A bid evaluation expert extraction method and system combining image matching and random extraction

Abstract

This invention relates to the field of expert selection technology, specifically a method and system for selecting expert bidders that integrates profile matching and random sampling. The method includes: a bid evaluation platform generating a project profile based on the requirements of the bidding project; each expert database node, within a trusted execution environment, generating a candidate expert index set containing expert indexes and corresponding weights based on the project profile, and simultaneously sending the node weights of the candidate expert index set and locally generated random numbers to the bid evaluation platform; the bid evaluation platform performing globally weighted random sampling through a multi-party secure computation protocol to determine the matched expert database nodes and their corresponding offsets; the expert database nodes determining the winning experts in their local candidate expert index sets based on the offsets, generating and returning encrypted tokens to the bid evaluation platform; and the bid evaluation platform jointly decrypting the encrypted tokens to obtain information on all winning experts. This invention protects the data sovereignty of each expert database while achieving a more impartial cross-regional expert selection process.

Description

Technical Field

[0001] This invention relates to the field of expert selection technology, and in particular to a method and system for selecting expert selection that integrates profile matching and random selection. Background Technology

[0002] In bidding processes, the selection of evaluation experts is a crucial step in ensuring the impartiality of the review. Traditionally, the bidding party or its agent randomly selects experts from a local expert database based on pre-defined criteria such as expertise and location, and then manually notifies the experts. This selection process relies on centralized systems or manual operation; the selection rules and the generation of the random seed are opaque, posing a risk of backend tampering or human intervention, making it difficult to guarantee the absolute fairness and auditability of the selection results.

[0003] With the increasing adoption of remote, off-site bidding, establishing a unified directory of shared expert resources has become essential. However, it is difficult to accurately verify the avoidance relationships among off-site experts under conditions of data isolation. Furthermore, existing systems often employ simple random sampling or weighted round-robin, leading to uneven utilization of expert resources.

[0004] Therefore, a more equitable method for selecting cross-regional bidding experts is needed, while protecting the data sovereignty of each expert database. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides a method and system for selecting evaluation experts that integrates profile matching and random sampling, achieving a more impartial cross-regional selection of evaluation experts while protecting the data sovereignty of each expert database.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for selecting evaluation experts that integrates profile matching and random sampling, comprising: The bidding platform generates a project profile based on the requirements of the bidding project and broadcasts the project profile to various expert database nodes. Within the trusted execution environment, each expert database node uses a hierarchical sensitive hash algorithm to map the locally stored expert avoidance association information into a privacy feature bit vector, and performs privacy matching with the project profile to generate a candidate expert index set containing expert indexes and corresponding weights. At the same time, the node weights of the candidate expert index set and the locally generated random number fragments are sent to the bidding platform. The evaluation platform uses a multi-party secure computation protocol to perform global weighted random sampling based on the node weights returned by all expert database nodes and random number sharding to determine the hit expert database node and its corresponding offset, and then sends the offset to the hit expert database node. If the expert database node receives the offset, it determines the winning expert in the local candidate expert index set based on the offset, and generates and returns an encrypted token containing the winning expert's information to the evaluation platform. The evaluation platform jointly decrypts the encrypted tokens returned by each expert database node to obtain information on all winning experts.

[0007] The beneficial effects of this invention are as follows: A project profile is generated by the bidding evaluation platform and broadcast to each expert database node. Each node, within a trusted execution environment, uses a hierarchical sensitive hashing algorithm to perform privacy feature mapping on local expert avoidance information and matches it with the profile. After generating a candidate expert set, only the node weights and random number shards are sent to the bidding evaluation platform. Without disclosing any original plaintext expert data, collaborative extraction across multiple regional expert database nodes is achieved, completely eliminating the possibility of human intervention in centralized systems. Simultaneously, the bidding evaluation platform performs globally weighted random sampling through a multi-party secure computation protocol, sending the offset only to the winning node. The winning node locally determines the winning expert and returns an encrypted token. Finally, joint decryption yields all expert information. This globally weighted random sampling ensures the balanced use of expert resources, thereby achieving a more equitable cross-regional expert extraction while protecting the data sovereignty of each expert database.

