Product recommendation method and device based on cross-industry data fusion, medium and equipment

Abstract

The invention discloses a product recommendation method and device based on cross-industry data fusion, a storage medium, electronic equipment and a product. Determining a product recommendation model corresponding to the product recommendation request, a financial encryption identifier of a target user and an encrypted financial feature corresponding to the target user; requesting an auxiliary participant associated with the product recommendation model to feed back an encryption auxiliary feature corresponding to the target user based on the financial encryption identifier of the target user; determining a recommendation prediction result through the product recommendation model based on the encrypted financial features and the encrypted auxiliary features corresponding to the target user, and feeding back the recommendation prediction result to the financial participant; the product recommendation model is obtained by jointly training the encrypted financial features of the financial participants and the auxiliary industry features of the auxiliary participants. By executing the method, the product recommendation accuracy and the user experience can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, privacy computing and product recommendation, and can be applied to the financial technology scene. The present application specifically relates to a product recommendation method and device based on cross-industry data fusion, a storage medium, an electronic device and a product. BACKGROUND

[0002] With the advancement of the digital wave, data as a key production factor plays an important role in various fields. Cross-industry data fusion product recommendation is a technology that, under the premise of strictly protecting data privacy, breaks down industry data barriers, safely fuses and jointly analyzes user feature data from multiple industries, and thus provides highly precise and personalized product or service recommendations that cannot be discovered from a single industry perspective.

[0003] However, cross-industry data fusion still has significant bottlenecks. On the one hand, data silos between different fields lead to one-sided user portraits, making it difficult to support accurate recommendations. On the other hand, traditional data sharing methods such as plaintext transmission or static desensitization not only cannot resist privacy attacks such as association reasoning, but also seriously damage the usability and granularity of data. Although distributed technologies such as federated learning have emerged, there are still problems such as data semantic barriers and gradient leakage risks. SUMMARY

[0004] The present application provides a product recommendation method and device based on cross-industry data fusion, a storage medium, an electronic device and a product, which can achieve the purpose of improving the accuracy of product recommendation and user experience.

[0005] According to a first aspect of the present application, a product recommendation method based on cross-industry data fusion is provided, which is executed by a central coordinator, and the method comprises:

[0006] According to a product recommendation request of a financial participant for a target user, determining a product recommendation model corresponding to the product recommendation request, a financial encrypted identifier of the target user and encrypted financial features corresponding to the target user;

[0007] Based on the financial encrypted identifier of the target user, requesting the product recommendation model associated auxiliary participants to feedback encrypted auxiliary features corresponding to the target user;

[0008] Through the product recommendation model, based on the encrypted financial features and encrypted auxiliary features corresponding to the target user, determining a recommendation prediction result and feeding back the recommendation prediction result to the financial participant;

[0009] The product recommendation model is obtained by jointly training the encrypted financial features of the financial participant and the auxiliary industry features of the auxiliary participant.

[0010] According to a second aspect of this application, a product recommendation device based on cross-industry data fusion is provided, configured in a central coordinator, the device comprising:

[0011] The request information determination module is used to determine the product recommendation model corresponding to the product recommendation request, the financial encryption identifier of the target user, and the encrypted financial characteristics corresponding to the target user based on the financial participation policy for the product recommendation request of the target user;

[0012] The auxiliary feature request module is used to request the auxiliary participants associated with the product recommendation model to provide feedback on the cryptographic auxiliary features corresponding to the target user based on the financial cryptographic identifier of the target user.

[0013] The prediction result determination module is used to determine the recommendation prediction result based on the crypto-financial characteristics and crypto-aided characteristics corresponding to the target user through the product recommendation model, and to feed the recommendation prediction result back to the financial participant;

[0014] The product recommendation model is obtained by jointly training the crypto-financial characteristics of the financial participants and the auxiliary industry characteristics of the auxiliary participants.

[0015] According to a third aspect of the present invention, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the product recommendation method based on cross-industry data fusion as described in embodiments of this application.

[0016] According to a fourth aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the product recommendation method based on cross-industry data fusion as described in the embodiments of the present application.

[0017] According to a fifth aspect of this application, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the product recommendation method based on cross-industry data fusion as described in embodiments of this application.

[0018] This application's technical solution constructs a cross-industry collaborative privacy-preserving recommendation mechanism. When a financial participant initiates a product recommendation request, it can accurately trigger a product recommendation model associated with it, which has been jointly trained with cross-industry data. Based on the target user's encrypted financial identifier, it securely obtains the encrypted auxiliary features from the auxiliary participant. This ensures that, without the original data leaving the local environments of the participating industries, it achieves deep fusion and joint computation of financial industry features and auxiliary industry features in an encrypted state. This means that the final recommendation prediction results not only rely on the financial participant's own user profile but also incorporate user features provided by the auxiliary participant, thereby significantly improving the accuracy and personalization of the recommendation results. Simultaneously, all sensitive data is circulated and processed in encrypted form throughout the entire process, effectively preventing the risk of user privacy information leakage and constructing a robust data security barrier. This allows for more efficient, accurate, and personalized recommendation results while complying with data compliance requirements, significantly improving user experience and business conversion efficiency while ensuring data security.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a product recommendation method based on cross-industry data fusion provided in Embodiment 1;

[0022] Figure 2 This is a flowchart of a product recommendation method based on cross-industry data fusion provided in Embodiment 2;

[0023] Figure 3 This is a schematic diagram of the product recommendation device based on cross-industry data fusion provided in Embodiment 3 of this application;

[0024] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," "target," and "candidate," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Example 1

[0028] Figure 1 This is a flowchart of a product recommendation method based on cross-industry data fusion provided in Embodiment 1. This embodiment can be applied to situations where product recommendations are made to users based on cross-industry data. The method is executed by a central coordinator, which is equipped with a product recommendation device based on cross-industry data fusion. The product recommendation device based on cross-industry data fusion is implemented in hardware and / or software.

