Identification information processing method and device, electronic equipment, medium and program product

Abstract

The application provides a kind of identification information processing method, device, electronic equipment, medium and program product, can be applied to privacy computing technical field.The method comprises: the hash processing of each identification information in the first identification information set stored locally is carried out, and the first hash identification set is obtained;Random number is generated for each hash identification and is saved, each hash identification is encrypted based on asymmetric encryption public key and the random number of each hash identification, and the encrypted confusion identification set is obtained;The decrypted identification set and the second hash identification set returned by the second participant are received;The random number elimination of corresponding decrypted identification in decrypted identification set is carried out based on the random number of each hash identification, and the identification set to be compared is obtained;The identification set to be compared is compared with the second hash identification set, and the common identification information set is determined.

Description

Technical Field

[0001] This application relates to the field of privacy computing technology and can be used in the field of fintech or other fields. More specifically, it relates to a method, apparatus, electronic device, medium, and program product for processing identification information. Background Technology

[0002] In cross-institutional data collaboration scenarios, in order to complete customer profiles or conduct joint risk control, it is often necessary to determine shared identification information (such as user ID, mobile phone number, etc.) while protecting privacy.

[0003] Existing conventional solutions typically rely on one party directly sending its plaintext identification information to the other, or the receiving party performing a hash operation on the received plaintext identification information and then comparing it. While these methods are simple to implement, the identification information is exposed in plaintext or as a single hash during data transmission and processing, lacking an effective privacy protection layer.

[0004] This demonstrates that the identification information processing methods in cross-organizational data collaboration scenarios have insufficient security. Summary of the Invention

[0005] In view of the above problems, embodiments of this application provide a method, apparatus, electronic device, medium, and program product for processing identification information.

[0006] According to a first aspect of this application, a method for processing identification information is provided, applied to a first participant. The method includes: in response to receiving an asymmetric encryption public key sent by a second participant, hashing each identification information in a locally stored first identification information set to obtain a first hash identification set; generating and storing a random number for each hash identification in the first hash identification set; encrypting each hash identification in the first hash identification set based on the asymmetric encryption public key and the random number of each hash identification to obtain an encrypted obfuscated identification set, and sending the encrypted obfuscated identification set to the second participant; receiving a decrypted identification set and a second hash identification set returned by the second participant, wherein the decrypted identification set is obtained by the second participant decrypting the encrypted obfuscated identification set based on the private key corresponding to the asymmetric encryption public key, and the second hash identification set is obtained by the second participant hashing its locally stored second identification information set; performing random number elimination on the corresponding decrypted identification in the decrypted identification set based on the random number of each hash identification to obtain a set of identifications to be compared; comparing the set of identifications to be compared with the second hash identification set to determine a common identification information set, wherein the common identification information set represents the set of common identification information existing in the first identification information set and the second identification information set.

[0007] According to an embodiment of this application, each hash identifier in the first hash identifier set is encrypted based on the asymmetric encryption public key and the random number of each hash identifier to obtain an encryption obfuscation identifier set, including: performing encryption operations based on the modulus of the asymmetric encryption public key and the public key exponent, based on each hash identifier and the random number of each hash identifier, to generate an encryption obfuscation identifier corresponding to each hash identifier; and constructing an encryption obfuscation identifier set based on the encryption obfuscation identifier corresponding to each hash identifier.

[0008] According to an embodiment of this application, random number elimination is performed on the corresponding decryption identifiers in the decryption identifier set based on the random number of each hash identifier to obtain a set of identifiers to be compared, including: determining the random number associated with each hash identifier in the first hash identifier set corresponding to each decryption identifier in the decryption identifier set; performing an inverse operation on the decryption identifier based on the random number to obtain the identifier to be compared corresponding to each decryption identifier; and constructing a set of identifiers to be compared based on the identifier to be compared corresponding to each decryption identifier.

[0009] According to an embodiment of this application, after comparing the set of identifiers to be compared with the second hash identifier set to determine the common identifier information set, the method further includes: determining the first transaction data of the owner corresponding to the identifier information in the common identifier information set; inputting the first transaction data into a trained vertical federated learning model to perform default prediction, and obtaining the default risk of the owner corresponding to the identifier information in the common identifier information set output by the vertical federated learning model.

[0010] According to an embodiment of this application, the vertical federated learning model includes a first feature extraction layer located in the first participant, a second feature extraction layer located in the second participant, an interaction fusion layer located in the first participant, and a classification prediction layer located in the first participant. Inputting first transaction data into the trained vertical federated learning model for default prediction, and obtaining the default prediction result output by the vertical federated learning model, includes: inputting the first transaction data into the first feature extraction layer for feature transformation to obtain a first feature vector; receiving a second feature vector sent by the second participant, wherein the second feature vector is obtained by the second participant inputting the second transaction data determined by a common identifier information set into the second feature extraction layer for feature transformation; inputting the first feature vector and the second feature vector into the interaction fusion layer for feature fusion to obtain a fused feature; and inputting the fused feature into the classification prediction layer for classification prediction to obtain the default risk.

[0011] According to an embodiment of this application, the vertical federated learning model includes a first feature extraction layer, a second feature extraction layer, an interactive fusion layer, and a classification prediction layer. The vertical federated learning model is trained in the following manner: a training sample set is constructed based on a common identifier information set, the training sample set including: a first transaction data sample and a corresponding real default risk label; the first transaction data sample is input into the first feature extraction layer for feature transformation to obtain a first feature training vector; a second feature training vector is received from a second participant, the second feature training vector being obtained by the second participant inputting the second transaction data sample determined by the common identifier information set into the second feature extraction layer for feature transformation; the first feature training vector and the second feature training vector are input into the interactive fusion layer for feature fusion to obtain a fused training feature; the fused training feature is input into the classification prediction layer for classification prediction to obtain a predicted default risk label; based on the predicted default risk label and the real default risk label, a preset loss function is used to determine the prediction loss of the vertical federated learning model.

[0012] According to an embodiment of this application, after determining the predicted loss of the longitudinal federated learning model using a preset loss function based on the predicted default risk label and the actual default risk label, the method further includes: determining the gradient of the predicted loss relative to the parameters of each level in the longitudinal federated learning model; adding random noise conforming to a Laplace distribution to the gradient of each level parameter to obtain a noisy gradient set; and updating the parameters of each level in the longitudinal federated learning model based on the noisy gradient set to obtain a trained longitudinal federated learning model.

