Intelligent credit bank system and cross-school credit mutual recognition method

Abstract

The invention discloses an intelligent credit bank system and a cross-school credit mutual recognition method, and the architecture of the intelligent credit bank system is composed of a data perception layer, a federal learning layer, a block chain evidence storage layer, and an application service layer. The data perception layer is used for carrying out collection, edge preprocessing and desensitization processing on structured data and unstructured achievement data with diversified sources, and the federal learning layer is used for constructing a global model so as to carry out score evaluation and equivalent credit calculation on the same class of courses of each school and realize cross-school achievement mutual recognition. The block chain certificate storage layer adopts a main chain and side chain collaborative double-chain architecture, the main chain is used for storing core certificates including credit hash and token transaction, the side chain is used for processing high-frequency services, and the application service layer provides functional interfaces for users and institutions and is in butt joint with an external system. According to the method, data privacy protection is realized by utilizing federated learning and block chain security cooperative calculation, and the adaptability of the model to different scenes is improved while privacy protection is realized.

Description

Technical Field

[0001] This invention relates to a credit bank system for storing and circulating academic credits, and more particularly to an intelligent credit bank system and a method for cross-institutional credit recognition built using federated learning and blockchain collaborative technologies. Background Technology

[0002] A credit bank is a storage center that measures credits and serves as a bridge for the communication and conversion of learning outcomes at all levels and of all types. Traditional centralized management platforms suffer from problems such as data loss, tampering of results, and difficulties in certificate verification and credit conversion. Blockchain, a decentralized data processing method, solves many of these problems by applying its asymmetric encryption, consensus mechanism, and smart contract technology to credit banks. However, some issues still remain:

[0003] Because learners' sensitive personal information and original learning outcomes still need to be uploaded to the blockchain in plaintext, there is a risk of leakage. At the same time, updates such as conversion rules need to be approved by on-chain voting according to the admission rules, resulting in long response times and poor dynamic adaptability. Furthermore, the storage format requirements for storing results in the credit bank are high, only supporting structured data such as courses and certificates, and cannot intelligently process unstructured data such as videos and code.

[0004] The existing credit bank system does not allow for the free transfer of credits between schools, and it is limited to simple storage in enterprise skills certification. It also lacks intelligent recommendation and guidance for learners' personal growth paths and career planning. Summary of the Invention

[0005] The purpose of this invention is to provide a model for an intelligent credit bank system that utilizes federated learning and blockchain collaborative technologies to achieve data privacy protection, intelligent storage, conversion, and personalized recommendations for learners, thus filling the gap in research on intelligent credit bank systems.

[0006] To address this, the present invention provides an intelligent credit bank system, comprising a data perception layer, a federated learning layer, a blockchain storage layer, and an application service layer. The data perception layer collects diverse structured and unstructured data, performs edge preprocessing, and desensitizes sensitive information using homomorphic encryption algorithms. The federated learning layer uploads only encrypted model parameters from sub-models trained on local data by various educational institutions to the aggregation node, generates a global model through weighted averaging, and uses this global model to evaluate the performance of similar courses across schools, calculate equivalent credits, and achieve cross-institutional recognition of achievements. The blockchain storage layer adopts a dual-chain architecture with a main chain and side chains. The main chain stores core credentials, including credit hashes and token transactions, while the side chains handle high-frequency transactions, including credit exchange requests. During credit conversion, smart contracts drive automatic rule execution, enabling one-click credit conversion. The application service layer provides functional interfaces to users and institutions, connects to external systems, establishes lifelong student profiles for learner full-cycle achievement storage, and automatically generates multi-dimensional competency graphs.

[0007] This invention also provides a method for cross-institutional credit recognition based on an intelligent credit bank system. The intelligent credit bank system includes the following participants: students, universities, federated learning nodes, a blockchain network, and smart contracts. The method comprises the following steps: S1, students upload learning outcome data, submit a credit recognition application to the credit-granting university, and specify the target university and target courses for credit recognition; S2, the credit-granting university processes the uploaded learning outcome data and uploads encrypted learning features to the federated learning nodes; S3, the federated learning nodes use a global model to evaluate performance based on the encrypted learning features, generate a feature vector containing equivalent credits, and use this vector to request credit recognition from the smart contract. Evaluation; S4. The smart contract executes the mutual recognition rules, calculates the course matching degree, and when the course matching degree exceeds the threshold, it automatically executes the smart contract conversion rules, mints credit conversion NFT tokens, and causes the blockchain network to upload the hash value of the credit source data to the chain, while sending synchronization authentication information to the credit recipient; S5. The credit recipient verifies the validity of the token to the blockchain network, and the blockchain network requests zero-knowledge proofs from the smart contract for verification. If the verification is successful, the smart contract confirms the equivalent credits to the credit recipient, who then parses the NFT metadata to obtain the converted credits. The credit recipient updates the student file database and notifies the student, and completes blockchain data synchronization through the HTLC protocol; S6. The student confirms and responds to the information notified by the credit recipient.

