Call scene identification method and system based on multiple features

Abstract

The invention discloses a call scene recognition method and system based on multiple features, and belongs to the technical field of communication. The method comprises the following steps of data processing, feature construction, model training, model evaluation, service establishment and data identification. The system comprises a data generation module, a cloud storage module, a task management module and a data processing module. According to the method, the multiple features are constructed based on the call list data, the machine learning model is used for modeling according to the multiple features, whether the application scene and the report scene are consistent or not is recognized through the constructed recognition model, and the problems that in the prior art, the recognition speed is low, the deployment cost is high, and the manpower demand is large can be effectively solved.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and in particular relates to a call scene recognition method and system based on multiple features. Background Art

[0002] With the development of Internet technology, many mobile applications have the function of voice calls. After selecting the business or reporting related usage scenarios during registration, users can use this function to carry out related business in the reported scenario. However, there are also cases where this function is used for other non-reported businesses, such as the reported takeaway business, but the user uses this function to carry out recruitment business, so there is a situation where the usage scenario and the reported scenario are inconsistent. For this case, the conventional method is to use the voice-to-text method to perform voice quality inspection, and then determine whether the content is consistent with the reported scenario. If it is inconsistent, it is judged as a violation, but this method has the defects of large workload, slow recognition speed and low efficiency. Since there is a large amount of audio content in daily life, and the amount of voice-to-text conversion is limited every day, a large number of manual labor is required to perform quality inspection on the content after voice-to-text conversion. In order to increase the amount of conversion, there are two conventional methods, one is to increase manpower, and the other is to deploy multiple voice-to-text services. The former will increase the cost of manpower, and the latter will increase the cost of hardware, both of which will increase the cost of quality inspection. Summary of the invention

[0003] The present invention provides a call scene recognition method and system based on multiple features, constructs multiple features based on call list data, and uses a machine learning model to model according to the multiple features, and then uses the constructed recognition model to identify whether the application scenario and the reporting scenario are consistent, so as to solve the problems of slow recognition speed, high deployment cost and large manpower demand in the prior art.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] The present invention proposes a call scene recognition method based on multiple features, comprising the following steps:

[0006] Step 1: Process the call list data of different systems to obtain the relevant feature fields of each caller every day, including: the number of calls made by the caller on the same day, the number of called persons, the number of calls between the same caller and the same called person, the number of calls in the morning time period, the number of calls in the afternoon time period, the number of calls in the evening time period, the number of calls in the early morning time period, the average call duration of the caller, the total call duration of the caller, the variance value of the caller call duration, the average call connection duration, the variance value of the call connection duration, and the number of called number regions. By integrating the data, the data set D is obtained. L ;

[0007] Step 2: For the data set D obtained in step 1 L Perform feature construction, perform standardization operations and principal component analysis in turn, divide the processed data into training sets and test sets, and finally obtain the training data set D t And the test dataset D v ;

[0008] Step 3: Select the initial model and use the training data set D obtained in step 2 t Train the initial model to obtain a recognition model;

[0009] Step 4: Test data set D obtained in step 2 v Test the recognition model obtained in step 3 and evaluate the recognition model;

[0010] Step 5: Build the service and use the recognition model evaluated in step 4 to recognize the data uploaded by the user to the cloud to obtain the recognition result.

[0011] Furthermore, the step 1 comprises the following steps:

[0012] Step 1.1: Unify the call order data from different systems, unify different fields with the same meaning into the same field, and then obtain the original data D 0 , the original data D 0 The basic fields included are: dialing time, connection time, end time, calling party, called party, calling area, called party area;

[0013] Step 1.2: For the original data D 0 Analyze the fields in the table. For numeric fields, if there are null values ​​in the data, fill them with 0. For categorical fields, if there are null values ​​in the data, fill them with the mode of the field to get D. 1 ;

[0014] Step 1.3: Calculate the connection time of the called party based on the dialing time and the connection time;

[0015] Step 1.4: Calculate the call duration of the caller based on the connection time and end time;

[0016] Step 1.5: Calculate the call time period according to the dialing time, where the call time period has the following values: morning, afternoon, evening, and early morning;