[0013] According to an embodiment of this application, updating the parameters of each layer in a longitudinal federated learning model based on a noisy gradient set to obtain a trained longitudinal federated learning model includes: determining a first noise gradient corresponding to a first feature extraction layer and a second noise gradient corresponding to a classification prediction layer in the noisy gradient set; updating the first feature extraction layer based on the first noise gradient and updating the classification prediction layer based on the second noise gradient; sending the first noise gradient to a second participant, and the second participant updating its second feature extraction layer based on the received first noise gradient.

[0014] According to a second aspect of this application, an identification information processing apparatus is provided, comprising: a public key receiving module, configured to, in response to receiving an asymmetric encryption public key sent by a second participant, perform hash processing on each identification information in a locally stored first identification information set to obtain a first hash identification set; a random encryption module, configured to generate and store a random number for each hash identification in the first hash identification set, encrypt each hash identification in the first hash identification set based on the asymmetric encryption public key and the random number for each hash identification to obtain an encrypted and obfuscated identification set, and send the encrypted and obfuscated identification set to the second participant; and a set receiving module, configured to receive a decrypted identification set returned by the second participant. The system comprises a first hash identifier set and a second hash identifier set. The decryption identifier set is obtained by the second participant decrypting the encrypted and obfuscated identifier set using the private key corresponding to the asymmetric encryption public key. The second hash identifier set is obtained by the second participant hashing the second identifier information set stored locally. The random elimination module is used to eliminate the corresponding decryption identifier in the decryption identifier set based on the random number of each hash identifier, thereby obtaining the identifier set to be compared. The identifier determination module is used to compare the identifier set to be compared with the second hash identifier set to determine the common identifier information set. The common identifier information set represents the set of common identifier information that exists in the first identifier information set and the second identifier information set.

[0015] According to a third aspect of this application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] According to a fourth aspect of this application, a computer-readable storage medium is also provided, on which a computer program or instructions are stored, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0017] According to a fifth aspect of this application, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0018] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 This illustration schematically depicts an application scenario of the identification information processing method according to an embodiment of this application.

[0020] Figure 2 A flowchart illustrating an identification information processing method according to an embodiment of this application is shown schematically.

[0021] Figure 3 The diagram illustrates the interactive flowchart of the identification information processing method according to an embodiment of this application.

[0022] Figure 4 A model architecture diagram of a vertical federated learning model according to an embodiment of this application is shown;

[0023] Figure 5 This illustration schematically shows a structural diagram of the interaction fusion layer according to an embodiment of the present application;

[0024] Figure 6 This schematic diagram illustrates a structural block diagram of an identification information processing apparatus according to an embodiment of the present application;

[0025] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing identification information processing according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0030] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all provide users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process enters the expert decision-making process.

[0031] As used herein, the term "model" refers to a model that learns the relationship between inputs and outputs from training data, thereby generating a corresponding output for a given input after training. Model generation can be based on machine learning techniques. Deep learning is a machine learning algorithm that processes inputs and provides corresponding outputs using multiple layers of processing units. A neural network model is an example of a deep learning-based model. In this application, "model" may also be referred to as a "machine learning model," "learning model," "machine learning network," or "learning network," and these terms are used interchangeably throughout this application.

[0032] Figure 1 The illustration shows an application scenario diagram of the identification information processing method according to an embodiment of this application. For example... Figure 1 As shown, application scenario 100 according to an embodiment of this application may include a first participating device 101, a second participating device 102, and a network 103. The network 103 serves as a medium for providing a communication link between the first participating device 101 and the second participating device 102. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables. A user can use the first participating device 101 to interact with the second participating device 102 through the network 103 to receive or send information, etc.

[0033] The first participating device 101 and the second participating device 102 can be electronic devices such as smartphones, wearable devices, personal computers, intelligent voice interaction devices, smart home appliances, intelligent vehicles, in-vehicle terminals, aircraft, unmanned vending terminals, and extended reality devices. Extended reality devices can include virtual reality devices, augmented reality devices, and mixed reality devices. The terminal device can install and run a client application for the target application, which can include, but is not limited to, financial transaction applications, payment applications, shopping applications, web browser applications, search applications, instant messaging tools, email clients, and social media platform software (only examples). Furthermore, this application embodiment does not limit the form of the target application, including but not limited to applications, mini-programs, etc., installed on the terminal device, and can also be in web page form.

[0034] It should be noted that the identification information processing method provided in this application embodiment can generally be executed by the first participating party device 101. Accordingly, the identification information processing device provided in this application embodiment can generally be disposed in the first participating party device 101.

[0035] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0036] First, the technical terms described in this application are explained and clarified as follows.

[0037] Vertical federated learning: a machine learning method that allows multiple participants to jointly train a model without sharing their original data. It is suitable for scenarios where participants have the same samples but different features.

[0038] Differential privacy: A privacy protection technique that adds random noise to data or model parameters, making it impossible for attackers to determine whether a specific piece of data was used in training. The amount of noise is controlled by a privacy budget parameter.

[0039] Encrypted Data Identity Alignment: A privacy-preserving data matching method that finds a common set of users among multiple data holders without revealing the identity information of non-intersecting users.

[0040] Existing cross-institutional default risk prediction solutions mainly fall into two categories. The first type of solution requires directly aggregating the original customer data from all parties to a central node for modeling. This approach cannot meet data security compliance requirements and poses a high information security risk. The second type of solution employs vertical federated learning without privacy protection. Although it does not directly transmit the original data, attackers may still infer sensitive features through gradients or model updates during model interaction, resulting in insufficient privacy protection.

[0041] The aforementioned approach leads to a significant "data silo" problem. Banks and wealth management companies serve the same customer base, and their data is highly complementary. Banks focus on data related to customers' debt repayment ability, while wealth management companies focus on data related to customers' investment attributes. However, due to privacy concerns, the two sides cannot effectively integrate their data. Relying solely on one party's characteristics for modeling results in a single dimension of customer risk assessment, failing to comprehensively reflect the customer's true asset and liability situation, easily leading to misjudgments, and affecting the accuracy of credit decisions and the appropriateness of wealth management product recommendations.