[0008] Compared with the prior art, the present invention has the following beneficial effects:

[0009] (1) Federated learning and blockchain secure collaborative computing achieve data privacy protection.

[0010] (2) By adopting a dynamic optimization and privacy enhancement mechanism for the global model, the model’s adaptability to different scenarios is improved while protecting privacy, thus avoiding the problem of reduced data utility caused by static noise in traditional federated learning.

[0011] (3) Establish a credit circulation mechanism to realize cross-school and cross-province certification and conversion, and reduce the delay of manual operation.

[0012] (4) Dynamically calculate the course similarity driven by the knowledge graph to solve the problem of outdated recommendations caused by static weights in traditional course matching, and improve the accuracy and practicality of cross-school credit transfer.

[0013] (5) Establish a multimodal data fusion and real-time feedback mechanism for smart contracts to solve the problem of unstructured data being difficult to quantify and evaluate, while improving the system's flexibility and user participation.

[0014] (6) The collaborative evidence storage optimization of federated learning and blockchain achieves a balance between data privacy and system performance, and improves processing efficiency by more than 40% compared with a single chain structure.

[0015] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0017] Figure 1 This is the overall architecture diagram of the credit bank system of the present invention;

[0018] Figure 2 This is a technical roadmap for the cross-school credit recognition scenario of the credit bank system of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] like Figure 1 As shown, the architecture of the intelligent credit bank system of the present invention consists of a data perception layer, a federated learning layer, a blockchain storage layer, and an application service layer.

[0021] The data awareness layer is the platform's infrastructure, responsible for the collection, classification, and preliminary processing of learning outcome data. It needs to address the core issues of data source diversity and privacy protection. This includes two aspects: multimodal data collection and privacy-preserving preprocessing mechanisms. It integrates structured and unstructured data from formal education, non-formal education, and informal learning outcomes. Employing Local Differential Privacy (LDP) technology supported by federated learning, it adds noise to sensitive information such as ID numbers and learning trajectories at the data collection end to ensure that the original data does not leave the local machine.

[0022] The federated learning layer optimizes credit assessment models through decentralized collaboration, addressing data silos and privacy breaches while incentivizing participants. It primarily comprises horizontal federated learning, vertical federated learning, global model aggregation, and contribution auditing. Horizontal federated learning is used for credit assessment, where each educational institution trains its sub-model based on local data, uploading only encrypted model parameters to the aggregation node. A weighted average is then used to generate a global model, such as for cross-institutional credit equivalence assessment. Vertical federated learning is used to discover credit conversion rules. For example, in cross-domain credit conversion scenarios like exchanging vocational skills certificates for theoretical course credits, it uses secure multi-party computation to align the entity characteristics of different institutions, such as skill points and course objectives, and jointly trains the conversion rule model, avoiding the exposure of original data relationships.

[0023] The blockchain-based evidence storage layer ensures the security and trustworthiness of academic credit data. It automates the execution of business logic through smart contracts. The credit data is stored using a Merkle tree structure to protect privacy and enable rapid verification. During credit conversion, smart contracts drive the automatic execution of rules, and if a user disagrees with the authentication result, the contract is promptly invoked to handle the issue according to the rules.

[0024] The application service layer provides functional interfaces for users and institutions, building an open ecosystem. It can establish lifelong learning profiles, facilitate cross-institutional credit recognition, and connect to university and corporate HR systems through standardized APIs to achieve "one-click credit transfer," reducing manual review costs.

[0025] The data awareness layer includes the following callable modules: multimodal terminal module, data acquisition SDK module, and privacy computing protocol module.

[0026] The multimodal terminal module mainly performs heterogeneous data acquisition and edge preprocessing, supports real-time storage of multimodal learning results such as text, video, and images, and performs preliminary data cleaning and format standardization on the terminal device, solving the problem of single data source in traditional systems.