[0017] Step 1.6: Calculate the dial date based on the dial time;

[0018] Step 1.7: Group the data according to the calling date and calling party, and obtain the calling data D of each calling party every day. 2 ;

[0019] Step 1.8: D 2 Each group of data in the above data is counted, including: the number of calls made by the caller on the same day, the number of called persons, the number of calls between the same caller and the same called person, the number of calls made by the caller in each call time period, the average call duration of the caller, the total call duration of the caller, the variance of the caller call duration, the average call connection duration, the variance of the caller connection duration, and the number of called number regions;

[0020] Step 1.9: Through processing, the relevant feature fields of each caller every day are obtained, and then the data set D is obtained. L .

[0021] Furthermore, the step 2 comprises the following steps:

[0022] Step 2.1: Standardization operation processing, the data set D L The data features are standardized so that the features are scaled to a range of mean 0 and variance 1, and the relevant parameters are saved to obtain the standardized model M s ;

[0023] Step 2.2: Principal component analysis: perform principal component analysis on the features after standardization, calculate the covariance matrix, eigenvalues ​​and eigenvectors of the data, rearrange the data features according to the eigenvalues ​​and eigenvectors, save the relevant parameters, and obtain the principal component analysis model M. p ;

[0024] Step 2.3: Divide the data into training set and test set. Randomly divide 70% of the data processed by principal component analysis into training data and the remaining 30% into test data.

[0025] Step 2.4: Get the training data set D through the above operations t And the test dataset D v .

[0026] Furthermore, the step 3 comprises the following steps:

[0027] Step 3.1: Define the initial model. The initial model uses the xgboost model.

[0028] Step 3.2: Determine the parameters that need to be searched for the initial model, use the grid parameter search method to determine the optimal parameters, and use the 5-fold cross-validation method in the process of grid parameter search;

[0029] Step 3.3: Based on the optimal parameters determined in step 3.2 and the 5-fold cross-validation method, use the training data set D t Train the xgboost model to obtain the recognition model M xgb .

[0030] Furthermore, in step 4, the recognition model M xgb When evaluating, we count relevant indicators to observe the performance of the model. The statistical indicators are: accuracy, recall and F 1 Fraction.

[0031] Further, the step 5 comprises the following steps:

[0032] Step 5.1: Use the producer-consumer design pattern to build services;

[0033] Step 5.2: The user uploads the call sheet data to the cloud storage, the system automatically generates a task, and assembles the task ID, the data address corresponding to the uploaded call sheet data, and the result callback address into a recognition task and sends it to the service. The service puts the recognition task into the task queue; the task queue is implemented using RocketMQ;

[0034] Step 5.3: Monitor the task queue and obtain the recognition task from the task queue;

[0035] Step 5.4: According to the data address provided by the obtained recognition task, call the corresponding call list data stored in the cloud;

[0036] Step 5.5: Use the recognition model evaluated in step 4 to process the called call record data, obtain the recognition result, and call back the recognition result to the user.

[0037] Further, step 5.5 includes the following steps:

[0038] Step 5.5.1: Data processing: The processing method is the same as step 1, and the predicted data D is finally obtained. init ;

[0039] Step 5.5.2: Feature construction, the data to be predicted D init After the feature standardization processing model M s And the principal component analysis model M p Processing is performed to obtain the data to be predicted D p ;

[0040] Step 5.5.3: Model prediction, using the evaluated recognition model M xgb Treat the predicted data D p Make predictions and get the recognition result R 0 ;

[0041] Step 5.5.4: Result processing: the recognition result R 0 It is encapsulated with the corresponding task ID and sent to the user callback service according to the corresponding result callback address.

[0042] Furthermore, steps 5.5.1, 5.5.2, 5.5.3, and 5.5.4 are all processed using a multi-threaded approach.

[0043] The present invention also proposes a call scene recognition system based on multiple features, including a data generation module, a cloud storage module, a task management module and a data processing module;

[0044] The data generation module is used to generate the user's call list data, and send the call list data to the cloud storage module, and initiate the recognition task at the same time;

[0045] The cloud storage module is used to store the user's call list data;

[0046] The task management module is used to put the recognition task into the task queue, monitor the task queue, obtain the recognition task from the task queue and send it to the data processing module;

[0047] The data processing module is used to call the call order data corresponding to the recognition task stored in the cloud storage module according to the received recognition task, and use the recognition model constructed above to identify the called call order data, and return the recognition result to the user.