[0008] Optionally, the evaluation platform uses a multi-party secure computation protocol to perform globally weighted random sampling based on the sum of node weights returned by all expert database nodes and random number sharding to determine the hit expert database node and its corresponding offset, including: Within the encrypted domain, the random number fragments returned by each expert database node are aggregated to generate a global random seed. The node weights returned by each expert database node are aggregated to generate a global total weight. The global random seed is modulo the global total weight to obtain the target landing point value. Within the encrypted domain, the interval boundaries of each expert database node are determined based on the sum of the node weights of each expert database node. The expert database node whose interval boundary the target landing point value falls into is determined as the hit, and the offset is determined based on the interval boundary of the target landing point value and the hit expert database node.

[0009] As described above, this invention completes all calculations from random number shard generation to hit node determination within the encrypted domain. Each expert database node cannot know the weight sum of other expert database nodes, nor can it predict the value of the random number shard, thus ensuring both the randomness and fairness of the extraction results.

[0010] Optionally, the step of performing privacy matching with the project profile to generate a candidate expert index set containing expert indexes and corresponding weights includes: The privacy feature bit vector is compared with the feature bit vector corresponding to the avoidance condition in the project profile by performing bit operations, and the bit operation result is used to filter experts with avoidance relationships. A non-linear scoring function based on expert credit score and professional matching degree scores other filtered experts, generating a candidate expert index set containing expert indexes and corresponding weights.

[0011] As described above, this invention transforms complex unstructured avoidance logic into efficient computable feature vectors, achieves rapid judgment of avoidance relationships through bitwise operations, and introduces a multi-dimensional scoring mechanism to ensure the professional quality of candidate experts, significantly improving the accuracy and efficiency of avoidance rule verification in cross-regional extraction.

[0012] Optionally, determining the winning expert in the local candidate expert index set based on the offset includes: The expert database node constructs a cumulative weight sequence locally according to the weights of the candidate experts, and determines the winning expert based on the position of the offset in the cumulative weight sequence.

[0013] Optionally, the global random seed The generating function is: ; In the formula, n is the number of expert database nodes, and r k The random number partition is the k-th expert database node, and L is the preset security parameter.

[0014] Optionally, the corresponding weights of the candidate experts are: Each expert database node assigns a weight to each candidate expert based on at least one of the following dimensions: the candidate expert's historical evaluation score, years of professional experience, and regional balance requirements.

[0015] Optionally, the encryption and joint decryption of the encryption token includes: The evaluation platform secretly shares the global decryption private key into private key fragments equal to the number of winning experts, and sends the private key fragments encrypted with the node public key and the global public key to the matched expert database nodes when sending the offset. The expert database node that hits the target node uses the global public key to encrypt the information of the winning expert, generates an encrypted token, and returns the encrypted token and its own private key fragment to the evaluation platform. After the evaluation platform collects at least a threshold number of private key fragments, it recovers the global decryption private key using the Lagrange interpolation algorithm, and then decrypts all encrypted tokens using the global decryption private key.

[0016] As described above, this invention implements a threshold decryption mechanism that prevents the private key from leaving the node. Only through multi-party collaboration can a sufficient number of private key fragments be collected to form a global decryption private key, thereby ensuring the transparency and auditability of the extraction results while protecting the security of expert information.

[0017] Optionally, it also includes: The global random seed, the weights of each node, the node identifier and offset of the hit expert database node, and the hash value of all winning expert information are synchronously uploaded to the blockchain.

[0018] Optionally, the expert avoidance association information includes the expert's affiliated unit, the company where his / her immediate family members work, and information on related projects he / she has participated in within a preset number of years.

[0019] Secondly, the present invention provides a system for selecting evaluation experts that integrates profile matching and random selection, comprising: The evaluation platform is used to generate project profiles based on the requirements of the bidding project and broadcast the project profiles to various expert database nodes; through a multi-party secure computation protocol, it performs global weighted random sampling based on the sum of node weights returned by all expert database nodes and random number sharding to determine the hit expert database nodes and their corresponding offsets, and sends the offsets to the hit expert database nodes; it also jointly decrypts the encrypted tokens returned by each expert database node to obtain information on all winning experts. Multiple expert database nodes are used within a trusted execution environment to map locally stored expert avoidance association information into privacy feature bit vectors using a hierarchical sensitive hash algorithm. These vectors are then matched with the project profile for privacy purposes, generating a candidate expert index set containing expert indexes and corresponding weights. Simultaneously, the node weights of the candidate expert index set and locally generated random number fragments are sent to the bidding platform. If the offset is received, the winning expert is determined in the local candidate expert index set based on the offset, and an encrypted token containing the winning expert's information is generated and returned to the bidding platform.

