Call bill data conversion method, electronic device, storage medium and program product

Abstract

The invention discloses a call ticket data conversion method, an electronic device, a storage medium and a program product, and relates to the technical field of mobile communication, and the call ticket data conversion method comprises the steps: inputting vector call ticket data corresponding to original call ticket data into a preset target call ticket conversion model, each ticket feature vector in the time sequence of the vector ticket data is extracted through a feature extraction layer of a target ticket conversion model, and the target ticket conversion model is used for predicting and obtaining corresponding standard ticket data according to the input vector ticket data; inputting each call ticket feature vector into a self-attention mechanism network of a target call ticket conversion model, and determining an attention weight between the call ticket feature vectors; and determining standard call ticket data corresponding to the original call ticket data according to the output layer and the attention weight of the target call ticket conversion model. According to the technical scheme, the flexibility and generalization ability of the call ticket data conversion process are improved.

Description

Technical Field

[0001] The present application relates to the field of mobile communication technology, and in particular to a call record data conversion method, an electronic device, a computer-readable storage medium, and a program product. Background Art

[0002] In the field of mobile communications, cloud service billing bills include more than 30 different types of bills, including mobile cloud non-SMS bills, WLAN (Wireless Local Area Network) bills, Internet of Vehicles bills, Internet of Things fixed-fee bills, Internet of Things GPRS (general packet radio service) / EPC (Evolved Packet Core) / LTE (Long Term Evolution) roaming bills, etc. In order to uniformly perform statistics and batch price processing on bill data, it is necessary to convert it into a standard bill in a unified format.

[0003] Currently, the commonly used call record conversion solution is usually to standardize the heterogeneous call records based on predefined rules and specifications. However, this processing solution can only be applied to known data formats and processing requirements. Moreover, when adding new types of call records, it is necessary to reconfigure the call record conversion matching rules, call record content verification rules, and call record content conversion rules, etc. The modification workload is large, so the current call record data conversion solution has poor flexibility and weak generalization ability.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a call record data conversion method, electronic device, computer-readable storage medium and program product, aiming to solve the technical problem that the current call record data conversion solution has weak generalization ability.

[0006] To achieve the above purpose, the present application proposes a call record data conversion method, the call record data conversion method comprising: Inputting the vector bill data corresponding to the original bill data into a preset target bill conversion model, extracting each bill feature vector in the time series of the vector bill data through the feature extraction layer of the target bill conversion model, wherein the target bill conversion model is trained by historical bill data and corresponding real bill data, and is used to predict the corresponding standard bill data according to the input vector bill data; Inputting each of the call record feature vectors into the self-attention mechanism network of the target call record conversion model to determine the attention weight between each of the call record feature vectors; According to the output layer of the target call record conversion model and the attention weight, the standard call record data corresponding to the original call record data is determined and output.

[0007] In one embodiment, before the step of inputting the vector bill data corresponding to the original bill data into a preset target bill conversion model, the method further includes: Preprocessing the original call record data, wherein the preprocessing includes at least one of deduplication, removal of invalid values ​​and abnormal values, and filling in missing values; The pre-processed original call record data is serialized to obtain vector call record data.

[0008] In one embodiment, after the step of determining and outputting the standard call record data corresponding to the original call record data, the method further includes: Querying the customer identifier in the standard call record data in a preset customer table; If the query is successful, the price is determined based on the standard call bill data to obtain the price information corresponding to the standard call bill data; If the query fails, the standard call record data is sent to the error record queue.

[0009] In one embodiment, before the step of inputting the vector bill data corresponding to the original bill data into a preset target bill conversion model, the method further includes: Acquire multiple types of historical call record data and corresponding real call record data, and train a pre-built initial call record conversion model based on each of the historical call record data and the real call record data; During the training process of the initial call record conversion model, updating the model parameters of the initial call record conversion model; When the loss function value of the initial call record conversion model meets the preset conditions, the training is stopped and the initial call record conversion model is set as the target call record conversion model.

[0010] In one embodiment, the step of training the pre-built initial call record conversion model based on each of the historical call record data and the real call record data includes: Dividing each of the historical call record data into a training set and a test set based on the time series of each of the historical call record data, wherein the data lengths of the historical call record data corresponding to the training set and the test set are different; Inputting the historical call record data in the training set into the initial call record conversion model, performing calculations through each network layer in the initial call record conversion model, and outputting predicted call record data; Calculating the loss function value of the initial call record conversion model according to the predicted call record data and the real call record data corresponding to the predicted call record data; Based on the change of the loss function value, the model parameters of the initial call record conversion model are iteratively updated.

[0011] In one embodiment, the step of dividing the historical call record data into a training set and a test set based on the time series of the historical call record data includes: The number of time steps of the training set is calculated according to the total number of time steps corresponding to the time series and the preset division ratio; Taking the start time step as the starting point, dividing the call record data corresponding to the continuous time steps of the number of training set time steps in the historical call record data into a training set; The call record data in the historical call record data except the training set is divided into a test set.

[0012] In one embodiment, the model parameters include a learning rate, and the step of iteratively updating the model parameters of the initial call record conversion model includes: Obtaining the current gradient of the initial call record conversion model and initializing the value of the optimization parameter to 0; Inputting the current gradient and the optimization parameter into a preset model parameter optimization function to obtain updated optimization parameters; Calculating a corresponding smoothing bandwidth according to the updated optimization parameters, and adjusting a learning rate of the initial call record conversion model based on the smoothing bandwidth; Return to the step of inputting the current gradient and the optimization parameters into a preset model parameter optimization function to obtain updated optimization parameters until model training is completed.

[0013] In addition, to achieve the above-mentioned purpose, the present application also proposes an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the call record data conversion method as described above.

[0014] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the call record data conversion method described above are implemented.

[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the call record data conversion method described above.

[0016] The present application proposes a method for converting call bill data, firstly inputting vector call bill data corresponding to the original call bill data into a preset target call bill conversion model, extracting each call bill feature vector in the time series of the vector call bill data through the feature extraction layer of the target call bill conversion model, wherein the target call bill conversion model is trained by historical call bill data and corresponding real call bill data, and is used to predict the corresponding standard call bill data according to the input vector call bill data, and then inputting each of the call bill feature vectors into the self-attention mechanism network of the target call bill conversion model, determining the attention weight between each of the call bill feature vectors, and finally determining and outputting the standard call bill data corresponding to the original call bill data according to the output layer of the target call bill conversion model and the attention weight. The technical solution of the present application utilizes the continuous features in the time series of the call bill data, so that the target call bill conversion model can capture the time relationship and context information in the time series of the call bill, and the conversion rules between the original call bill data and the standard call bill data learned by the target call bill conversion model during the training process can be used for call bill data conversion. Furthermore, the target call sheet conversion model uses a self-attention mechanism to determine the attention weights between call sheet features, thereby achieving effective conversion processing of various types of original call sheet data. Therefore, the target call sheet conversion model trained by deep learning in the technical solution of this application can adapt to heterogeneous call sheet data of various lengths and convert them into tabular call sheet data of a unified standard. It has strong generalization ability and does not need to configure corresponding conversion rules for different types of heterogeneous call sheet data, which improves the flexibility and adaptability of call sheet conversion and effectively reduces the workload. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 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.