[0048] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0049] The present invention constructs a variety of features based on call sheet data, and uses a machine learning model to build a model based on the multiple features, and then uses the constructed recognition model to identify whether the application scenario and the reporting scenario are consistent. Compared with the method of speech-to-text + manual quality inspection, the method of using a machine learning model for prediction and recognition can reduce the GPU resources and human resources required for deploying speech-to-text services, reduce hardware costs and labor costs, and thus greatly reduce the cost of deployment. The method of using a machine learning model for recognition has the advantage of fast recognition speed. At the same time, using multi-threading for recognition can greatly increase the amount of daily detection, improve the quality inspection speed, and transform the quality inspection from random inspection to full coverage quality inspection, which can effectively solve the problems of slow recognition speed, high deployment cost, and large manpower demand in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of the call scene recognition method provided by the present invention;

[0051] Figure 2 This is a business flow chart of the call scene identification method provided by the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] like Figure 1-2 As shown, the present invention proposes a call scene recognition method based on multiple features, including steps 1-5.

[0054] Step 1: Process the call list data of different systems to obtain the relevant feature fields of each caller every day, including: the number of calls made by the caller on the same day, the number of called persons, the number of calls between the same caller and the same called person, the number of calls in the morning time period, the number of calls in the afternoon time period, the number of calls in the evening time period, the number of calls in the early morning time period, the average call duration of the caller, the total call duration of the caller, the variance value of the caller call duration, the average call connection duration, the variance value of the call connection duration, and the number of called number regions. By integrating the data, the data set D is obtained. L .

[0055] Specifically, the step 1 includes steps 1.1-1.9.

[0056] Step 1.1: Unify the call order data from different systems, unify different fields with the same meaning into the same field, and then obtain the original data D 0 , the original data D 0 The basic fields included are: dialing time, connection time, end time, calling number, called number, calling area, called area.

[0057] Step 1.2: For the original data D 0 Analyze the fields in the table. For numeric fields, if there are null values ​​in the data, fill them with 0. For categorical fields, if there are null values ​​in the data, fill them with the mode of the field to get D. 1 .

[0058] Step 1.3: Calculate the connection time of the called party based on the dialing time and the connection time.

[0059] Step 1.4: Calculate the call duration of the caller based on the connection time and end time.

[0060] Step 1.5: Calculate the call time period according to the dialing time, where the values ​​of the call time period are: morning, afternoon, evening, and early morning.

[0061] Step 1.6: Calculate the dial date based on the dial time.

[0062] Step 1.7: Group the data according to the calling date and calling party, and obtain the calling data D of each calling party every day. 2 .

[0063] Step 1.8: D 2 Statistics are performed on each group of data, including: the number of calls made by the caller on the same day, the number of called persons, the number of calls between the same caller and the same called person, the number of calls made by the caller in each call time period, the average call duration of the caller, the total call duration of the caller, the variance of the caller call duration, the average connection duration of the called person, the variance of the connection duration of the called person, and the number of called number regions.

[0064] Step 1.9: Through processing, the relevant feature fields of each caller every day are obtained, and then the data set D is obtained. L .

[0065] Step 2: For the data set D obtained in step 1 L Perform feature construction, perform standardization operations and principal component analysis in turn, divide the processed data into training sets and test sets, and finally obtain the training data set D t And the test dataset D v .

[0066] Specifically, the step 2 includes steps 2.1-2.4.

[0067] Step 2.1: Standardization operation processing, the data set D L The data features are standardized so that the features are scaled to a range of mean 0 and variance 1, and the relevant parameters are saved to obtain the standardized model M s After the standardization operation, the performance of the subsequent model can be improved, and the gradient disappearance and gradient explosion can be effectively prevented.

[0068] Step 2.2: Principal component analysis: perform principal component analysis on the features after standardization, calculate the covariance matrix, eigenvalues ​​and eigenvectors of the data, rearrange the data features according to the eigenvalues ​​and eigenvectors, save the relevant parameters, and obtain the principal component analysis model M. p This step can effectively reduce the noise of the data.