[0020] The technical effects of the bid evaluation expert selection system that integrates profile matching and random sampling provided in the second aspect are described in the relevant description of the bid evaluation expert selection method that integrates profile matching and random sampling provided in the first aspect. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the main process of a method for selecting evaluation experts that integrates profile matching and random sampling according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a bid evaluation expert extraction system that integrates profile matching and random sampling according to an embodiment of the present invention.

[0022] Explanation of reference numerals in the attached figures: 1. A system for selecting evaluation experts that integrates profile matching and random selection; 2. Bidding evaluation platform; 3. Expert database node. Detailed Implementation

[0023] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0024] Example 1 Please refer to Figure 1 A method for selecting evaluation experts that integrates profile matching and random sampling includes the following steps: S1. The bidding platform generates a project profile based on the requirements of the bidding project and broadcasts the project profile to various expert database nodes.

[0025] In this embodiment, the bidding evaluation platform of the transaction center initiates a request to select experts at the bidding site, and then generates a project profile according to the requirements of the bidding project. The project profile is then encrypted and broadcast to each expert database node.

[0026] S2. Within the trusted execution environment, each expert database node uses a hierarchical sensitive hash algorithm to map the locally stored expert avoidance association information into a privacy feature bit vector, and performs privacy matching with the project profile to generate a candidate expert index set containing expert indexes and corresponding weights. At the same time, the node weights of the candidate expert index set and the locally generated random number shards are sent to the bidding platform.

[0027] The Trusted Execution Environment (TEE) can be Kunpeng TrustZone, Hygon CSV (China Security Virtualization), etc.

[0028] Among them, the Hierarchical-Sensitive Hashing (LSH) algorithm is an algorithm that can map logical topology algorithms into fixed-dimensional privacy feature bits. Expert avoidance association information includes the expert's affiliated units, the companies where their immediate family members work, and information on related projects they have participated in within a preset number of years. This can form a multi-dimensional expert avoidance conflict graph, which is then mapped into fixed-dimensional privacy feature bits. This not only eliminates the need for plaintext expert information but also significantly reduces the computational overhead of cross-node comparisons, achieving more private and faster expert avoidance. Therefore, in step S2, privacy matching is performed with the project profile to generate a candidate expert index set containing expert indices and corresponding weights, including: Step S21: Perform bitwise operations on the privacy feature bit vector and compare it with the feature bit vector corresponding to the avoidance condition in the project profile. Instruct the experts with avoidance relationships to filter the results of the bitwise operations.

[0029] Step S22: Based on the nonlinear scoring function of expert credit score and professional matching degree, score the other filtered experts to generate a candidate expert index set containing expert index and corresponding weight.

[0030] Therefore, this embodiment not only uses the Private Set Intersection (PSI) technique, but also ensures the professional quality of candidate experts by filtering through hard avoidance before continuing the matching process, which significantly improves the accuracy and efficiency of avoidance rule verification in cross-regional extraction.

[0031] The evaluation criteria for selected experts can be flexibly set according to different bidding platforms and different bidding projects. In one example of this invention, the corresponding weights of the candidate experts are: Each expert database node assigns a weight to each candidate expert based on at least one of the following dimensions: the candidate expert's historical evaluation score, years of professional experience, and regional balance requirements.

[0032] Among them, the node weights and W of the candidate expert index set k for: ; In the formula, w k,i Let be the weight of the i-th candidate expert.

[0033] S3. The evaluation platform uses a multi-party secure computation protocol to perform global weighted random sampling based on the node weights and random number shards returned by all expert database nodes to determine the hit expert database node and its corresponding offset, and then sends the offset to the hit expert database node.

[0034] The Multi-Party Computation (MPC) protocol allows multiple participants to collaboratively compute a function and obtain the correct result without disclosing their private input data. Therefore, in step S3 of this embodiment, the evaluation platform uses the MPC protocol to perform globally weighted random sampling based on the node weights returned by all expert database nodes and random number sharding to determine the matched expert database node and its corresponding offset, including: S31. Within the encrypted domain, aggregate the random number fragments returned by each expert database node to generate a global random seed. Aggregate the node weights returned by each expert database node to generate a global total weight. Perform a modulo operation on the global random seed and the global total weight to obtain the target landing point value.