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0019] Figure 1 This is a flow chart of the standardized processing of heterogeneous call records in the traditional solution; Figure 2 This is a flow chart of a method for converting call log data in an embodiment of the present application; Figure 3 This is a schematic diagram of the correspondence between subscripts and definitions in the CSMP original call record; Figure 4This is a schematic diagram of the correspondence between subscripts and definitions in the CDN original call list; Figure 5 A schematic diagram of the correspondence between subscripts and definitions in a standard call list in an embodiment of the present application; Figure 6 Schematic diagram of the subscript conversion rules between the CSMP original call list and the standard call list in the embodiment of the present application; Figure 7 This is a schematic diagram of the subscript conversion rules between the CDN original call list and the standard call list in the embodiment of the present application; Figure 8 A schematic diagram of a CSMP original call record sample before conversion and a standard call record after conversion in an embodiment of the present application; Fig. 9 A schematic diagram of a CDN original call bill sample before conversion and a standard call bill after conversion in an embodiment of the present application; Fig.10 A schematic diagram of the expected training result of the CSMP original call record in an embodiment of the present application; Fig.11 This is a schematic diagram of the expected training results of the CDN original call list in the embodiment of the present application; Fig.12 This is a schematic diagram showing that an error exists in the product ID during the training of the initial call record conversion model in an embodiment of the present application; Fig.13 This is a schematic diagram of data before and after the conversion of the CSMP original call record after the model parameters are adjusted in the embodiment of the present application; Fig.14 This is a data schematic diagram of the CDN original call record before and after conversion after the model parameters are adjusted in the embodiment of the present application; Fig.15 The standard bill data is the data obtained by converting a CDN bill with a bill length of 125 bits and a CDN bill with a bill length of 127 bits in the embodiment of the present application; Fig.16 This is a schematic diagram of the device structure of the hardware operating environment involved in the call record data conversion method in the embodiment of the present application.

[0020] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0022] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, a server, etc., or an electronic device or a control device capable of realizing the above functions. The following takes the server as an execution subject as an example to illustrate this embodiment and the following embodiments.

[0024] The current limitations of the standardized processing solutions for heterogeneous call records are mainly due to their poor adaptability to data changes and difficulty in processing complex heterogeneous data and large-scale data processing requirements. With the explosive growth of cloud services, a large amount of call record data of different structural types will be connected. If a new type of call record is added each time (changes in call record length, call record mapping rules, call record conversion rules), it is necessary to configure new conversion rules in the database and modify the call record conversion process to be compatible with the conversion of this type of call record. For unknown or frequently changing data models, it is necessary to frequently update and adjust the rules, which will be very complicated and cumbersome and the workload will be too large.

[0025] like Figure 1 As shown, after the original call billing is connected to the gateway process, it is necessary to perform call bill type matching, call bill content verification, and call bill content conversion. Each time a new type of call bill is added, it is necessary to reconfigure the call bill matching rules, call bill content verification rules, and call bill content conversion rules. At the same time, the program of the billing offline access process needs to be modified to be compatible with the call bill conversion.

[0026] Since traditional standardized processing solutions for heterogeneous call records usually process known call records based on established rules and specifications, they have poor adaptability, flexibility and generalization capabilities, and cannot better cope with the complexity and challenges of processing multiple heterogeneous data.

[0027] In order to overcome the technical problems and defects in the above-mentioned prior art, the present application embodiment provides a call data conversion method, referring to Figure 2 , Figure 2 This is a flow chart of the first embodiment of the call record data conversion method of the present application, and the call record data conversion method includes: Step S10, inputting the vector bill data corresponding to the original bill data into a preset target bill conversion model, extracting each bill feature vector in the time series of the vector bill data through the feature extraction layer of the target bill conversion model, wherein the target bill conversion model is trained by historical bill data and corresponding real bill data, and is used to predict the corresponding standard bill data according to the input vector bill data; The call bill data conversion method proposed in the embodiment of the present application mainly utilizes a pre-trained target call bill conversion model, which can learn long-term dependencies based on the time series of the learned call bill data, where the time series refers to the sequence formed by the attribute fields in the original call bill data sorted by time. It can memorize and utilize the conversion rules between historical call bill data and real call bill data, and use the previously learned knowledge as context information to assist the current prediction or analysis. Specifically, the feature extraction layer in the target call bill conversion model is used to extract the call bill feature vector of the vector call bill data of the original call bill, so as to facilitate further identification and calculation by other network layers in the target call bill conversion model.

[0028] Exemplarily, the information extracted by the feature extraction layer includes the user's ordering information, call duration, and traffic volume, etc.

[0029] It should be noted that the original call bill data may include various types of call bill data, which may be types processed by the target call bill conversion model during training, or new types. Vector call bill data is a type of data with array subscripts, hidden layer dimension definition variables, and output feature dimension definition variables. For example, the first feature vector array subscript in a vector call bill data is [0], the hidden layer dimension definition variable is 64, and the output feature dimension definition variable is 1. In addition, the real call bill data refers to the corresponding standard call bill data obtained by converting the original call bill data according to the conversion rules between the original call bill data and the standard call bill data. The real call bill data may be obtained by manual conversion or by conversion according to a pre-set conversion rule for the original call bill data type.

[0030] For example, taking the mobile cloud non-SMS call record (CSMP) as an example, each line in the call record is a sample (for example, there are the following heterogeneous original call records -20,2024041001050000000485982685, 20240410010513,E7712024031851891153 ,1010402,111716,0005,10034264537,32,2000cdda-2e20-4da0-aebb-b6201e,20240410000000,2024041001000 0,0000000000,0000000000,0000000000000000,0000000000000000,0000,00000000,00000000,111716000000). It is understandable that the processing of the call record only involves a CSMP original call record consisting of 20 attribute fields. In this example, only these 20 attribute fields need to be processed. Figure 3 As shown, the CSMP original call bill can have up to 25 attribute fields (0-24), and the definition corresponding to each subscript is different. In another feasible embodiment, taking the mobile CDN (Content Delivery Network) call bill as an example, the present application can at least process different types of heterogeneous call bills such as CSMP original call bills and CDN original call bills. Each line in the CDN original call bill is also a sample (for example, there is the following CDN original call bill (20,2024041013450100000002,20240410134500,01,10017079643,100,0,0,0,1,1,2,591,1). It can be understood that, with reference to Figure 4 , the bill processing only involves a CDN original bill consisting of 13 attribute fields (0-12). In this example, only these 13 attribute fields need to be processed.

