TWI728636B - Intelligent identity verification system and method - Google Patents

Abstract

The invention provides an intelligent identity verification system and method, the main components of which are an intelligent verification evaluation module and an identity verification module. The intelligent verification evaluation module aggregates relevant information of users, and determines whether intelligent verification can be used through the intelligent verification evaluation model. The identity verification module captures user characteristics from a multimedia file provided by the user according to one or more available verification methods, and compares with previously registered user models to obtain verification scores for each verification method, and integrates the verification score into the verification confidence value through a verification confidence integration model. The intelligent identity verification system also includes an intelligent verification evaluation model training module and a verification confidence value integration model training module. The intelligent verification evaluation model training module captures first training data for training the intelligent verification evaluation model. The verification confidence value integration model training module captures second training data for training the verification confidence integration model.

Description

Smart identity verification system and method

The present invention relates to an identity verification technology. In detail, it relates to a smart identity verification system and method.

In recent years, due to the rapid advancement of biometric identification technology, various commercial services have sought to use biometric identification to verify user identity through automatic verification, such as: digital finance, online banking, smart customer service, smart speakers, cloud services, online Insurance etc. Moreover, since the contents of these commercial services all involve the vital interests of users, it is extremely necessary to rely on safe and accurate identity verification technology.

The biometrics currently available for identification include fingerprints, faces, iris, voice, etc., but most identity verification systems only adopt a single verification method. That is, with the different identification technologies used, various identity verification systems either only use fingerprint comparison to confirm identity, or only use iris recognition to confirm identity. However, it is relatively easy to break through using a single verification method. In particular, various biological characteristics have different advantages and disadvantages in verification effects. If they can be properly combined and applied, they will obviously have more advantages in verification effects.

Furthermore, most identity verification systems can only make judgments based on fixed conditions, and cannot dynamically adjust based on the feedback of the user’s actual connection status. This makes the judgment procedure of identity verification excessively rigid and reduces the security and safety of identity verification. Practicality.

Therefore, how to provide an identity verification technology that can support multi-factor verification and cut the advantages and disadvantages of various verification methods to achieve higher security and practicability of the verification system, and through feedback from actual services, Smartly advancing with the times, in order to show the best results of automated smart verification, has become an important topic.

The present invention provides a smart identity verification system, including: a smart verification evaluation module, which integrates the user's historical incoming information, personal information, and current incoming information to evaluate through a smart verification evaluation model, and then determine the user Whether smart verification can be used; the identity verification module selects the available verification methods based on the user's registered method and the file type of the multimedia file provided by the current verification, and retrieves the multimedia file with the available verification method to retrieve the user Features, to compare the registered user model based on the user characteristics to obtain the verification score obtained after verification according to the available verification method, and then integrate the verification score into the verification confidence value through a verification confidence value integration model to obtain The verification confidence value provides the identity verification module to determine whether the identity verification is passed.

In one embodiment, the smart identity verification system further includes: a smart verification evaluation model training module, which extracts the first training data from the user's personal information and historical incoming information, so as to train the wisdom based on the first training data Verify the evaluation model; and verify the confidence value integration The model training module extracts the second training data from the historical incoming information of the user to train and verify the confidence value integration model based on the second training data.

In one embodiment, the smart verification evaluation model training module includes: a training data acquisition unit from a personal information database that records the user's personal information and incoming log data that records the user's historical incoming information The first training data is retrieved from the library; and the model training unit is used to train the wisdom verification evaluation model.

In one embodiment, the verification confidence value integration model training module includes: a smart verification confidence value training data acquisition unit that extracts second training data from an incoming log database; and a smart verification confidence value model training unit , Is used to train and verify the confidence value integration model.

In one embodiment, the smart identity verification system further includes: a personal information database, which records the user’s personal information; an entry log database, which records the user’s historical entry information; and a verification method database, which records the history of the user. Define the available verification methods and related information; and the user model database, which records the user model and related feature vectors registered by the user.

In one embodiment, the smart verification evaluation module further includes: a predictive data acquisition unit, which collects the user's historical incoming information in the incoming log database, personal information in the personal information database, and the current incoming online information. The incoming line information of the client is aggregated to generate a prediction data; and the incoming prediction unit receives the prediction data to use the smart verification evaluation model to evaluate the prediction data to generate a smart verification evaluation result, which is used to evaluate whether smart verification can be used.

In one embodiment, the identity verification module further includes: a user verification method screening unit, which filters out available verification methods from the verification method database based on the method that the user has registered and the file type of the multimedia file; The user feature extraction unit is based on the available verification methods to extract user features from the multimedia file; the user feature comparison unit is based on the verification method The feature comparison method specified in the method database compares the user characteristics with the registered user model in the user model database to obtain the verification score obtained after verification according to the available verification method; and the wisdom verification confidence value The calculation unit integrates the verification scores into the verification confidence value based on the verification confidence value integration model.

In one embodiment, the smart identity verification system further includes a process management module, which includes: a client interface unit, which accepts user verification or registration requests; and an account data access unit, which accesses personal information data The database data is used to confirm whether it is a valid account; the process control unit is used to control the verification process; the back-end core client unit is used to query the smart verification evaluation model training module, the identity verification module and an identity registration module Group; and the incoming line record unit, which records the incoming line log of the user in the incoming line log database.

