A federal model contribution evaluation method and system for government-enterprise cooperation

Abstract

Embodiments of the present disclosure provide a federal model contribution evaluation method and system for government-enterprise cooperation. Applied to the technical field of privacy computing and federal learning, the method comprises: a government subject party and an enterprise subject party respectively train a federal model based on their local data to generate model update data, and generate first and second encrypted differential data using a verifiable encryption method. In a preset encryption domain, the encrypted differential data is subjected to secret feature fusion to obtain secret feature results of each party, and first and second encrypted contribution results are generated based on encryption performance evaluation and fair allocation rules. Subsequently, first and second contribution degree proofs are generated in combination with their respective public parameters. The present scheme generates independently verifiable contribution degree proofs by encrypted differential data and secret feature processing under the premise of protecting the privacy of local data of each party, providing a reliable basis for automated incentives and fair allocation.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of privacy computing and federated learning, in particular to a federated model contribution evaluation method and system for government-enterprise cooperation. BACKGROUND

[0002] With the development of artificial intelligence and the marketization of data elements, the demand for collaborative modeling of federated learning in the fields of smart cities and public security between governments and enterprises is increasing. After local training, a fair and verifiable contribution evaluation mechanism is needed to support incentive distribution and trust establishment, and to promote the safe circulation of data value.

[0003] The prior art uploads local model updates (such as gradients or weight differences) to the center server by each participant; the server calculates indicators such as feature importance, gradient similarity or Shapley value in the clear domain or semi-encrypted state, and preliminarily evaluates the contributions of each party; and finally the center node generates a contribution score and notifies the participants.

[0004] However, the model update data in the prior art needs to be decrypted during aggregation or evaluation, which poses a risk of sensitive information leakage; the evaluation process is not transparent, and participants cannot independently verify the fairness of the contribution score, which can easily lead to trust disputes; and there is a lack of verifiable credentials, and the evaluation results are only internal scores, which cannot be used as a trusted basis for on-chain incentives, audits or compliance. SUMMARY

[0005] To solve the problems of the prior art, the present disclosure provides a federated model contribution evaluation method and system for government-enterprise cooperation. The present disclosure solves the problem that the model update data in the prior art needs to be decrypted, which leads to exposure of sensitive information, the problem that the evaluation process is black-boxed, which makes it impossible for each party to verify fairness, and the problem that there is a lack of on-chain verifiable credentials, and the evaluation results are difficult to use as a trusted basis for incentives, audits or compliance.

[0006] According to a first aspect of the present disclosure, a federated model contribution evaluation method for government-enterprise cooperation is provided, comprising: obtaining first model update data generated by a government subject when training a federated model based on first local data, and second model update data generated by an enterprise subject when training the federated model based on second local data, encrypting the first model update data and the second model update data using a verifiable encryption method to obtain first encrypted difference data and second encrypted difference data;

[0007] Performing a ciphertext feature fusion process on the first encrypted difference data and the second encrypted difference data in a preset encryption domain to obtain a first ciphertext feature result of the government subject and a second ciphertext feature result of the enterprise subject;

[0008] The first encrypted contribution result of the government subject and the second encrypted contribution result of the enterprise subject are generated by performing encryption performance influence evaluation and fair distribution rule quantization on the first encrypted feature result and the second encrypted feature result in a preset encryption domain.

[0009] The first contribution degree proof of the government subject is generated based on the first encrypted contribution result and the first public parameter corresponding to the first encrypted contribution result, and the second contribution degree proof of the enterprise subject is generated based on the second encrypted contribution result and the second public parameter corresponding to the second encrypted contribution result.

[0010] According to a second aspect of the present disclosure, a federal model contribution evaluation system for government-enterprise cooperation is provided for performing the method as described in the first aspect, comprising: a data encryption module configured to obtain first model update data generated by a government subject when training a federal model based on first local data, and second model update data generated by an enterprise subject when training the federal model based on second local data, and encrypt the first model update data and the second model update data by using a verifiable encryption method to obtain first encrypted differential data and second encrypted differential data;

[0011] A secret feature fusion module is configured to perform secret feature fusion processing on the first encrypted differential data and the second encrypted differential data in a preset encryption domain to obtain a first secret feature result of the government subject and a second secret feature result of the enterprise subject.

[0012] An encrypted contribution result generation module is configured to perform encryption performance influence evaluation and fair distribution rule quantization on the first secret feature result and the second secret feature result in a preset encryption domain to generate a first encrypted contribution result of the government subject and a second encrypted contribution result of the enterprise subject.

[0013] A contribution degree proof generation module is configured to obtain a first public parameter corresponding to the first encrypted contribution result and a second public parameter corresponding to the second encrypted contribution result, generate a first contribution degree proof of the government subject based on the first encrypted contribution result and the first public parameter, and generate a second contribution degree proof of the enterprise subject based on the second encrypted contribution result and the second public parameter.

[0014] According to a third aspect of the present disclosure, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method as described above when executing the program.

[0015] In the federated model contribution evaluation method and system for government-enterprise cooperation provided above, the embodiments of the present disclosure generate independently verifiable contribution degree proofs by encrypted differential data and ciphertext feature processing under the premise of protecting the local data privacy of each party, and provide a trusted basis for automatic incentive and fair distribution. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0017] Figure 1 A flowchart of a federated model contribution evaluation method for government-enterprise cooperation according to an embodiment of the present disclosure is shown;

[0018] Figure 2 A flowchart of a federated model contribution evaluation method for government-enterprise cooperation according to an embodiment of the present disclosure is shown;

[0019] Figure 3 A schematic block diagram of a federated model contribution evaluation system for government-enterprise cooperation according to an embodiment of the present disclosure is shown;

[0020] Figure 4 A block diagram of an exemplary electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0021] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0022] Those skilled in the art can understand that the terms "first", "second" and the like in the embodiments of the present disclosure are only used to distinguish different steps, devices or modules, and do not represent any specific technical meaning, nor indicate their logical order. It should also be understood that in the embodiments of the present disclosure, "multiple" can mean two or more, and "at least one" can mean one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present disclosure, without explicit limitation or in the context of the preceding and following text, it can be understood as one or more. In addition, the term "and / or" in the present disclosure is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B can represent the existence of A alone, the existence of A and B together, and the existence of B alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the front and rear associated objects. It should also be understood that the description of each embodiment of the present disclosure emphasizes the differences between each embodiment, and the same or similar parts can be referred to each other, and for the sake of brevity, they will not be repeated.

[0023] It should also be understood that the dimensions of the various parts shown in the drawings are not necessarily to scale. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application or uses. Techniques, methods, and devices known to those of ordinary skill in the relevant art can not be discussed in detail, but where appropriate, they can be regarded as part of the specification. It should be noted that like reference numerals and letters refer to like items in the following drawings, and therefore, once an item is defined in one drawing, it need not be discussed further in subsequent drawings.

[0024] In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present disclosure.

[0025] Figure 1 A flowchart of a federal model contribution evaluation method for government-enterprise cooperation provided by the embodiments of the present disclosure is shown. The method of the embodiments of the present disclosure aims to realize accurate detection of large and small goals for pictures.

[0026] S101, obtain first model update data generated by the government subject party when training the federated model based on the first local data, and second model update data generated by the enterprise subject party when training the federated model based on the second local data, encrypt the first model update data and the second model update data using a verifiable encryption method to obtain first encrypted differential data and second encrypted differential data.

[0027] The government subject party can refer to a government agency or a government department with data resources and business application scenarios in a government-enterprise cooperation scenario, which is one of the participants of federated learning, holds training data related to government business locally, and participates in the training process of the federated model.

[0028] The enterprise subject party can refer to an enterprise agency with industry data or computing resources in a government-enterprise cooperation scenario, which is another participant of federated learning, holds training data related to enterprise business locally, and participates in the training process of the federated model.

[0029] The first local data can refer to a set of original data related to government business held by the government subject party in its local data environment. The data can include government transaction records, administrative approval data, public service interaction logs, etc. The first local data is not uploaded or shared, and is only used to train the federated model in the local computing environment of the government subject party.

[0030] The second local data can refer to a set of original data related to enterprise business held by the enterprise subject party in its local data environment. The data includes enterprise operation data, customer interaction data, industry production data, etc. The second local data is only used to train the federated model in the local computing environment of the enterprise subject party.

[0031] The federated model can refer to a machine learning model trained by multiple data participants without sharing original data through a federated learning framework. In this application, the federated model can integrate the data characteristics of the government subject party and the enterprise subject party, and realize the modeling capability of cross-domain collaboration.

[0032] The first model update data can refer to the parameter gradient or weight difference information generated by the government subject party for updating the global model after iterative calculation of the federated model parameters based on local data in the local training process.

[0033] The second model update data can refer to the parameter gradient or weight difference information generated by the enterprise subject party for updating the global model after iterative calculation of the federated model parameters based on local data in the local training process.

[0034] The verifiable encryption method can refer to a type of cryptographic method that can support third-party correctness verification of the encrypted result while encrypting data. For example, zero-knowledge proof, homomorphic encryption combined with commitment mechanism, etc. can be used to realize the verifiability of the encrypted model update data.

[0035] The first encrypted differential data can refer to the ciphertext differential information obtained by processing the first model update data generated by the government subject party by the verifiable encryption method.

[0036] The second encrypted differential data can refer to the ciphertext differential information obtained by processing the second model update data generated by the enterprise subject party by the verifiable encryption method.

[0037] The government subject party performs local training on the shared federated model based on the first local data held by it. In the training process, the government subject party updates the model parameters iteratively by using the gradient descent algorithm, and obtains the first model update data corresponding to the distribution characteristics of the local data. The first model update data is usually expressed as a parameter gradient vector or a weight differential matrix, which can reflect the contribution of the government subject party data to the global model.