[0031] In order to facilitate model recognition and format unification, each feature (including null) is separated by a separator (such as a comma), and each line of the converted standard call record is a label (for example, the standard call record after CSMP conversion—a,20,,,ZB20231129101005AUTO1001010,,,,10034264537,20240402,100000,20240402,105439,,,,,0001 ,1010402,,,,,,,,,,,,,94,,,,0,,,,,,,0000,,,,,,,,,,,,,,,,,,,,,,123454,123453,CSMP20231 12914305000009200410,,,,E7712024031851891153,9200553,123451,,11,9200553001,123452,0 And the CDN converted bill a,20,,01,2024041013450100000002,,,,10017079643,,20240416,080000,20240416,080000,,,,,,,,,,,,100,,,,,09,,1,,,,,,,,1,,,,,,,,,,,,,,,,,,,,,1,2,,,CDN_ECLOUD_20240416.0003,,,,,10017079643,1367,,,,,,,,,,,,). It can be found that Figure 5As shown, in this example, the call record processing only involves the standard call record consisting of 78 attribute fields (0-77). This means that in this example, only these 78 attribute fields need to be processed. It should be noted that this application is not limited to the number of fields in the standard call record and the original call record. The above is only an example.

[0032] Furthermore, the conversion between the CSMP original call record and the corresponding real call record data (i.e., standard call record data) is mainly the conversion between subscripts. After the subscript in the CSMP original call record is converted to the standard call record data, it is updated to a new subscript. Exemplarily, the conversion rule between the CSMP original call record and the standard call record can be represented by a function F: X ->Y to represent a one-to-one correspondence from the CSMP original call record set X to the standard call record set Y. The subscript conversion rule between the CSMP original call record and the standard call record is as follows: Figure 6 Similarly, the conversion rules between CDN original bills and standard bills use the function F:Z ->Y to represent a one-to-one correspondence between the CDN original bill set Z and the standard bill set Y. The subscript conversion rules between CDN original bills and standard bills are as follows: Figure 7 shown.

[0033] It should be noted that the definition of the above standard call record takes into account the attribute fields required in the technical problem to be solved by this application. Therefore, all attribute fields of the standard call record are determined according to the technical problem to be solved, and there are no unnecessary attribute fields. Figure 5 The standard call record shown also reserves extension fields from subscripts 49 to 53 to facilitate the expansion of fields according to business needs in the future. It can be understood that Figure 5 Only five extension fields are reserved in the model. In actual applications, staff can reserve more extension fields according to needs to ensure the generalization ability of the model.

[0034] For ease of understanding, the historical call data (CSMP original call data or CDN original call data) used in the model training has been converted to the real call data. For example, the CSMP original call data sample before conversion and the standard call data after conversion are as follows: Figure 8 As shown in the figure, it includes the CSMP original call bill (sample) and the converted standard call bill (labeled), where the standard call bill can be understood as the real label of the original call bill; the CDN original call bill sample before conversion and the converted standard call bill are shown in the figure. Fig. 9 As shown in the figure, it includes the CDN original call record (sample) and the converted standard call record (labeled). It can be understood that the meaning of the attribute field and its field value are two related and corresponding information. Therefore, even if the same field value appears, such as 20240509, it represents different meanings. Figure 3 , Figure 4 The meaning of different attribute fields must be reflected in the text. Therefore, in the processing of the technical solution of this application, the meaning of the attribute field and its field value need to be examined at the same time, and there is no situation of incorrect conversion. Figure 3 The corresponding CSMP original call is one-way Figure 5 Defined standard call record conversion, Figure 4 Towards Figure 5 The defined standard call record conversion is subject to the constraints of its conversion rules, and will not cause the situation where fields with the same field value but different meanings are converted incorrectly.

[0035] Step S20, inputting each call record feature vector into the self-attention mechanism network of the target call record conversion model, and determining the attention weight between each call record feature vector; The target call bill conversion model in the embodiment of the present application applies the self-attention mechanism, which is a technical means for capturing the relationship between different parts of the input data. The self-attention mechanism is widely used in processing sequence data, such as natural language processing and time series analysis. When processing heterogeneous call bill data, the self-attention mechanism network in the target call bill conversion model can help the model understand the temporal relationship, feature association and contextual information in the data. Specifically, the self-attention mechanism network is used to calculate the attention weights corresponding to the traffic usage, traffic timestamp, traffic type, etc. in the call bill.

[0036] Step S30, determining and outputting standard call record data corresponding to the original call record data according to the output layer and attention weight of the target call record conversion model.

[0037] It should be noted that the forward propagation method is used in the embodiment of the present application to generate the final prediction result, that is, the standard call record data. Forward propagation (FORWARD PROPAGATION) refers to the process of passing the input data from the input layer of the model to the output layer. In this process, the neurons in each layer receive the output of the previous layer, and after weighted summation, obtain the output of this layer through the activation function, and then pass it to the next layer. This process continues until it reaches the output layer, and the output layer generates the final prediction result. In the forward propagation process, there is no update of parameters involved, only the input is converted according to the current network weights and biases to generate the prediction result.

[0038] For example, the user ordering information extracted in the feature extraction layer is Figure 5 The mobile phone number with subscript 8 in the feature extraction layer corresponds to the call duration extracted from Figure 8 The call time information in subscripts 10 to 14 (including call start date, call start time, call end date, call end time) corresponds to, where, Figure 5The call duration in subscript 14 corresponds to the call duration extracted in the feature extraction layer, and the traffic extracted in the feature extraction layer corresponds to Figure 5 The uplink traffic and the downlink traffic in subscripts 63 to 64 correspond to each other.

[0039] Taking the original call bill of Mobile Cloud CSMP as an example, in the process of converting the input vector call bill data to generate the corresponding standard call bill data, the model uses the forward propagation function (FORWARD) to take the output of the previous layer in the model (the number of traffic hops in the call bill) as the input of the next layer in the model, and calculate the output result of the next layer (traffic usage) until the operation reaches the output layer. The definition of the forward propagation function can be expressed as DEF FORWARD (64, 1), and then the input call bill data is passed through the various layers of the target call bill conversion model through the forward propagation function, and then operated and converted in sequence.