In one embodiment, the identity registration module includes: a registration method screening unit, which screens out available registration methods from a verification method database according to the file type of the multimedia file; and a user feature extraction unit, which is based on the verification method data The specified feature extraction method in the library retrieves the registered features from the multimedia file; and the user model update unit is used to update the user model database.

The present invention also provides a smart identity verification method, which includes: receiving client information; checking the validity of an account; using a smart verification and evaluation process to consolidate the user's historical incoming information, personal information, and current incoming information, Evaluate through a smart verification evaluation model to determine whether smart verification can be performed; and if smart verification is determined, perform an identity verification process, where the identity verification process is based on the method that the user has registered and the multimedia provided The file type of the file filters out the available verification methods and extracts the multimedia files The user characteristics are compared with the user model to obtain the verification score obtained after verification according to the available verification method, and then the verification score is integrated into the verification confidence value through a verification confidence value integration model.

In one embodiment, before the step of verifying the validity of the account, it further includes the following steps: judging whether it is designated as a registered service; if the judgment result is designated as a registered service, an identity registration process is performed. The identity registration process includes the following steps: searching for available registration methods; retrieving user characteristics; updating the user model; and returning the registration result.

In one embodiment, when performing the smart verification evaluation process, it further includes performing a smart verification evaluation model training process to train the smart verification evaluation model. The smart verification evaluation model training process includes: capturing first training data, including capturing personal information characteristics, incoming record summary characteristics, current incoming information characteristics, and correct answers expected by the model; and training with the first training data Wisdom verification evaluation model.

In one embodiment, when performing the identity verification process, it further includes performing a verification confidence value integration model training process to train the verification confidence value integration model. The verification confidence value integration model training process includes: capturing second training data, including capturing verification scores and correct answers expected by the model; and training the verification confidence value integration model using the second training data.

The smart identity verification system and method of the present invention have considerable advantages, which are described as follows.

First, different from the general traditional manual verification system, the smart identity verification system of the present invention is a fully automatic identity verification system, which can save labor costs on the enterprise side and provide more accurate verification results.

Second, the smart identity verification system of the present invention improves the limitation that the past system can only use one algorithm or the same type of algorithm. In the framework of the system and method proposed by the present invention Below, it can integrate a variety of different types of identity verification methods, such as voiceprint verification, face verification, fingerprint verification, etc., and build an integrated model by deep learning or other machine learning methods to achieve a multi-factor and multi-model verification architecture to complement A single factor causes verification security issues, and improves the usability and scalability of the client.

Furthermore, the improved conventional technology of the present invention can only judge whether the evaluation criteria of the smart verification service are applicable according to fixed conditions. By using the past incoming line records and user data, as well as the information reported by the client whether the service has been terminated normally, the deep learning training model can be used to evaluate whether the new incoming line is suitable for smart verification. Through this mechanism, different services can dynamically adjust the evaluation criteria implicit in the back-end model according to their service characteristics, and customize the security and availability requirements of different services.

1‧‧‧Smart Identity Verification System

010‧‧‧Client

020‧‧‧Incoming line information

021‧‧‧device serial number

022‧‧‧Incoming account

023‧‧‧Service Number

025‧‧‧Whether the service ended normally

030‧‧‧Multimedia files

100‧‧‧Process Management Module

110‧‧‧User terminal interface interface unit

120‧‧‧Incoming recording unit

130‧‧‧Account Data Access Unit

140‧‧‧Back-end core client unit

150‧‧‧Process Control Unit

200‧‧‧Smart verification evaluation model training module

201‧‧‧Smart verification evaluation model

210‧‧‧Training Data Acquisition Unit

211‧‧‧First training data

212‧‧‧Personal information characteristics

213‧‧‧Incoming record summary characteristics

214‧‧‧Characteristics of incoming line information

215‧‧‧The correct answer expected by the model

220‧‧‧Model Training Unit

300‧‧‧Smart Verification Evaluation Module

301‧‧‧Smart verification evaluation results

310‧‧‧Prediction Data Acquisition Unit

311‧‧‧Forecast data

312‧‧‧Personal information characteristics

313‧‧‧Incoming record summary characteristics

314‧‧‧Characteristics of incoming line information

320‧‧‧Incoming line prediction unit

400‧‧‧Verify the confidence value integration model training module

401‧‧‧Verify the confidence value integration model

410‧‧‧Smart verification confidence value training data acquisition unit

411‧‧‧Second training data

412‧‧‧Verification score

413‧‧‧The correct answer expected by the model

420‧‧‧Smart verification confidence value model training unit

500‧‧‧Identity Verification Module

501‧‧‧Verify the confidence value

510‧‧‧User Feature Extraction Unit

511‧‧‧User Features

520‧‧‧User feature comparison unit

521‧‧‧Verification score

530‧‧‧Smart verification confidence value calculation unit

540‧‧‧User Verification Method Screening Unit

541‧‧‧Available verification methods

600‧‧‧Identity Registration Module

610‧‧‧Registration Method Screening Unit

611‧‧‧Available registration method

620‧‧‧User Feature Extraction Unit

621‧‧‧Registration features

630‧‧‧User Model Update Unit

901 to 907‧‧‧Main Process of Smart Identity Verification

910‧‧‧Identity Registration Process

920‧‧‧Smart verification evaluation process

930‧‧‧Smart verification evaluation model training process

940‧‧‧ Identity Verification Process

950‧‧‧Verify the confidence value integration model training process

A00‧‧‧Personal Information Database

B00‧‧‧Incoming log database

C00‧‧‧Verification Method Database

D00‧‧‧User Model Database

Figure 1 shows the architecture of the smart identity verification system of the present invention;