[0038] At the same time, the enterprise subject party performs local training on the same federated model based on the second local data held by it. In the training process, the enterprise subject party also updates the model parameters by using the optimization algorithm, and obtains the corresponding second model update data. The second model update data is consistent with the first model update data in the data feature dimension, but the numerical result reflects the differential characteristics of the enterprise subject party data.

[0039] Then, the verifiable encryption method is performed on the first model update data and the second model update data respectively. The verifiable encryption method can adopt the way of homomorphic encryption combined with zero-knowledge proof: first, the homomorphic encryption algorithm (such as Paillier encryption) is used to perform ciphertext mapping on the first model update data and the second model update data, to obtain the first ciphertext update vector and the second ciphertext update vector, to ensure that the encrypted data can still support subsequent additive aggregation operation without decryption; second, the corresponding zero-knowledge proof circuit is constructed to generate correctness proof for the ciphertext update vector submitted by each party, so that the verifier can confirm that the encrypted data indeed comes from the legal model training update without knowing the plaintext, instead of being forged or tampered. Through the above processing, the first encrypted differential data and the second encrypted differential data are finally obtained. The first encrypted differential data corresponds to the ciphertext form of the first model update data of the government subject party, combined with its zero-knowledge correctness proof; the second encrypted differential data corresponds to the ciphertext form of the second model update data of the enterprise subject party, combined with its zero-knowledge correctness proof.

[0040] S102, performing a ciphertext feature fusion process on the first encrypted differential data and the second encrypted differential data in a preset encryption domain to obtain a first ciphertext feature result of the government subject party and a second ciphertext feature result of the enterprise subject party.

[0041] The preset encryption domain can refer to a mathematical operation space defined based on a specific cryptography algorithm, which is used to calculate data in a ciphertext state. The encryption domain is usually defined by a homomorphic encryption algorithm or a multi-party secure computation protocol, such as a modulus operation domain based on the Paillier homomorphic encryption system or a finite field based on elliptic curve encryption.

[0042] The first ciphertext feature result can refer to a result obtained after performing a feature fusion operation on the first encrypted differential data in the preset encryption domain. The result is still in a ciphertext state and does not expose any plaintext information, but its numerical value represents the feature expression of the first model update data of the government subject party in a ciphertext condition.

[0043] The second ciphertext feature result can refer to a result obtained after performing a feature fusion operation on the second encrypted differential data in the preset encryption domain.

[0044] First, a preset encryption domain is determined, which can be constructed by a homomorphic encryption algorithm, such as a Paillier homomorphic encryption system based on a large integer modulus space. In this encryption domain, all calculation operations are performed in a ciphertext form, ensuring that the calculation process does not leak any plaintext information. Subsequently, the first encrypted differential data generated by the government subject party and the second encrypted differential data generated by the enterprise subject party are input into the ciphertext feature fusion process. The encrypted differential data is essentially a ciphertext vector obtained by encrypting the differential result of the model parameter update in the local training process.

[0045] In the process of secret feature fusion, firstly, the first encrypted differential data and the second encrypted differential data are uniformly formatted to be in the same encrypted calculation space to ensure the executability of subsequent fusion. Then, without decryption, weighted combination operations are performed on each component of the two sets of encrypted differential data. For example, a corresponding weight factor can be preset for different data sources, and the weight factor is embedded in the fusion operation in an encrypted state, so that the weight adjustment of the contribution difference is completed in the ciphertext state. Through the ciphertext scalar multiplication operation supported by the homomorphic encryption algorithm, the encrypted differential data is weighted without decryption. Specifically, the system uses the power operation characteristics of encrypted data to directly act on each component of the ciphertext vector with the preset weight factor, thereby completing the differentiated adjustment of the contribution of different data sources in the ciphertext space. After completing the weighted combination, the fusion result is further normalized in the encrypted domain. The purpose of this normalization is to keep the numerical result after fusion within a comparable range, so that it can be directly used in subsequent on-chain verification or global optimization. The whole process of this processing is based on homomorphic operations in the encrypted domain, without any plaintext exposure. Based on the approximate calculation method supported by the homomorphic encryption, the encrypted ciphertext vector after fusion is securely normalized. The system adjusts the numerical range of the ciphertext result through a preset reference value and a linear scaling mechanism in an encrypted state, so that it meets the standardization requirements required by subsequent on-chain verification or global optimization, and the whole process does not need to be decrypted. In the process of secret feature fusion, the system retains the independent processing path of each participant's data to ensure that while completing collaborative calculation, the dominant contribution components of each subject can still be traced and separated. This design enables the fusion result to be mapped into a secret output that reflects the respective feature expressions of the government and enterprise parties. Finally, the system outputs a first secret feature result generated by the first encrypted differential data of the government subject party as the dominant contribution, and a second secret feature result generated by the second encrypted differential data of the enterprise subject party as the dominant contribution.

[0046] On the basis of the above technical solutions, optionally, in the preset encrypted domain, the first encrypted differential data and the second encrypted differential data are subjected to secret feature fusion processing to obtain the first secret feature result of the government subject party and the second secret feature result of the enterprise subject party, comprising:

[0047] In the preset encrypted domain, the first encrypted differential data is subjected to shallow nonlinear transformation to obtain the first preliminary secret feature representation of the government subject party, and in the preset encrypted domain, the second encrypted differential data is subjected to shallow nonlinear transformation to obtain the second preliminary secret feature representation of the enterprise subject party;

[0048] perform affine transformation on the first preliminary secret state feature representation and the second preliminary secret state feature representation based on preset global statistical parameters in a preset encryption domain, to obtain a first normalized secret state feature representation of the government subject party and a second normalized secret state feature representation of the enterprise subject party;

[0049] project the first normalized secret state feature representation and the second normalized secret state feature representation to a unified secret state semantic space through a preset linear mapping matrix in the preset encryption domain, to obtain a first mapped secret state feature representation of the government subject party and a second mapped secret state feature representation of the enterprise subject party;

[0050] perform dimension coordination processing on the first mapped secret state feature representation and the second mapped secret state feature representation in the preset encryption domain, to obtain a first aligned secret state feature representation of the government subject party and a second aligned secret state feature representation of the enterprise subject party;

[0051] perform weighted combination processing on the first aligned secret state feature representation and the second aligned secret state feature representation based on a preset fusion weight rule in the preset encryption domain, to obtain a first secret state feature result of the government subject party and a second secret state feature result of the enterprise subject party.

[0052] In the scheme, the first preliminary secret state feature representation can refer to a secret state feature vector or matrix representation obtained by shallow nonlinear transformation on the first encrypted differential data of the government subject party in the preset encryption domain, which retains the core feature information of the original encrypted data, but enhances the feature expression ability through nonlinear mapping while maintaining the encryption security.

[0053] The second preliminary secret state feature representation can refer to a secret state feature vector or matrix representation obtained by shallow nonlinear transformation on the second encrypted differential data of the enterprise subject party in the preset encryption domain.

[0054] The preset global statistical parameters can refer to a parameter set obtained by the system according to historical or global data statistics in the process of federal modeling or secret state feature processing, which is used for standardization or normalization processing of the preliminary feature representations of each party, including mean, variance, minimum value, maximum value, etc.

[0055] The first normalized secret state feature representation can refer to a secret state feature representation obtained by performing affine transformation (standardization or normalization) on the first preliminary secret state feature representation of the government subject party based on the preset global statistical parameters, which is used to eliminate the dimensional difference and offset influence between different feature dimensions.

[0056] The second normalized secret state feature representation can refer to a secret state feature representation obtained by performing affine transformation on the second preliminary secret state feature representation of the enterprise subject party based on the preset global statistical parameters.

[0057] The preset linear mapping matrix can refer to a reversible or reducible matrix used in the process of mapping in the ciphertext semantic space, for mapping the normalized ciphertext features to a unified ciphertext semantic space, so that the features of multiple parties are compared and fused in the same vector space or feature subspace.

[0058] The unified ciphertext semantic space can refer to a common representation space of the normalized ciphertext features of multiple parties after linear mapping, for ensuring that the ciphertext features of different subjects are interoperable and aligned in the same semantic scale, so as to perform dimension coordination, weighted fusion and contribution evaluation.

[0059] The first mapped ciphertext feature representation can refer to the result of projecting the first normalized ciphertext feature representation of the government subject party to the unified ciphertext semantic space through the preset linear mapping matrix, for maintaining the original feature information while making it comparable with the feature of the enterprise subject party in the same space.

[0060] The second mapped ciphertext feature representation can refer to the result of projecting the second normalized ciphertext feature representation of the enterprise subject party to the unified ciphertext semantic space through the preset linear mapping matrix.

[0061] The first aligned ciphertext feature representation can refer to the ciphertext feature representation obtained by performing dimension coordination processing on the first mapped ciphertext feature representation in the unified ciphertext semantic space, for adjusting the feature scale and variance to eliminate the imbalance between different feature dimensions, facilitating weighted fusion.

[0062] The second aligned ciphertext feature representation can refer to the ciphertext feature representation obtained by performing dimension coordination processing on the second mapped ciphertext feature representation in the unified ciphertext semantic space.

[0063] The preset fusion weight rule can refer to a rule set used to calculate the contribution proportion of each party to the final ciphertext feature result in the process of ciphertext feature fusion, including weight coefficient distribution, weighted summation, regularization processing and other methods, for ensuring that the features of each party are preserved after fusion both independently contribute and generate a global unified feature representation.

[0064] The first ciphertext feature result can refer to the ciphertext feature result obtained by weighted combination of the first aligned ciphertext feature representation of the government subject party under the action of the fusion weight rule.

[0065] The second ciphertext feature result can refer to the ciphertext feature result obtained by weighted combination of the second aligned ciphertext feature representation of the enterprise subject party under the action of the fusion weight rule.