[0040] For example, for the input layer to the first hidden layer, the expression of the forward propagation function can be:

[0041] For l Layer to l +1 layer ( l =1,2,…, L -1), the expression of the forward propagation function can be:

[0042] Wherein, X in the above formula represents the input of the model, which represents the original CDN call record data; W represents the weight matrix and b represents the bias, neither of which has any actual physical meaning; z is an intermediate result and has no actual physical meaning; f is the activation function and has no actual physical meaning; a is the activation value and has no actual physical meaning; L is the total number of layers. The above variables and functions are all vectors and functions that process vectors.

[0043] The forward propagation function formula is described in detail below with its formula and specific examples.

[0044] Input x is the original CDN bill data, for example (20,2024041013450100000002, 20240410134500,01, 10017079643,100,0, 20,0,1,1,2,591,1).

[0045] Weight Matrix : Each weight corresponds to the strength of the connection from an input field to a hidden layer node. For example, the weight for connecting subscript [0] call record type (value 20) to hidden layer node 1 is .

[0046] The values ​​of the weight matrix are exemplarily: [0.12345678 0.23456787 0.3456788 0.45678901 0.56789012 0.67890123 0.78901234 0.89012345 0.90123456 0.012345670.01234567 0.01234568 0.01234569 0.01234561].

[0047] Bias : The bias of hidden layer node 1 is .

[0048] The value of the bias vector is exemplarily: [ 0.12345677 0.23456789 0.3456789 0.456789030.56789012 0.67890123 0.78901235 0.89012345 0.90123456 0.01234563 0.012345540.01234532 0.01234521 0.01234569 ].

[0049] Intermediate results : The sum of weighted input and bias, hidden layer node 1 = *20+ .

[0050] The values ​​of the intermediate results are exemplarily: [ 0.12345611 0.23456715 0.3456717 0.45678912 0.56789021 0.67890133 0.78901255 0.89012385 0.90123459 0.01234560 0.01234559 0.01234562 0.01234591 0.01234534 ].

[0051] Activation Function : For each Apply nonlinear changes to get activation values , the subscripts of the first subsequent row start from 1, and so on.

[0052] The activation value a is exemplarily: [ 0.124456263 0.234567292 0.34567390.45678780.5678549 0.6712323 0.78901447 0.89013455 0.90126789 0.012341270.0123412245 0.01234582 0.01234503 0.01234589 ].

[0053] Exemplarily, the activation function f may be a ReLU function or a Sigmoid function, etc.

[0054] Further, the above forward propagation function formula is further exemplified from layer 3 to layer 3+1: Weight Matrix : Each weight corresponds to the strength of the connection from an input field to a hidden layer node. For example, the weight for connecting subscript [0] call record type (value 20) to hidden layer node 1 is .

[0055] The values ​​of the weight matrix are exemplarily: [[ 0.12345678 0.23456789 0.3456789 0.45678901 0.567890120.678901230.78901234 0.89012345 0.90123456 0.01234567 0.01234120 0.0123412190.01234572 0.01234593]; [-0.12345678 0.23456789 -0.3456789 0.45678901 0.567890120.678901230.78901234 0.89012345 0.90123456 0.01234567 0.01234120 0.0123412190.01234572 0.01234593]; [ 0.45678901 -0.3456789 0.23456789 0.12345678 0.567890120.678901230.78901234 0.89012345 0.90123456 0.01234567 0.01234120 0.0123412190.01234572 0.01234593]; [ 0.23456789 0.12345678 -0.45678901 -0.3456789 0.567890120.678901230.78901234 0.89012345 0.90123456 0.01234567 0.01234120 0.0123412190.01234572 0.01234593]].

[0056] Bias : The bias of hidden layer node 1 is The value of the bias vector is exemplarily: [[ 0.12345679 0.23456789 0.3456789 0.45678901 0.56789012 0.67890123 0.78901234 0.89012345 0.90123456 0.01234567 0.01234122 0.012341211 0.01234572 0.01234593]; [-0.12345670 0.23456789 -0.3456789 0.45678901 0.567890120.678901230.78901234 0.89012345 0.90123455 0.01234567 0.01234123 0.012341210.0123457250.01234594]; [ 0.4567890 -0.3456789 0.23456789 0.12345678 0.567890120.678901230.78901234 0.89012345 0.90123453 0.01234567 0.01234124 0.012341190.01234572 0.01234595]; [ 0.2345678 0.12345678 -0.45678901 -0.3456789 0.567890120.678901230.78901234 0.89012345 0.90123456 0.01234567 0.01234125 0.01234110.01234574 0.01234596]].

[0057] Intermediate results : The sum of weighted input and bias, hidden layer node 1 = * + ; The values ​​of the intermediate results are exemplarily: [[ 0.123456 0.234567 0.34567 0.456789 0.567890 0.6789010.7892340.890145 0.90123456 0.012347 0.012341 0.0123411 0.01234572 0.012593]; [0.12345673 0.23456789 -0.34567890.45678901 0.567890120.678901230.78901234 0.89012345 0.90123455 0.01234567 0.01234123 0.012341210.0123457250.01234594]; [ 0.45678901 -0.3456789 0.23456789 0.12345678 0.56789012 0.67890123 0.789012343 0.890123453 0.901234535 0.012345678 0.012341249 0.012341194 0.01234572 0.01234595 ]; [ 0.1345678 0.123456783 -0.45678901 -0.3456789 0.567890120.678901230.78901234 0.89012345 0.90123456 0.01234567 0.01234125 0.01234110.01234574 0.01234598]].

[0058] Activation Function : For each Apply nonlinear changes to get activation values , the subscripts of the first subsequent row start from 1, and so on.

[0059] The activation values ​​are exemplarily: [[ 0.323459 0.534567 0.34567 0.4567890.5678900.678901 0.789234 0.890145 0.90123456 0.012347 0.012341 0.0123410.01234572 0.012593]; [0.12345679 0.23456789 -0.3456789 0.45678901 0.56789012 0.67890123 0.78901234 0.89012345 0.90123455 0.0123456 0.01234123 0.01234121 0.012345725 0.01234594]; [ 0.45678902 -0.3456789 0.23456789 0.12345678 0.56789012 0.67890123 0.789012343 0.890123453 0.901234535 0.012345678 0.01234124 0.012341194 0.01234572 0.01234595]; [ 0.1345611 0.123456783 0.45678901 0.3456789 0.56789012 0.6789013 0.78901234 0.89012345 0.90123456 0.01234567 0.0123412 0.0123410.0123574 0.01234598]].