Figure 2 shows the structure of the process management module of the present invention;

Figure 3 shows the architecture of the smart verification evaluation model training module of the present invention;

Figure 4 shows the architecture of the smart verification evaluation module of the present invention;

Figure 5 shows the architecture of the verification confidence value integration model training module of the present invention;

Figure 6 shows the architecture of the identity verification module of the present invention;

Figure 7 shows the structure of the identity registration module of the present invention;

Figure 8 shows the main flow of the smart identity verification method of the present invention;

Figure 9 shows the detailed steps of the identity registration process of the present invention;

Figure 10 shows the detailed steps of the smart verification evaluation process of the present invention;

Figure 11 shows the detailed steps of the smart verification evaluation model training process of the present invention;

Figure 12 shows the detailed steps of the identity verification process of the present invention; and

Figure 13 shows the detailed steps of the training process of the verification confidence value integration model of the present invention.

The following specific examples illustrate the implementation of the present invention. Those familiar with the art can easily understand the advantages and effects of the present invention from the contents disclosed in this specification.

Please refer to Figure 1, which shows the architecture of the smart identity verification system (1) of the present invention. The smart identity verification system (1) includes four databases (A00) ~ (D00) and six modules (100) ~ (600). The four databases are personal information database (A00), incoming log database (B00), verification method database (C00) and user model database (D00), and the six modules are process management modules. (100), Smart Verification Evaluation Model Training Module (200), Smart Verification Evaluation Module (300), Verification Confidence Value Integration Model Training Module (400), Identity Verification Module (500) and Identity Registration Module (600) ). In addition to the aforementioned database and modules, the smart identity verification system (1) further includes a smart verification evaluation model (201) and a verification confidence value integration model (401).

The personal information database (A00) is used to record the user's personal information, such as account number, usage status, name, ID number, age, gender, contact number, address and time stamp and other personal information.

The incoming line log database (B00) is used to record the user's historical incoming line information and current incoming line information, such as incoming line serial number, device serial number, verification of multimedia file path, and registration Multimedia file path, account number, service number, verification confidence value, whether it passes smart verification, whether the verification process is terminated abnormally, time stamp, and whether the client service ends normally and other related information.

The verification method database (C00) is used to record the available verification methods and related information defined by the system, such as verification method number, verification multimedia type, feature extraction module name, feature comparison module name, time stamp, etc. News.

The user model database (D00) is used to record the user model and related feature vectors of the user registration, such as account number, verification method number, incoming serial number, registered user characteristics, whether to include user model, time stamp and other information.

Please also refer to Figure 2 which shows the structure of the process management module (100) of the present invention. The components of the process management module (100) include a client interface unit (110), an incoming record unit (120), an account data access unit (130), a back-end core client unit (140), and a process control unit ( 150). The client interface interface unit (110) can accept verification or registration requests from the client (010); access the data in the personal information database (A00) through the account data access unit (130) to confirm whether it is a valid account; Use the process control unit (150) to control the verification process; use the back-end core client unit (140) to query the aforementioned smart verification evaluation module (300), identity verification module (500), and identity registration module (600); and Record the user's incoming log into the incoming log database (B00) through the incoming log unit (120).

In one embodiment, the incoming line information of the user end (010) received by the user end interface unit (110) includes a device serial number (021), an incoming line account number (022), and a service number (023). In addition, the user terminal interface unit (110) can also receive the multimedia file (030) provided by the user terminal (010), and receive the message of whether the service has ended normally (025).

Please refer to Figure 3, which shows the architecture of the smart verification evaluation model training module (200) of the present invention. The components of the smart verification evaluation model training module (200) include a training data acquisition unit (210) and a model training unit (220). The training data acquisition unit (210) records the personal information of the user. Extract the first training data (211) from the database (A00) and the incoming log database (B00) that records the historical incoming information of the user, and train the smart verification evaluation model (201) with the model training unit (220) . In one embodiment, the first training data (211) includes personal information characteristics (212), incoming record summary characteristics (213), current incoming information characteristics (214), and correct answers expected by the model (215).

Please refer to Figure 4, which shows the structure of the smart verification evaluation module (300) of the present invention. The components of the smart verification evaluation module (300) include a prediction data acquisition unit (310) and an incoming prediction unit (320). Through the predictive data acquisition unit (310), the historical incoming information of the incoming user in the incoming log database (B00), the personal information of the personal information database (A00) and the incoming user information of the current incoming user can be collected (020) Collect and generate forecast data (311). Furthermore, the incoming prediction unit (320) receives the prediction data 311 to use the smart verification evaluation model (201) to evaluate the prediction data to generate a smart verification evaluation result (301), which is used to evaluate whether smart verification can be used. In one embodiment, the forecast data (311) includes personal information characteristics (312), incoming record summary characteristics (313), and current incoming information characteristics (314).