[0066] In the preset encrypted domain, first, the first encrypted differential data of the government subject party is subjected to shallow nonlinear transformation processing. The process includes: applying a ciphertext activation function such as a ciphertext Sigmoid, a ciphertext ReLU or a ciphertext Tanh to the encrypted value of each feature dimension to maintain homomorphic computability in the data encryption state, while mapping the linear relationship of the original differential data to a nonlinear space to enhance the feature expression capability, generating the first preliminary ciphertext feature representation of the government subject party. Similarly, the same nonlinear transformation is performed on the second encrypted differential data of the enterprise subject party to generate the second preliminary ciphertext feature representation of the enterprise subject party, ensuring consistency in multi-party data processing.

[0067] Subsequently, the preliminary ciphertext feature representation is subjected to affine transformation based on preset global statistical parameters. The global statistical parameters include the mean, variance and maximum and minimum values of each feature dimension, which are used to realize standardization processing in the encrypted domain. The specific operation is: subtracting the corresponding mean from the preliminary ciphertext feature value, and then dividing by the standard deviation or scaling to a fixed interval to form the first normalized ciphertext feature representation of the government subject party and the second normalized ciphertext feature representation of the enterprise subject party, so as to eliminate the dimensional differences of different features.

[0068] After completing the normalization processing, the normalized ciphertext feature representation is projected to a unified ciphertext semantic space through a preset linear mapping matrix. The linear mapping matrix can realize rotation, scaling and orthogonal projection operations through ciphertext matrix multiplication, so that the features of different participants remain semantically consistent in the same vector space, and the effectiveness of homomorphic operation is guaranteed in the encrypted state. After mapping processing, the first mapped ciphertext feature representation of the government subject party and the second mapped ciphertext feature representation of the enterprise subject party are generated.

[0069] Next, the mapped ciphertext feature representation is subjected to dimension coordination processing. The processing includes scaling adjustment, variance matching, offset correction and feature value range unification of each dimension feature value, generating the first aligned ciphertext feature representation of the government subject party and the second aligned ciphertext feature representation of the enterprise subject party. The purpose of dimension coordination is to ensure that the numerical distribution of multi-party features is consistent in the unified semantic space.

[0070] Finally, the aligned ciphertext feature representation is weighted and combined based on a preset fusion weight rule. The fusion weight rule dynamically allocates according to the contribution degree of each party, the historical performance of the participants or the preset proportion, and the final feature value can be calculated by weighted summation, exponential weighting or regularization weighting method, to obtain the first ciphertext feature result of the government subject party and the second ciphertext feature result of the enterprise subject party. The ciphertext feature result not only retains the independent feature information of each party, but also generates a fused feature representation.

[0071] In the scheme, not only the data privacy is protected, but also the verifiability and fairness of subsequent contribution evaluation and incentive distribution are guaranteed, and the overall system security and multi-party collaboration efficiency are improved.

[0072] S103, in the preset encryption domain, the first and second encrypted performance impact evaluation and fair allocation rule quantification are performed on the first and second encrypted feature results, to generate the first encrypted contribution result of the government subject and the second encrypted contribution result of the enterprise subject.

[0073] The first encrypted contribution result can refer to a quantization result generated in the preset encryption domain based on the first encrypted feature result of the government subject after the encrypted performance impact evaluation and fair allocation rule quantification. The result exists in an encrypted form and is used to represent the effective contribution degree of the government subject in the multi-party joint calculation of the global model update or task completion. The result not only contains the calculation contribution size of the government subject, but also contains the weighted evaluation of the performance overhead and fairness constraint in the encryption environment.

[0074] The second encrypted contribution result can refer to a quantization result generated in the preset encryption domain based on the second encrypted feature result of the enterprise subject after the same encrypted performance impact evaluation and fair allocation rule quantification. The result also exists in an encrypted form.

[0075] The system counts the number of encrypted operations of the first and second ciphertext feature results when performing homomorphic addition and homomorphic multiplication, and estimates the time consumption of each operation in combination with the key length of the encryption algorithm, to obtain the corresponding calculation complexity index. If Paillier encryption is used, the bit operation overhead of single addition and multiplication is determined according to the modulus length. If polynomial homomorphic encryption is used, the calculation complexity is calculated by the number of polynomial expansion and the size of the coefficient. According to the ciphertext length (e.g. 2048 bits, 4096 bits) of the ciphertext feature result and the required number of interactions, the system estimates the bandwidth occupation and delay required in the network transmission process, to obtain the communication overhead index. The system counts the file size of the ciphertext feature result in the encrypted state, calculates the space cost required for storage in the distributed ledger, and obtains the storage overhead index. After the above steps, the encryption performance evaluation vector for the first ciphertext feature result and the encryption performance evaluation vector for the second ciphertext feature result are obtained respectively. To ensure the comparability between different indicators, the system normalizes the calculation complexity, communication overhead and storage overhead, and maps them to standardized values in a unified interval. For example, each indicator is scaled to the [0, 1] interval. This normalization process is implemented in the encryption domain through a preset linear transformation mechanism, ensuring that the entire process does not require decryption. In this way, the performance evaluation results of the government subject party and the enterprise subject party can be used as input for subsequent fair distribution calculation. The encryption performance evaluation vector is also encrypted by the homomorphic encryption algorithm after generation, ensuring that it is in the same encryption domain as the first and second ciphertext feature results.

[0076] The system adopts a preset fair distribution strategy, for example: based on the proportional distribution rule: the contribution degree and the performance overhead are weighted and combined to ensure that a subject with high contribution but excessive performance consumption will not be excessively rewarded; or based on the Shapley value distribution rule: the marginal contribution of each participant to the global performance improvement in collaborative training is calculated one by one. In the preset encryption domain, the system homomorphically accumulates the first ciphertext feature result and its corresponding performance evaluation value to form the first encrypted contribution result, and homomorphically accumulates the second ciphertext feature result and its corresponding performance evaluation value to form the second encrypted contribution result. For the government subject party, the first encrypted contribution result reflects the weighted balance result of its feature contribution value and performance overhead; for the enterprise subject party, the second encrypted contribution result reflects the weighted balance result of its feature contribution value and performance overhead. Finally, the system obtains the first encrypted contribution result of the government subject party in the encrypted state, whose value contains the comprehensive quantification of its ciphertext feature contribution and performance consumption; and the second encrypted contribution result of the enterprise subject party, whose value also represents its contribution in an encrypted form.

[0077] On the basis of the above technical solutions, optionally, the first and second encrypted characteristic results are evaluated in the preset encryption domain according to the preset performance evaluation index, to obtain the first performance impact score of the government subject party and the second performance impact score of the enterprise subject party.

[0078] On the basis of the above technical solutions, optionally, the first and second encrypted characteristic results are evaluated in the preset encryption domain according to the preset performance evaluation index, to obtain the first performance impact score of the government subject party and the second performance impact score of the enterprise subject party.

[0079] On the basis of the above technical solutions, optionally, the first and second encrypted characteristic results are evaluated in the preset encryption domain according to the preset performance evaluation index, to obtain the first performance impact score of the government subject party and the second performance impact score of the enterprise subject party.

[0080] In this scheme, the preset performance evaluation index can refer to a measurement standard for measuring the performance impact that may be generated when the encrypted characteristic result is calculated, analyzed or transmitted in the encryption domain, including but not limited to calculation delay, encryption calculation complexity, ciphertext size, data transmission load and resource consumption, etc.

[0081] The first performance impact score can refer to a quantitative score calculated according to the preset performance evaluation index based on the first encrypted characteristic result of the government subject party, which is used to reflect the influence degree of the encrypted characteristic result on the system performance in the encryption processing and calculation process.

[0082] The second performance impact score can refer to a quantitative score calculated according to the preset performance evaluation index based on the second encrypted characteristic result of the enterprise subject party.

[0083] The preset fair distribution rule can refer to a rule set used to determine the distribution proportion or weight of the encrypted contribution result of each participant in a multi-party cooperation environment, including but not limited to distribution according to contribution ratio, distribution according to performance score adjustment, weight adjustment according to historical contribution, etc.

[0084] The preset game theory distribution model can refer to a mathematical model established based on the game theory, which is used to calculate the encrypted contribution result according to the encrypted characteristic contribution, performance impact score and fair distribution rule of each party in a multi-party participation environment, including Nash equilibrium analysis, optimal strategy calculation and benefit distribution strategy generation, etc. Through the model, the first encrypted contribution result of the government subject party and the second encrypted contribution result of the enterprise subject party can be generated, to realize safe, fair and verifiable multi-party incentive distribution.

[0085] In the preset encryption domain, first, the first ciphertext characteristic result of the government subject party and the second ciphertext characteristic result of the enterprise subject party are obtained. According to the preset performance evaluation index, the ciphertext characteristic results are respectively evaluated in terms of encryption performance. The specific steps include: performing encryption calculation and transmission simulation on the ciphertext characteristic results in the encryption domain, quantifying the calculation delay, ciphertext expansion size, resource occupation and encryption operation complexity generated in the processing process, and standardizing the indexes; then, the first performance influence score of the government subject party and the second performance influence score of the enterprise subject party are generated according to the standardized results, which are used to reflect the calculation efficiency and resource consumption level of each party under encryption processing.