[0042] Therefore, there is a need for a method for processing identification information that can predict default risk across participating parties without disclosing the original data.

[0043] Figure 2 A flowchart illustrating an identification information processing method according to an embodiment of this application is shown. Figure 2 As shown, the identification information processing method 200 according to the embodiments of this application may include steps S210 to S250.

[0044] This application is applied to a default risk prediction scenario between a bank (i.e., the first participant / initiator) and a wealth management agency (i.e., the second participant / partner). Both parties hold different dimensions of data on the same customer group, and the aim is to determine a set of common identification information without disclosing the original data through privacy computing technology.

[0045] In step S210, in response to receiving the asymmetric encryption public key sent by the second participant, each identifier in the locally stored first identifier information set is hashed to obtain the first hash identifier set.

[0046] In embodiments of this application, user consent or authorization can be obtained before acquiring the first set of locally stored identification information. For example, before step S210, a request can be sent to the user to acquire the first set of locally stored identification information. If the user consents or authorizes the acquisition of user information, step S210 is executed.

[0047] The first participant receives an asymmetric encryption public key from the second participant. This asymmetric encryption public key is pre-generated by the second participant and is used to encrypt data in subsequent processes, ensuring that only the second participant, who holds the corresponding private key, can decrypt it.

[0048] The first participating party retrieves a first set of identification information stored locally. The first set of identification information contains customer identification information held by the first participating party, such as customer ID card number, mobile phone number, or bank account number, which can uniquely identify the customer.

[0049] To prevent the original identification information from being leaked during subsequent interactions, the first participant invokes a cryptographic hash function to perform hash processing on each identification information in the first identification information set. Hash processing is a one-way transformation process that can convert input data of arbitrary length into a fixed-length string output, and the input cannot be deduced from the output. After this processing, the original identification information is converted into an irreversible hash value.

[0050] The first participating party performs hash transformation on all the identifiers in the first identifier information set one by one, and then summarizes them to form the first hash identifier set. The elements in this set are all hash values ​​that have been de-identified and no longer contain plaintext information that can directly identify the customer's identity, thus providing a basis for subsequent encrypted transmission and matching operations.

[0051] In step S220, a random number is generated and saved for each hash identifier in the first hash identifier set. Based on the asymmetric encryption public key and the random number of each hash identifier, each hash identifier in the first hash identifier set is encrypted to obtain an encrypted and obfuscated identifier set, and the encrypted and obfuscated identifier set is sent to the second participant.

[0052] The first participant generates a corresponding random number (or mask value) for each hash identifier in the first hash identifier set, and associates and saves the random number with the first participant's local machine for subsequent elimination operations.

[0053] The first participant uses the asymmetric encryption public key, pre-received from the second participant, and the random number corresponding to each hash identifier to perform encryption operations on each hash identifier in the first hash identifier set. This encryption process obfuscates and binds the hash identifiers with the random numbers, generating an encrypted obfuscated identifier set. Subsequently, the first participant sends this encrypted obfuscated identifier set to the second participant via the network. At this point, the transmitted data, having undergone public key encryption and random number obfuscation, does not possess plaintext characteristics, effectively ensuring data security during transmission.

[0054] In step S230, the decryption identifier set and the second hash identifier set returned by the second participant are received.

[0055] Among them, the decryption identifier set is obtained by the second participant decrypting the encrypted obfuscated identifier set based on the private key corresponding to the asymmetric encryption public key, and the second hash identifier set is obtained by the second participant performing hash processing on the second identifier information set stored locally.

[0056] After receiving the set of encrypted obfuscated identifiers, the second participant uses its asymmetric encryption private key to decrypt the set and obtain the decrypted identifier set (which still contains the random number interference added by the first participant).

[0057] The second participant performs the same hashing process as in step S210 on the locally stored second set of identifier information (i.e., the customer identifiers held by the second participant) to generate a second hash identifier set. Subsequently, the second participant packages the decrypted identifier set and the second hash identifier set together and returns them to the first participant.

[0058] The first participant receives the decryption identifier set and the second hash identifier set returned by the second participant.

[0059] In step S240, the corresponding decryption identifiers in the decryption identifier set are randomized based on the random number of each hash identifier to obtain the identifier set to be compared.

[0060] After receiving the returned data, the first participant performs random number elimination processing on each data item in the decryption identifier set based on the random numbers stored locally in step S220 (i.e., removing random number interference through mathematical inverse operations). After the elimination processing, the hash identifiers that were originally covered by random numbers are restored, forming the identifier set to be compared.

[0061] Since the first participant does not possess the second participant's private key, it cannot decrypt the original data sent by the second participant. Therefore, this step only restores the obfuscated data (i.e., the set of encrypted obfuscated identifiers) sent by itself, further ensuring the privacy of the computation process.

[0062] In step S250, the set of identifiers to be compared is compared with the second hash identifier set to determine the set of common identifier information.

[0063] The common identification information set represents the set of common identification information that exists in the first identification information set and the second identification information set.

[0064] In the embodiments of this application, a corresponding operation entry can be provided to the user, allowing the user to choose to agree to or reject the automated decision result. That is, before comparing the set of identifiers to be compared with the second hash identifier set to determine the common identifier information set, the user can obtain an instruction to agree to or reject the processing / decision input through the corresponding operation entry. If the user agrees to the processing / decision, the set of identifiers to be compared with the second hash identifier set is compared to determine the common identifier information set, i.e., step S250 is executed. If the user refuses to process / decision, the expert decision-making process is entered.

[0065] The first participant compares the restored set of identifiers to be compared with the second set of hashed identifiers returned by the second participant. Since the elements in both sets are irreversible strings processed by the same hash algorithm, the first participant only needs to compare the consistency of the strings to determine whether there is common identifier information. The matching elements in the comparison result are the identifier information jointly held by both parties, and the set of these matching elements is the common identifier information set.

[0066] The first and second participants can index the customer information shared by both parties based on the common identification information set, thereby providing an aligned sample basis for subsequent joint model training and achieving accurate customer identity matching without exchanging original data.