[0029] like Figure 1 As shown, the method includes:

[0030] S110. Based on the financial participation policy for the product recommendation request of the target user, determine the product recommendation model corresponding to the product recommendation request, the financial encryption identifier of the target user, and the encrypted financial characteristics corresponding to the target user.

[0031] S120. Based on the financial encryption identifier of the target user, request the auxiliary participant associated with the product recommendation model to provide feedback on the encryption auxiliary features corresponding to the target user.

[0032] S130. Based on the crypto-financial characteristics and crypto-aided characteristics corresponding to the target user, the product recommendation model determines the recommendation prediction result and feeds the recommendation prediction result back to the financial participant.

[0033] A product recommendation request is an instruction or query generated by a financial participant to trigger cross-industry collaborative computing to generate personalized product recommendations for a target user. The target user is the specific customer being recommended. Optionally, the product recommendation request includes the target user's encrypted financial identifier, the recommendation model identifier, and the target user's corresponding encrypted financial characteristics.

[0034] The central coordinator is a neutral computing entity independent of both financial and auxiliary participants. It coordinates the multi-party computation process without directly holding any participant's raw feature data. The central coordinator parses product recommendation requests to obtain a recommendation model identifier, the target user's encrypted financial identifier, and the target user's corresponding encrypted financial features. The target user's encrypted financial identifier is obtained by encrypting the target user's local identity identifier by the financial participants. The recommendation model identifier is used by the central coordinator to determine the product recommendation model corresponding to the product recommendation request. The target user's encrypted financial features are obtained by encrypting the target user's local feature data using a homomorphic encryption algorithm, resulting in ciphertext. Homomorphic encryption is a special encryption method that allows direct arithmetic operations to be performed on the ciphertext.

[0035] The central coordinator determines the product recommendation model corresponding to a product recommendation request based on the recommendation model identifier. This model is jointly trained using encrypted financial features from financial participants and auxiliary industry features from auxiliary participants. The product recommendation model is jointly trained in a homomorphic encryption environment and can simultaneously understand both financial and auxiliary industry features. Joint training of the product recommendation model in a homomorphic encryption environment protects the privacy and security of local feature data and prevents information leakage. Joint training refers to a distributed machine learning paradigm where multiple industry participants, without needing to share local feature data, jointly train a unified model by exchanging encrypted model parameter updates, such as encrypted gradients. This significantly improves the model's recognition and generalization ability of cross-industry features while completely avoiding the out-of-domain transmission of local feature data. It overcomes the problem of one-sided user profiles caused by data silos and ensures data privacy.

[0036] Both financial participants and auxiliary participants are industry stakeholders involved in the joint training of the model. Financial participants are institutional entities within the financial industry that hold financial industry data of the target user. Auxiliary participants are institutional entities that establish a collaborative relationship with financial participants through privacy computing protocols, and they hold auxiliary industry data of the target user. Auxiliary participants belong to industries other than finance. Encrypted auxiliary features refer to the ciphertext of the target user's local feature data after the auxiliary participant has encrypted it using a homomorphic encryption algorithm. The local feature data is related to the industry type of the participating entity and is structured data extracted from specific industry data that describes the target user's behavior or attributes within that industry. For example, the financial participant's local feature data is structured data extracted from financial industry data that describes the target user's behavior or attributes within the financial industry.

[0037] The central coordinator performs cross-industry user matching based on the target user's encrypted financial identifier. This helps participating parties determine the target user's industry-specific encrypted identifier, and then requests feedback from these parties on the corresponding encrypted auxiliary features. Both encrypted financial features and encrypted auxiliary features serve as the data basis for the product recommendation model's predictions. Both encrypted financial features and encrypted auxiliary features can be used in calculations in encrypted form.

[0038] The central coordinator inputs the target user's cryptographic financial characteristics and cryptographic auxiliary characteristics into the product recommendation model, which then determines the recommendation prediction result in encrypted form. This prediction result is a single fragment corresponding to the product recommendation request. The central coordinator sends this prediction result as feedback to the financial participants.

[0039] This application's technical solution constructs a cross-industry collaborative privacy-preserving recommendation mechanism. When a financial participant initiates a product recommendation request, it can accurately trigger a product recommendation model associated with it, which has been jointly trained with cross-industry data. Based on the target user's encrypted financial identifier, it securely obtains the encrypted auxiliary features from the auxiliary participant. This ensures that, without the original data leaving the local environments of the participating industries, it achieves deep fusion and joint computation of financial industry features and auxiliary industry features in an encrypted state. This means that the final recommendation prediction results not only rely on the financial participant's own user profile but also incorporate user features provided by the auxiliary participant, thereby significantly improving the accuracy and personalization of the recommendation results. Simultaneously, all sensitive data is circulated and processed in encrypted form throughout the entire process, effectively preventing the risk of user privacy information leakage and constructing a robust data security barrier. This allows for more efficient, accurate, and personalized recommendation results while complying with data compliance requirements, significantly improving user experience and business conversion efficiency.

[0040] In an optional embodiment, the step of requesting the auxiliary participant associated with the product recommendation model to provide feedback on the encrypted auxiliary features corresponding to the target user based on the target user's financial encrypted identifier includes: determining the target user's industry encrypted identifier for the auxiliary participant based on the encrypted identifier mapping relationship and the target user's financial encrypted identifier; wherein, the encrypted identifier mapping relationship is calculated by the central coordinator performing a global private set intersection protocol with the financial participant and the auxiliary participant; generating an industry feature request based on the target user's industry encrypted identifier and sending the industry feature request to the corresponding auxiliary participant; obtaining the encrypted auxiliary features by the auxiliary participant encrypting the target user's local feature data using a homomorphic encryption algorithm locally, and feeding the encrypted auxiliary features back to the central coordinator.

[0041] The encrypted identifier mapping relationship is calculated by the central coordinator in conjunction with financial and auxiliary participants using a global private set intersection protocol. This protocol is a cryptographic method that securely computes a shared set of users without exposing the original identities of multiple participants. Based on this encrypted identifier mapping relationship, the central coordinator can securely associate the encrypted identifiers of the same user across different participants, thus helping auxiliary participants determine the industry-specific encrypted identifiers of target users.