Please refer to Figure 5, which shows the architecture of the verification confidence value integration model training module (400) of the present invention. The components of the verification confidence value integration model training module (400) include a smart verification confidence value training data acquisition unit (410) and a smart verification confidence value model training unit (420). Through the wisdom verification confidence value training data acquisition unit (410), the second training data (411) can be retrieved from the incoming log database (B00), that is, the wisdom verification confidence value model training data; At the same time, the wisdom verification confidence value model training unit (420) is used to train the verification confidence value integration model (401). In one embodiment, the second training data (411) includes the verification score (412) and the correct answer (413) expected by the model.

Please refer to Figure 6, which shows the structure of the identity verification module (500) of the present invention. The components of the identity verification module (500) include a user feature extraction unit (510), a user feature comparison unit (520), a smart verification confidence value calculation unit (530), and a user verification method screening unit (540). The user authentication method screening unit (540) can filter the available authentication methods (541) in the authentication method database (C00) according to the method that the user has registered and the file type of the multimedia file (030); then, Use the user feature extraction unit (510) to extract the user features (511) from the multimedia file (030) according to the feature extraction method specified by the available verification method (541); furthermore, use the user feature comparison unit (520) Use the feature comparison method specified in the verification method database (C00) to compare the user feature (511) with the registered user model in the user model database (D00) to obtain the verification method according to the available verification method The obtained verification score (521); finally, the smart verification confidence value calculation unit (530) integrates the verification score (521) into a verification confidence value (501) according to the verification confidence value integration model (401).

Please refer to Figure 7, which shows the structure of the identity registration module (600) of the present invention. The components of the identity registration module (600) include a registration method screening unit (610), a user feature extraction unit (620), and a user model update unit (630). First, use the registration method screening unit (610) to filter the available registration methods (611) suitable for use from the verification method database (C00) according to the file type of the multimedia file (030); then, the user feature extraction unit ( 620) Then according to the feature extraction method specified in the available registration method (611), retrieve the registered feature (621) from the multimedia file (030); finally, update the user model database (630) with the user model update unit (630) D00).

According to the related descriptions of Figs. 1 to 7, it can be understood that the main components of the smart identity verification system (1) of the present invention can be the smart verification evaluation module (300) and the identity verification module (500). The smart verification evaluation module (300) can aggregate the user's historical incoming information, personal information, and current incoming information to evaluate through the smart verification evaluation model (201), and then determine whether the user can use smart verification. The identity verification module (500) can filter out the available verification methods (541) based on the method that the user has registered and the file type of the multimedia file (030) provided in the current verification, and the multimedia file (030) can be used The verification method extracts the user characteristics (511), compares the registered user model based on the user characteristics, and obtains the verification score (521) obtained after verification by the available verification method, and then uses the verification confidence value integration model (401) ) The verification score (521) is integrated into the verification confidence value (501), and the obtained verification confidence value (501) is used to provide an identity verification module to determine whether the identity verification is passed.

As mentioned above, the smart identity verification system (1) also includes a smart verification evaluation model training module (200) and a verification confidence value integration model training module (400). The smart verification evaluation model training module (200) can extract the first training data (211) from the user's personal information and historical incoming information to train the smart verification evaluation model (201) based on the first training data (211). The verification confidence value integration model training module (400) can extract the second training data (411) from the user's historical incoming information to train the verification confidence value integration model (401) based on the second training data (411).

Please refer to Figure 8. This figure shows the main flow of the smart identity verification method of the present invention. The implementation steps are described as follows.

Receive client information (901): Please refer to Figure 2 at the same time, the client interface unit (110) can receive the client (010) incoming line and its incoming information, such as device serial number (021), incoming account number (022) ) And the service number (023), and a set of incoming data is generated by the smart identity verification system (1) The line serial number is used as a unique identification code for incoming lines. This part of the information can be shown in Table 1 below, where it is assumed that the device serial number of the incoming line is "SERIAL1", the incoming account number is "12345", the service number is "001", and the system gives the incoming line of this incoming line The serial number is "789".

Table 1. Client incoming line information

Determine whether it is designated as a registered service (902): Please also refer to Figure 2. The flow control unit (150) accepts the service number (023) designated by the client (010). If the service number (023) represents a registered service, proceed to the identity registration process (910); otherwise, proceed to the process of verifying the validity of the account (903).

Check the validity of the account (903): Use the account data access unit (130) to inquire whether the account in the personal information database (A00) is the same as the incoming account (022) is still in use. If yes, proceed to the next step; if no, then return a message that the client (010) refuses to verify and end the service. Take the form of the personal information database shown in Table 2 below as an example. If the account number is "12345", the value of the field of whether it is in use is "Yes", so you can directly enter the smart verification evaluation process (920); if it is "No" ", the client (010) refuses to verify the message and ends the service.