[0060] In summary, through the conversion between layers, the fields in the input original call record are gradually converted and filled into the standard call record, and (a, 20,, 01, 2024041013450100000002,,,, 10017079 643,,20240416,080000,20240416,080000,,,,,,,,,,,,,100,,,,,09,,1,,,,,,,,1,,,,,,, ,,,,,,,,,,,,,1,2,,20,CDN_ECLOUD_20240416.0003,,,,,10017079643,1367,,,,,,,,,,,).

[0061] It should be noted that the target call bill conversion model includes a self-attention mechanism network to calculate the weights of traffic usage, traffic timestamp, and traffic type in the call bill. For example, the weight calculation function can be defined as ,in and are two vectors in the input sequence, and f is a similarity measure function. , where the traffic is 1024KB and the hop number is 2, the result is 2048KB.

[0062] Reference Fig.10 , the 3rd customer ID (E7712024031851891153) of the original CSMP bill should correspond to the 69th digit of the standard bill after model processing and the spaces need to be removed. The 7th subscription ID (10034264537) of the original CSMP bill should correspond to the 8th digit of the standard bill after model processing and the spaces need to be removed.

[0063] Taking the original call record of Mobile Cloud CDN as an example, the weight calculation function can be defined for the weights of the input layer bandwidth and product ID in the model: ,in and are two vectors in the input sequence, and f is a similarity measure function. , the bandwidth is 20M, and the hop number is 1, then the result is 20M.

[0064] Reference Fig.11 , the 4th product ID (10017079643) of the original CDN bill should correspond to the 70th of the standard bill after model processing and the spaces need to be removed. The 7th bandwidth (20) of the original CDN bill should correspond to the 64th of the standard bill after model processing and the spaces need to be removed.

[0065] The embodiment of the present application provides a method for converting heterogeneous call records to obtain unified standard call record data, which fully utilizes the continuity of the time series data of mobile cloud call records, and allows the call record data to be continuously transmitted and calculated within the model, and learns long-term dependencies from the sequence of call records by effectively capturing the time relationship and context information between the sequence data. The target call record conversion model is trained in advance with a large amount of historical call record data, so that the adaptability and robustness of the model are more perfect, thereby realizing the automatic conversion processing of call records of different types and lengths. The technical solution of the embodiment of the present application avoids the situation where the program and configuration must be modified every time a new call record type is added, which brings a lot of workload, improves the flexibility, adaptability and generalization ability of call record conversion, can more accurately convert to obtain standard call record data in order to predict user behavior or trends, effectively simplifies the call record conversion process, and reduces the workload of manual intervention by developers.

[0066] In the field of mobile communications, staff can use the technical solutions of the embodiments of the present application to process complex data, including the identification and processing of abnormal mobile cloud call records and the optimization of network resources. The target call record conversion model can automatically learn the user's ordering patterns in the call record data, so as to make intelligent decisions and predict the user's ordering behavior based on the learned knowledge, which is conducive to the dynamic adjustment and expansion of marketing strategies according to the complexity of the call record data and the requirements of the task.

[0067] In a feasible implementation manner, before the step of inputting the vector bill data corresponding to the original bill data into a preset target bill conversion model, the method further includes: Step S11, preprocessing the original call record data, wherein the preprocessing includes at least one of deduplication, removal of invalid values ​​and abnormal values, and filling of missing values; Step S12, serializing the pre-processed original call record data to obtain vector call record data.

[0068] It is understandable that after obtaining the original call record data that needs to be converted, certain data cleaning (i.e. preprocessing) is required to remove duplicate data, invalid data, and abnormal data, and to fill in missing values ​​to standardize its format for easy conversion into corresponding vector call record data.

[0069] After data cleaning is completed, because the target call bill conversion model can only recognize and process data in vector form, the original call bill data needs to be serialized and converted into vector form. Among them, the vector call bill data includes at least three parameters, one is the feature vector array subscript (for example, [0]), the second is the hidden layer dimension definition variable (for example, 64), and the third is the output feature dimension definition variable (for example, 1).

[0070] The embodiment of the present application provides a method for preprocessing and serializing original call record data to obtain vector call record data for facilitating identification preprocessing of a target call record conversion model.

[0071] In a feasible implementation manner, after the step of determining and outputting the standard call record data corresponding to the original call record data, the method further includes: Step S40, querying the customer identifier in the standard call record data in the preset customer table; Step S50, if the query is successful, the price is determined according to the standard call bill data to obtain the price information corresponding to the standard call bill data; Step S60: If the query fails, the standard call record data is sent to the error record queue.

[0072] It is understandable that if the customer identifier (such as customer ID) and the subscript of the subscription ID of the original call sheet in the formal generation environment are wrong, then the order of the standard call sheet customer ID and subscription ID obtained after model processing will also be wrong. In order to improve the robustness of the call sheet data conversion method program in the embodiment of the present application and avoid errors in the converted standard call sheet data leading to errors in the pricing process, after the target call sheet conversion model outputs the standard call sheet data, the customer ID in the standard call sheet data is first queried in the preset customer table to determine whether the customer ID is accurate. If it is accurate, the query will be successful, that is, the customer ID exists in the customer table; if it is inaccurate, the query fails, indicating that the customer ID is wrong, and pricing cannot be performed based on the standard call sheet data, and it is divided into the wrong order queue. The wrong order queue is used to store the erroneous standard call sheet data, which is convenient for the staff to query the erroneous standard call sheet data and the cause of the error later.

[0073] Specifically, after the query is successful, it means that the customer ID and standard call bill data are accurate, and then the batch pricing process is further executed. Batch pricing refers to calculating the price information involved in each service based on the customer's call bill data, thereby counting the customer's tariff information for mobile communication service settlement.

[0074] In a feasible embodiment, before the step of inputting the vector call record data corresponding to the original call record data into a preset target call record conversion model, the method further includes: Step A10, obtaining various types of historical call record data and corresponding real call record data, and training a pre-built initial call record conversion model based on each type of historical call record data and the real call record data; An embodiment of the present application also provides a method for training a target call bill conversion model. Specifically, a large amount of heterogeneous historical call bill data of multiple different types is first collected, and these historical call bill data are converted using a traditional call bill conversion scheme based on conversion rules to obtain corresponding real call bill data. The real call bill data refers to the standard call bill data obtained after these different types of heterogeneous historical call bill data are uniformly standardized and converted, and is used as the real label corresponding to the historical call bill data used for training for supervised learning.