[0067] In this embodiment, the first participant hashes the first set of identifier information, transforming sensitive identifiers into an irreversible first hash identifier set. Then, each hash identifier is bound to a random number and encrypted using an asymmetric encryption public key, generating an encrypted and obfuscated identifier set that cannot be reversed and sent to the second participant, thus eliminating the risk of original identifier leakage during transmission. The second participant uses its private key to decrypt and obtain a decrypted identifier set containing random numbers, which it returns. The first participant performs an elimination operation based on pre-stored random numbers to restore a clean identifier set to be compared. Finally, by comparing this set with the second hash identifier set provided by the second participant, the common identifier information set of both parties is accurately determined. This achieves cross-partner identifier information intersection without exposing the original identifier information of either party, balancing data security and privacy protection with the accuracy of the calculation results.

[0068] In some embodiments, each hash identifier in the first hash identifier set is encrypted based on the asymmetric encryption public key and the random number of each hash identifier to obtain an encryption obfuscation identifier set, including: performing encryption operations based on each hash identifier and the random number of each hash identifier according to the modulus of the asymmetric encryption public key and the public key exponent to generate an encryption obfuscation identifier corresponding to each hash identifier; and constructing an encryption obfuscation identifier set based on the encryption obfuscation identifier corresponding to each hash identifier.

[0069] In this embodiment of the application, the first participant retrieves a locally stored asymmetric encryption public key (e.g., an RSA public key), which includes a modulus for limiting the range of numerical calculations and a public key exponent for exponential operations.

[0070] For each hash identifier in the first hash identifier set, the first participant performs encryption processing one by one. When processing a single hash identifier, the first participant calls the preset encryption logic to combine the value of the hash identifier with the random number corresponding to the hash identifier in a specific form (i.e., using the modular exponentiation rule), and performs calculations using the modulus and public key exponent in the asymmetric encryption public key to obtain the encrypted obfuscated identifier corresponding to the hash identifier.

[0071] Because of the introduction of random numbers and the use of asymmetric encryption algorithms, even with the same hash identifier, the ciphertext generated each time will be different, and third parties cannot deduce the original hash value from the ciphertext.

[0072] After the above encryption operation is completed, the first participant summarizes the encrypted obfuscated identifiers generated from all hash identifiers in the first hash identifier set. These encrypted obfuscated identifiers together constitute a complete encrypted obfuscated identifier set.

[0073] Through the embodiments of this application, the randomness of encryption is enhanced by using random numbers, which effectively prevents hash identifiers from being correlated and analyzed, and significantly improves the privacy and security of data transmission.

[0074] In some embodiments, random number elimination is performed on the corresponding decryption identifiers in the decryption identifier set based on the random number of each hash identifier to obtain a set of identifiers to be compared, including: determining the random number associated with each hash identifier in the first hash identifier set corresponding to each decryption identifier in the decryption identifier set; performing an inverse operation on the decryption identifier based on the random number to obtain the identifier to be compared corresponding to each decryption identifier; and constructing a set of identifiers to be compared based on the identifier to be compared corresponding to each decryption identifier.

[0075] In this embodiment of the application, when the first participant generates the set of encrypted obfuscation identifiers, it generates and saves an associated random number for each hash identifier in the first set of hash identifiers, and the association between the hash identifier and the random number is stored locally in order or marked by an index.

[0076] The first participant can determine the hash identifier corresponding to each decryption identifier one by one based on the sequential position or index information of the decryption identifier in the decryption identifier set, and retrieve the random number associated with the hash identifier from the local storage based on the association relationship between the hash identifier and the random number.

[0077] Because the random number is mixed and bound with the hash identifier through specific operational rules during the encryption phase, the decrypted data still contains the influence of that random number. The first participant removes the random number from the decryption identifier by performing the inverse mathematical operation of the encryption phase, thereby restoring the identifier to be compared.

[0078] The first participant aggregates all the restored identifiers to be compared, constructing a set of identifiers to be compared. The identifiers in this set have had random number interference eliminated and have the same format and processing logic as the second hash identifier set returned by the second participant, and can be used for subsequent intersection comparison operations.

[0079] In this embodiment, a random number is determined and an inverse operation is performed on the decrypted identifier to eliminate the random number and restore the identifier to be compared, thereby constructing a set of identifiers to be compared. This process effectively removes the randomness interference introduced by encryption, ensures the accuracy of identifier restoration, and provides a foundation for subsequent data comparison.

[0080] refer to Figure 3 , Figure 3 The diagram illustrates the interactive flowchart of the identification information processing method according to an embodiment of this application.

[0081] The first participant (PartyA) stores the first set of identification information. ,in, These represent the first, second, third, and fourth identification information in the first set of identification information, respectively.

[0082] The second participant (PartyB) stores a second set of identification information. ,in, These represent the first, second, third, and fifth identifiers in the second identifier information set, respectively. The second participant also holds an asymmetric encryption key pair, such as an RSA key pair. ),in, Represents the modulus. Represents the public key. This represents the private key.

[0083] The second participant sends the public key from the asymmetric encryption key pair to the first participant.

[0084] The first participant extracts the first set of identifier information. Each identifier information in Then, each identifier is hashed using a hash function to obtain the first set of hash identifiers. .

[0085] The first participant generates a random number for each identifier in the first hash identifier set. Using the modulus of the received asymmetric encryption public key and the public key exponent The first participant calculates the set of encrypted obfuscation identifiers. :

[0086]

[0087] in, Represents a set of encryption obfuscation identifiers. Indicates the relationship with the first hash identifier The corresponding random number, The public key exponent represents the public key of an asymmetric encryption public key. The modulus representing the public key for asymmetric encryption. This represents the first set of identification information.

[0088] The first participant will encrypt the set of obfuscated identifiers. Send to the second participant.

[0089] After receiving the set of encrypted obfuscated identifiers, the second participant decrypts the set using the private key from the asymmetric encryption key pair, obtaining the decrypted identifier set. Based on the properties of the RSA algorithm, each received element is decrypted:

[0090]

[0091] in, Represents the set of decryption identifiers. Indicates the first A hash identifier, Indicates the relationship with the first hash identifier The corresponding random number, The public key exponent represents the public key of an asymmetric encryption public key. This represents the private key in an asymmetric encryption key pair. This represents a set of encryption obfuscation identifiers.