[0086] Then, the first performance influence score, the second performance influence score and the preset fair allocation rule are input into the preset game theory allocation model. Based on the principles of payoff allocation and strategy optimization in game theory, the model calculates the allocation strategy of multi-party encryption contribution. Specifically, the game theory allocation model performs the following processing steps: (1) constructing the participant payoff function: based on the input performance influence score and fair allocation rule, the model defines the payoff function for the government subject party and the enterprise subject party, respectively, which is used to represent the incentive value or payoff quantitative index obtained by each participant under different encryption contribution allocation schemes; (2) solving the optimal allocation weight: under the premise of meeting the fairness constraint, encryption performance efficiency requirement and system security, the optimal allocation weight is calculated by solving the Nash equilibrium or other optimization algorithm, so that the overall incentive allocation reaches the Pareto optimal or near-optimal allocation weight, thereby determining the weight ratio of the government subject party and the enterprise subject party in the final contribution result; (3) generating the encryption contribution result: according to the obtained optimal allocation weight, the encryption contribution of the government subject party and the enterprise subject party is quantitatively calculated in the encryption domain, and the first encryption contribution result of the government subject party and the second encryption contribution result of the enterprise subject party are generated. The first encryption contribution result and the second encryption contribution result not only reflect the characteristic contribution size of each participant, but also integrate the performance overhead and fairness adjustment factors in the encryption processing process, which can be directly used as the trusted input of the subsequent contribution proof generation and incentive allocation smart contract. In the whole process, all calculations are completed in the preset encryption domain, ensuring that the original data, intermediate parameters and final results are in an encrypted state.

[0087] The training process of the preset game theory allocation model is as follows:

[0088] Firstly, the internal parameters of the model are initialized, including the benefit function structure of each participant, the allocation weight parameter and the fairness constraint coefficient. The benefit function is used to quantify the expected income of the government subject and the enterprise subject under different encryption contribution allocation schemes, considering the indicators such as encrypted feature contribution value, encryption performance overhead, resource occupation and computing efficiency. Subsequently, historical encryption contribution data, performance impact score and fair allocation rules are obtained to form an encryption training sample set. The basic value of the encrypted feature contribution, the computing delay, the ciphertext expansion size and the resource consumption of each sample are standardized in the encryption domain for subsequent income calculation. In each training iteration, the model simulates multi-party allocation schemes in the encryption domain, combines different allocation strategies as candidate strategies, calculates the benefit function value under each strategy combination, and evaluates whether it meets the Nash equilibrium condition and Pareto improvement constraint to ensure the incentive equilibrium and fairness of participants under different allocation weights. Based on the iteration results, the model adjusts the allocation weight parameter using optimization algorithms to make the overall income optimal under the conditions of meeting the fairness constraint, encryption performance efficiency requirement and system security. During the training process, the performance overhead of each allocation scheme is dynamically evaluated through encryption calculation and transmission, and this performance indicator is fed back to the benefit function update to enhance the model's adaptability to encryption efficiency and resource consumption. After training, the model can calculate the optimal allocation weight in the encryption domain according to the real-time input performance impact score and fair allocation rules, and generate the first encryption contribution result and the second encryption contribution result based on the basic value of each party's encrypted feature contribution, realizing the quantification of the government subject and the enterprise subject's contribution, integrating encryption performance and fairness adjustment factors, and providing trusted, end-to-end encrypted input for subsequent contribution proof generation and incentive allocation smart contract.

[0089] In this scheme, the independent feature contribution of each participant is effectively preserved, and the secure alignment and unified representation of multi-party features are realized, providing trusted and privacy-protected basic data for subsequent contribution quantification and incentive allocation.

[0090] S104, obtaining the first public parameter corresponding to the first encryption contribution result and the second public parameter corresponding to the second encryption contribution result, generating the first contribution degree proof of the government subject based on the first encryption contribution result and the first public parameter, and generating the second contribution degree proof of the enterprise subject based on the second encryption contribution result and the second public parameter.

[0091] The first public parameter can refer to the verification auxiliary information corresponding to the first encrypted contribution result of the government subject party and disclosed in the encryption domain. The public parameter is usually generated by a key generation center or a zero-knowledge proof protocol, and includes a public key in a homomorphic encryption algorithm, a generator, a homomorphic operation reference value, a randomness proof auxiliary quantity, etc., which is used to ensure that the correctness and source of the first encrypted contribution result can be publicly verified without leaking the original contribution data.

[0092] The second public parameter can refer to the verification auxiliary information corresponding to the second encrypted contribution result of the enterprise subject party and disclosed in the encryption domain. The second public parameter is similar to the first public parameter and mainly includes a public key, a group parameter, a randomness commitment value, etc., which is used to support the verifiable check of the second encrypted contribution result.

[0093] The first contribution degree proof can refer to the encryption proof generated based on the first encrypted contribution result and the corresponding first public parameter, which can prove the real contribution degree of the government subject party in the federated model training or collaborative calculation. The proof is generated by using a zero-knowledge proof, a verifiable encryption or a commitment mechanism, which ensures that any verifier can confirm the validity of the contribution of the government subject party without decrypting the original contribution data.

[0094] The second contribution degree proof can refer to the encryption proof generated based on the second encrypted contribution result and the corresponding second public parameter, which can prove the real contribution degree of the enterprise subject party in the federated model training or collaborative calculation.

[0095] The system calls a preset encryption parameter generation mechanism to obtain the first public parameter corresponding to the first encrypted contribution result and also obtains the second public parameter corresponding to the second encrypted contribution result. For the first encrypted contribution result, the system constructs a commitment value by using the first public parameter. For example, in the homomorphic encryption domain, the first encrypted contribution result is combined with a randomness auxiliary quantity to generate an unforgeable commitment. The commitment ensures that an external verifier can confirm that the contribution result of the government subject party is consistent with the pre-designed calculation logic. For the second encrypted contribution result, the system performs the same commitment construction operation by using the second public parameter, thereby ensuring the verifiability of the encrypted contribution result of the enterprise subject party.

[0096] The system constructs a first contribution degree proof based on the first encrypted contribution result and the first public parameter. Specifically, the system constructs a statement logic of zero-knowledge proof in the encryption domain, takes the encrypted contribution result as an input statement of the proof, applies a zero-knowledge proof protocol (such as Schnorr proof or Bulletproofs proof), proves that "the contribution result of the government subject party is indeed generated by the preset encryption calculation process, and meets the fair distribution rule", without revealing the plaintext value of the contribution degree. The system constructs a second contribution degree proof based on the second encrypted contribution result and the second public parameter, which is used to ensure that the contribution result of the enterprise subject party is true and reliable. This proof also relies on the zero-knowledge protocol and supports any verifier to verify the result on the chain. The contribution degree proof mechanism is introduced to solve the "lack of trust" problem in multi-party cooperation, to ensure that each participant cannot fake or exaggerate the contribution, and to avoid the privacy leakage risk caused by centralized auditing. Finally, the system obtains: the first contribution degree proof of the government subject party; and the second contribution degree proof of the enterprise subject party. The contribution degree proof is written into the distributed ledger as an unalterable on-chain record.

[0097] In the embodiments of the present application, under the premise of protecting the privacy of local data of each party, through encrypted differential data and ciphertext feature processing, an independently verifiable contribution degree proof is generated, which provides a credible basis for automatic incentive and fair distribution.

[0098] On the basis of the above technical solutions, before generating the first contribution degree proof of the government subject party based on the first encrypted contribution result and the first public parameter, and generating the second contribution degree proof of the enterprise subject party based on the second encrypted contribution result and the second public parameter, the method further includes:

[0099] obtaining a first identity of the government subject party, a first participation data size, and first task context information, performing consistency verification on the first identity, the first participation data size, and the first task context information to obtain a first verification result, and obtaining a second identity of the enterprise subject party, a second participation data size, and second task context information, performing consistency verification on the second identity, the second participation data size, and the second task context information to obtain a second verification result;

[0100] Correspondingly, generating the first contribution degree proof of the government subject party based on the first encrypted contribution result and the first public parameter, and generating the second contribution degree proof of the enterprise subject party based on the second encrypted contribution result and the second public parameter, include:

[0101] If the first check result and the second check result meet a preset check condition, a first contribution commitment value bound to the government subject party is generated based on the first encrypted contribution result and first public parameters, and a second contribution commitment value bound to the enterprise subject party is generated based on the second encrypted contribution result and second public parameters.

[0102] Based on the preset contribution calculation model, the first contribution commitment value and the first public parameters, a first verifiable proof is generated by calling a preset zero-knowledge proof protocol, and based on the preset contribution calculation model, the second contribution commitment value and the second public parameters, a second verifiable proof is generated by calling the preset zero-knowledge proof protocol.

[0103] The first verifiable proof and the second verifiable proof are subjected to integrity encapsulation to obtain a first contribution degree proof of the government subject party and a second contribution degree proof of the enterprise subject party.

[0104] In the scheme, the first identity identifier can be identity information for uniquely identifying the government subject party, and can include a user ID, public key information or other unique authentication credentials allocated by a government system.

[0105] The first participation data size can be a local data quantity index provided by the government subject party in federated learning or multi-party collaboration tasks, such as the number of samples, data feature dimensions or the total number of data entries, for evaluating the size of the contribution.

[0106] The first task context information can be background information describing the participation of the government subject party in the task, including the type of task, the training round, the start time and the end time of the task, etc., for assisting consistency checking and contribution calculation.

[0107] The first check result can be a verification result obtained after consistency checking of the first identity identifier, the first participation data size and the first task context information, for judging whether the information submitted by the government subject party is complete, true and in accordance with the preset rules.

[0108] The second identity identifier can be identity information for uniquely identifying the enterprise subject party, and can include a user ID, public key information or other unique authentication credentials allocated by an enterprise system.

[0109] The second participation data size can be a local data quantity index provided by the enterprise subject party in federated learning or multi-party collaboration tasks, such as the number of samples, data feature dimensions or the total number of data entries, for evaluating the size of the contribution.

[0110] The second task context information can be background information describing the participation of the enterprise subject party in the task, including the type of task, the training round, the start time and the end time of the task, etc., for assisting consistency checking and contribution calculation.

[0111] The second verification result can be a verification result obtained after performing consistency checking on the second identity, the second participation data size, and the second task context information, and is used to determine whether the information submitted by the enterprise subject party is complete, true, and consistent with the preset rules.