[0027] The data acquisition SDK module supports cross-platform access and provides a lightweight SDK that can connect to third-party data sources such as school administration systems, corporate training platforms, and online education networks with one click, ensuring the structured compatibility of multi-source data.

[0028] The privacy computing protocol module is responsible for localized privacy protection. It uses integrated differential privacy (DP) to de-identify sensitive information at the data collection end, providing a secure data pipeline for the upper federated learning layer.

[0029] The federated learning layer includes the following callable modules: horizontal federated learning module, vertical federated learning module, global model aggregation module, and contribution auditing module.

[0030] The lateral federated learning module is used for joint modeling of homogeneous data across institutions, primarily applicable to scenarios where multiple educational institutions have overlapping sample spaces and identical feature spaces. Examples include evaluating grades for similar courses across schools, calculating equivalent credits, and achieving cross-institutional recognition of results. This requires training the model locally and encrypting the uploaded parameters to increase the training sample size while protecting privacy.

[0031] The vertical federated learning module fuses features from heterogeneous data and is applied to institutions with different feature spaces but overlapping sample IDs. For example, it calculates job matching scores by combining university course features with enterprise skill tags, providing suitable career development suggestions. Throughout the process, homomorphic encryption is used to achieve feature cross-calculation, ensuring that each party cannot obtain the original features of the others.

[0032] The global model aggregation module fuses parameters from multiple sources and uses a suitable algorithm to perform a weighted average of the encrypted gradients uploaded horizontally and vertically in federation, where the weight = local data volume / global total volume.

[0033] The contribution audit module is used to quantify the value of data. It uses algorithms to accurately calculate the contribution of each organization to the global model and maps the contribution to token rewards, driving organizations to continuously participate in the federated ecosystem.

[0034] The blockchain evidence storage layer includes the following callable modules: dual-chain architecture module, smart contract library, zero-knowledge proof module, and cross-chain protocol module.

[0035] A dual-chain architecture module refers to the collaboration between a main chain and side chains. The main chain stores core credentials such as credit hashes and token transactions, ensuring immutability and global trust. Side chains are mainly used to handle high-frequency transactions, such as credit redemption requests, ensuring timely responses.

[0036] The smart contract library module is used to establish dynamic rules, pre-configurable contract templates, and supports automatic updates of credit conversion rules and token allocation policies.

[0037] The zero-knowledge proof module is used for anonymized credit verification. Its main function is to generate zero-knowledge proofs, allowing companies to verify a job applicant's qualifications without needing to obtain specific grades.

[0038] The cross-chain protocol module supports multi-chain interoperability and dynamic routing strategies, such as building relay chains to connect regional credit chains and enable cross-domain credit transfer. Furthermore, it can automatically adapt conversion coefficients based on regional policy differences.

[0039] In one embodiment, when the contract is triggered, a video analysis model (such as 3D-CNN) and a code auditing engine (such as an AST parser) are automatically invoked to extract features from unstructured data (such as experimental videos and project code); 2) the feature hashes are cross-validated with the federated evaluation results stored on the chain to generate zero-knowledge proofs (ZKP); 3) a real-time feedback mechanism is introduced, allowing students / enterprises to adjust matching rules and update smart contract logic through on-chain voting.

[0040] The application service layer includes the following callable modules: Open API Gateway module, Credit Trading Market module, Intelligent Path Planning module, and Lifelong Learning Archive module.

[0041] The Open API Gateway module serves as a unified service entry point, providing standardized interfaces to connect with external systems such as school academic affairs systems, enterprise recruitment platforms, and government regulatory bodies, thereby improving integration efficiency.

[0042] The credit trading marketplace module is used for multilateral trading platforms, such as between educational institutions and individuals, or between enterprises and individuals. The exchange types are also diverse, such as exchanging credits for courses.

[0043] The intelligent path planning module is a personalized learning map for learners. By inputting information such as skill profiles, career goals, and industry demands, it can output dynamically recommended course sequences. It can monitor learning progress and automatically adjust the path.

[0044] The lifelong learning portfolio module is used to document learners' achievements throughout their entire learning lifecycle and automatically generates multi-dimensional competency maps. It can store multi-modal learning data, including formal education, non-formal education, and practical achievements, in a structured manner.