[0035] Among them, the global total weight W total The generating function is: .

[0036] Among them, global random seed The generating function is: ; In the formula, n is the number of expert database nodes, and r k The random number partition is the k-th expert database node, and L is the preset security parameter.

[0037] In a simple example, suppose there are expert database nodes A, B, and C. Expert database node A has experts a1 and a2 with weights of 1 and 6 respectively. Expert database node B has experts b1, b2, and b3 with weights of 2, 3, and 6 respectively. Expert database node C has experts c1 and c2 with weights of 1 and 2 respectively. The sum of the node weights of expert database nodes A, B, and C is 7, 11, and 3, respectively, resulting in a total global weight of 21. If the global random seed R is 334, then taking 334 modulo 21 yields a target landing point value of 19.

[0038] S32. Within the encrypted domain, determine the interval boundaries of each expert database node based on the sum of the node weights of each expert database node, identify the expert database node whose interval boundary the target landing point value falls into as the hit, and determine the offset based on the target landing point value and the interval boundary of the hit expert database node.

[0039] The function for determining whether a match has occurred is as follows: f(T,k)=IsGreaterEqual(T,S k-1 )∧IsLess(T,S k ); In the formula, T is the target landing point value, and S... k and Sk-1 Let be the interval boundary of the k-th expert database node, where, since the modulo operation starts from 0, the interval boundary is also accumulated starting from 0.

[0040] Based on the example above, the interval boundaries of expert database nodes A, B, and C are [0,6], [7,17], and [18,20], respectively. The target value 19 falls within the interval boundary of expert database node C, so expert database node C is hit, and thus the offset is confirmed to be 1.

[0041] Therefore, this embodiment completes all calculations from random number shard generation to hit node determination within the encrypted domain. Each expert database node cannot know the weight sum of other expert database nodes, nor can it predict the value of the random number shard, thus ensuring both the randomness and fairness of the extraction results.

[0042] S4. If the expert database node receives the offset, it determines the winning expert in the local candidate expert index set based on the offset, and generates and returns an encrypted token containing the winning expert's information to the evaluation platform.

[0043] Among these, determining the winning expert based on the offset in the local candidate expert index set includes: The expert database nodes construct a cumulative weight sequence locally according to the weights of the candidate experts, and determine the winning expert based on the position of the offset in the cumulative weight sequence.

[0044] In this embodiment, the cumulative weight sequence of expert database node C can be {0, 1, 2}, where 0 represents expert c1, and 1 and 2 represent experts c2. With an offset of 1, expert c2 is the winning expert. In other embodiments, the cumulative weight sequence of expert database node C can be {c1, c2, c3}, in which case the expert with an offset of 1 is c2.

[0045] This includes generating and returning an encrypted token containing the information of the winning expert to the bid evaluation platform, including: The evaluation platform secretly shares the global decryption private key into private key fragments equal to the number of winning experts, and sends the private key fragments encrypted with the node public key and the global public key to the matched expert database nodes when sending the offset. The expert database node that hits the target node uses the global public key to encrypt the information of the winning expert, generates an encrypted token, and returns the encrypted token and its own private key fragment to the evaluation platform.

[0046] The bidding evaluation platform generates a corresponding number of private key fragments based on the number of experts needed, but the platform itself does not store the complete private keys. The node public keys refer to the public keys of each expert database node stored on the bidding evaluation platform, used for encrypted communication between them.

[0047] Therefore, the other missing nodes are completely unaware of the global sampling trajectory.

[0048] S5. The evaluation platform jointly decrypts the encrypted tokens returned by each expert database node to obtain information on all winning experts.

[0049] Step S5 includes: After the evaluation platform collects at least the threshold number of private key fragments, it recovers the global decryption private key using the Lagrange interpolation algorithm. It then uses the global decryption private key to decrypt all encrypted tokens and obtain all the information of the winning experts.

[0050] Therefore, the evaluation platform can only form a global decryption private key by collecting enough private key fragments, thus ensuring the openness, transparency and auditability of the extraction results while protecting the security of expert information.