[0042] Financial encrypted identifiers are obtained by encrypting the target user's local identity identifier by financial participants. Industry encrypted identifiers refer to the encrypted identity identifier of the same target user within the assisting participants.

[0043] The industry feature request refers to the encrypted feature acquisition instruction issued by the central coordinator to the corresponding auxiliary participant based on a determined industry encryption identifier. The auxiliary participant holds auxiliary industry data of the target user. After receiving the industry feature request, the auxiliary participant determines the target user's local feature data and encrypts it locally using a homomorphic encryption algorithm to obtain encrypted auxiliary features. Then, it sends the encrypted auxiliary features back to the central coordinator.

[0044] The aforementioned technical solution, by pre-establishing an encrypted identifier mapping relationship based on a global private set intersection protocol, enables the central coordinator to accurately and securely determine the corresponding industry encrypted identifier of the financial participant based on the financial encrypted identifier provided by the financial participant. This ensures that user identity information remains encrypted throughout the cross-system transfer process, effectively preventing the risk of leakage of original identity information. Through industry feature requests generated based on industry encrypted identifiers, the auxiliary participant only performs homomorphic encryption on the local feature data of the specified user locally, generating encrypted auxiliary features and feeding them back to the central coordinator. This not only protects the data sovereignty of the auxiliary participant but also ensures the confidentiality of the feature data during transmission and subsequent calculations.

[0045] In an optional embodiment, before determining the industry encrypted identifier of the target user for the auxiliary participant based on the encrypted identifier mapping relationship and the target user's financial encrypted identifier, the method further includes: blinding the local user identifier through the financial participant and the auxiliary participant respectively, and feeding back the obtained financial blinded identifier and auxiliary blinded identifier; digitally signing the financial blinded identifier and the auxiliary blinded identifier using an identifier encryption private key, and feeding back the identifier signature results of the financial blinded identifier and the auxiliary blinded identifier to the financial participant and the auxiliary participant respectively; and digitally signing the identifier signature results of the financial blinded identifier and the auxiliary blinded identifier using an identifier encryption public key through the financial participant and the auxiliary participant respectively. The signature result of the assisted blinding identifier is verified. If the verification passes, the financial participant and the auxiliary participant respectively perform deblinding processing on the signature results of the financial blinding identifier and the signature result of the assisted blinding identifier to obtain the financial signature identifier and the auxiliary signature identifier, and determine the common user identifier based on the financial signature identifier and the auxiliary signature identifier. The financial participant and the auxiliary participant generate a global user identifier for the common user identifier and construct an identifier mapping relationship between the common user identifier and the global user identifier. The identifier mapping relationship fed back by the financial participant and the auxiliary participant is subjected to a set intersection operation, and the obtained intersection operation result is encrypted using a homomorphic encryption algorithm to obtain the encrypted identifier mapping relationship.

[0046] Among them, the local user identifier refers to the original identity information used to uniquely identify users within each participating party. Blinding the local user identifier refers to obfuscating it by introducing a random factor. The financial blinding identifier refers to the irreversible obfuscated identifier obtained after the financial participants blinded the local user identifier. The auxiliary blinding identifier refers to the irreversible obfuscated identifier obtained after the auxiliary participants blinded the local user identifier.

[0047] Financial participants and auxiliary participants respectively feed back the financial blinding identifier and the auxiliary blinding identifier to the central coordinator. The central coordinator uses the identifier encryption private key to digitally sign the financial blinding identifier and the auxiliary blinding identifier to prove their authenticity and integrity. The identifier encryption private key refers to the asymmetric encryption private key generated and stored by the central coordinator for digital signatures. The identifier signature result of the financial blinding identifier refers to the cryptographic proof obtained after digitally signing the financial blinding identifier; similarly, the identifier signature result of the auxiliary blinding identifier refers to the cryptographic proof obtained after digitally signing the auxiliary blinding identifier.

[0048] The central coordinator sends the signature result of the financial blinded identifier to the financial participants and the signature result of the auxiliary blinded identifier to the auxiliary participants. The financial participants verify the signature result of the financial blinded identifier using their identifier encryption public key to confirm the authenticity of the digital signature, obtaining a verification result. Similarly, the auxiliary participants verify the signature result of the auxiliary blinded identifier using their identifier encryption public key to confirm the authenticity of the digital signature, obtaining a verification result. If both the financial and auxiliary participants' verification results are successful, the financial participants perform deblinding processing on the signature result of the financial blinded identifier to remove the blinding factor, obtaining the financial signature identifier. The auxiliary participants perform deblinding processing on the signature result of the auxiliary blinded identifier to remove the blinding factor, obtaining the auxiliary signature identifier.

[0049] Among them, the financial signature identifier refers to the user identifier with the coordinator's signature obtained after the financial participants deblind the signature result. The auxiliary signature identifier refers to the user identifier with the coordinator's signature obtained after the auxiliary participants deblind the signature result. The shared user identifier refers to the user identifier jointly owned, determined by comparing the financial signature identifier and the auxiliary signature identifier. The global user identifier refers to the unified identifier generated for the shared user identifier. The identifier mapping relationship refers to the correspondence between the shared user identifier and the global user identifier.

[0050] Financial participants and auxiliary participants each send their constructed identifier mapping relationships to the central coordinator. The central coordinator performs a set intersection operation on the identifier mapping relationships, obtaining the intersection result. Then, the central coordinator uses a homomorphic encryption algorithm to encrypt the intersection result, obtaining an encrypted identifier mapping relationship. The set intersection operation refers to the process of calculating the intersection of the identifier sets submitted by different participants.

[0051] The aforementioned technical solution involves financial and auxiliary participants first blinding local user identifiers to ensure the original identifiers are not exposed. A central coordinator then digitally signs the blinded identifiers to certify data authenticity. After verification by all parties, a deblinding process is performed to obtain verifiable signature identifiers, thereby identifying shared users and generating a unified global user identifier. Finally, encrypted identifier mapping relationships are obtained through set operations and homomorphic encryption. While protecting the privacy of the original data throughout the process, a secure identity mapping foundation for cross-industry data collaboration is established. This not only solves the privacy leakage risk in user identifier matching but also provides necessary support for subsequent joint computation, achieving an effective balance between data value mining and privacy protection.