[0075] It should also be noted that the pre-built initial call conversion model refers to a pre-built neural network model that has completed the initialization of model parameters. The model may include multiple network layers such as input layer, feature extraction layer, RNN (Recurrent Neural Network), self-attention mechanism, generation layer, normalization layer, output layer, etc., and uses forward propagation and back propagation methods to process data and adjust and update model parameters.

[0076] Step A20, in the training process of the initial call record conversion model, updating the model parameters of the initial call record conversion model; In the process of model training, it mainly involves the calculation of the gradient of the loss function for the output call sheet model and the propagation of the error from the output layer to the hidden layer and the updating and adjustment of the parameters. For example, the hidden layer is updated and adjusted to 128, because in the process of training, the gradient may disappear, that is, the gradient of the underlying parameters of the call sheet model is almost zero, resulting in the inability of the underlying network to learn effective feature representations, and the gradient explodes, that is, the gradient of some parameters becomes very large, resulting in excessive parameter updates. Therefore, in the embodiment of the present application, the back propagation method can be used to timely update the hidden layers and parameters of the training, thereby improving the accuracy of the training results.

[0077] Among them, back propagation refers to the process of propagating the gradient from the output layer to the input layer according to the error value (i.e., the loss function value) calculated by the loss function. Specifically, back propagation calculates the gradient of the parameters of each layer to the loss function through the chain rule, and then uses this gradient information to update the parameters in the network to minimize the loss function value and maximize the prediction accuracy of the model.

[0078] In addition, the model parameters also include the learning rate. In machine learning (ML), the learning rate is a hyperparameter that determines the step size of updating model parameters during training. The larger the learning rate, the greater the adjustment of model parameters in each training. The historical call data has sufficient training data sources, so it can support a large number of iterative training. The learning rate can be set relatively small, such as between 0.001 and 0.01. The batch size in the model parameters can be set to 64, and the training round can be set to 20.

[0079] Assume that the training process fails to meet the training standards. For example, when the training call record data is small (1,000 original call records), the difference or error between the prediction result of the call record model and the result (label) after the actual call record conversion is large, and the loss function value on the Loss test set is 8.34 (the closer to 0, the higher the prediction accuracy of the model). Fig.12 As shown in the figure, after analysis, it is found that due to the abnormal conversion of the product id (111716), the product id corresponding to the standard call record should be the 70th, but due to the small sample data for training, it was converted to the 75th, resulting in a large error in this training. In this case, you can adjust the parameters of the model, for example: increase the number of training calls (increase to 1,000,000), adjust the batch size (adjust the batch size to 100), and adjust the training rounds (adjust the training rounds to 50).

[0080] Step A30, when the loss function value of the initial call record conversion model meets the preset conditions, stop training and set the initial call record conversion model as the target call record conversion model.

[0081] In an embodiment of the present application, a condition for terminating model training is provided, that is, whether to stop training is determined based on the loss function value. It can be understood that the smaller the loss function value, the better the performance of the model and the higher the prediction accuracy. Regarding the preset condition, it can be a threshold or convergence. For example, a threshold of the loss function value is set. After the loss function value of the initial call sheet conversion model is lower than the threshold at a certain time, it can be determined that the initial call sheet conversion model has met the preset condition, and the training can be stopped and the initial call sheet conversion model can be set as the target call sheet conversion model. On the other hand, the preset adjustment can also be set to the convergence of the loss function value. For example, during the model training process, the loss function value changes in a smaller direction and continues to decrease until it reaches a minimum value. It can be determined that the loss function value has converged. At this time, the training can be stopped and the initial call sheet conversion model can be set as the target call sheet conversion model.

[0082] Further, in a feasible embodiment, the step of training the pre-built initial call record conversion model based on each of the historical call record data and the real call record data includes: Step A11, dividing each historical call record data into a training set and a test set based on the time series of each historical call record data, wherein the data lengths of the historical call record data corresponding to the training set and the test set are different; It should be noted that in the process of model training, in order to improve the generalization and adaptability of the model, it is necessary to divide the historical call data used for training into corresponding training sets and test sets to complete the training and testing steps in model training. Among them, the call data in the test set is different in structure, length or type from the call data in the training set to test the model's conversion ability when dealing with different types of original call data. The purpose of the test set is to add some unconventional types of call sample data (such as an overlong field - the fourth digit of the original call is the customer id - E7712024031851891153, the default length is 30 characters, which can be adjusted to 40 characters) to improve the adaptability of the model.

[0083] Step A12, inputting the historical call record data in the training set into the initial call record conversion model, performing calculations through each network layer in the initial call record conversion model, and outputting predicted call record data; During the model training process, the historical call bill data in the training set (equivalent to the original call bill data) can be grouped and input into the constructed initial call bill conversion model. The initial call bill conversion model includes input layer, feature extraction layer, RNN layer, self-attention mechanism network, generation layer, normalization layer, output layer and other network parameters, all of which are initialized. Through the operation of these network layers, a predicted call bill data corresponding to the input historical call bill data can be predicted. The predicted call bill data is the standard call bill data obtained after converting the historical call bill data through the initial call bill conversion model.

[0084] It is understandable that the predicted call record data is not necessarily accurate and is used to calculate the loss function value of the model together with the real call record data during the model training process.

[0085] Step A13, calculating the loss function value of the initial call record conversion model according to the predicted call record data and the real call record data corresponding to the predicted call record data; In an embodiment of the present application, a mean square error loss function may be used to calculate the loss function value of the model.

[0086] For example, for a regression problem with n samples, the model's prediction result is , the corresponding true label is , then the mean square error function can be expressed as: .

[0087] For example, if the true value is y=[3.0,4.5,2.7,6.0,5.5]; The expected value is y=[2.8,4.7,3.0,6.2,5.4]; Compute the squared error for each sample:

[0088] Then sum the squared errors: 0.04+0.04+0.09+0.04+0.01=0.220.04+0.04+0.09+0.04+0.01=0.22; Divide the sum by the number of samples = 0.044, which means the mean squared error and loss function value is 0.044.

[0089] Step A14: iteratively update the model parameters of the initial call record conversion model based on the change of the loss function value.

[0090] Finally, some model parameter optimization algorithms commonly used in model training are used to iteratively update the model parameters of the initial call conversion model based on the loss function value. The optimization goal is to minimize the loss function value, such as gradient descent algorithm, adaptive learning rate method, grid search algorithm, random search algorithm, genetic algorithm, etc.

[0091] It should be noted that steps A12 to A14 need to be performed in a loop before the loss function value meets the preset conditions (refer to step A30) to continuously iterate and optimize the model parameters of the initial call record conversion model.

