, so as to notify the work order processing personnel at the user's location to handle the process; The data transceiver unit includes a data transceiver module, a data cleaning module, and a data input module; the data audit unit includes a manual audit module and a machine learning audit module; wherein, The data transceiver module is used to obtain the user's communication service data and the service category to which the communication service data belongs from at least one data collection device; and use the audit set to determine whether the service category is the first time to be audited; The data cleaning module is used to clean the communication service data in the following manner to obtain corresponding clean data: Conflicting data is identified on the communication service data, and if no conflicting data is identified in the communication service data, the communication service data is determined to be primary cleansing data; the communication service data includes an actual outgoing value and service parameter values ​​corresponding to each of a plurality of service parameters; the conflicting data is a plurality of different service parameter values ​​corresponding to the same service parameter; the primary cleansing data is secondary cleansed to obtain pure data; the secondary cleansing includes data standardization conversion, missing data completion, and invalid data elimination; The data input module is used to input the clean data whose business category is the first audited business category into the manual audit module to obtain the audit result based on the business category to which the clean data belongs; or to input the clean data whose business category is not the first audited business category to obtain the corresponding multidimensional feature vector, and input the multidimensional feature vector into the machine learning audit module to obtain the audit result; the multidimensional feature vector P is composed of the actual payment feature value q and multiple business parameter feature values ​​Pi; wherein i is a natural number; The manual audit module is used to audit the input clean data using a preset manual audit model to obtain an audit result; The machine learning audit module is used to audit the input clean data using a multi-layer feedforward neural network model to obtain an audit result; the multi-layer feedforward neural network model includes an input layer, at least one hidden layer, a normalization layer and an output layer; the node activation function of the hidden layer is a linear rectification function; the activation function of the normalization layer is a normalized exponential softmax function.

2. The auditing device according to claim 1, characterized in that: The auditing device also includes a data correction unit; The data transceiver unit is further used to send the communication service data that the audit result shows needs to be reconfirmed to the data correction unit; The data correction unit is used to correct the received communication service data to obtain corrected service data, and send the corrected service data to the data transceiver unit; The data transceiver unit is further used to send the modified business data to the data audit unit; The data audit unit is also used to audit the modified business data based on the business category corresponding to the modified business data, using a preset manual audit model or the machine learning audit model to obtain a corresponding audit result.

3. A communication service data audit method, characterized in that: The communication service data auditing method is applied to the communication service data auditing device as claimed in claim 1 or 2, and the method comprises: Acquire the user's communication service data and the service category to which the communication service data belongs; Using an audit set, determining whether the business category is a business category being audited for the first time, the audit set including all business categories that have not been audited; If the service category is a service category that is audited for the first time, a preset manual audit model is used to audit the communication service data to obtain an audit result; If the service category is not the service category being audited for the first time, auditing the communication service data using a machine learning audit model to obtain an audit result; The machine learning audit model is trained using communication service data corresponding to multiple audited service categories.

4. The method according to claim 3, characterized in that: The communication service data includes actual outgoing value and multiple service parameter values; the audit result is any one of the following: outgoing normal, requiring reconfirmation, and outgoing abnormal; The adopting of a machine learning audit model to audit the communication service data to obtain an audit result includes: The communication service data is input and processed to obtain a multidimensional feature vector P consisting of an actual outgoing account feature value q and a plurality of service parameter feature values ​​Pi; Input the multidimensional feature vector into the machine learning audit model to obtain an audit probability value γ of normal payment; Based on the audit probability value γ, determining an audit result corresponding to the audit probability value γ; Wherein, i is a natural number.

5. The method according to claim 4, characterized in that The step of determining the audit result corresponding to the audit probability value γ based on the audit probability value γ includes: The audit probability value γ is compared with the preset threshold value δ0, and the audit result is determined as follows: If δ0<γ, the audit result is determined to be normal; If γ≤δ0, the audit result is determined to be an abnormal payment.

6. The method according to claim 4, characterized in that The step of determining the audit result corresponding to the audit probability value γ based on the audit probability value γ includes: The audit probability value γ and the preset lower limit threshold δ L and the upper threshold δ H The comparison is performed and the audit results are determined as follows: If δ H <γ, the audit result is determined to be normal; If δ L ≤γ≤δ H , then the audit result is determined to require reconfirmation; If γ<δ L , then the audit result is determined to be abnormal payment.

7. The method according to any one of claims 3 to 6, characterized in that: Before the preset manual audit model or the machine learning audit model is used to audit the communication service data to obtain the audit result, the method further includes: The communication service data is cleansed to obtain corresponding pure data.

8. The method according to any one of claims 3, 4 and 6, characterized in that: The audit result includes an audit result that needs to be reconfirmed; after determining that the audit result needs to be reconfirmed, the method further includes: Performing correction processing on the communication service data to obtain corrected service data; Based on the business category corresponding to the modified business data, a preset manual audit model or the machine learning audit model is used to audit the modified business data to obtain a corresponding audit result.

9. The method according to claim 7, characterized in that: After obtaining the audit result, the method further includes: Based on the communication business data corresponding to the audit result of the abnormal billing, an abnormal user work order for the user corresponding to the audit result is generated to notify the work order processing personnel in the user's location to handle it.

10. A communication service data auditing device, characterized in that: include: Processor and memory; The memory stores executable instructions executable by the processor; The processor executes the executable instructions stored in the memory, so that the processor executes the method according to any one of claims 3 to 9.

Images