[0052] In an optional embodiment, the step of determining a recommendation prediction result based on the encrypted financial features and encrypted auxiliary features corresponding to the target user through the product recommendation model and feeding the recommendation prediction result back to the financial participant includes: concatenating the encrypted financial features and encrypted auxiliary features corresponding to the target user to obtain the encrypted joint features corresponding to the target user, and inputting the encrypted joint features into the product recommendation model; outputting a recommendation prediction vector based on the encrypted joint features through the product recommendation model, and sending the recommendation prediction vector to the financial participant and the auxiliary participant; partially decrypting the recommendation prediction vector through the financial participant and the auxiliary participant, and feeding back the partial decryption result of the recommendation prediction vector; combining the partial decryption results fed back by the financial participant and the auxiliary participant to obtain the prediction vector plaintext; decomposing the prediction vector plaintext to obtain a single result fragment corresponding to the financial participant, and feeding back the single result fragment as the recommendation prediction result to the financial participant; wherein, the prediction vector plaintext is cleared from memory after the recommendation prediction result is fed back to the financial participant.

[0053] Here, the encrypted joint feature refers to a joint feature vector formed by concatenating encrypted financial features and encrypted auxiliary features in ciphertext. The central coordinator inputs the encrypted joint feature into the product recommendation model, which then calculates and outputs a recommendation prediction vector based on the encrypted joint feature. This recommendation prediction vector includes the recommendation results for multiple participating parties.

[0054] The central coordinator sends the recommended prediction vector to the financial and auxiliary participants. These participants then partially decrypt the vector, obtaining a partially decrypted result. Partial decryption refers to the partial decryption operation performed by each industry participant using their respective private key shares. The partially decrypted result is an intermediate result obtained after the participants have partially decrypted the vector, requiring cooperation from other parties to recover the plaintext. The private key shares used by each industry participant are cryptographic shares collaboratively generated and distributed during the system initialization phase through a one-time distributed key generation protocol.

[0055] The central coordinator combines the partial decryption results from financial and auxiliary participants to obtain the predicted plaintext vector. This combination process refers to the use of cryptographic algorithms to synthesize the partial decryption results from multiple participants into the complete plaintext. The predicted plaintext vector is the plaintext vector obtained through this combination process, containing the recommendations from all participants.

[0056] The central coordinator decomposes the predicted plaintext vector into single result fragments corresponding to financial participants, and feeds these single result fragments back to the financial participants as recommended prediction results. Decomposition refers to the operation of separating the predicted plaintext vector according to business affiliation. A single result fragment refers to the recommended result separated from the predicted plaintext vector that is only relevant to the specific participant's business. The recommended prediction result is the single result fragment ultimately fed back to the financial participant.

[0057] After the prediction vector plaintext is decomposed and only the single result fragment belonging to the financial participant, namely the recommended prediction result, is sent out, it is immediately and completely cleared from memory. This ensures that the prediction vector plaintext exists only for a very short time and serves only the specific purpose of result separation and distribution, and is irrecoverable afterward.

[0058] The aforementioned technical solution achieves deep fusion of cross-industry data in encrypted form by concatenating encrypted financial features and encrypted auxiliary features to form encrypted joint features. Next, a product recommendation model calculates a recommendation prediction vector in encrypted form, ensuring the privacy and security of the model's inference process. Subsequently, financial and auxiliary participants partially decrypt the recommendation prediction vector and provide feedback on the partial decryption results, establishing a weighted decryption mechanism to effectively prevent any single party from independently obtaining the complete recommendation result. Then, the prediction vector plaintext is obtained by combining the partial decryption results, completing the secure decryption process. Finally, the prediction vector plaintext is decomposed into single result fragments and fed back to the financial participants, while the prediction vector plaintext is cleared from memory, implementing the data minimization principle and an immediate destruction mechanism to ensure that all parties only obtain necessary business information and that complete and sensitive data is not persisted. While ensuring recommendation accuracy, a complete privacy protection closed loop is constructed, meeting both the business needs of cross-industry data collaboration and complying with stringent data protection regulations, achieving an effective unification of data value mining and privacy protection.

[0059] Example 2

[0060] Figure 2 This is a flowchart of a product recommendation method based on cross-industry data fusion, provided in Embodiment 2. This embodiment further optimizes the above embodiments. Specifically, it describes the steps used to train the product recommendation model through a central coordinator.

[0061] like Figure 2 As shown, the method includes:

[0062] S210. Initialize the global model parameters of the product recommendation model, and encrypt the initialized global model parameters using a homomorphic encryption algorithm to obtain encrypted global parameters.

[0063] In this context, global model parameters refer to the set of all learnable parameters of the product recommendation model, such as weights and biases, maintained by the central coordinator. These global model parameters represent the current knowledge state of the product recommendation model. Initialization refers to the process of setting initial values ​​for the learnable parameters of the product recommendation model before training begins; random initialization is typically used to provide a starting point for training. Encrypted global parameters refer to global model parameters processed using a homomorphic encryption algorithm, transforming their content into unreadable ciphertext to protect the confidentiality of model information during transmission. Homomorphic encryption algorithms allow direct mathematical operations on encrypted data without decryption.

[0064] S220. Distribute the encrypted global parameters to at least two industry participants participating in the training of the product recommendation model; wherein the industry participants include at least a financial participant and an auxiliary participant.

[0065] Industry participants refer to organizations that actually own and manage industry data. The specific number of auxiliary participants is not limited here and will be determined based on actual business needs.

[0066] S230. The industry participants use the encrypted global parameters and their respective local feature data to calculate the local model gradient for the product recommendation model, and then add noise and encrypt the local model gradient to obtain the encrypted local gradient.