[0069] Step 2.3: Divide the data into training set and test set. Randomly divide 70% of the data processed by principal component analysis into training data and the remaining 30% into test data.

[0070] Step 2.4: Get the training data set D through the above operations t And the test dataset D v .

[0071] Step 3: Select the initial model and use the training data set D obtained in step 2 t The initial model is trained to obtain the recognition model.

[0072] Specifically, the initial model is trained using a convex optimization method, and step 3 includes steps 3.1-3.3.

[0073] Step 3.1: Define the initial model, which uses the xgboost model. xgboost is an optimized gradient boosting decision tree algorithm.

[0074] Step 3.2: Determine the parameters that need to be searched for the initial model, use the grid parameter search method to determine the best parameters, and use the 5-fold cross validation method in the process of grid parameter search. For example, determine the parameters that need to be searched for the model: {'min_child_weight':[2,3,4,5], 'max_depth':[2,3,4,5,6,7]}, and use the grid parameter search method to determine the best parameters as {'max_depth':3, 'min_child_weight':3}.

[0075] Step 3.3: Based on the optimal parameters determined in step 3.2 and the 5-fold cross-validation method, use the training data set D t Train the xgboost model to obtain the recognition model M xgb .

[0076] Step 4: Test data set D obtained in step 2 v Test the recognition model obtained in step 3 and evaluate the recognition model. xgb When evaluating, we count relevant indicators to observe the performance of the model. The statistical indicators are: accuracy, recall and F 1 The score also counts the contribution of each feature to obtain the importance matrix of each feature.

[0077] Step 5: Service construction, and use the recognition model evaluated in step 4 to recognize the data uploaded by the user to the cloud to obtain the recognition result. Specifically, step 5 includes steps 5.1-5.5.

[0078] Step 5.1: Use the producer-consumer design pattern to build the service. The producer-consumer pattern is a classic multi-threaded design pattern, which mainly involves the following two roles: Producer: Responsible for creating data or production tasks, and putting them into a shared data structure (usually a queue, etc.). For example, in the service, it can be a module that receives external requests and performs preliminary processing and generates corresponding business tasks. Consumer: Take out data or tasks from the shared data structure and perform subsequent processing operations, such as executing specific business logic in the service, performing substantial calculations and storage on the request. Its core advantage is that it decouples the production and consumption process of data, realizes a certain degree of asynchronous processing, and improves the overall concurrency and resource utilization of the system. In the present invention, the producer uses fastapi to encapsulate a task receiving interface.

[0079] Step 5.2: The user uploads the call sheet data to the cloud storage, the system automatically generates a task, and assembles the task ID, the data address corresponding to the uploaded call sheet data, and the result callback address into a recognition task and sends it to the service. The service puts the recognition task into the task queue; the task queue is implemented using RocketMQ.

[0080] Step 5.3: Monitor the task queue and obtain the recognition task from the task queue.

[0081] Step 5.4: According to the data address provided by the obtained recognition task, call the corresponding call order data stored in the cloud.

[0082] Step 5.5: Use the recognition model evaluated in step 4 to process the called call record data, obtain the recognition result, and call back the recognition result to the user.

[0083] Specifically, step 5.5 includes steps 5.5.1-5.5.4.

[0084] Step 5.5.1: Data processing: The processing method is the same as step 1, and the predicted data D is finally obtained. init Since there are multiple systems at this stage and the data volume of each system is in the tens of millions, a parallel method is used to process the data to obtain the relevant feature fields of each caller every day. The relevant feature fields include: the number of calls made by the caller on the same day, the number of called persons, the number of calls between the same caller and the same called person, the number of calls in the morning during the call time period, the number of calls in the afternoon during the call time period, the number of calls in the evening during the call time period, the number of calls in the early morning during the call time period, the average call duration of the caller, the total call duration of the caller, the variance value of the call duration of the caller, the average value of the called person's connection duration, the variance value of the called person's connection duration, and the number of called number regions. The data to be predicted D is obtained through data integration. init .