Table 2. Forms of personal information database

Evaluate whether smart identity verification is possible (904): Please refer to Figure 2 and Figure 4 at the same time, call the smart verification evaluation module (300) with the back-end core client unit (140), and use the smart The smart verification evaluation model training process (930) trains the smart verification evaluation model (201), and then the smart verification evaluation process (920) obtains the smart verification evaluation result (301) to evaluate whether smart verification can be used. If the wisdom verification evaluation result (301) is yes, then the verification confidence value is calculated (905); on the contrary, if the wisdom verification evaluation result (301) is no, then the return result is rejection verification.

Calculate the verification confidence value (905): Please refer to Figure 2 and Figure 5 at the same time, use the back-end core client unit (140) to call the identity verification module (500) for the identity verification process (940), and integrate the verification confidence value The verification confidence value integration model (401) trained by the model training process (950) calculates the verification confidence value (501); otherwise, the returned result is unable to be verified.

Receive information on whether the client service has ended normally (906): Please refer to Figure 2 at the same time. After the client service process ends, the user must actively return whether the service has ended normally (025) and its incoming serial number (024) To the user interface interface unit (110), the system needs to record for model training.

Record the incoming line information to the incoming line log database (907): Please refer to Figure 2, Figure 4 and Figure 6 at the same time, and use the incoming line recording unit (120) to record the device serial number (021) and the incoming line account number (022) ), service number (023), incoming serial number (024), whether the service ends normally (025), verification confidence value (501), and smart verification evaluation result (301) are recorded in the incoming log database (B00). The data field.

Please refer to Figure 9, which shows the detailed steps of the identity registration process (910) of the present invention, which are described as follows.

Search for available registration methods (911): Please refer to Figure 7 at the same time, use the registration method screening unit (610) to find all the applicable features of the verification method database (C00) based on the file type of the multimedia file (030) The module name becomes the available registration method (611). For example In other words, if the verified multimedia type is an audio file, you can use the method provided in the verification method database (C00), as shown in Table 3 below, to find the applicable feature extraction modules, and organize them into the following Table 4 Show the available registration methods.

Table 3. Methods provided by the verification method database

Table 4. Available registration methods

Retrieve user features (912): Please refer to Figure 7 at the same time, use the user feature capture unit (620) to extract the registered features (from the multimedia file (030) with the feature capture module specified by the available registration method (611) 621). The registration feature (621) can be shown in Listing 5 below.

Table 5. Registration features

Update user model (913): Please refer to Figure 7 at the same time, use the user model update unit (630) to add the registered feature (621) to the user model database (D00), and determine whether to include the update in the user model column . For example, after new user features are added, whether the variability change ratio is less than 50% can be used to evaluate whether to include the update in the user model field, as shown in Table 6 below; as for the form in the updated user model database, it is as follows Shown in Listing 7.

Table 6. Whether to be included in the user model judgment form

Table 7. The updated form of the user model database

Return the registration result (914): Please refer to Fig. 2 and Fig. 7 at the same time, the process control unit (150) obtains the registration result, and sends the registration result back to the client through the user terminal interface unit (110).

Please refer to Figure 10, which shows the detailed steps of the smart verification evaluation process (920) of the present invention, which are described as follows.

Retrieve forecast data (921): Please refer to Figures 2 and 4 at the same time, and use the forecast data acquisition unit (310) to retrieve the personal information characteristics (312) of the current incoming account (022) and the summary characteristics of incoming records ( 313) and the current incoming line information feature (314) to combine to form prediction data (311). For detailed feature extraction methods, please refer to the description of the Smart Verification Evaluation Model Training Process (930).

Predict the evaluation result of smart verification (922): Please refer to Figure 4 at the same time, use the incoming prediction unit (320) to use the prediction data (311) to use the smart verification evaluation model (201) to predict the smart verification evaluation result of the incoming line (301) ), for example, the wisdom verification evaluation results shown in Table 8 below.

Table 8. Wisdom verification evaluation results

Please refer to Figure 11, which shows the detailed steps of the smart verification evaluation model training process (930) of the present invention, which are described as follows.

Retrieve the first training data (931): Please refer to Figure 3 at the same time, use the training data acquisition unit (210) to collect the first training from the personal information database (A00) and the incoming log database (B00) Training data (211), where the first training data (211) includes personal information characteristics (212), entry record summary characteristics (213), current entry information characteristics (214), and correct answers expected by the model (215).

Retrieving personal information features (932): Please also refer to Figure 3, where the personal information features (212) are extracted from information such as age, gender, and user model variability, and coded into feature vectors that can be used in deep learning. For example, the personal information feature extraction process shown in Table 9 below.

Table 9. Personal information feature extraction process

Retrieving the incoming log summary feature (933): Please refer to Figure 3 at the same time, where the retrieval method of the incoming log summary feature (213) is described as follows: the training data acquisition unit (210) uses the incoming log database (B00) ), count the user’s past online history, whether the user passed smart verification during a certain period in the past, the time of the last registration and the record of abnormal termination of the verification process during a certain period in the past, etc., and count the number of times by statistical means, or Encode as a time series as the in-line record summary feature (213). In one embodiment, for the record of each incoming serial number, the average confidence value of the previous three verifications of the same account can be obtained, whether the previous three times passed the smart verification evaluation feature vector, whether the previous three times passed the smart verification feature vector, and whether the previous three times failed The normal termination verification process feature vector and the number of days since the previous normal termination of the service, such as the incoming log database record form shown in Table 10 below, and the summary feature of incoming log records shown in Table 11 below. For example, the summary feature of the incoming line record with the incoming line serial number 005 is [0.7,1,1,1,0,1,1,1,1,0,16].