[0067] Local model gradient refers to the direction and magnitude of model parameter updates calculated by industry participants during local training based on locally held local feature data. Local model gradient includes feature information about the distribution of locally held local feature data.

[0068] Encrypted local gradients refer to the results obtained by industry participants through adding noise and encryption to the local model gradients, in order to prevent information leakage during the process of returning the local model gradients to the central coordinator.

[0069] S240. Aggregate the encrypted local gradients obtained from the industry participants to obtain an encrypted aggregated gradient, and update the encrypted global parameters based on the encrypted aggregated gradient until the product recommendation model converges.

[0070] In this context, aggregation refers to the operation where the central coordinator receives encrypted local gradients from multiple industry participants and merges them according to specific rules to form a more general model update direction. The encrypted aggregated gradient refers to the final gradient result, which remains encrypted after the aggregation operation.

[0071] The central coordinator uses encrypted aggregated gradients to adjust and optimize encrypted global parameter values, thereby improving model performance. After multiple rounds of parameter updates, the performance improvement of the product recommendation model tends to stabilize, and the loss function value no longer decreases significantly. At this point, the product recommendation model can be considered to have converged, marking the completion of model training.

[0072] This application's technical solution combines federated learning with encryption technology to ensure that the local feature data of each industry participant remains locally, eliminating the need to upload or share it to a central coordinator and significantly reducing the risk of data leakage. During training, all transmitted model parameters and gradients are encrypted, effectively preventing man-in-the-middle attacks or malicious parties from inferring the original data through reverse engineering. Simultaneously, by aggregating encrypted local gradients from multiple industry participants to update the global model, the final trained product recommendation model can integrate knowledge from different domains, obtaining a more comprehensive and accurate user profile, thereby significantly improving recommendation performance. This distributed training paradigm not only strictly adheres to the data privacy regulations of various industries but also provides a practical technical path for breaking down data silos and achieving cross-industry data value integration.

[0073] In an optional embodiment, the step of adding noise and encrypting the local model gradient to obtain an encrypted local gradient includes: performing gradient clipping on the local model gradient based on a preset clipping threshold by the industry participant to obtain a gradient clipping result; generating a random noise vector based on a noise scale parameter and a noise distribution type; wherein the noise scale parameter is calculated according to a preset privacy budget and the preset clipping threshold; adding the random noise vector to the gradient clipping result to obtain a noisy local gradient, and encrypting the noisy local gradient using a homomorphic encryption algorithm to obtain an encrypted local gradient.

[0074] The preset clipping threshold refers to a pre-defined upper limit for constraining the gradient range. Gradient clipping refers to the operation of norm-constraining the local model gradient based on the preset clipping threshold. The gradient clipping result is the gradient vector after norm constraint processing. Gradient clipping is performed by industry stakeholders.

[0075] The noise scale parameter refers to the key numerical parameter that controls the intensity of the added noise. The noise scale parameter is determined based on the privacy budget and a preset pruning threshold. The privacy budget is a differential privacy parameter used to quantify the strength of privacy protection. When determining the noise scale parameter, the preset pruning threshold is used to define the upper limit of global sensitivity; essentially, the preset pruning threshold is a specific quantified value of global sensitivity.

[0076] Noise distribution type refers to the probability distribution form upon which random noise is generated. Random noise vector refers to a random numerical sequence generated based on noise distribution type and noise scale parameters. Noisy local gradient refers to the privacy-preserving gradient obtained after adding the random noise vector to the gradient clipping result. After obtaining the noisy local gradient, industry participants use homomorphic encryption algorithms to encrypt the noisy local gradient, obtaining the encrypted local gradient.

[0077] The above technical solution involves industry participants first pruning local model gradients based on a preset pruning threshold, limiting the gradient norm within a predetermined range and defining the actual upper limit of global sensitivity. Then, noise scale parameters are calculated based on a preset privacy budget and the preset pruning threshold, and random noise vectors are generated according to the noise distribution type. This mathematically provable approach achieves an optimal balance between privacy protection strength and model training utility. Next, the random noise vector is added to the gradient pruning result to obtain a noisy local gradient, ensuring that the impact of individual data on model updates is strictly limited to the range of differential privacy protection, effectively defending against privacy inference attacks based on gradient analysis. Finally, the noisy local gradient is encrypted to obtain an encrypted local gradient, maintaining secure aggregation capabilities within the federated learning framework while ensuring gradient privacy. This solution addresses the privacy leakage problem that may occur during gradient propagation in federated learning, while maintaining model training efficiency and accuracy through precise noise control, providing a verifiable privacy guarantee foundation for cross-institutional collaborative learning.

[0078] In an optional embodiment, the method further includes: determining the common fields and industry-specific fields included by the industry participants, and determining industry-common data and industry-specific data in the original industry data held by the industry participants based on the common fields and the industry-specific fields; processing the industry-common data and industry-specific data using a unified data model and an industry-extended model respectively to obtain local feature data; wherein the unified data model can carry the commonalities between different industries for defining cross-industry common fields, and the industry-extended model is used to define industry-specific fields; the industry-extended model is interconnected with the unified data model.

[0079] In this context, common fields refer to data fields that exist in data structures across different industries and have the same or similar semantics. The specific content of common fields is not limited here; it is determined based on actual business needs. For example, common fields could be basic fields such as user identifiers, time, and geographic location, which can be used for cross-industry alignment.

[0080] Industry-specific fields refer to specialized data fields that exist only within a specific industry or exhibit significant semantic differences across different industries. Both industry-common data and industry-specific data are extracted from raw industry data. Industry-common data is constructed based on common fields and possesses cross-industry comparability and consistency. Industry-specific data consists of industry-specific fields and reflects the professional characteristics of the industry.

[0081] A unified data model is a standardized data framework used to define and describe common fields across industries. An industry-specific extended model is a custom data framework built upon the unified data model to describe industry-specific fields. Taking user identification as an example, a global identifier is used across all industries under the unified data model. Optionally, the industry type can be marked in the unified data model using preset fields, and the industry extended model can be associated with the unified data model via foreign keys, thus ensuring cross-industry compatibility while preserving the characteristics of different industries.