[0045] like Figure 2 As shown, the cross-school credit recognition method of the credit bank system of the present invention includes the following steps S1-S6.

[0046] S1. Students submit credit transfer applications.

[0047] S11. Students log in to the Smart Credit Bank Platform and fill out the cross-school credit recognition application form;

[0048] S12. Upload structured data such as course transcripts and skill certificates, as well as unstructured output data such as experimental operation videos, project code, and research reports, to the credit bank platform. The platform automatically preprocesses the data, standardizes the format, and automatically calls the 3D-CNN video analysis model and AST parser for code auditing through smart contracts. It also extracts features from unstructured data such as experimental videos and project code, and uses homomorphic encryption algorithms to de-identify sensitive information such as names and ID numbers.

[0049] S13. Select the target school and target course for which credits need to be mutually recognized. For example, Machine Learning (School A) → Introduction to Artificial Intelligence (School B).

[0050] S2, Evaluation of Federated Learning Model

[0051] This step utilizes a global model to evaluate performance based on the uploaded encrypted learning features, generating a feature vector containing equivalent credits, and then requesting credit recognition assessment from the smart contract.

[0052] The method for obtaining the global model includes the following steps:

[0053] S21. Each participating university (such as University A and University B) node uses local data to train the course evaluation model, such as data such as grades in "Machine Learning", lab reports and student behavior logs. When training the model, the focus can be on knowledge point dimensions, such as the quantification of neural network and data processing capabilities, and horizontal federated learning can be used to complete the local model training.

[0054] S22. Each participating university node will homomorphically encrypt the model gradient parameters. During global model parameter updates, an adaptive noise injection strategy will be introduced. The noise variance will be dynamically adjusted based on course popularity and node contribution. For example, the "Introduction to Artificial Intelligence" course will automatically enhance noise to prevent overfitting. The global model calculation formula is as follows:

[0055]

[0056] In formula (1), W global Represents global model parameters, N represents the number of nodes, and W... i N(0,σ) represents the encryption gradient uploaded by the i-th node. 2 ) represents the noise value added during the aggregation process to protect model privacy and security. In formula (2), α and β are weighting coefficients, and Contribution i CourseHotness is calculated using the gradient contribution value recorded by the blockchain to represent the real-time access volume of the course.

[0057] S23. By using blockchain smart contracts to record the gradient contribution values ​​of each participating school, the gradient of high-contribution nodes with large amounts of high-quality course data is preserved more completely.

[0058] S3, Matching Degree Calculation and Automatic Smart Contract Verification

[0059] S31. Extract knowledge points from the course syllabus, and use NLP technology to analyze unstructured text such as "course description" and "lab manual" to generate semantic relationships, constructing a semantic knowledge graph. For example, extract "neural network" and "data processing" from University A's "Machine Learning" and University B's "Introduction to Artificial Intelligence" to calculate the matching degree of knowledge points. The smart contract calls the knowledge graph to calculate the similarity of the courses, introducing course timeliness factors and industry demand factors to dynamically adjust the weight of knowledge graph nodes. If the matching degree is ≥50%, automatic conversion is triggered. The matching degree is calculated using the following formula:

[0060] Matching = Sim × ∑(w) A,i ×w B,i )--------------------------------(3)

[0061]

[0062] w i =γ·Recency(c i )+δ·IndustryDemand(c i )----------------------(5)

[0063] In formula (3), Matching represents the matching degree, Sim represents the course similarity, and w A,i and w B,i The weights of the i-th knowledge points of schools A and B are respectively represented by w in formula (4). i G represents the weight of the node in the i-th knowledge graph. embed c1 and c2 represent the embeddings of the knowledge graph, respectively, and represent the embedding vectors of the first and second courses. In formula (5), Recency decays according to the course update time, and IndustryDemand is obtained in real time through the labor market API.

[0064] S32. If the calculation result of the above steps satisfies the matching degree ≥ 0.5, the smart contract conversion rule will be automatically executed. If the weight of "neural network" in school A is 0.8, the weight in school B is 0.7, and the semantic similarity is 0.9, then the matching degree is 0.8 × 0.7 × 0.9 = 0.504, which exceeds the threshold of 0.5 and can be converted.

[0065] S4, Blockchain-based Evidence Storage and Token Minting

[0066] S41. Hash the matched course knowledge points (such as "neural network") to generate a unique identifier (such as SHA-256) and store it;

[0067] S42. The smart contract pre-sets rules that trigger the standard contract and mint "credit conversion NFTs" when the matching degree is ≥0.5.