Table 10. Record form of incoming log database

Table 11. Summary characteristics of incoming records

Retrieve the current incoming line information feature (934): Please also refer to Figure 3, where the current incoming line information feature (214) includes device serial number, service number and time stamp, etc. In one embodiment, The above-mentioned information characteristics are respectively converted into the brand characteristic vector, service number characteristic vector and line-in time characteristic vector, as shown in Table 12 to Table 15 below.

Table 12. Label feature vector conversion table

Table 13. Service number feature vector conversion table

Table 14. Feature vector conversion table for incoming time

Table 15. Current incoming line information characteristics

Retrieve the correct answer expected by the model (935): Please refer to Figure 2 and Figure 3 at the same time. The correct answer expected by the model (215) is based on whether each service ends normally (025) as the standard answer, as shown in Table 16 below Show.

Table 16. Correct answers expected by the model

Use the training data to train the smart verification evaluation model (936): Please refer to Figure 3 at the same time, use the model training unit (220) to pass the first training data (211) through the fully connected layer, CNN, LSTM or various deep learning model architectures, The wisdom verification evaluation model (201) can be learned. For example, the first training data (211) in the following list 17 can be input to the model training unit (220), and then the smart verification evaluation model (201) can be learned through the deep learning model architecture of the fully connected layer.

Table 17. The first training data

Please refer to Figure 12, which shows the detailed steps of the identity verification process (940) of the present invention, which are described as follows.

Search for available verification methods (941): Please refer to Figure 6 at the same time, and use the user verification method filter unit (540) to verify that all users have registered according to the file type of the multimedia file (030) and the user model database (D00) Method number, find all applicable verification methods in the verification method database (C00) and become available verification methods (541). For example, the user verification method screening unit (540) searches the user model database (D00) based on the incoming account number "12345" to obtain all the verification methods available to the user, as shown in Table 18 below. All verification methods can be found from the verification method database (C00), as shown in Table 19 below. And from these methods, find out the available verification methods (541) that are available to the user and verify that the multimedia type is an audio file, as shown in Table 20 below. In this embodiment, the verification method numbers are 1 and 3 in order to find the available verification methods (541).

Table 18. User model database

Table 19. Verification methods in the verification method database

Table 20. Available verification methods

Retrieving user characteristics (942): Please refer to Figure 6 at the same time, using the user feature extraction unit (510) to extract the user characteristics from the multimedia file (030) according to the feature extraction module specified by the available verification method (541) (511). For example, using the available verification methods, numbered 1 and 3, the specified feature extraction module names speaker_verify_feature_extract_1 and stt_verify_feature_extract_1, the user feature extraction unit (510) is used to extract the user features (511), as shown in Table 21 below.

Table 21. User characteristics

Compare the user model and get the verification score (943): Please refer to Figure 6 at the same time, take out the user model with the user feature comparison unit (520), and compare it with the feature comparison module specified by the available verification method (541) The user model and user characteristics of the same verification method number are obtained through the score set obtained by the user characteristic comparison unit (520) to obtain the verification score (521). For example, verify the same The user feature (511) extracted by the method number is compared with the registered user model, and the feature comparison module names speaker_verify_feature_compare_1 and stt_verify_feature_compare_1 specified by the verification method number are compared with the user feature comparison unit (520) to verify each available The verification score (521) is obtained by the set of scores obtained by the method (541), as shown in Table 22 below.

Table 22, verification score

Calculate the verification confidence value (944): Please refer to Figure 6 at the same time, and use the smart verification confidence value calculation unit (530) to calculate the verification confidence value (501) from the verification score (521) through the verification confidence value integration model (401).

Return the verification result (945): Please refer to Figure 2 and Figure 6 at the same time, and return the verification confidence value (501) to the client through the client interface unit (110).

Please refer to Figure 13, which shows the detailed steps of the verification confidence value integration model training process (950) of the present invention, which are described as follows.

Acquire training data (951): Please also refer to Figure 5 to use the wisdom verification confidence value training data acquisition unit (410) to collect the wisdom verification confidence value model training data from the incoming log database (B00), which is the second Training materials (411). The second training data (411) includes at least the verification score (412) and the correct answer (413) expected by the model. The correct answer (413) expected by the model is based on whether each client service ends normally as the standard answer. In one embodiment, the correct answer (413) expected by the model is calculated as shown in the following table 23, which is based on whether each client service ends normally and whether Passing wisdom verification is the basis for the standard answer. If both columns are yes, the correct answer (413) expected by the model is 1; otherwise, it is 0. Users can also be asked to provide actual data sets for testing, and combine them into a registration and verification comparison data set to perform user model registration and verification one by one. In this case, the expected correct answer (413) of the model is the user model And verify whether the audio files are the same person, the same is 1 and the difference is 0. As for, the following table 24 is the second training data (411) captured here.