[0082] Local characteristic data is obtained by processing industry-common data and industry-specific data using a unified data model and an industry-extended model, respectively. Local characteristic data includes the common components processed by the unified data model and the specific components processed by the industry-extended model.

[0083] The aforementioned technical solution addresses the fundamental challenges of semantic alignment and structuring in cross-industry data fusion, providing foundational support for efficient and secure data collaboration. By using a unified data model, common fields across industries are effectively defined, ensuring comparability and interoperability of data from different sources across core dimensions. Furthermore, by fully preserving and defining the unique and complex data fields of each industry through industry-specific extended models, information loss or semantic distortion caused by forced format unification is avoided, thus improving the efficiency of data alignment and gradient aggregation in subsequent federated learning and other joint modeling processes.

[0084] Example 3

[0085] Figure 3 This is a schematic diagram of the product recommendation device based on cross-industry data fusion provided in Embodiment 3 of this application. This embodiment can be applied to the situation of recommending products to users based on cross-industry data. The device can be configured in a central coordinator and is implemented by software and / or hardware.

[0086] like Figure 3 As shown, the product recommendation device 300 based on cross-industry data fusion may include:

[0087] The request information determination module 310 is used to determine the product recommendation model corresponding to the product recommendation request, the financial encryption identifier of the target user, and the encrypted financial characteristics corresponding to the target user based on the financial participation policy for the product recommendation request of the target user;

[0088] The auxiliary feature request module 320 is used to request the auxiliary participants associated with the product recommendation model to provide feedback on the encrypted auxiliary features corresponding to the target user based on the financial encryption identifier of the target user.

[0089] The prediction result determination module 330 is used to determine the recommendation prediction result based on the crypto-financial characteristics and crypto-aided characteristics corresponding to the target user through the product recommendation model and to feed the recommendation prediction result back to the financial participant;

[0090] The product recommendation model is obtained by jointly training the crypto-financial characteristics of the financial participants and the auxiliary industry characteristics of the auxiliary participants.

[0091] This application's technical solution constructs a cross-industry collaborative privacy-preserving recommendation mechanism. When a financial participant initiates a product recommendation request, it can accurately trigger a product recommendation model associated with it, which has been jointly trained with cross-industry data. Based on the target user's encrypted financial identifier, it securely obtains the encrypted auxiliary features from the auxiliary participant. This ensures that, without the original data leaving the local environments of the participating industries, it achieves deep fusion and joint computation of financial industry features and auxiliary industry features in an encrypted state. This means that the final recommendation prediction results not only rely on the financial participant's own user profile but also incorporate user features provided by the auxiliary participant, thereby significantly improving the accuracy and personalization of the recommendation results. Simultaneously, all sensitive data is circulated and processed in encrypted form throughout the entire process, effectively preventing the risk of user privacy information leakage and constructing a robust data security barrier. This allows for more efficient, accurate, and personalized recommendation results while complying with data compliance requirements, significantly improving user experience and business conversion efficiency while ensuring data security.

[0092] Optionally, the auxiliary feature request module 320 includes: an industry identifier determination submodule, used to determine the industry encrypted identifier of the target user for the auxiliary participant based on the encrypted identifier mapping relationship and the financial encrypted identifier of the target user; wherein, the encrypted identifier mapping relationship is calculated by the central coordinator, the financial participant, and the auxiliary participant through a global private set intersection protocol; a feature request generation submodule, used to generate an industry feature request based on the industry encrypted identifier of the target user and send the industry feature request to the corresponding auxiliary participant; and an auxiliary feature feedback submodule, used to encrypt the local feature data of the target user locally using a homomorphic encryption algorithm to obtain encrypted auxiliary features, and feed the encrypted auxiliary features back to the central coordinator.

[0093] Optionally, the apparatus further includes: an identifier blinding module, configured to, before determining the industry encrypted identifier of the target user for the auxiliary participant based on the encrypted identifier mapping relationship and the target user's financial encrypted identifier, perform blinding processing on the local user identifier by the financial participant and the auxiliary participant respectively, and return the obtained financial blinded identifier and auxiliary blinded identifier; a digital signature module, configured to digitally sign the financial blinded identifier and the auxiliary blinded identifier using an identifier encryption private key, and return the identifier signature result of the financial blinded identifier and the identifier signature result of the auxiliary blinded identifier to the financial participant and the auxiliary participant respectively; and a result verification module, configured to, using an identifier encryption public key by the financial participant and the auxiliary participant, verify the identifier signature result of the financial blinded identifier and the identifier signature result of the auxiliary blinded identifier respectively. The system includes: a signature verification module; a shared user identification module, which, if the verification passes, obtains a financial signature identifier and an auxiliary signature identifier by deblinding the signature results of the financial blind identifier and the auxiliary blind identifier, respectively, and determines a shared user identifier based on the financial signature identifier and the auxiliary signature identifier; a mapping relationship construction module, which generates a global user identifier for the shared user identifier through the financial participant and the auxiliary participant, and constructs an identifier mapping relationship between the shared user identifier and the global user identifier; and a set intersection operation module, which performs a set intersection operation on the identifier mapping relationship fed back by the financial participant and the auxiliary participant, and encrypts the obtained intersection operation result using a homomorphic encryption algorithm to obtain the encrypted identifier mapping relationship.

[0094] Optionally, the prediction result determination module 330 includes: a feature splicing submodule, used to splice the encrypted financial features and encrypted auxiliary features corresponding to the target user to obtain the encrypted joint features corresponding to the target user, and input the encrypted joint features into the product recommendation model; a prediction vector sending submodule, used to output a recommendation prediction vector based on the encrypted joint features through the product recommendation model, and send the recommendation prediction vector to the financial participant and the auxiliary participant; a partial decryption submodule, used to partially decrypt the recommendation prediction vector through the financial participant and the auxiliary participant, and feed back the partial decryption result of the recommendation prediction vector; a vector plaintext acquisition submodule, used to combine the partial decryption results fed back by the financial participant and the auxiliary participant to obtain the prediction vector plaintext; and a prediction result feedback submodule, used to decompose the prediction vector plaintext to obtain a single result fragment corresponding to the financial participant, and feed back the single result fragment as the recommendation prediction result to the financial participant; wherein, the prediction vector plaintext will be cleared from memory after the recommendation prediction result is fed back to the financial participant.