[0112] The preset verification condition can be a set of rules or thresholds for determining the legality of the participant information, including identity information uniqueness, data size and task context integrity, format specification, and consistency with system records.

[0113] The first contribution commitment value can be an encrypted commitment bound to the government subject party generated based on the first encrypted contribution result of the government subject party and the first public parameter, and is used to prove that the contribution of the government subject party in the task is quantified and tamper-proof.

[0114] The second contribution commitment value can be an encrypted commitment bound to the enterprise subject party generated based on the second encrypted contribution result of the enterprise subject party and the second public parameter.

[0115] The preset contribution calculation model can be a mathematical or calculation model for mapping the contribution commitment value and the public parameter to a verifiable proof, supporting quantitative calculation, proof generation, and verification consistency, and can include a weighting function, a mapping rule, and a hash constraint.

[0116] The preset zero-knowledge proof protocol can be a cryptographic protocol that enables participants to generate a verifiable proof without exposing the actual data content, proving the correctness of their contribution or operation while maintaining privacy and security.

[0117] The first verifiable proof can be a verifiable proof generated by the government subject party based on the first contribution commitment value and the preset contribution calculation model using the zero-knowledge proof protocol, proving the authenticity and consistency of the quantified result of its contribution.

[0118] The second verifiable proof can be a verifiable proof generated by the enterprise subject party based on the second contribution commitment value and the preset contribution calculation model using the zero-knowledge proof protocol.

[0119] In the preset encryption domain, first, the first identity of the government subject party, the first participation data scale, and the first task context information are acquired. The first identity is composed of a unique user ID, a digital certificate, or a public key generated by the government system, which is used to uniquely identify the subject in the system; the first participation data scale includes the number of data samples provided by the government subject party in the federated model training, the dimension of each data, and the total number of data entries; the first task context information includes task type, training round, participation time window, and task phase information. After acquiring the above information, consistency verification is performed in the encryption domain. The specific steps are as follows: first, the first identity is compared with the user ID and public key of the registration center to verify its uniqueness and legality; then, the first participation data scale is compared with the minimum data amount, data format, and dimension constraint in the task requirement to confirm that the data amount and structure meet the specification; next, the first task context information is matched in time, task type, and phase to ensure that it is consistent with the task plan and historical record; finally, the comparison result is recorded as the first verification result, whose value is used to indicate whether the preset verification condition is met (for example, “valid” if all verification items pass, otherwise “invalid”). Correspondingly, the second identity of the enterprise subject party, the second participation data scale, and the second task context information are acquired, and the same comparison and verification steps are performed in the encryption domain to obtain the second verification result.

[0120] When the first verification result and the second verification result both meet the preset verification condition, the first contribution commitment value bound to the government subject party is generated in the encryption domain based on the first encrypted contribution result and the first public parameter. The specific operation is as follows: using an encryption commitment algorithm (such as Pedersen commitment), the first encrypted contribution result is combined with the first public parameter to calculate and generate an unforgeable commitment value, while recording the generation timestamp and adding a random blind factor to ensure zero-knowledge property; similarly, based on the second encrypted contribution result and the second public parameter, the second contribution commitment value bound to the enterprise subject party is generated, and the same algorithm is used to ensure security and unforgeability. Subsequently, the contribution commitment values are quantitatively calculated in the encryption domain using a preset contribution calculation model. The model includes: a contribution amount mapping function that maps the commitment value to a standardized contribution range; a weighted quantization function that generates a quantitative value based on the contribution weight and encryption performance index of each party; a hash constraint used to generate a verifiable digest to prevent tampering.

[0121] After the calculation is completed, a preset zero-knowledge proof protocol is called to perform a commitment-challenge-response process: the system generates a random challenge, the participant generates a response using the commitment value of the contribution, and the first verifiable proof and the second verifiable proof are generated by combining the zero-knowledge proof algorithm to ensure that the authenticity of the contribution can be verified without leaking the original contribution data. Finally, the generated verifiable proof is integrity-encapsulated, including calculating a hash digest of the proof content, generating a digital signature, attaching a timestamp, and packing into the final output of the first contribution degree proof of the government subject and the second contribution degree proof of the enterprise subject.

[0122] The training process of the preset contribution calculation model is as follows:

[0123] First, collect multi-party encrypted contribution data in historical federal training tasks, including encrypted contribution results of each party, corresponding public parameters, actual participation data size, task context information, and historical contribution degree results verified by zero-knowledge proof; then, feature encoding is performed on these data in the encrypted domain, and the contribution result is mapped to a numerical feature vector, and the public parameter, data size, and task context information are mapped to a standardized auxiliary vector; then, the initial parameters of the quantization mapping function and the weighted quantization function are constructed and embedded into the model structure in the encrypted domain; subsequently, the model parameters are iteratively updated in the encrypted domain by gradient descent or other optimization algorithms to minimize the error between the predicted contribution quantization value of the model and the historical verified contribution degree, while adding regularization constraints to ensure the robustness of the model under different data sizes and different task contexts; in each iteration, the mapping output of the contribution commitment value under the model is simulated by encrypted operation to generate a preliminary quantization value; then, the quantization value is subjected to a hash constraint (such as calculating H(quantization value||context)), and consistency comparison is performed with the historical verification record to ensure that the output is tamper-proof and verifiable; at the same time, combined with the zero-knowledge proof protocol, the model output is verified to satisfy the preset logical rules without leaking the plaintext. After the training is completed, the model can input new contribution commitment values and public parameters in the encrypted domain to automatically generate quantized contribution values, while maintaining compatibility with the zero-knowledge proof protocol to ensure that the generated quantized contribution can be used for subsequent contribution degree proof generation and incentive allocation, while ensuring that the original data and intermediate calculations are encrypted throughout to meet the end-to-end privacy protection and verifiability requirements.

[0124] In this scheme, the authenticity and tamper resistance of the contribution are ensured, and end-to-end privacy protection and verifiable incentive allocation are supported, providing a reliable basis for subsequent smart contract execution and auditing.

[0125] Figure 2 A flowchart of a method for federal model contribution evaluation for government-enterprise cooperation provided by the embodiments of the present disclosure is provided. The method can include the following steps:

[0126] S201, obtain first model update data generated by the government subject party when training the federated model based on the first local data, and second model update data generated by the enterprise subject party when training the federated model based on the second local data, encrypt the first model update data and the second model update data using a verifiable encryption method to obtain first encrypted differential data and second encrypted differential data.

[0127] S202, perform secret feature fusion processing on the first encrypted differential data and the second encrypted differential data in a preset encryption domain to obtain a first secret feature result of the government subject party and a second secret feature result of the enterprise subject party.

[0128] S203, perform encryption performance influence evaluation and fair allocation rule quantization on the first secret feature result and the second secret feature result in a preset encryption domain to generate a first encryption contribution result of the government subject party and a second encryption contribution result of the enterprise subject party.

[0129] S204, obtain first public parameters corresponding to the first encryption contribution result and second public parameters corresponding to the second encryption contribution result, generate a first contribution degree proof of the government subject party based on the first encryption contribution result and the first public parameters, and generate a second contribution degree proof of the enterprise subject party based on the second encryption contribution result and the second public parameters.

[0130] S205, submit the first contribution degree proof and the second contribution degree proof to a blockchain network respectively to obtain first on-chain record and second on-chain record stored in a distributed ledger.

[0131] The blockchain network can refer to a decentralized distributed computing network composed of multiple nodes, and the nodes maintain the consistency of the ledger based on a consensus algorithm (such as PoS, PBFT, Raft, etc.), ensuring that the written data cannot be tampered with and is traceable, and usually uses cryptographic hash, digital signature and chain data structure to realize secure storage and verification.

[0132] The distributed ledger can refer to a data storage structure maintained by multiple nodes in the blockchain network, which is characterized by synchronized storage of ledger copies in all participating nodes, and any addition or modification needs to be confirmed by consensus, and tamper-proofing and traceability are achieved through block timestamps and hash pointers.

[0133] The first on-chain record can refer to the chain data record generated after the first contribution degree proof submitted by the government subject party is packaged into a block and consensus-confirmed by the blockchain network. This record contains the hash digest of the contribution degree proof, the submission timestamp, the submitter's identity, and the unique block index number generated by the blockchain.

[0134] The second on-chain record can refer to a data record on the chain generated after the second contribution degree proof submitted by the enterprise subject party is packaged into a block by the blockchain network and consensus confirmed.

[0135] First, the first contribution degree proof and the second contribution degree proof are respectively subjected to hash calculation, and a corresponding hash digest is generated using a secure hash algorithm (such as SHA-256 or SM3) for subsequent on-chain storage verification of data integrity. Next, the digital signature of the submitter is attached to each proof and its hash digest, and the digital signature is generated by combining the private key of the submitter with an elliptic curve signature algorithm (such as ECDSA or SM2) to ensure the authenticity and non-repudiation of the digital signature. Subsequently, the system packs the signed data into a blockchain transaction request, and the transaction request structure includes: transaction initiator identity, transaction timestamp, contribution degree proof hash, contribution degree proof ciphertext digest, and digital signature. The transaction request is broadcast to the entire network through the node broadcast mechanism of the blockchain network, and all nodes receive the transaction and verify its validity through a preset consensus algorithm (such as PBFT or Raft), which includes: checking whether the transaction signature is legal, whether the contribution degree proof hash is unique, and whether the submission time is within the preset time limit. The verified transaction is packed into a new block, and the block timestamp, previous block hash pointer, and Merkle tree root hash are attached to ensure the integrity and tamper resistance of the blockchain structure. After the new block is constructed, the system generates a unique transaction index number for each transaction in the block and establishes a hash mapping relationship between the transaction index number and the transaction content for subsequent quick retrieval and verification. For the transaction request containing the first contribution degree proof submitted by the government subject, the entry in the new block forms the first on-chain storage record; for the transaction request containing the second contribution degree proof submitted by the enterprise subject, the entry in the new block forms the second on-chain storage record. Each on-chain storage record is stored in the distributed ledger in the form of an unalterable data entry, including but not limited to: transaction index number, submitter identity, contribution degree proof hash digest, contribution degree proof ciphertext digest, digital signature verification result, block height, block timestamp, and Merkle path information. The transaction index number is used to uniquely identify the storage entry; the submitter identity ensures the traceability of the storage source; the contribution degree proof hash digest and ciphertext digest are used for subsequent integrity verification; the digital signature verification result indicates the authenticity status of the storage entry; the block height and block timestamp are used to determine the ledger location and time sequence of the storage entry; and the Merkle path information ensures the verifiability of the transaction within the block. When the new block is officially written into the distributed ledger, the first on-chain storage record and the second on-chain storage record are permanently fixed in the blockchain data structure, and any subsequent node can verify the integrity, authenticity, and tamper resistance of the storage record based on the hash chain, Merkle tree verification path, and digital signature.