[0068] S43. The metadata hash value of the credit is put on the chain, and the feature hash is cross-validated with the federated evaluation results stored on the chain. The authenticity of the data is verified by zero-knowledge proof (ZKP) to protect student privacy.

[0069] S44. The target school sends a request to verify the validity of the token. Once the verification is successful, the credits are credited to the target school's system.

[0070] S5. Credit Transfer and Results Synchronization

[0071] S51. The target school's blockchain node parses the NFT metadata to obtain the converted credits, such as "0.9 credits";

[0072] S52, target school and new student file database, and record the source school and course of the conversion;

[0073] S53. Complete blockchain data synchronization through the HTLC protocol;

[0074] S54. For data with a matching degree of less than 50%, manual review will be automatically triggered, which is required to be completed within 3 working days. After the review is passed, the smart contract will be manually triggered.

[0075] S6. Student Confirmation and Feedback

[0076] S61. The federated learning model generates a personalized learning path based on the matching degree recorded in the NFT and pushes the path notification to the student's end, such as: "Your 'Machine Learning' has been converted into 0.9 credits of 'Introduction to Artificial Intelligence' at School B" or "Insufficient matching degree, it is recommended to supplement the 'Deep Learning' course and reapply."

[0077] S62. Based on the hierarchical relationship of the knowledge graph (such as "data processing" being a prerequisite for "neural network"), the recommended optimal learning order is: School A's "data cleaning" -> School B's "feature engineering" -> School A's "CNN practice".

[0078] S63. Appeals can be made by submitting additional materials such as project results and supervisor recommendations, and the system will re-evaluate the case.

[0079] S64. The smart credit bank system will dynamically update the smart contract rule base based on the global feedback of the federated model. For example, it will generate scoring data such as students' satisfaction with the conversion results, or adjust the matching threshold, add knowledge point tags, etc., and feed the feedback to the federated learning model through on-chain voting to dynamically adjust the matching rules and optimize the course matching algorithm. The model will automatically update the data and the smart contract logic every month.

[0080] The above steps complete the cross-school credit transfer scenario.

[0081] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent credit bank system, characterized in that, It includes a data perception layer, a federated learning layer, a blockchain evidence storage layer, and an application service layer. The data perception layer is used to collect structured data and unstructured output data from diverse sources. The data undergoes edge preprocessing, and sensitive information is de-identified using a homomorphic encryption algorithm. The federated learning layer is used to upload only the encrypted model parameters of the sub-models trained by each educational institution based on local data to the aggregation node. A global model is generated through weighted averaging, and this global model is used to evaluate the performance of similar courses in each school, calculate equivalent credits, and achieve cross-school mutual recognition of achievements. The blockchain evidence storage layer adopts a dual-chain architecture with the main chain and side chains working together. The main chain is used to store core credentials, including credit hashes and token transactions, while the side chains are used to handle high-frequency transactions, including credit redemption requests. During credit conversion, smart contracts are used to drive the automatic execution of rules, enabling one-click credit conversion. The application service layer provides functional interfaces to users and institutions, connects to external systems, establishes lifelong student files for the full-cycle record-keeping of learners' achievements, and automatically generates multi-dimensional competency maps.

2. The intelligent credit bank system according to claim 1, characterized in that, The data perception layer includes a multimodal terminal module, which is used for heterogeneous data acquisition and edge preprocessing, and at the same time performs preliminary data cleaning and format standardization operations on the terminal device; The data acquisition SDK module provides an SDK toolkit to enable one-click access to third-party data sources and ensures the structured compatibility of multi-source data. The privacy computing protocol module utilizes integrated differential privacy technology to de-identify sensitive information at the data collection end.

3. The intelligent credit bank system according to claim 1, characterized in that, The federated learning layer includes: a horizontal federated learning module for cross-institutional joint modeling of homogeneous data, wherein each educational institution trains a sub-model based on local data and uploads the model parameters in encrypted form; The global model aggregation module is used to fuse the parameters of multiple source sub-models and generate a global model through weighted averaging. The contribution audit module is used to calculate the contribution of each organization to the global model and map the contribution to token rewards.

4. The intelligent credit bank system according to claim 3, characterized in that, The federated learning layer also includes a vertical federated learning module, which is used to fuse features of heterogeneous data. Throughout the process, feature cross-computation is achieved through homomorphic encryption to ensure that each party cannot obtain the original features of the other party.