[0051] After the selection is confirmed, the evaluation platform sends a text message notification to the winning experts through a preset encrypted gateway, or conducts biometric authentication when the winning experts arrive at the remote evaluation room.

[0052] S6. Synchronously upload the global random seed, the sum of the weights of each node, the node identifier and offset of the hit expert database node, and the hash value of all winning expert information to the blockchain.

[0053] The computational fingerprint of the entire extraction process is uploaded to the blockchain in real time, ensuring that the extraction logic is auditable but the data content is not visible.

[0054] In summary, this embodiment, under the premise that the original plaintext data of experts never leaves the local node, enables each transaction center to complete cross-regional and cross-domain expert resource sharing and distributed consensus extraction without changing data ownership or disclosing expert privacy. At the same time, it provides hardware-level isolation through a trusted execution environment and achieves transparency, auditability and immutability of the extraction process through multi-party secure computation and blockchain notarization. Thus, while protecting the data sovereignty of each expert database, it achieves a more equitable cross-regional expert selection for bidding.

[0055] Example 2 Please refer to Figure 2 A system for selecting evaluation experts that integrates profile matching and random sampling 1, comprising: The evaluation platform 2 is used to generate a project profile based on the requirements of the bidding project and broadcast the project profile to each expert database node 3; through a multi-party secure computation protocol, it performs global weighted random sampling based on the node weights and random number shards returned by all expert database nodes 3 to determine the hit expert database node 3 and its corresponding offset, and sends the offset to the hit expert database node 3; it jointly decrypts the encrypted tokens returned by each expert database node 3 to obtain the information of all winning experts; Multiple expert database nodes 3 are used in a trusted execution environment to map locally stored expert avoidance association information into privacy feature bit vectors using a hierarchical sensitive hash algorithm, and perform privacy matching with project profiles to generate a candidate expert index set containing expert indexes and corresponding weights. At the same time, the node weights of the candidate expert index set and the locally generated random number fragments are sent to the bidding evaluation platform 2. If an offset is received, the winning expert is determined in the local candidate expert index set according to the offset, and an encrypted token containing the winning expert information is generated and returned to the bidding evaluation platform 2.

[0056] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0057] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0058] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0059] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0060] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0061] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0062] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A method for selecting evaluation experts that integrates profile matching and random sampling, characterized in that, include: The bidding platform generates a project profile based on the requirements of the bidding project and broadcasts the project profile to various expert database nodes. Each expert database node, within a trusted execution environment, uses a hierarchical sensitive hashing algorithm to map locally stored expert avoidance association information into privacy feature bit vectors, and performs privacy matching with the project profile to generate a candidate expert index set containing expert indices and corresponding weights. Simultaneously, the node weights of the candidate expert index set and locally generated random number shards are sent to the evaluation platform. The privacy matching with the project profile to generate the candidate expert index set containing expert indices and corresponding weights includes: The privacy feature bit vector is compared with the feature bit vector corresponding to the avoidance condition in the project profile by performing bit operations, and the bit operation result is used to filter experts with avoidance relationships. A non-linear scoring function based on expert credit score and professional matching degree scores other filtered experts, generating a candidate expert index set containing expert index and corresponding weight; The evaluation platform uses a multi-party secure computation protocol to perform globally weighted random sampling based on the sum of node weights returned by all expert database nodes and random number sharding. This determines the matched expert database node and its corresponding offset, and the offset is then sent to the matched expert database node. This process includes: Within the encrypted domain, the random number fragments returned by each expert database node are aggregated to generate a global random seed. The node weights returned by each expert database node are aggregated to generate a global total weight. The global random seed is modulo the global total weight to obtain the target landing point value. Within the encrypted domain, the interval boundaries of each expert database node are determined based on the sum of the node weights of each expert database node. The expert database node whose interval boundary the target landing point value falls into is determined as the hit, and the offset is determined based on the interval boundary of the target landing point value and the hit expert database node. If the expert database node receives the offset, it determines the winning expert in the local candidate expert index set based on the offset, and generates and returns an encrypted token containing the winning expert's information to the evaluation platform. The evaluation platform jointly decrypts the encrypted tokens returned by each expert database node to obtain information on all winning experts. The function for determining whether a match has occurred is as follows: f(T,k)=IsGreaterEqual(T,S k-1 )∧IsLess(T,S k ); In the formula, T is the target landing point value, and S... k and S k-1 Let be the interval boundary of the k-th expert database node.