[0136] S206, performing a secure aggregation operation based on the first summary information of the first on-chain record and the second summary information of the second on-chain record to obtain a joint commitment value, and writing the joint commitment value into the distributed ledger to obtain a joint record.

[0137] The first summary information can refer to a set of summary fields extracted from the first on-chain record for subsequent aggregation calculation, including but not limited to: hash summary of the first contribution proof, ciphertext summary, transaction index number, and Merkle path summary. Its role is to provide a lightweight representation form, so that aggregation calculation and verifiability can be supported without exposing the complete contribution proof content.

[0138] The second summary information refers to a set of summary fields extracted from the second on-chain record for subsequent aggregation calculation, including but not limited to: hash summary of the second contribution proof, ciphertext summary, transaction index number, and Merkle path summary.

[0139] The joint commitment value can refer to a unified commitment result generated in a predetermined encryption domain based on the secure aggregation operation of the first summary information and the second summary information. This commitment value is usually calculated through homomorphic encryption addition operation (such as Paillier homomorphic encryption) or commitment superposition algorithm (such as Pedersen commitment aggregation), with tamper-proof and zero-knowledge properties, which can uniformly bind the record summaries of government and enterprise subjects without revealing the original summary information, for subsequent trusted verification of joint records.

[0140] In a predetermined encryption domain, first extract the first summary information and the second summary information from the first on-chain record and the second on-chain record respectively, including the hash summary, ciphertext summary, transaction index number, and corresponding Merkle path summary of the respective contribution proofs. In order to ensure the security of the summary information in the aggregation calculation, the extracted data needs to be calculated again through a secure hash function (such as SHA-256 or SM3) to obtain a second summary value, which is used to ensure the uniqueness and collision resistance of the input data.

[0141] Next, a secure aggregation operation is performed in the encryption domain. Specifically, if the Paillier homomorphic encryption algorithm is used, the system first encrypts the first and second digest information using a public encryption key to obtain two encrypted digest information; then performs homomorphic addition in the ciphertext space to directly add the encrypted first and second digest information to obtain a joint commitment value in the encrypted state; finally, the system performs partial decryption or zero-knowledge verification on the joint commitment value through a private key or a decentralized decryption authorization mechanism to generate a publicly verifiable joint commitment value. Another way is to use the Pedersen commitment aggregation algorithm: generate a single commitment value using the digest information of each party and an independent random blind factor, and then combine the two commitment values through group operation (such as addition on an elliptic curve group) to obtain a joint commitment value, which naturally has binding and hiding properties, ensuring that the original digest information cannot be forged or reverse-engineered.

[0142] After generating the joint commitment value, the system constructs a new blockchain transaction request, which includes the joint commitment value, the joint commitment generation timestamp, the set of participant identities, the aggregation algorithm identifier, and the verification random factor hash. After the transaction request is broadcast to the blockchain network by the node, all nodes verify its validity through a consensus algorithm (such as PBFT or Raft), which includes checking whether the joint commitment value is generated from the legal digest information, whether the random factor hash matches, and whether the set of participant identities is legal. The verified transaction is packaged into a new block, with a block timestamp, a previous block hash pointer, and a Merkle tree root hash, ensuring data integrity and consistency across the chain. Finally, the new transaction entry written to the distributed ledger constitutes the joint evidence record.

[0143] S207, based on the first contribution degree proof, call the preset incentive distribution smart contract to generate the first incentive settlement record of the government subject party, and based on the second contribution degree proof, call the preset incentive distribution smart contract to generate the second incentive settlement record of the enterprise subject party.

[0144] The preset incentive distribution smart contract can refer to a contract program deployed in the blockchain network in advance, written in a smart contract programming language (such as Solidity or Rust) and running on a blockchain virtual machine (such as Ethereum EVM or Substrate VM). The contract has built-in contribution degree calculation rules, reward distribution functions, settlement trigger conditions, and exception handling mechanisms, which are used to automatically calculate and settle the incentive amount, points, or resources of the participants according to the predefined algorithm rules after receiving the contribution degree proof.

[0145] The first incentive settlement record can refer to an on-chain settlement result generated by the incentive distribution smart contract based on the first contribution degree proof of the government subject party, and the record content includes the identity of the government subject party, the corresponding contribution degree proof hash, the incentive amount or integral value, the generation timestamp, and the contract transaction hash, etc.

[0146] The second incentive settlement record can refer to an on-chain settlement result generated by the incentive distribution smart contract based on the second contribution degree proof of the enterprise subject party, and the record content includes the identity of the enterprise subject party, the corresponding contribution degree proof hash, the incentive amount or integral value, the generation timestamp, and the contract transaction hash, etc.

[0147] The system first performs hash verification on the contribution degree proof before submitting it to the smart contract, calculates the proof digest using a secure hash algorithm (such as SHA-256 or SM3), and compares it with the stored digest record on the chain to ensure that the proof content has not been tampered with. At the same time, the digital signature of the submitter is verified using an elliptic curve signature algorithm (such as ECDSA or SM2) combined with the public key to ensure the authenticity of the source. The system calls the distribution function of the incentive distribution smart contract through a blockchain transaction request. The transaction request structure includes: contribution degree proof hash, submitter identity, transaction timestamp and signature credentials. The transaction request is broadcasted through the blockchain network and distributed to all network nodes, and the contract function is executed by the virtual machine environment of each node. In the execution process of the smart contract, the commitment value and contribution degree parameter in the contribution degree proof are first read, and the incentive distribution rule defined in the contract is called. This rule can include: contribution weight function: weighted calculation according to the contribution degree of the participants and the task setting weight; reward mapping function: mapping the weighted contribution value to a specific reward amount (such as tokens, points or resource shares); threshold checking mechanism: ensuring that participants with a contribution degree below the minimum threshold are not allocated or are allocated a fixed compensation. After the contract calculation is completed, the settlement result is output, including the participant's identity, the reward amount and the hash binding of the corresponding contribution degree proof. After the reward calculation is completed, the contract automatically generates a chain transaction result, which is the incentive settlement record. For the government subject party, the generated result is the first incentive settlement record, which contains the government subject party's identity, the contribution degree proof hash, the reward amount, the settlement timestamp and the transaction hash; for the enterprise subject party, the generated result is the second incentive settlement record, which contains the enterprise subject party's identity, the contribution degree proof hash, the reward amount, the settlement timestamp and the transaction hash. After executing the smart contract, all nodes confirm the validity of the transaction result through a consensus algorithm (such as PBFT or Raft), package the settlement record into a new block, and add the block timestamp, Merkle tree root hash and previous block hash pointer to ensure that the settlement record is tamper-proof and traceable in the distributed ledger. Finally, the government subject party and the enterprise subject party obtain the first incentive settlement record and the second incentive settlement record respectively, which are bound to their own identity and contribution degree proof. These records serve as chain-based trusted credentials.

[0148] In this embodiment, the independent on-chain storage, secure aggregation verification and smart contract automatic incentive of government and enterprise contributions can be realized, taking into account privacy protection and collaborative credibility, and improving the transparency, security and automation level of cross-organizational collaboration.

[0149] On the basis of the above technical solution, optionally, after generating the second incentive settlement record of the enterprise subject party, the method further comprises:

[0150] In the preset incentive distribution smart contract, the first contribution index of the first contribution degree proof and the second contribution index of the second contribution degree proof are analyzed by the built-in rule engine to make an incentive decision, and the first incentive parameter of the government subject party and the second incentive parameter of the enterprise subject party are determined.

[0151] Based on the preset incentive distribution smart contract, the first on-chain incentive voucher of the government subject party and the second on-chain incentive voucher of the enterprise subject party are generated according to the first incentive parameter and the second incentive parameter.

[0152] In the present scheme, the built-in rule engine can refer to a decision calculation component preset in the incentive distribution smart contract, which is used for parameter analysis and rule matching of the input contribution degree proof. Its specific implementation is usually based on rule matching algorithm (such as conditional judgment tree, Boolean logic table, or rule matching calculation based on weight matrix), which can automatically analyze the contribution indexes of different subjects and output corresponding incentive parameters.

[0153] The first contribution index can refer to a set of core parameters extracted from the first contribution degree proof for measuring the actual contribution of the government subject party. For example: data sample quantity, data quality score, task completion degree, training precision improvement rate, etc.

[0154] The second contribution index can refer to a set of core parameters extracted from the second contribution degree proof for measuring the actual contribution of the enterprise subject party, such as: the number of provided computing resources, data coverage rate, model convergence speed improvement rate, etc.

[0155] The first incentive parameter can refer to a numerical value or weight parameter generated by the built-in rule engine after analyzing and rule matching the first contribution index, which can be used for incentive distribution, such as reward points, token quantity or resource allocation share.