[0092] Further, in a feasible embodiment, the step of dividing the historical call record data into a training set and a test set based on the time series of the historical call record data includes: Step A11, calculating the number of time steps of the training set according to the total number of time steps corresponding to the time series and the preset division ratio; Step A12, taking the start time step as the starting point, dividing the call record data corresponding to the continuous time steps of the number of training set time steps in the historical call record data into a training set; Step A13: divide the call record data in the historical call record data except the training set into a test set.

[0093] In an embodiment of the present application, a data set may be divided into a training set and a test set according to a time series, wherein the number of samples in the training set and the test set is determined according to the length of the time series and a preset division ratio, wherein the division ratio includes a ratio corresponding to the training set and a ratio corresponding to the test set, and the sum of the two ratios is 1. For example, if there are T time steps (i.e., the total number of time steps) in the time series of a user's historical call data, and the division ratio is P, then the number of time steps in the training set is T *P, and the number of time steps in the test set is T * (1 - P), that is, of all the historical call data collected, 80% (i.e., the ratio corresponding to the training set) of the data is the training set, and 20% (i.e., the ratio corresponding to the test set) of the data is the test set, wherein the starting time step is the earliest time step.

[0094] In the embodiment of the present application, the time step is used as the basis for dividing the training set and the test set. Since the call record conversion model in the present application mainly uses the continuity of time series data for prediction, the time steps of the time series in the training set and the test set are continuous. Among them, the historical call record data in the training set is used to input into the initial call record conversion model during the training process to further determine the predicted call record data and thus determine the loss function value of the model, while the test set uses heterogeneous call record data of different lengths from the test set to test the model's adaptability, accuracy and generalization ability when facing new data, further enhancing the performance and stability of the model, so that the model can process call record sequence data of indefinite length, and is more suitable for conversion scenarios of heterogeneous call records such as data of non-fixed length, and does not require fixed-length padding or truncation operations.

[0095] The target call bill conversion model trained in the embodiment of the present application can memorize and utilize the correlation relationship of the time series in the historical call bill data through the deep learning and training process, and use the previous records as context information to assist the current prediction or analysis. With the access to massive call bill data, the model will become more and more perfect, and the prediction ability and data processing effect will be greatly improved.

[0096] It should be noted that experiments have shown that the target call bill conversion model in the embodiment of the present application has a good conversion effect on call bill data of different lengths. For example, due to the CDN live broadcast source station and overseas value-added service business requirements of a mobile communication network operator, the call bill length changes and is compared with the previous Figure 3 CSMP call records and Figure 4 The CDN call bill has additional attribute fields. If the traditional call bill conversion solution is followed, the program code needs to be modified to be compatible with the processing of CDN call bills with lengths of 125 and 127 bits. The configuration data also needs to be modified, and testers need to run different types of CDN call bills in batches to verify the robustness of the program, which will bring a lot of workload. The call bill data conversion method provided by the application embodiment does not require the above-mentioned cumbersome operations. Since the embodiment of the application trains a call bill conversion model, and a large amount of different types of call bill data are used in the early training process, for this type of call bill that has only changed its length and all fields do not exceed the field range of the standard call bill, the conversion rules itself have not changed, and the model can directly perform the conversion process without additional workload.

[0097] For example, the length of the bill with 125 bits before the added attribute field is (20,2024050210200 100000001,20240509102000,01,10017079643,551, 0, 0, ,1,1,2,591; the length of the bill after the new attribute field is 127 bits (20,2024050210200100000001,20240509102000, 01,10017079643,551, 0,,1,1,2,591,3). When converting in this example: when the conversion rules of examples such as F (X, Y) and F (Z, Y) remain unchanged, since all attribute fields of the final standard call record are as high as 78 (such as steps), and are related to the data length of the 78 attribute fields and their respective fields, and the 78 attribute fields have taken into account the attribute fields required for the conversion process of all known types of call records in the current mobile communication network services, and have fully considered the attribute fields required for call record conversion processing in the future, such as 5.5G and 6G mobile communication network technologies, even if a new type of call record suddenly involves additional attribute fields, the call record data conversion method of the embodiment of the present application can still achieve conversion. For example, after converting the call record before and after the type containing the newly added attribute fields through the target call record conversion model, the following is obtained. Fig.15 The conversion results of the CDN bill with a length of 125 bits and the CDN bill with a length of 127 bits are both good. This is because the extension of the field is taken into consideration when designing the standard bill data. The 49th to 53rd bits of the standard bill are used to store the content of the extension field (refer to Figure 5 ), so the target call record conversion model in the embodiment of the present application can process call record sequence data of uncertain length and has strong generalization ability.

[0098] In a feasible embodiment, the model parameters include a learning rate, and the step of iteratively updating the model parameters of the initial call record conversion model includes: Step A21, obtaining the current gradient of the initial call record conversion model, and initializing the value of the optimization parameter to 0; Step A22, inputting the current gradient and the optimization parameter into a preset model parameter optimization function to obtain updated optimization parameters; Step A23, calculating the corresponding smoothing bandwidth according to the updated optimization parameters, and adjusting the learning rate of the initial call record conversion model based on the smoothing bandwidth; Step A24, returning to the step of inputting the current gradient and the optimization parameters into a preset model parameter optimization function to obtain updated optimization parameters, until the model training is completed.

[0099] Since the call list parameters of the model during training are in a constantly changing state, the result of the loss function calculation also changes dynamically. In the embodiment of the present application, a model parameter optimization function is introduced to adaptively adjust the learning rate. The model parameter optimization function can be constructed by one of the ADAM (Adaptive Moment Estimation) algorithm, RMSProp (Root Mean Square Propagation, deep learning gradient calculation) algorithm, and Adadelta (gradient weighted average) algorithm.

[0100] Taking the ADAM optimization function as an example, the expression of the ADAM optimization function is: ; in, Through the previous Calculation, each v starts from 0, no physical meaning, Used to determine the smoothing bandwidth and learning rate; represents the attenuation rate, preferably 0.9; Represents the gradient, which is the intermediate value; The above ADAM optimization function is further explained by the following exemplary values.

[0101] For example, for the original CDR: (20,2024041013450100000002,20240410134500,01, 10017079643,100,0, 20,0,1,1,2,591,1); Taking the original bandwidth value of the seventh digit of the original bill as 20 as an example, the calculation based on the above ADAM optimization function is as follows: ; = 40; = 40.9; =41.8; Taking the original value of the content center traffic of the eighth digit of the original bill as 10 as an example, the calculation based on the above ADAM optimization function is as follows: ; = 10; = 10.9; =11.8; After practice, after ADAM optimization, the loss function value on the Loss test set was 1.98, which was in line with expectations (lower than the preset threshold), and the model training process could be ended.