[0092] The second participant calculates the hash value of each element in the stored second identifier set and uses the private key from the asymmetric encryption key pair. Perform an exponential operation to obtain the second set of hash identifiers. The calculation formula is as follows:

[0093]

[0094] in, Represents the second set of hash identifiers. Indicates hash processing, Indicates the first A hash identifier, This represents the private key in the encryption key pair. This represents the second set of customer identifiers.

[0095] The second participant returns the decryption identifier set and the second hash identifier set to the first decryptor.

[0096] After receiving the decryption identifier set and the second hash identifier, the first participant performs random number elimination on each element in the decryption identifier set to obtain the identifier set to be compared, thus removing the influence of random numbers:

[0097]

[0098] in, This represents the set of identifiers to be compared. Indicates hash processing, Indicates the first A hash identifier, This represents the private key in the encryption key pair. Indicates the relationship with the first hash identifier The corresponding random number, This represents the set of decryption identifiers.

[0099] The first participant compares the set of identifiers to be compared with the second set of hash identifiers. The elements that match are the common identifiers of both parties, calculated using the following formula:

[0100]

[0101] in, Represents a set of common identification information. This represents the set of identifiers to be compared. Represents the second set of hash identifiers. Indicates hash processing, Indicates the first A hash identifier, where , This represents the private key in the encryption key pair.

[0102] Based on the shared identifier set, both the first participant and the second participant can deduce the identifier information they jointly possess: .

[0103] In some embodiments, after comparing the set of identifiers to be compared with the second hash identifier set to determine the common identifier information set, the method further includes: determining the first transaction data of the owner corresponding to the identifier information in the common identifier information set; inputting the first transaction data into a trained longitudinal federated learning model to perform default prediction, and obtaining the default risk of the owner corresponding to the identifier information in the common identifier information set output by the longitudinal federated learning model.

[0104] In this embodiment of the application, the first participant traverses each identifier in the common identifier information set, and based on the identifier, the first participant indexes and extracts the first transaction data of the owner corresponding to the identifier in the locally stored customer data.

[0105] For example, the first transaction data could be financial attribute data held by the bank, including customer credit scores, repayment behavior, credit card statements, and account transaction history.

[0106] The first participant inputs the first transaction data into a pre-trained longitudinal federated learning model. This model is constructed through multiple rounds of joint training, with the first participant as the initiator and the second participant as the collaborator, based on the alignment of encrypted data identifiers. The longitudinal federated learning model receives the first transaction data as input, performs calculations and feature fusion within its internal neural network layers, and outputs the probability value of the entity defaulting within a predetermined future time period, i.e., the default risk.

[0107] Through the embodiments of this application, by extracting the first transaction data corresponding to the common identifier and inputting it into a vertical federated learning model, default risk prediction is achieved; while protecting the data privacy of all parties, the prediction efficiency of default risk of the attributing entity corresponding to the common identifier information is improved.

[0108] In some embodiments, the longitudinal federated learning model includes a first feature extraction layer located in the first participant, a second feature extraction layer located in the second participant, an interaction fusion layer located in the first participant, and a classification prediction layer located in the first participant.

[0109] The first transaction data is input into a trained longitudinal federated learning model for default prediction, and the default prediction result output by the longitudinal federated learning model is obtained. This includes: inputting the first transaction data into a first feature extraction layer for feature transformation to obtain a first feature vector; receiving a second feature vector sent by a second participant, wherein the second feature vector is obtained by the second participant inputting the second transaction data determined by the common identification information set into a second feature extraction layer for feature transformation; inputting the first feature vector and the second feature vector into an interactive fusion layer for feature fusion to obtain a fused feature; and inputting the fused feature into a classification prediction layer for classification prediction to obtain the default risk.

[0110] In this embodiment, the first participant inputs the first transaction data indexed by a common identifier information set into a first feature extraction layer deployed locally on the first participant's premises. This first feature extraction layer consists of multiple neural networks and is used to perform feature transformation on the input first transaction data, outputting a first feature vector representing the customer's financial attributes.

[0111] The second participant determines the second transaction data (e.g., financial data, including investment preferences, holdings, etc.) of the corresponding subject stored locally based on the common identification information set, and inputs it into the second feature extraction layer deployed in the second participant to obtain the second feature vector, and then sends it to the first participant.

[0112] After obtaining the second feature vector, the first participant inputs both the first and second feature vectors into an interactive fusion layer deployed locally on its own machine. This interactive fusion layer concatenates or performs a weighted summation operation on the feature vectors from the two sources to generate a fused feature that integrates the data features from both parties.

[0113] The first participant inputs the generated fused features into a classification prediction layer deployed locally on the first participant's machine. This classification prediction layer consists of a fully connected neural network, which performs logistic regression or classification operations on the fused features, and finally outputs the default risk corresponding to the attributing entity.

[0114] This application's embodiments utilize a longitudinal federated learning model for classification prediction based on local feature extraction and multi-party feature vector fusion. This enables multi-party feature complementarity while ensuring data privacy, significantly improving the comprehensiveness and accuracy of default risk prediction.

[0115] In some embodiments, the longitudinal federated learning model includes a first feature extraction layer, a second feature extraction layer, an interaction fusion layer, and a classification prediction layer.

[0116] The longitudinal federated learning model is trained as follows: A training sample set is constructed based on a common identifier information set. This training sample set includes: a first transaction data sample and its corresponding real default risk label; the first transaction data sample is input into a first feature extraction layer for feature transformation to obtain a first feature training vector; a second feature training vector is received from a second participant, which is obtained by the second participant inputting the second transaction data sample determined by the common identifier information set into a second feature extraction layer for feature transformation; the first and second feature training vectors are input into an interactive fusion layer for feature fusion to obtain fused training features; the fused training features are input into a classification prediction layer for classification prediction to obtain a predicted default risk label; based on the predicted default risk label and the real default risk label, a preset loss function is used to determine the prediction loss of the longitudinal federated learning model.

[0117] refer to Figure 4 , Figure 4 A model architecture diagram of a vertical federated learning model according to an embodiment of this application is shown, which includes: a first feature extraction layer (located in the first participant, including the first layer to the nth layer), a second feature extraction layer (located in the second participant, including the first layer to the nth layer), an interactive fusion layer (including the first feature vector and the second feature vector), and a classification prediction layer (including the first layer to the nth layer, used to determine the prediction loss Loss).