[0156] The second incentive parameter can refer to a numerical value or weight parameter generated by the built-in rule engine after analyzing and rule matching the second contribution index, which can be used for incentive distribution.

[0157] The first on-chain incentive voucher can refer to the on-chain incentive record of the government subject party generated according to its first contribution degree proof and the built-in rule engine of the smart contract in the preset incentive distribution smart contract, which includes the identity of the government subject party, the hash value of the corresponding first contribution degree proof, the first incentive parameter value, the incentive distribution amount or points value, the transaction timestamp and the smart contract transaction hash, etc.

[0158] The second on-chain incentive certificate can refer to an on-chain incentive record generated by the enterprise subject party in the preset incentive distribution smart contract according to the second contribution degree proof and a built-in rule engine of the smart contract. The on-chain incentive record includes the enterprise subject party identity, the hash value of the corresponding second contribution degree proof, the second incentive parameter value, the incentive distribution amount or integral value, the transaction timestamp, and the smart contract transaction hash, and the like.

[0159] In the preset incentive distribution smart contract, the first contribution index of the first contribution degree proof and the second contribution index of the second contribution degree proof are first parsed into a data format recognizable by the smart contract, and a built-in rule engine is called in the contract to perform rule matching and logical reasoning processing on the parsed indexes. Specifically, the Boolean logic and conditional judgment are used to compare each contribution index with the preset incentive threshold, and a multi-dimensional scoring function (such as weighted integral calculation or standardized contribution ratio calculation) is used to generate a quantitative incentive decision basis. At the same time, the priority queue and the weighted scheduling algorithm are used to process the conflicts or deviations between different contribution dimensions, so as to ensure that the contribution value of each subject can be comprehensively evaluated in terms of fairness, efficiency and security. According to the evaluation result, the first contribution index and the second contribution index are mapped to the corresponding first incentive parameter and the second incentive parameter, wherein the incentive parameter can include the incentive amount, the integral value or the token distribution ratio, and the timestamp, the random blind factor and the smart contract call record are attached to ensure the on-chain non-tamperability and traceability. After determining the incentive parameter, the smart contract performs integrity verification on each incentive distribution in the encryption domain using the signature mechanism: the incentive parameter is bound with the participant identity, the contribution degree proof hash and the current blockchain state hash to generate a transaction object, and the participant public key is used for encryption signature to ensure the transaction authenticity and non-repudiation; then, the smart contract constructs an on-chain transaction according to the incentive parameter and the bound data, sends the transaction to the blockchain network node for broadcast, and the node verifies the transaction legality, signature validity and contribution proof hash consistency using the consensus algorithm (such as PBFT or Raft). After verification, the smart contract writes the transaction data into a new block, and generates the first on-chain incentive certificate and the second on-chain incentive certificate, each of which includes the subject identity, the contribution degree proof hash, the incentive parameter value, the distribution amount or integral value, the generation timestamp, the transaction hash and the contract call record, so as to form a traceable, transparent and tamper-proof on-chain incentive record in the distributed ledger.

[0160] In the present scheme, the incentive distribution is transparent, fair and traceable, the generated on-chain certificate is efficient and reliable, and the contribution data privacy is protected.

[0161] On the basis of the above technical scheme, after the first on-chain incentive certificate of the government subject party and the second on-chain incentive certificate of the enterprise subject party are generated, the method further includes:

[0162] extracting a first incentive parameter of the first on-chain incentive credential, performing difference analysis on the first incentive parameter and a first contribution indicator to obtain a first difference analysis result;

[0163] extracting a second incentive parameter of the second on-chain incentive credential, performing difference analysis on the second incentive parameter and a second contribution indicator to obtain a second difference analysis result;

[0164] adjusting parameters of a built-in rule engine in a preset incentive distribution smart contract based on the first difference analysis result and the second difference analysis result.

[0165] In the present scheme, the first incentive parameter can refer to a parameter value extracted from the first on-chain incentive credential of the government subject party, which is used to quantify the incentive distribution amount of the government subject party. It usually includes incentive amount, integral number or other reward quantification indicators, and is used to reflect the actual incentive distribution of the government subject party in the contribution proof.

[0166] The first difference analysis result can refer to the difference quantification result obtained by comparing and analyzing the first incentive parameter and the corresponding first contribution indicator, which is used to evaluate the deviation between the actual incentive distribution and the expected contribution, including absolute difference, relative deviation or standardized error, etc.

[0167] The second incentive parameter can refer to a parameter value extracted from the second on-chain incentive credential of the enterprise subject party, which is used to quantify the incentive distribution amount of the enterprise subject party.

[0168] The second difference analysis result can refer to the difference quantification result obtained by comparing and analyzing the second incentive parameter and the corresponding second contribution indicator.

[0169] The parameters of the built-in rule engine can refer to adjustable numerical values or weights in the preset incentive distribution smart contract, which are used to control the incentive decision logic and distribution strategy, including threshold setting, contribution weight, proportion coefficient and incentive upper / lower limit, etc. By adjusting these parameters, the incentive distribution strategy can be optimized and the incentive deviation can be corrected.

[0170] In the preset incentive distribution smart contract, first, the first incentive parameter is parsed and extracted from the incentive certificate on the first chain, and the specific operation is to read the stored smart contract transaction data on the chain, including the identity of the government subject party, the corresponding contribution proof hash and the incentive distribution value field, and map the incentive amount or integral value field to a standardized numerical vector to obtain the quantized first incentive parameter. Then, the first incentive parameter and the corresponding first contribution index of the government subject party are analyzed. The specific method is as follows: first, the first incentive parameter and the first contribution index are normalized to eliminate dimensional differences; then, the difference between the two is calculated by using the difference calculation or relative deviation formula, for example, the absolute difference or relative error of each index dimension is calculated to form the first difference analysis result, which is used to quantify the deviation between the actual distribution of incentives and the expected contribution. Similarly, the second incentive parameter is parsed and extracted from the incentive certificate on the second chain, the identity of the enterprise subject party, the contribution proof hash and the incentive distribution value field are read, and after the incentive value is mapped to a standardized vector, the second difference analysis result is obtained by analyzing the difference between the corresponding second contribution index through normalization, difference and relative deviation calculation. After obtaining the first difference analysis result and the second difference analysis result, the parameters of the built-in rule engine in the preset incentive distribution smart contract are adjusted, and the specific operation includes: identifying the incentive deviation mode according to the difference analysis result, calculating the correction coefficient of each rule weight or threshold, mapping the absolute difference or relative error to the adjustment amount of the contribution weight, proportion coefficient or upper and lower limit of the incentive in the rule engine, and updating the parameter value stored in the smart contract to realize the adaptive optimization of the incentive distribution strategy, so that the future incentive distribution is closer to the actual contribution of each participant, and the distribution fairness and system sustainability are ensured. The whole process is executed in the encryption environment of the blockchain ledger and the smart contract, ensuring the integrity of the data on the chain and the verifiability of the operation.

[0171] In the present scheme, the incentive distribution rules can be dynamically adjusted, the incentive parameters can be more accurately matched with the actual contribution, the distribution fairness and transparency can be improved, and the safety and verifiability of the data on the chain can be maintained.

[0172] Figure 3 A federal model contribution evaluation system for government-enterprise cooperation provided by the embodiment of the present disclosure is shown in the schematic block diagram. The system comprises:

[0173] The data encryption module 301 is configured to obtain first model update data generated by the government subject party when training the federal model based on the first local data, and second model update data generated by the enterprise subject party when training the federal model based on the second local data, and encrypt the first model update data and the second model update data by using a verifiable encryption method to obtain first encrypted difference data and second encrypted difference data.

[0174] The secure state feature fusion module 302 is configured to perform secure state feature fusion processing on the first encrypted differential data and the second encrypted differential data in a preset encryption domain to obtain a first secure state feature result of the government subject and a second secure state feature result of the enterprise subject.

[0175] The encrypted contribution result generation module 303 is configured to perform encrypted performance influence evaluation and fair allocation rule quantization on the first secure state feature result and the second secure state feature result in the preset encryption domain to generate a first encrypted contribution result of the government subject and a second encrypted contribution result of the enterprise subject.

[0176] The contribution degree proof generation module 304 is configured to obtain a first public parameter corresponding to the first encrypted contribution result and a second public parameter corresponding to the second encrypted contribution result, generate a first contribution degree proof of the government subject based on the first encrypted contribution result and the first public parameter, and generate a second contribution degree proof of the enterprise subject based on the second encrypted contribution result and the second public parameter.

[0177] As shown in Figure 4 The electronic device 400 includes a processor 401, a memory 402, and a program or instruction stored in the memory 402 and executable on the processor 401. The program or instruction is executed by the processor 401 to implement each process of the above-mentioned federal model contribution evaluation method for government-enterprise cooperation, and achieve the same technical effect. To avoid repetition, details are not repeated here.

[0178] It should be noted that the electronic device in the present embodiment includes the mobile electronic device and the non-mobile electronic device described above.

[0179] The present embodiment further provides a readable storage medium having a program or instruction stored thereon. The program or instruction is executed by a processor to implement each process of the above-mentioned federal model contribution evaluation system for government-enterprise cooperation, and achieve the same technical effect. To avoid repetition, details are not repeated here.

[0180] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes a computer readable storage medium, such as a computer readable memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, etc.

[0181] It should be noted that, in the present document, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or system. An element proceeded by "comprises a", "comprising", or "comprises" does not, without more constraints, preclude the existence of additional identical elements in the process, method, article, or system that comprises the element. Additionally, it should be noted that the scope of the methods and systems of the present embodiments are not limited by the order of the steps or the order of the functions thereof, as shown or discussed, and that the present embodiments reside in the method and system themselves in their various forms.