[0095] Optionally, the device further includes a model training module for training the product recommendation model; wherein the model training module includes: a global parameter determination submodule, used to initialize the global model parameters of the product recommendation model and encrypt the initialized global model parameters using a homomorphic encryption algorithm to obtain encrypted global parameters; a global parameter distribution submodule, used to distribute the encrypted global parameters to at least two industry participants participating in the training of the product recommendation model; wherein the industry participants include at least a financial participant and an auxiliary participant; a local gradient determination submodule, used by the industry participants to calculate the local model gradient for the product recommendation model using the encrypted global parameters and their respective local feature data, and to add noise and encrypt the local model gradient to obtain an encrypted local gradient; a local gradient aggregation submodule, used to aggregate the encrypted local gradients obtained from the industry participants to obtain an encrypted aggregate gradient, and to update the encrypted global parameters based on the encrypted aggregate gradient until the product recommendation model converges.

[0096] Optionally, the local gradient determination submodule includes: a gradient clipping unit, used to clip the local model gradient based on a preset clipping threshold by the industry participant to obtain a gradient clipping result; a noise generation unit, used to generate a random noise vector based on a noise scale parameter and a noise distribution type; wherein the noise scale parameter is calculated according to a preset privacy budget and the preset clipping threshold; and a gradient noise-adding unit, used to add the random noise vector to the gradient clipping result to obtain a noisy local gradient, and to encrypt the noisy local gradient using a homomorphic encryption algorithm to obtain an encrypted local gradient.

[0097] Optionally, the apparatus further includes: a field type determination module, used to determine the common fields and industry-specific fields included by the industry participants, and to determine industry common data and industry-specific data in the original industry data held by the industry participants based on the common fields and the industry-specific fields; and a feature data determination module, used to process the industry common data and industry-specific data using a unified data model and an industry extended model respectively to obtain local feature data; wherein the unified data model can carry the commonalities between different industries to define cross-industry common fields, and the industry extended model is used to define industry-specific fields; the industry extended model and the unified data model are interconnected.

[0098] The product recommendation device based on cross-industry data fusion provided in the embodiments of the invention can execute the product recommendation method based on cross-industry data fusion provided in any embodiment of this application, and has the corresponding performance modules and beneficial effects for executing the product recommendation method based on cross-industry data fusion.

[0099] In the technical solution of this application, the user data involved in the product recommendation based on cross-industry data fusion is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, 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.

[0100] Example 4

[0101] According to embodiments of this application, this application also provides an electronic device, a readable storage medium, and a computer program product.

[0102] Figure 4 A schematic diagram of an electronic device 410, which can be implemented using an embodiment, is shown. The electronic device 410 includes at least one processor 411 and a memory, such as a read-only memory 412 or a random access memory 413, communicatively connected to the at least one processor 411. The memory stores computer programs executable by the at least one processor. The processor 411 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 412 or loaded from storage unit 418 into the random access memory 413. The random access memory 413 may also store various programs and data required for the operation of the electronic device 410. The processor 411, read-only memory 412, and random access memory 413 are interconnected via a bus 414. An input / output interface 415 is also connected to the bus 414.

[0103] Multiple components in electronic device 410 are connected to input / output interface 415, including: input unit 416, such as keyboard, mouse, etc.; output unit 417, such as various types of monitors, speakers, etc.; storage unit 418, such as disk, optical disk, etc.; and communication unit 419, such as network card, modem, wireless transceiver, etc. Communication unit 419 allows electronic device 410 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0104] Processor 411 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 411 performs the various methods and processes described above, such as product recommendation methods based on cross-industry data fusion.

[0105] In some embodiments, the product recommendation method based on cross-industry data fusion can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 418. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 410 via read-only memory 412 and / or communication unit 419. When the computer program is loaded into random access memory 413 and executed by processor 411, one or more steps of the product recommendation method based on cross-industry data fusion described above can be performed. Alternatively, in other embodiments, processor 411 can be configured to perform the product recommendation method based on cross-industry data fusion by any other suitable means (e.g., by means of firmware).

[0106] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), complex programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0107] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to the processor of a general-purpose computer, special-purpose computer, or other programmable product recommendation device based on cross-industry data fusion, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0108] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory or flash memory, optical fiber, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0109] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a cathode ray tube or liquid crystal display monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0110] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., a product recommendation server based on cross-industry data fusion), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0111] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to address the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0112] This application also discloses a computer program product, which includes a computer program that, when executed by a processor, implements the product recommendation method based on cross-industry data fusion provided in any embodiment of this application. This program product shares the same inventive concept as the product recommendation method based on cross-industry data fusion disclosed in the embodiments of this application, and therefore will not be described further here.

[0113] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0114] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A product recommendation method based on cross-industry data fusion, characterized in that, The method, executed by the central coordinator, includes: Based on the financial participation policy regarding the product recommendation request of the target user, determine the product recommendation model corresponding to the product recommendation request, the financial encryption identifier of the target user, and the encrypted financial characteristics corresponding to the target user; Based on the target user's financial encryption identifier, request the auxiliary participants associated with the product recommendation model to provide feedback on the encryption auxiliary features corresponding to the target user; The product recommendation model determines the recommendation prediction result based on the crypto-financial characteristics and crypto-aided characteristics corresponding to the target user, and feeds the recommendation prediction result back to the financial participant. The product recommendation model is obtained by jointly training the crypto-financial characteristics of the financial participants and the auxiliary industry characteristics of the auxiliary participants.