[0118] In this embodiment of the application, the first participating party extracts the corresponding first transaction data sample and its true default risk label from local storage based on a common identification information set.

[0119] For example, the first transaction data sample includes the customer's credit score, repayment behavior, and account transaction history. The true default risk label includes the objective fact that the customer has defaulted within a historical period. The first participant uses the first transaction data sample and the true default risk label to construct a training sample set.

[0120] During the forward propagation phase of training, the first participant inputs the first transaction data sample into the first feature extraction layer deployed locally. The first feature extraction layer performs feature transformation on the input data and outputs the first feature training vector.

[0121] For example, feature processing of transaction data samples can be expressed by the following formula:

[0122]

[0123] in, Indicates the first Feature training vectors of the participants This represents the activation function. Indicates the first The weight matrix of the feature extraction layer of the participating parties. Indicates the first Sample transaction data from the participants. Indicates the first Bias terms of the feature extraction layer of the participating parties.

[0124] The first participant receives the second feature training vector sent by the second participant. The second feature training vector is generated by the second participant based on the second transaction data sample corresponding to the common identifier information set (e.g., financial data including investment preferences, holdings, etc.), and after feature transformation by inputting it into the second feature extraction layer deployed on the second participant.

[0125] refer to Figure 5 , Figure 5 The schematic diagram illustrates the structure of the interactive fusion layer according to an embodiment of this application. The first feature training vector and the second feature training vector are input into fully connected layer A and fully connected layer B, respectively, followed by summation, activation function processing, and regularization, and finally output.

[0126] The first participant inputs the received second feature training vector along with its locally generated first feature training vector into the locally deployed interactive fusion layer. The interactive fusion layer performs a fusion operation on the two feature vectors to generate fused training features.

[0127] For example, fusing the feature training vectors of multiple participants can be expressed by the following formula:

[0128]

[0129] in, Indicates the first Feature training vectors of the participants Indicates the number of participants. This represents the aggregation result. The aggregation result is then... Perform activation and regularization, and output fused training features. .

[0130] The first participant inputs the fused training features into a locally deployed classification prediction layer. The classification prediction layer calculates the fused training features and outputs a predicted default risk label.

[0131] For example, prediction based on fused training features using the parameters of a classification prediction layer can be expressed by the following formula:

[0132]

[0133] in, Indicates a label representing the predicted risk of default. Indicates the characteristics of fusion training. The parameters represent the classification prediction layer. This represents the mapping relationship / prediction function of the classification prediction layer.

[0134] During the error calculation phase, the first participant compares the predicted default risk labels output by the classification prediction layer with the actual default risk labels in the training sample set. Using a pre-defined loss function, the first participant calculates the difference between the predicted default risk labels and the actual default risk labels, determining the prediction loss of the longitudinal federated learning model. This prediction loss guides subsequent updates to the model parameters.

[0135] Through the embodiments of this application, by combining multi-party samples with real labels, training the model using a hierarchical feature extraction and fusion mechanism, and optimizing parameters based on prediction loss, joint training of cross-institutional data is achieved, while ensuring data privacy and security.

[0136] In some embodiments, after determining the prediction loss of the longitudinal federated learning model using a preset loss function based on the predicted default risk label and the actual default risk label, the method further includes: determining the gradient of the prediction loss relative to the parameters of each level in the longitudinal federated learning model; adding random noise conforming to a Laplace distribution to the gradient of each level parameter to obtain a noisy gradient set; and updating the parameters of each level in the longitudinal federated learning model based on the noisy gradient set to obtain a trained longitudinal federated learning model.

[0137] In this embodiment, the first participant calculates the rate of change of each level parameter in the longitudinal federated learning model relative to the error, i.e., the gradient of each level parameter, based on the prediction loss. This gradient reflects the direction and magnitude of adjustment required for each level parameter to reduce the prediction loss.

[0138] After obtaining the gradients of the parameters at each level, random noise conforming to a Laplace distribution is generated according to a predetermined privacy budget parameter. The first participant then adds this random noise to the gradients of each level parameter, perturbing the original gradients to obtain a set of noisy gradients. This process ensures that subtle features of individual samples are not leaked during gradient updates.

[0139] To prevent back-inference of the original data through model gradients, this application introduces a differential privacy mechanism during parameter updates.

[0140] The first participant (initiator) calculates the predicted loss. The gradient of the parameters of each layer is used to predict the parameters of the classification layer. For example, the gradient of the classification prediction layer is Inject noise conforming to a Laplace distribution into the parameter update formula:

[0141]

[0142] in, This represents the parameters of the noisy classification prediction layer. The parameters represent the classification prediction layer. This represents the gradient of the classification prediction layer. This indicates that it conforms to a Laplace distribution ( Random noise, Indicates the learning rate. ( (To control noise intensity for privacy budget parameters).

[0143] In addition to using the Laplace distribution to generate random noise, the first participant can also choose other types of statistical distributions to generate random noise, such as exponential distribution or uniform distribution, in order to protect against gradient perturbations.

[0144] Subsequently, the first participant updates the parameters of each layer in the longitudinal federated learning model using a noisy gradient set. Based on the adjustment direction and magnitude indicated by the noisy gradient, the first participant performs numerical updates on all locally stored model parameters. Simultaneously, the first participant transmits the noisy gradient information to the second participant, which then synchronously updates its own second feature extraction layer parameters based on this gradient information.

[0145] Through the embodiments of this application, the vertical federated learning model maintains the privacy and security of the data of the participating parties while completing the training.

[0146] In some embodiments, updating the parameters of each layer in the longitudinal federated learning model based on a noisy gradient set to obtain a trained longitudinal federated learning model includes: determining a first noise gradient corresponding to a first feature extraction layer and a second noise gradient corresponding to a classification prediction layer in the noisy gradient set; updating the first feature extraction layer based on the first noise gradient and updating the classification prediction layer based on the second noise gradient; sending the first noise gradient to a second participant, and the second participant updating its second feature extraction layer based on the received first noise gradient.