[0102] Furthermore, after the model is trained, a more accurate result customer ID can be obtained in actual testing, and the converted subscripts corresponding to the product ID and order ID are correct, such as Fig.13 The data before and after the conversion of the CSMP original call record after the model parameters are adjusted (CSMP original call record and the corresponding standard call record) and Fig.14 The data before and after the conversion of the CDN original call bill after the model parameters are adjusted (CDN original call bill and the corresponding standard call bill).

[0103] In the embodiment of the present application, the ADAM optimization function can optimize and update the call log parameters of the model to minimize the loss function. By combining the ideas of momentum optimization and adaptive learning rate, the embodiment of the present application can effectively adjust the learning rate of each model parameter and adaptively adjust the direction and size of parameter updates during the training process, thereby ensuring that the model training meets the standards and improving the performance of the target call log conversion model obtained through training.

[0104] The present application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the call record data conversion method in the above-mentioned embodiment one.

[0105] Reference below Fig.16 , which shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present application. The electronic devices in the embodiments of the present application may include but are not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players: portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.16 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0106] like Fig.16As shown, the electronic device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the electronic device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an electronic device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0107] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0108] The electronic device provided by the present application adopts the call data conversion method in the above embodiment, which can solve the technical problem that the generalization ability of the current call data conversion solution is weak. Compared with the prior art, the beneficial effects of the electronic device provided by the present application are the same as the beneficial effects of the call data conversion method provided by the above embodiment, and the other technical features in the electronic device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0109] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0110] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0111] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the call record data conversion method in the above-mentioned embodiment.

[0112] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0113] The computer-readable storage medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0114] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by an electronic device, the electronic device: inputs the vector call bill data corresponding to the original call bill data into a preset target call bill conversion model, and extracts each call bill feature vector in the time series of the vector call bill data through the feature extraction layer of the target call bill conversion model, wherein the target call bill conversion model is trained by historical call bill data and corresponding real call bill data, and is used to predict the corresponding standard call bill data based on the input vector call bill data; inputs each of the call bill feature vectors into the self-attention mechanism network of the target call bill conversion model, and determines the attention weight between each of the call bill feature vectors; and determines and outputs the standard call bill data corresponding to the original call bill data based on the output layer of the target call bill conversion model and the attention weight.

[0115] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0116] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0117] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0118] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned call data conversion method, and can solve the technical problem that the generalization ability of the current call data conversion solution is weak. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the call data conversion method provided in the above-mentioned embodiment, and will not be repeated here.

[0119] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned call record data conversion method when executed by a processor.

[0120] The computer program product provided by this application can solve the technical problem that the generalization ability of the current call data conversion solution is weak. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the call data conversion method provided by the above embodiment, which will not be repeated here.

[0121] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for converting call data, characterized in that: The call record data conversion method comprises: Inputting the vector bill data corresponding to the original bill data into a preset target bill conversion model, extracting each bill feature vector in the time series of the vector bill data through the feature extraction layer of the target bill conversion model, wherein the target bill conversion model is trained by historical bill data and corresponding real bill data, and is used to predict the corresponding standard bill data according to the input vector bill data; Inputting each of the call record feature vectors into the self-attention mechanism network of the target call record conversion model to determine the attention weight between each of the call record feature vectors; According to the output layer of the target call record conversion model and the attention weight, the standard call record data corresponding to the original call record data is determined and output.

2. The method for converting call data according to claim 1, characterized in that: Before the step of inputting the vector call record data corresponding to the original call record data into the preset target call record conversion model, the method further includes: Preprocessing the original call record data, wherein the preprocessing includes at least one of deduplication, removal of invalid values ​​and abnormal values, and filling in missing values; The pre-processed original call record data is serialized to obtain vector call record data.

3. The method for converting call data according to claim 1, characterized in that: After the step of determining and outputting the standard call record data corresponding to the original call record data, the method further includes: Querying the customer identifier in the standard call record data in a preset customer table; If the query is successful, the price is determined based on the standard call bill data to obtain the price information corresponding to the standard call bill data; If the query fails, the standard call record data is sent to the error record queue.

4. The method for converting call data according to claim 1, characterized in that: Before the step of inputting the vector call record data corresponding to the original call record data into the preset target call record conversion model, the method further includes: Acquire multiple types of historical call record data and corresponding real call record data, and train a pre-built initial call record conversion model based on each of the historical call record data and the real call record data; During the training process of the initial call record conversion model, updating the model parameters of the initial call record conversion model; When the loss function value of the initial call record conversion model meets the preset conditions, the training is stopped and the initial call record conversion model is set as the target call record conversion model.

5. The method for converting call data according to claim 4, characterized in that: The step of training the pre-built initial call record conversion model based on the historical call record data and the real call record data includes: Dividing each of the historical call record data into a training set and a test set based on the time series of each of the historical call record data, wherein the data lengths of the historical call record data corresponding to the training set and the test set are different; Inputting the historical call record data in the training set into the initial call record conversion model, performing calculations through each network layer in the initial call record conversion model, and outputting predicted call record data; Calculating the loss function value of the initial call record conversion model according to the predicted call record data and the real call record data corresponding to the predicted call record data; Based on the change of the loss function value, the model parameters of the initial call record conversion model are iteratively updated.

6. The method for converting call data according to claim 5, characterized in that: The step of dividing the historical call record data into a training set and a test set based on the time series of the historical call record data comprises: The number of time steps of the training set is calculated according to the total number of time steps corresponding to the time series and the preset division ratio; Taking the start time step as the starting point, dividing the call record data corresponding to the continuous time steps of the number of training set time steps in the historical call record data into a training set; The call record data in the historical call record data except the training set is divided into a test set.

7. The method for converting call data according to claim 5, characterized in that: The model parameters include a learning rate, and the step of iteratively updating the model parameters of the initial call record conversion model includes: Obtaining the current gradient of the initial call record conversion model and initializing the value of the optimization parameter to 0; Inputting the current gradient and the optimization parameter into a preset model parameter optimization function to obtain updated optimization parameters; Calculating a corresponding smoothing bandwidth according to the updated optimization parameters, and adjusting a learning rate of the initial call record conversion model based on the smoothing bandwidth; Return to the step of inputting the current gradient and the optimization parameters into a preset model parameter optimization function to obtain updated optimization parameters until model training is completed.

8. An electronic device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the call record data conversion method as described in any one of claims 1 to 7.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the call record data conversion method according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the call record data conversion method according to any one of claims 1 to 7 are implemented.

Images