[0085] Step 5.5.2: Feature construction, the data to be predicted D init After the feature standardization processing model M s And the principal component analysis model M p Processing is performed to obtain the data to be predicted D p .

[0086] Step 5.5.3: Model prediction, using the evaluated recognition model M xgb Treat the predicted data D p Make predictions and get the recognition result R 0 .

[0087] Step 5.5.4: Result processing: the recognition result R 0 It is encapsulated with the corresponding task ID and sent to the user callback service according to the corresponding result callback address.

[0088] Among them, steps 5.5.1, 5.5.2, 5.5.3, and 5.5.4 are all processed using a multi-threaded approach.

[0089] In addition, the present invention also proposes a call scene recognition system based on multiple features, including a data generation module, a cloud storage module, a task management module and a data processing module.

[0090] The data generation module is used to generate the user's call list data, and send the call list data to the cloud storage module, and initiate the identification task at the same time.

[0091] The cloud storage module is used to store the user's call list data.

[0092] The task management module is used to put the recognition task into the task queue, monitor the task queue, obtain the recognition task from the task queue and send it to the data processing module.

[0093] The data processing module is used to call the call order data corresponding to the recognition task stored in the cloud storage module according to the received recognition task, and use the recognition model constructed above to identify the called call order data, and return the recognition result to the user.

[0094] The present invention constructs multiple features based on call list data, and uses a machine learning model to build a model according to the multiple features, and then uses the constructed recognition model to identify whether the application scenario and the reporting scenario are consistent, which can greatly reduce the deployment cost and has a fast recognition speed. At the same time, using multi-threading for recognition can greatly increase the daily detection volume, thereby effectively solving the problems of slow recognition speed, high deployment cost and large manpower demand in the prior art.

[0095] The above description is a detailed description of the preferred feasible embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modified changes completed under the technical spirit suggested by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for identifying call scenes based on multiple features, characterized in that: The following steps are involved: Step 1: Process the call list data of different systems to obtain the relevant feature fields of each caller every day, including: the number of calls made by the caller on the same day, the number of called persons, the number of calls between the same caller and the same called person, the number of calls in the morning, the number of calls in the afternoon, the number of calls in the evening, the number of calls in the early morning, the average call duration of the caller, the total call duration of the caller, the variance value of the caller's call duration, the average call connection duration, the variance value of the call connection duration, and the number of called number regions. By integrating the data, the data set D is obtained. L ; Step 2: For the data set D obtained in step 1 L Perform feature construction, perform standardization operations and principal component analysis in turn, divide the processed data into training sets and test sets, and finally obtain the training data set D t And the test dataset D v ; Step 3: Select the initial model and use the training data set D obtained in step 2 t Train the initial model to obtain a recognition model; Step 4: Test data set D obtained in step 2 v Test the recognition model obtained in step 3 and evaluate the recognition model; Step 5: Build the service and use the recognition model evaluated in step 4 to recognize the data uploaded by the user to the cloud to obtain the recognition result.

2. The method for identifying call scenes based on multiple features according to claim 1, characterized in that: The step 1 comprises the following steps: Step 1.1: Unify the call order data of different systems, unify different fields with the same meaning into the same field, and then obtain the original data D0. The original data D0 contains the following basic fields: dialing time, connection time, end time, calling party, called party, calling area, called party area; Step 1.2: Analyze the fields in the original data D0. For numeric fields, if there are null values ​​in the data, fill them with 0. For categorical fields, if there are null values ​​in the data, fill them with the mode of the field to obtain D1. Step 1.3: Calculate the connection time of the called party based on the dialing time and the connection time; Step 1.4: Calculate the call duration of the caller based on the connection time and end time; Step 1.5: Calculate the call time period according to the dialing time, where the call time period has the following values: morning, afternoon, evening, and early morning; Step 1.6: Calculate the dial date based on the dial time; Step 1.7: Group the data according to the dialing date and the caller, and obtain the call data D2 of each caller every day; Step 1.8: Perform statistics on each group of data in D2, including: the number of calls made by the caller on the same day, the number of called parties, the number of calls between the same caller and the same called party, the number of calls made by the caller in each call time period, the average call duration of the caller, the total call duration of the caller, the variance of the caller call duration, the average call connection duration, the variance of the caller connection duration, and the number of called number regions; Step 1.9: Through processing, the relevant feature fields of each caller every day are obtained, and then the data set D is obtained. L .