Table 23, the correct answer expected by the model

Table 24, the second training data

Training verification confidence value integration model (952): Please also refer to Figure 5 to use the wisdom verification confidence value model training unit (420) to learn from the second training data (411) through a fully connected layer or various deep learning model architectures Develop a verification confidence value integration model (401).

From the descriptions in Figures 8 to 13 above, it can be understood that the main steps of the smart identity verification method of the present invention include: receiving client information (901); verifying the validity of the account (903) to verify the user’s account Validity; evaluate whether smart identity verification can be performed (904), through the smart verification evaluation process (920), the user’s historical incoming information, personal information, and current incoming information are aggregated, and the smart verification evaluation model ( 930) perform an evaluation to determine whether smart verification can be performed; and if it is determined that smart verification can be performed, perform an identity verification process (940), wherein the identity verification process is based on the method that the user has registered and the provided multimedia file (030 ) File type to screen out the available verification methods, extract the user characteristics from the multimedia file to compare with the user model to obtain the verification score obtained after verification by the available verification method, and then use a verification confidence value integration model to integrate The verification score is integrated to calculate the verification confidence value (905).

In addition, as mentioned above, before proceeding to the step of verifying the validity of the account, the following steps can be included: judging whether it is designated as a registered service (902); and if the judgment result is designated as a registered service, proceed to the identity registration process (910) .

The smart identity verification system and method of the present invention have considerable advantages, which are described as follows.

First, different from the general traditional manual verification system, the smart identity verification system of the present invention is a fully automatic identity verification system, which can save labor costs on the enterprise side and provide more accurate verification results.

Second, the smart identity verification system of the present invention improves the limitation that the past system can only use one algorithm or the same type of algorithm. Under the framework of the system and method proposed in the present invention, a variety of different types of identity verification methods can be integrated, such as voiceprint verification, face verification, and fingerprint verification. Verification, etc., use deep learning or other machine learning methods to establish an integrated model to achieve a multi-factor and multi-model verification architecture to complement the verification security problems caused by a single factor and improve the usability and scalability of the client.

Furthermore, the improved conventional technology of the present invention can only judge whether the evaluation criteria of the smart verification service are applicable according to fixed conditions. By using the past incoming line records and user data, as well as the information reported by the client whether the service has been terminated normally, the deep learning training model can be used to evaluate whether the new incoming line is suitable for smart verification. Through this mechanism, different services can dynamically adjust the evaluation criteria implicit in the back-end model according to their service characteristics, and customize the security and availability requirements of different services.

The above-mentioned embodiments are merely illustrative to illustrate the technical principles, features and effects of the present invention, and are not intended to limit the scope of the present invention. Anyone familiar with this technology can do the same without departing from the spirit and scope of the present invention. The above embodiment is modified and changed. However, any equivalent modifications and changes made by using the teachings of the present invention should still be covered by the scope of the attached patent application. The scope of protection of the rights of the present invention shall be as listed in the scope of the attached patent application.

1‧‧‧Smart Identity Verification System

010‧‧‧Client

100‧‧‧Process Management Module

200‧‧‧Smart verification evaluation model training module

201‧‧‧Smart verification evaluation model

300‧‧‧Smart Verification Evaluation Module

400‧‧‧Verify the confidence value integration model training module

401‧‧‧Verify the confidence value integration model

500‧‧‧Identity Verification Module

600‧‧‧Identity Registration Module

A00‧‧‧Personal Information Database

B00‧‧‧Incoming log database

C00‧‧‧Verification Method Database

D00‧‧‧User Model Database

Claims (13)