2. The method for selecting evaluation experts by integrating profile matching and random sampling according to claim 1, characterized in that, The step of determining the winning expert in the local candidate expert index set based on the offset includes: The expert database node constructs a cumulative weight sequence locally according to the weights of the candidate experts, and determines the winning expert based on the position of the offset in the cumulative weight sequence.

3. The method for selecting evaluation experts by integrating profile matching and random sampling according to claim 1, characterized in that, The global random seed The generating function is: ； In the formula, n is the number of expert database nodes, and r k The random number partition is the k-th expert database node, and L is the preset security parameter.

4. A method for selecting evaluation experts by integrating profile matching and random sampling according to any one of claims 1 to 3, characterized in that, The corresponding weights of the candidate experts are: Each expert database node assigns a weight to each candidate expert based on at least one of the following dimensions: the candidate expert's historical evaluation score, years of professional experience, and regional balance requirements.

5. A method for selecting evaluation experts by integrating profile matching and random sampling according to any one of claims 1 to 3, characterized in that, The encryption and joint decryption of the encryption token include: The evaluation platform secretly shares the global decryption private key into private key fragments equal to the number of winning experts, and sends the private key fragments encrypted with the node public key and the global public key to the matched expert database nodes when sending the offset. The expert database node that hits the target node uses the global public key to encrypt the information of the winning expert, generates an encrypted token, and returns the encrypted token and its own private key fragment to the evaluation platform. After the evaluation platform collects at least a threshold number of private key fragments, it recovers the global decryption private key using the Lagrange interpolation algorithm, and then decrypts all encrypted tokens using the global decryption private key.

6. A method for selecting evaluation experts by integrating profile matching and random sampling according to any one of claims 1 to 3, characterized in that, Also includes: The global random seed, the weights of each node, the node identifier and offset of the hit expert database node, and the hash value of all winning expert information are synchronously uploaded to the blockchain.

7. A method for selecting evaluation experts by integrating profile matching and random sampling according to any one of claims 1 to 3, characterized in that, The expert avoidance information includes the expert's affiliated organizations, the companies where their immediate family members work, and information on related projects they have participated in within a preset timeframe.

8. A system for selecting evaluation experts that integrates profile matching and random selection, comprising: The evaluation platform is used to generate project profiles based on the requirements of the bidding project and broadcast the project profiles to various expert database nodes; Using a multi-party secure computation protocol, a globally weighted random sampling is performed based on the sum of node weights and random number fragments returned by all expert database nodes to determine the hit expert database nodes and their corresponding offsets. The offsets are then sent to the hit expert database nodes. The encrypted tokens returned by each expert database node are jointly decrypted to obtain all winning expert information. Specifically, within the ciphertext domain, random number fragments returned by each expert database node are aggregated to generate a global random seed. The sum of node weights returned by each expert database node is aggregated to generate a global total weight. The global random seed is modulo the global total weight to obtain the target landing point value. Within the ciphertext domain, the interval boundaries of each expert database node are determined based on the sum of node weights. Expert database nodes whose target landing point value falls within the interval boundary are considered hits. The offset is determined based on the target landing point value and the interval boundary of the hit expert database nodes. The hit determination function is as follows: f(T,k)=IsGreaterEqual(T,S k-1 )∧IsLess(T,S k ); In the formula, T is the target landing point value, and S... k and S k-1 Let k be the interval boundary of the k-th expert database node; Multiple expert database nodes are used within a trusted execution environment to map locally stored expert avoidance association information into privacy feature bit vectors using a hierarchical sensitive hash algorithm. These vectors are then matched with the project profile for privacy purposes, generating a candidate expert index set containing expert indices and corresponding weights. Simultaneously, the node weights of the candidate expert index set and locally generated random number fragments are sent to the bidding platform. If the offset is received, the winning expert is determined from the local candidate expert index set based on the offset. An encrypted token containing the winning expert's information is generated and returned to the bidding platform. Specifically, this involves performing bitwise operations to compare the privacy feature bit vector with the feature bit vector corresponding to the avoidance condition in the project profile, and using the bitwise operation result to filter experts with avoidance relationships. The filtered experts are then scored based on a non-linear scoring function of expert credit score and professional matching degree, generating a candidate expert index set containing expert indices and corresponding weights.

Images