[0147] In this embodiment of the application, the first participant extracts the gradient associated with the first feature extraction layer deployed locally, which is denoted as the first noise gradient; at the same time, it extracts the gradient associated with the classification prediction layer deployed locally, which is denoted as the second noise gradient.

[0148] During the local update phase, the first participant adjusts the weight and bias parameters in the first feature extraction layer using a first noise gradient, and adjusts the weight and bias parameters in the classification prediction layer using a second noise gradient. Through these operations, the first participant completes the parameter optimization of the local model structure.

[0149] During the cross-party update phase, the first participant sends the first noisy gradient to the second participant via a network communication link. Upon receiving this gradient information, the second participant uses it to synchronously adjust the weight and bias parameters in the locally deployed second feature extraction layer. Thus, the first and second participants collaboratively complete the full training of the longitudinal federated learning model while protecting data privacy.

[0150] In this embodiment, the first participant updates the classification prediction layer and the locally stored first feature extraction layer of the vertical federated model, and transmits the noisy gradient information to the collaborator (the second participant). The collaborator completes the update of the second feature extraction layer locally without exposing the original data.

[0151] The beneficial effects of the technical solution in this application are mainly reflected in the following three aspects:

[0152] First, it breaks down the data silos between banks and wealth management companies, achieving deep integration of credit and investment data without the need to share raw data. This significantly improves the accuracy and comprehensiveness of customer risk assessment models and solves the problem of the two parties being unable to cooperate due to privacy concerns.

[0153] Secondly, it provides a strict dual privacy protection mechanism. RSA encryption and hash obfuscation techniques ensure that the identity information of non-intersecting customers is not leaked; differential privacy technology is introduced into the training of federated learning models, and by adding noise to the gradient parameters, it effectively prevents attackers from inferring the original data through model updates, thus ensuring data compliance and security.

[0154] Third, it achieves a good balance between privacy protection and model utility. While providing strong privacy protection, the model performance degradation does not exceed 3%, which can meet the dual requirements of risk assessment accuracy and real-time performance in real-world business scenarios.

[0155] Based on the above-described identification information processing, embodiments of this application also provide an identification information processing apparatus. The following will be combined with... Figure 6 The device is described in detail.

[0156] Figure 6 A schematic block diagram of an identification information processing apparatus according to an embodiment of this application is shown.

[0157] like Figure 6 As shown, the identification information processing device 600 in this embodiment includes a public key receiving module 610, a random encryption module 620, a set receiving module 630, a random elimination module 640, and an identification determination module 650.

[0158] The public key receiving module 610 is used to perform hash processing on each identifier in the first identifier information set stored locally in response to receiving the asymmetric encryption public key sent by the second participant, so as to obtain the first hash identifier set. In one embodiment, the public key receiving module 610 can be used to perform the step S210 described above, which will not be repeated here.

[0159] The random encryption module 620 is used to generate and save a random number for each hash identifier in the first hash identifier set, encrypt each hash identifier in the first hash identifier set based on the asymmetric encryption public key and the random number of each hash identifier, obtain an encrypted and confused identifier set, and send the encrypted and confused identifier set to the second participant. In one embodiment, the random encryption module 620 can be used to perform the step S220 described above, which will not be repeated here.

[0160] The set receiving module 630 is used to receive the decryption identifier set and the second hash identifier set returned by the second participant. The decryption identifier set is obtained by the second participant decrypting the encrypted obfuscated identifier set based on the private key corresponding to the asymmetric encryption public key. The second hash identifier set is obtained by the second participant hashing the second identifier information set stored locally. In one embodiment, the set receiving module 630 can be used to execute the step S230 described above, which will not be repeated here.

[0161] The random elimination module 640 is used to perform random number elimination on the corresponding decryption identifier in the decryption identifier set based on the random number of each hash identifier to obtain the identifier set to be compared; in one embodiment, the random elimination module 640 can be used to perform the step S640 described above, which will not be repeated here.

[0162] The identifier determination module 650 is used to compare the identifier set to be compared with the second hash identifier set to determine the common identifier information set. The common identifier information set represents the set of common identifier information that exists in the first identifier information set and the second identifier information set. In one embodiment, the identifier determination module 650 can be used to execute the step S650 described above, which will not be repeated here.

[0163] According to embodiments of this application, any multiple modules among the public key receiving module 610, random encryption module 620, set receiving module 630, random elimination module 640, and identifier determination module 650 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the public key receiving module 610, random encryption module 620, set receiving module 630, random elimination module 640, and identifier determination module 650 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array, programmable logic array, system-on-a-chip, system-on-a-substrate, system-on-package, application-specific integrated circuit, or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the public key receiving module 610, random encryption module 620, set receiving module 630, random elimination module 640, and identifier determination module 650 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0164] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing identification information processing according to an embodiment of this application.

[0165] like Figure 7As shown, an electronic device 700 according to an embodiment of this application includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory 702 or a program loaded from a storage portion 708 into a random access memory 703. The processor 701 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0166] Random access memory 703 stores various programs and data required for the operation of electronic device 700. Processor 701, read-only memory 702, and random access memory 703 are interconnected via bus 704. Processor 701 executes various steps of the method flow according to embodiments of this application by executing programs stored in read-only memory 702 and / or random access memory 703. It should be noted that programs may also be stored in one or more memories other than read-only memory 702 and random access memory 703. Processor 701 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in one or more memories.

[0167] According to embodiments of this application, the electronic device 700 may further include an input / output interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the input / output interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube, liquid crystal display, etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card, such as a local area network card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0168] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0169] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include the read-only memory 702, and / or random access memory 703, and / or one or more memories other than read-only memory 702 and random access memory 703 described above.

[0170] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0171] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0172] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by processor 701, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0173] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0174] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for processing identification information, characterized in that, Applied to the first participant, the method includes: In response to receiving the asymmetric encryption public key sent by the second participant, each identifier in the first identifier information set stored locally is hashed to obtain the first hash identifier set; A random number is generated and saved for each hash identifier in the first hash identifier set. Based on the asymmetric encryption public key and the random number of each hash identifier, each hash identifier in the first hash identifier set is encrypted to obtain an encrypted obfuscated identifier set, and the encrypted obfuscated identifier set is sent to the second participant. The system receives a decryption identifier set and a second hash identifier set returned by the second participant. The decryption identifier set is obtained by the second participant decrypting the encryption obfuscation identifier set based on the private key corresponding to the asymmetric encryption public key. The second hash identifier set is obtained by the second participant performing hash processing on the second identifier information set stored locally. Based on the random number of each hash identifier, the corresponding decryption identifier in the decryption identifier set is randomly eliminated to obtain the identifier set to be compared; The set of identifiers to be compared is compared with the second set of hash identifiers to determine a set of common identifier information. The set of common identifier information represents the set of common identifier information that exists in the first set of identifier information and the second set of identifier information.