A smart identity verification system includes: a smart verification evaluation module, which evaluates the user's historical incoming information, personal information, and current incoming information through a smart verification evaluation model to generate a smart verification evaluation result, and then the smart verification evaluation result is generated by the smart verification evaluation model. The verification evaluation module is based on whether the smart verification evaluation result generated by the smart verification evaluation model is yes or no to determine whether to perform identity verification on the user; and the identity verification module is based on the wisdom verification evaluation module The smart verification evaluation result generated by the verification evaluation model is yes. When determining the identity verification of the user, the verification method is based on the verification method that the user has registered and the file type of the multimedia file provided in the current verification. The available verification methods are filtered out in the database, the multimedia file is extracted from the available verification methods filtered in the verification method database to extract user characteristics, and the identity verification module will be retrieved from the verification method database. The selected user characteristics captured by the available verification method are compared with the registered user model in the user model database to obtain the verification score obtained after verification by the available verification method, and then through a The verification confidence value integration model will be based on the comparison result of the user characteristics extracted from the available verification methods selected in the verification method database and the registered user model in the user model database. The obtained verification score is integrated into a verification confidence value, so that the verification confidence value obtained through the verification confidence value integration model is provided to the identity verification module to determine whether the user passes the identity verification. For example, the smart identity verification system described in item 1 of the scope of patent application further includes: a smart verification evaluation model training module, which extracts the first training data from the personal information of the user and the historical incoming information, based on The first training data trains the smart verification evaluation model; and The verification confidence value integration model training module extracts second training data from the historical incoming information of the user to train the verification confidence value integration model based on the second training data. For example, the smart identity verification system described in item 2 of the scope of patent application, wherein the smart verification evaluation model training module includes: a training data acquisition unit, which is derived from a personal information database and records that record the user’s personal information The first training data is retrieved from the incoming log database of the historical incoming information of the user; and the model training unit is used to train the smart verification evaluation model. For example, the smart identity verification system described in item 2 of the scope of patent application, wherein the verification confidence value integration model training module further includes: a smart verification confidence value training data acquisition unit, which retrieves the first data from the incoming log database 2. Training data; and a smart verification confidence value model training unit for training the verification confidence value integration model. For example, the smart identity verification system described in item 1 of the scope of the patent application further includes: a personal information database, which records the personal information of the user; an incoming log database, which records the historical incoming information of the user; the verification method The database records the defined available verification methods and related information; and the user model database records the user model and related feature vectors registered by the user. For example, the smart identity verification system described in item 1 of the scope of patent application, wherein the smart verification evaluation module further includes: a predictive data acquisition unit, which is the historical incoming information and personal information of the user in the incoming log database The personal information in the database is merged with the incoming information of the current incoming client to generate a forecast data; and The incoming prediction unit receives the prediction data and uses the smart verification evaluation model to evaluate the prediction data to generate the smart verification evaluation result, so that the smart verification evaluation result generated according to the smart verification evaluation model is yes or no, To determine whether to verify the identity of the user. For example, the smart identity verification system described in item 1 of the scope of patent application, wherein the identity verification module further includes: a user verification method screening unit, which is based on the method that the user has registered and the file type of the multimedia file, The available verification method is selected from the verification method database; the user feature extraction unit extracts the user characteristics from the multimedia file according to the available verification method; the user feature comparison unit uses the verification method data The feature comparison method specified in the library compares the user feature with the registered user model in the user model database to obtain the verification score obtained after verification according to the available verification method; and smart verification The confidence value calculation unit integrates the verification score into the verification confidence value according to the verification confidence value integration model. For example, the smart identity verification system described in item 1 of the scope of patent application further includes a process management module, including: a client interface unit, which accepts verification or registration requests from the user; an account data access unit, which stores Obtain data from the personal information database to confirm whether it is a valid account; the process control unit is used to control the verification process; the back-end core client unit is used to query the smart verification evaluation model training module and the identity verification module And an identity registration module; and an incoming log unit, which records the incoming log of the user in the incoming log database. For example, the smart identity verification system described in item 8 of the scope of patent application, wherein the identity registration module includes: a registration method screening unit, which filters out the available registration methods from the verification method database according to the file type of the multimedia file The user feature extraction unit is used to extract registered features from the multimedia file according to the feature extraction method specified in the verification method database; and the user model update unit is used to update the user model database. A smart identity verification method includes: a smart verification evaluation module evaluates the user's historical incoming information, personal information, and current incoming information through a smart verification evaluation model to generate a smart verification evaluation result, which is then verified by the wisdom The evaluation module determines whether the smart verification evaluation result generated by the smart verification evaluation model is yes or no to determine whether to verify the user's identity; and if the smart verification evaluation module generates the smart verification evaluation model based on the smart verification evaluation model. The result of the smart verification evaluation is yes, to determine the identity verification for the user, then the identity verification module is based on the verification method that the user has registered and the file type of the multimedia file provided by the current verification from the verification method database The available verification methods are screened out, the multimedia file is extracted from the available verification methods in the verification method database to extract user characteristics, and the identity verification module will be selected from the verification method database. The user characteristics captured by the available verification method are compared with the registered user model in the user model database to obtain the verification score obtained after verification by the available verification method, and then a verification confidence value is used The integrated model will be based on the comparison result of the user characteristics extracted from the available verification method selected in the verification method database and the registered user model in the user model database. The verification score is integrated into a verification confidence value, so that the verification confidence value obtained through the verification confidence value integration model is provided to the identity verification module to determine whether the user passes the identity verification. For example, the smart identity verification method described in item 10 of the scope of patent application further includes the following steps: the process control unit judges whether it is designated as a registered service; if the judgment result is designated as the registered service, an identity registration process is carried out. The identity registration process includes the following steps: the registration method screening unit searches for available registration methods; the user feature extraction unit captures the user features; the user model update unit updates the user model; and the user terminal interface unit returns the registration result. For example, the smart identity verification method described in item 10 of the scope of patent application further includes performing a smart verification evaluation model training process to train the smart verification evaluation model, and the smart verification evaluation model training process includes: training data acquisition unit acquisition The first training data includes the extracted personal information characteristics, the incoming record summary characteristics, the current incoming information characteristics and the correct answer expected by the model; and the model training unit uses the first training data to train the wisdom verification evaluation model. For example, the smart identity verification method described in item 10 of the scope of patent application further includes a verification confidence value integration model training process to train the verification confidence value integration model, and the verification confidence value integration model training process includes: smart verification confidence value The training data acquisition unit acquires the second training data, including acquiring the verification score and the correct answer expected by the model; and the smart verification confidence value model training unit trains the verification confidence value integration model using the second training data.

Images