2. The method according to claim 1, characterized in that, The step of requesting the auxiliary participants associated with the product recommendation model to provide feedback on the encrypted auxiliary features corresponding to the target user based on the target user's financial encrypted identifier includes: Based on the encrypted identifier mapping relationship and the financial encrypted identifier of the target user, the industry encrypted identifier of the target user is determined for the auxiliary participant; wherein, the encrypted identifier mapping relationship is calculated by the central coordinator performing a global private set intersection protocol with the financial participant and the auxiliary participant; An industry feature request is generated based on the industry encrypted identifier of the target user, and the industry feature request is sent to the corresponding auxiliary participant; The assisting party encrypts the target user's local feature data using a homomorphic encryption algorithm to obtain encrypted assisting features, and then feeds these encrypted assisting features back to the central coordinator.

3. The method according to claim 2, characterized in that, Before determining the industry-specific encrypted identifier of the target user for the auxiliary participant based on the encrypted identifier mapping relationship and the target user's financial encrypted identifier, the method further includes: The financial participant and the auxiliary participant respectively perform blinding processing on the local user identifier and return the financial blinding identifier and the auxiliary blinding identifier; The financial blinding identifier and the auxiliary blinding identifier are digitally signed using an identifier encryption private key, and the identifier signature results of the financial blinding identifier and the auxiliary blinding identifier are respectively fed back to the financial participant and the auxiliary participant; The financial participant and the auxiliary participant use the identifier encryption public key to verify the identifier signature results of the financial blinding identifier and the identifier signature results of the auxiliary blinding identifier, respectively. If the signature verification is successful, the financial signature identifier and the auxiliary signature identifier are obtained by the financial participant and the auxiliary participant respectively deblinding the signature results of the financial blinding identifier and the signature results of the auxiliary blinding identifier, and the common user identifier is determined based on the financial signature identifier and the auxiliary signature identifier. The financial participants and the auxiliary participants generate a global user identifier for the shared user identifier, and construct an identifier mapping relationship between the shared user identifier and the global user identifier; The encrypted identifier mapping relationship is obtained by performing a set intersection operation on the identifier mapping relationship fed back by the financial participant and the auxiliary participant, and then encrypting the result of the intersection operation using a homomorphic encryption algorithm.

4. The method according to claim 1, characterized in that, The step of determining a recommendation prediction result based on the target user's corresponding crypto-financial characteristics and crypto-aided characteristics through the product recommendation model and then feeding the recommendation prediction result back to the financial participant includes: The encrypted financial features and encrypted auxiliary features corresponding to the target user are concatenated to obtain the encrypted joint features corresponding to the target user, and the encrypted joint features are input into the product recommendation model; The product recommendation model outputs a recommendation prediction vector based on the encrypted joint features, and sends the recommendation prediction vector to the financial participant and the auxiliary participant. The financial participants and the auxiliary participants partially decrypt the recommendation prediction vector and return the partial decryption result of the recommendation prediction vector. The partial decryption results fed back by the financial participant and the auxiliary participant are combined and processed to obtain the predicted vector plaintext; The plaintext of the prediction vector is decomposed to obtain a single result fragment corresponding to the financial participant, and the single result fragment is fed back to the financial participant as the recommendation prediction result. The plaintext prediction vector will be cleared from memory after the recommended prediction result is fed back to the financial participant.

5. The method according to any one of claims 1 to 4, characterized in that, The product recommendation model is trained by a central coordinator using the following steps: The global model parameters of the product recommendation model are initialized, and the initialized global model parameters are encrypted using a homomorphic encryption algorithm to obtain encrypted global parameters; The encrypted global parameters are distributed to at least two industry participants involved in training the product recommendation model; wherein, the industry participants include at least a financial participant and an auxiliary participant. The industry participants use the encrypted global parameters and their respective local feature data to calculate the local model gradient for the product recommendation model, and then add noise and encrypt the local model gradient to obtain the encrypted local gradient. The encrypted local gradients obtained from the industry participants are aggregated to obtain an encrypted aggregate gradient, and the encrypted global parameters are updated based on the encrypted aggregate gradient until the product recommendation model converges.

6. The method according to claim 5, characterized in that, The step of adding noise and encryption to the local model gradient to obtain the encrypted local gradient includes: The gradient clipping result is obtained by the industry participants performing gradient clipping on the local model gradient based on a preset clipping threshold. A random noise vector is generated based on a noise scale parameter and a noise distribution type; wherein the noise scale parameter is calculated according to a preset privacy budget and a preset pruning threshold; The random noise vector is added to the gradient clipping result to obtain a noisy local gradient, and a homomorphic encryption algorithm is used to encrypt the noisy local gradient to obtain an encrypted local gradient.

7. The method according to claim 5, characterized in that, The method further includes: Identify the common fields and industry-specific fields included in the industry participants, and based on the common fields and industry-specific fields, determine the common industry data and industry-specific data in the original industry data held by the industry participants; The common industry data and industry-specific data are processed using a unified data model and an industry-extended model, respectively, to obtain local feature data; The unified data model can carry commonalities across different industries to define cross-industry common fields, while the industry extension model is used to define industry-specific fields; the industry extension model is interconnected with the unified data model.

8. A product recommendation device based on cross-industry data fusion, characterized in that, Configured in the central coordinator, the device includes: The request information determination module is used to determine the product recommendation model corresponding to the product recommendation request, the financial encryption identifier of the target user, and the encrypted financial characteristics corresponding to the target user based on the financial participation policy for the product recommendation request of the target user; The auxiliary feature request module is used to request the auxiliary participants associated with the product recommendation model to provide feedback on the cryptographic auxiliary features corresponding to the target user based on the financial cryptographic identifier of the target user. The prediction result determination module is used to determine the recommendation prediction result based on the crypto-financial characteristics and crypto-aided characteristics corresponding to the target user through the product recommendation model, and to feed the recommendation prediction result back to the financial participant; The product recommendation model is obtained by jointly training the crypto-financial characteristics of the financial participants and the auxiliary industry characteristics of the auxiliary participants.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the product recommendation method based on cross-industry data fusion as described in any one of claims 1-7.

10. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the product recommendation method based on cross-industry data fusion as described in any one of claims 1-7.

Images