3. The method for identifying call scenes based on multiple features according to claim 2, characterized in that: The step 2 comprises the following steps: Step 2.1: Standardization operation processing, the data set D L The data features are standardized so that the features are scaled to a range of mean 0 and variance 1, and the relevant parameters are saved to obtain the standardized model M s ; Step 2.2: Principal component analysis: perform principal component analysis on the features after standardization, calculate the covariance matrix, eigenvalues ​​and eigenvectors of the data, rearrange the data features according to the eigenvalues ​​and eigenvectors, save the relevant parameters, and obtain the principal component analysis model M. p ; Step 2.3: Divide the data into training set and test set. Randomly divide 70% of the data processed by principal component analysis into training data and the remaining 30% into test data. Step 2.4: Get the training data set D through the above operations t And the test dataset D v .

4. The method for identifying call scenes based on multiple features according to claim 3, characterized in that: The step 3 comprises the following steps: Step 3.1: Define the initial model. The initial model uses the xgboost model. Step 3.2: Determine the parameters that need to be searched for the initial model, use the grid parameter search method to determine the optimal parameters, and use the 5-fold cross-validation method in the process of grid parameter search; Step 3.3: Based on the optimal parameters determined in step 3.2 and the 5-fold cross-validation method, use the training data set D t Train the xgboost model to obtain the recognition model M xgb .

5. The method for identifying call scenes based on multiple features according to claim 4, characterized in that: In step 4, the recognition model M xgb When evaluating, relevant indicators are counted to observe the performance of the model. The statistical indicators are: accuracy, recall rate and F1 score.

6. The method for identifying call scenes based on multiple features according to claim 5, characterized in that: The step 5 comprises the following steps: Step 5.1: Use the producer-consumer design pattern to build services; Step 5.2: The user uploads the call sheet data to the cloud storage, the system automatically generates a task, and assembles the task ID, the data address corresponding to the uploaded call sheet data, and the result callback address into a recognition task and sends it to the service. The service puts the recognition task into the task queue; Step 5.3: Monitor the task queue and obtain the recognition task from the task queue; Step 5.4: According to the data address provided by the obtained recognition task, call the corresponding call list data stored in the cloud; Step 5.5: Use the recognition model evaluated in step 4 to process the called call record data, obtain the recognition result, and call back the recognition result to the user.

7. The method for identifying call scenes based on multiple features according to claim 6, characterized in that: In step 5.2, the task queue is implemented using RocketMQ.

8. The method for identifying call scenes based on multiple features according to claim 7, characterized in that: Step 5.5 includes the following steps: Step 5.5.1: Data processing: The processing method is the same as step 1, and the predicted data D is finally obtained. init ; Step 5.5.2: Feature construction, the data to be predicted D init After the feature standardization processing model M s And the principal component analysis model M p Processing is performed to obtain the data to be predicted D p ; Step 5.5.3: Model prediction, using the evaluated recognition model M xgb Treat the predicted data D p Make a prediction and then get the recognition result R0; Step 5.5.4: Result processing: encapsulate the recognition result R0 and the corresponding task ID, and send them to the user callback service according to the corresponding result callback address.

9. The method for identifying call scenes based on multiple features according to claim 8, characterized in that: Steps 5.5.1, 5.5.2, 5.5.3, and 5.5.4 are all processed using a multi-threaded approach.

10. A call scene recognition system based on multiple features, comprising a data generation module, a cloud storage module, a task management module and a data processing module; The data generation module is used to generate the user's call list data, and send the call list data to the cloud storage module, and initiate the recognition task at the same time; The cloud storage module is used to store the user's call list data; The task management module is used to put the recognition task into the task queue, monitor the task queue, obtain the recognition task from the task queue and send it to the data processing module; The data processing module is used to call the call record data corresponding to the identification task stored in the cloud storage module according to the received identification task, and use the identification model constructed by any one of claims 1-9 to identify the called call record data, and return the identification result to the user.

Images