2. The identification information processing method according to claim 1, characterized in that, The step involves encrypting each hash identifier in the first hash identifier set based on the asymmetric encryption public key and the random number for each hash identifier, to obtain an encrypted obfuscated identifier set, including: Based on the modulus and exponent of the asymmetric encryption public key, encryption operations are performed on each hash identifier and its random number to generate an encryption obfuscation identifier corresponding to each hash identifier; A set of cryptographic obfuscation identifiers is constructed based on the cryptographic obfuscation identifiers corresponding to each hash identifier.

3. The identification information processing method according to claim 1, characterized in that, The random number elimination of the corresponding decryption identifiers in the decryption identifier set based on the random number of each hash identifier, to obtain the identifier set to be compared, includes: Determine the random number associated with each hash identifier in the first hash identifier set for each decryption identifier in the decryption identifier set; Perform an inverse operation on the decryption identifier based on the random number to obtain the comparison identifier corresponding to each decryption identifier; Based on the comparison identifier corresponding to each decryption identifier, a set of comparison identifiers is constructed.

4. The identification information processing method according to claim 1, characterized in that, After comparing the set of identifiers to be compared with the second set of hash identifiers to determine the set of common identifier information, the method further includes: Determine the first transaction data of the entity to which the identification information in the common identification information set belongs; The first transaction data is input into a trained vertical federated learning model to predict default, and the default risk of the subject corresponding to the identifier information in the common identifier information set output by the vertical federated learning model is obtained.

5. The identification information processing method according to claim 4, characterized in that, The vertical federated learning model includes a first feature extraction layer located in the first participant, a second feature extraction layer located in the second participant, an interaction fusion layer located in the first participant, and a classification prediction layer located in the first participant. The step of inputting the first transaction data into a trained longitudinal federated learning model for default prediction, and obtaining the default prediction result output by the longitudinal federated learning model, includes: The first transaction data is input into the first feature extraction layer for feature transformation to obtain the first feature vector; The second feature vector sent by the second participant is received. The second feature vector is obtained by the second participant inputting the second transaction data determined by the common identification information set into the second feature extraction layer for feature transformation. The first feature vector and the second feature vector are input into the interactive fusion layer for feature fusion to obtain fused features; The fused features are input into the classification prediction layer for classification prediction to obtain the default risk.

6. The identification information processing method according to claim 4, characterized in that, The vertical federated learning model includes a first feature extraction layer, a second feature extraction layer, an interaction fusion layer, and a classification prediction layer; The longitudinal federated learning model is trained in the following manner: Based on the common identifier information set, a training sample set is constructed, which includes: a first transaction data sample and the corresponding real default risk label; The first transaction data sample is input into the first feature extraction layer for feature transformation to obtain the first feature training vector; The second feature training vector sent by the second participant is received. The second feature training vector is obtained by the second participant inputting the second transaction data sample determined by the common identification information set into the second feature extraction layer for feature transformation. The first feature training vector and the second feature training vector are input into the interactive fusion layer for feature fusion to obtain fused training features; The fused training features are input into the classification prediction layer for classification prediction to obtain the predicted default risk label; Based on the predicted default risk label and the actual default risk label, the prediction loss of the longitudinal federated learning model is determined using a preset loss function.

7. The identification information processing method according to claim 6, characterized in that, After determining the prediction loss of the longitudinal federated learning model using a preset loss function based on the predicted default risk label and the actual default risk label, the method further includes: Determine the gradient of the prediction loss with respect to the parameters of each layer in the longitudinal federated learning model; Random noise conforming to a Laplace distribution is added to the gradients of the parameters at each level to obtain a set of noisy gradients. The parameters of each level in the longitudinal federated learning model are updated based on the noisy gradient set to obtain the trained longitudinal federated learning model.

8. The identification information processing method according to claim 7, characterized in that, The step of updating the parameters of each layer in the longitudinal federated learning model based on a noisy gradient set to obtain a trained longitudinal federated learning model includes: Determine the first noise gradient corresponding to the first feature extraction layer and the second noise gradient corresponding to the classification prediction layer in the noisy gradient group; The first feature extraction layer is updated based on the first noise gradient, and the classification prediction layer is updated based on the second noise gradient. The first noise gradient is sent to the second participant, and the second participant updates its second feature extraction layer based on the received first noise gradient.

9. An identification information processing device, characterized in that, include: The public key receiving module is used to perform hash processing on each identifier in the first identifier information set stored locally in response to receiving the asymmetric encryption public key sent by the second participant, so as to obtain the first hash identifier set. The random encryption module is used to generate and save a random number for each hash identifier in the first hash identifier set, encrypt each hash identifier in the first hash identifier set based on the asymmetric encryption public key and the random number of each hash identifier, obtain an encrypted and obfuscated identifier set, and send the encrypted and obfuscated identifier set to the second participant. The set receiving module is used to receive the decryption identifier set and the second hash identifier set returned by the second participant. The decryption identifier set is obtained by the second participant decrypting the encryption and obfuscation identifier set based on the private key corresponding to the asymmetric encryption public key. The second hash identifier set is obtained by the second participant performing hash processing on the second identifier information set stored locally. The random elimination module is used to perform random number elimination on the corresponding decryption identifiers in the decryption identifier set based on the random number of each hash identifier, so as to obtain the identifier set to be compared; The identifier determination module is used to compare the set of identifiers to be compared with the second hash identifier set to determine a common identifier information set, wherein the common identifier information set represents the set of common identifier information that exists in the first identifier information set and the second identifier information set.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the identification information processing method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they implement the steps of the identification information processing method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the steps of the identification information processing method according to any one of claims 1 to 8.

Images