[0182] From the above description of the embodiments, it is clear that the above-described method of the embodiments can be realized by means of software and the necessary universal hardware platform, of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, or network device, etc.) execute the method described in each embodiment of the present application.

[0183] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative, not restrictive, and those of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the protection scope of the claims.

[0184] The above are only the preferred embodiments of the present application and the technical principles used. The present application is not limited to the specific embodiments described herein, and various obvious changes, re-adjustments and replacements made by those skilled in the art will not deviate from the scope of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without deviating from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for evaluating the contribution of a federal model for government-enterprise collaboration, characterized in that, The method includes: The system obtains the first model update data generated by the government entity when training the federated model based on the first local data, and the second model update data generated by the enterprise entity when training the federated model based on the second local data. The first model update data and the second model update data are encrypted using a verifiable encryption method to obtain the first encrypted differential data and the second encrypted differential data. Within a preset encryption domain, the first encrypted differential data and the second encrypted differential data are subjected to encrypted feature fusion processing to obtain the first encrypted feature result of the government entity and the second encrypted feature result of the enterprise entity; wherein, the first encrypted differential data is subjected to a shallow nonlinear transformation within the preset encryption domain to obtain the first preliminary encrypted feature representation of the government entity, and the second encrypted differential data is subjected to a shallow nonlinear transformation within the preset encryption domain to obtain the second preliminary encrypted feature representation of the enterprise entity; Within a preset encryption domain, an affine transformation is performed on the first preliminary encrypted feature representation and the second preliminary encrypted feature representation based on preset global statistical parameters to obtain the first normalized encrypted feature representation of the government entity and the second normalized encrypted feature representation of the enterprise entity. Within a preset encryption domain, the first normalized encrypted feature representation and the second normalized encrypted feature representation are projected onto a unified encrypted semantic space through a preset linear mapping matrix, respectively, to obtain the first mapped encrypted feature representation of the government entity and the second mapped encrypted feature representation of the enterprise entity. Within a preset encryption domain, the first mapping cryptographic feature representation and the second mapping cryptographic feature representation are subjected to dimensional coordination processing to obtain the first aligned cryptographic feature representation of the government entity and the second aligned cryptographic feature representation of the enterprise entity. Within a preset encryption domain, the first aligned encrypted feature representation and the second aligned encrypted feature representation are weighted and combined according to a preset fusion weighting rule to obtain the first encrypted feature result of the government entity and the second encrypted feature result of the enterprise entity. Within a preset encryption domain, the encryption performance impact of the first and second encrypted feature results is evaluated and the fair allocation rules are quantified to generate the first encryption contribution result of the government entity and the second encryption contribution result of the enterprise entity. Obtain the first public parameter corresponding to the first encrypted contribution result and the second public parameter corresponding to the second encrypted contribution result; generate a first contribution certificate for the government entity based on the first encrypted contribution result and the first public parameter; and generate a second contribution certificate for the enterprise entity based on the second encrypted contribution result and the second public parameter.

2. The method according to claim 1, characterized in that, in, Within a preset encryption domain, the encryption performance impact of the first and second encrypted feature results is evaluated, and fair allocation rules are quantified to generate the first encryption contribution result for government entities and the second encryption contribution result for enterprise entities, including: Within a preset encryption domain, the encryption performance impact of the first and second encrypted feature results is evaluated based on preset performance evaluation indicators, resulting in a first performance impact score for the government entity and a second performance impact score for the enterprise entity. Within a preset encrypted domain, the first performance impact score, the second performance impact score, and the preset fair allocation rules are input into a preset game theory allocation model to generate the first encrypted contribution result for the government entity and the second encrypted contribution result for the enterprise entity.

3. The method according to claim 1, characterized in that, in, Before generating a first contribution certificate for the government entity based on the first encrypted contribution result and the first public parameter, and before generating a second contribution certificate for the enterprise entity based on the second encrypted contribution result and the second public parameter, the method further includes: The system obtains the first identity identifier, the first participation data scale, and the first task context information of the government entity, performs a consistency check on the first identity identifier, the first participation data scale, and the first task context information, and obtains a first check result. The system also obtains the second identity identifier, the second participation data scale, and the second task context information of the enterprise entity, performs a consistency check on the second identity identifier, the second participation data scale, and the second task context information, and obtains a second check result. Accordingly, a first contribution certificate for the government entity is generated based on the first encrypted contribution result and the first public parameter, and a second contribution certificate for the enterprise entity is generated based on the second encrypted contribution result and the second public parameter, including: If the first verification result and the second verification result meet the preset verification conditions, a first contribution commitment value bound to the government entity is generated based on the first encrypted contribution result and the first public parameter, and a second contribution commitment value bound to the enterprise entity is generated based on the second encrypted contribution result and the second public parameter. Based on a preset contribution calculation model, a first contribution commitment value, and a first public parameter, a first verifiable proof is generated by calling a preset zero-knowledge proof protocol; and based on a preset contribution calculation model, a second contribution commitment value, and a second public parameter, a second verifiable proof is generated by calling a preset zero-knowledge proof protocol. Perform integrity encapsulation on the first verifiable proof and the second verifiable proof to obtain the first contribution proof of the government entity and the second contribution proof of the enterprise entity.

4. The method according to claim 1, characterized in that, in, After generating the second contribution proof of the enterprise entity based on the second encrypted contribution result and the second public parameters, the method further includes: The first contribution proof and the second contribution proof are submitted to the blockchain network respectively to obtain the first on-chain evidence record and the second on-chain evidence record stored in the distributed ledger. Based on the first digest information of the first on-chain evidence record and the second digest information of the second on-chain evidence record, a secure aggregation operation is performed to obtain a joint commitment value, and the joint commitment value is written into the distributed ledger to obtain a joint evidence record; Based on the first contribution proof, a preset incentive allocation smart contract is invoked to generate a first incentive settlement record for the government entity, and based on the second contribution proof, a preset incentive allocation smart contract is invoked to generate a second incentive settlement record for the enterprise entity.

5. The method according to claim 4, characterized in that, in, After generating the second incentive settlement record for the corporate entity, the method further includes: In the pre-set incentive allocation smart contract, the built-in rule engine performs incentive decision analysis on the first contribution index of the first contribution proof and the second contribution index of the second contribution proof to determine the first incentive parameters for the government entity and the second incentive parameters for the enterprise entity. Based on the preset incentive allocation smart contract, the incentive allocation operation is performed according to the first incentive parameter and the second incentive parameter to generate the first on-chain incentive certificate for the government entity and the second on-chain incentive certificate for the enterprise entity.

6. The method according to claim 5, characterized in that, in, After generating the first on-chain incentive certificate for the government entity and the second on-chain incentive certificate for the enterprise entity, the method further includes: Extract the first incentive parameters from the first chain incentive certificate, perform a difference analysis on the first incentive parameters and the first contribution index, and obtain the first difference analysis result. Extract the second incentive parameters from the incentive certificate on the second chain, perform a difference analysis on the second incentive parameters and the second contribution index, and obtain the second difference analysis results. Based on the results of the first and second difference analyses, the parameters of the built-in rule engine in the preset incentive allocation smart contract are adjusted.

7. A contribution evaluation system for a federal model of government-enterprise collaboration, characterized in that, The system includes: The data encryption module is used to obtain the first model update data generated by the government entity when training the federated model based on the first local data, and the second model update data generated by the enterprise entity when training the federated model based on the second local data. The first model update data and the second model update data are encrypted using a verifiable encryption method to obtain the first encrypted differential data and the second encrypted differential data. The encrypted feature fusion module is used to perform encrypted feature fusion processing on the first encrypted differential data and the second encrypted differential data within a preset encrypted domain to obtain the first encrypted feature result of the government entity and the second encrypted feature result of the enterprise entity; wherein, it includes performing a shallow nonlinear transformation on the first encrypted differential data within the preset encrypted domain to obtain the first preliminary encrypted feature representation of the government entity, and performing a shallow nonlinear transformation on the second encrypted differential data within the preset encrypted domain to obtain the second preliminary encrypted feature representation of the enterprise entity; Within a preset encryption domain, an affine transformation is performed on the first preliminary encrypted feature representation and the second preliminary encrypted feature representation based on preset global statistical parameters to obtain the first normalized encrypted feature representation of the government entity and the second normalized encrypted feature representation of the enterprise entity. Within a preset encryption domain, the first normalized encrypted feature representation and the second normalized encrypted feature representation are projected onto a unified encrypted semantic space through a preset linear mapping matrix, respectively, to obtain the first mapped encrypted feature representation of the government entity and the second mapped encrypted feature representation of the enterprise entity. Within a preset encryption domain, the first mapping cryptographic feature representation and the second mapping cryptographic feature representation are subjected to dimensional coordination processing to obtain the first aligned cryptographic feature representation of the government entity and the second aligned cryptographic feature representation of the enterprise entity. Within a preset encryption domain, the first aligned encrypted feature representation and the second aligned encrypted feature representation are weighted and combined according to a preset fusion weighting rule to obtain the first encrypted feature result of the government entity and the second encrypted feature result of the enterprise entity. The encryption contribution result generation module is used to evaluate the encryption performance impact of the first encryption state feature result and the second encryption state feature result and quantify the fair allocation rules within a preset encryption domain, and generate the first encryption contribution result of the government entity and the second encryption contribution result of the enterprise entity. The contribution proof generation module is used to obtain the first public parameter corresponding to the first encrypted contribution result and the second public parameter corresponding to the second encrypted contribution result, generate the first contribution proof of the government entity based on the first encrypted contribution result and the first public parameter, and generate the second contribution proof of the enterprise entity based on the second encrypted contribution result and the second public parameter.

8. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the federal model contribution assessment method for government-business collaboration as described in any one of claims 1-6.

9. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the federal model contribution assessment method for government-enterprise collaboration as described in any one of claims 1-6.

Images