5. The intelligent credit bank system according to claim 1, characterized in that, The blockchain evidence storage layer includes a dual-chain architecture module, a smart contract library module, a zero-knowledge proof module, and a cross-chain protocol module. The dual-chain architecture module comprises a main chain and side chains working together. The main chain stores credit hashes and token transactions, while the side chains handle high-frequency transactions. The smart contract library module is used to establish dynamic rules, has pre-set configurable contract templates, and supports automatic updates of credit conversion rules and token allocation policies. The zero-knowledge proof module is used to cross-validate the feature hash with the federated evaluation results stored on the blockchain to generate a zero-knowledge proof, thereby achieving anonymized credit verification. The cross-chain protocol module supports multi-chain interoperability and dynamic routing strategies, constructs a relay chain, and connects regional credit chains to enable cross-chain credit transfer.

6. The intelligent credit bank system according to claim 1, characterized in that, The application service layer includes an open API gateway module, a credit trading market module, and a lifelong learning archive module. The open API gateway module provides a unified service interface; the credit trading market module provides a multilateral trading platform to enable diversified credit trading; and the lifelong learning archive module is used for the full-cycle documentation of learners' achievements and automatically generates a multi-dimensional competency map.

7. The intelligent credit bank system according to claim 1, characterized in that, The application service layer also includes an intelligent path planning module, which is used to output dynamically recommended course sequences based on the input competency profile, career goals, and industry needs.

8. A method for cross-institutional credit transfer, characterized in that, The intelligent credit bank system according to any one of claims 1-7 includes the following participants: students, university participants, federated learning nodes, blockchain network, and smart contracts. The method for cross-institutional credit transfer includes the following steps: S1. Students upload their learning outcomes data, submit a credit transfer application to the credit-granting school, and specify the target school and target courses for credit transfer. S2. The credit-granting entity processes the learning outcome data uploaded by students on the terminal and uploads encrypted learning features to the federated learning nodes. S3. Federated learning nodes use a global model to evaluate performance based on cryptographic learning characteristics, generate feature vectors containing equivalent credits, and then request credit recognition evaluation from smart contracts. S4. The smart contract executes the mutual recognition rules, calculates the course matching degree, and when the course matching degree exceeds the threshold, it automatically executes the smart contract conversion rules, mints credit conversion NFT tokens, and causes the blockchain network to put the credit source data hash value on the chain, while sending synchronous authentication information to the credit recipient. S5. The credit recipient verifies the validity of the token on the blockchain network. The blockchain network requests a zero-knowledge proof from the smart contract for verification. If the verification is successful, the smart contract confirms the equivalent credits to the credit recipient. The credit recipient then parses the NFT metadata to obtain the converted credits. The credit recipient updates the student file database and notifies the student, and completes the blockchain data synchronization through the HTLC protocol. S6. Students confirm and respond to the information notified by the credit recipient.

9. The method for cross-institutional credit recognition according to claim 8, characterized in that, The calculation method for the global model is as follows: Each participating university will homomorphically encrypt the model gradient parameters. When updating the global model parameters, an adaptive noise injection strategy will be introduced. The noise variance will be dynamically adjusted according to the course popularity and node contribution to prevent the model from overfitting. The global model calculation formula is as follows: In equation (1) above, W global Represents the global model parameters, N represents the number of nodes, Wi represents the encryption gradient uploaded by the i-th node, and N(0,σ) 2 ) represents the noise value added during the aggregation process to protect the privacy and security of the model. In (2) above, α and β are weighting coefficients. Contribution i CourseHotness is calculated using the gradient contribution value recorded on the blockchain, representing the real-time access volume of a course.

10. The method for cross-institutional credit recognition according to claim 8, characterized in that, The matching degree calculation method includes the following steps: S31. Extract knowledge points from the course syllabus, use NLP technology to parse unstructured text to generate semantic relationships, construct a semantic knowledge graph, and use a smart contract to call the knowledge graph to calculate the similarity of the courses. Introduce course timeliness factors and industry demand factors to dynamically adjust the weight of knowledge graph nodes. The matching degree is calculated using the following formula: Matching=Sim×∑(w A,i ×w B,i )-----------------------------(3) w i =γ·Recency(c i )+δ·IndustryDemand(c i )--------------------(5) In equation (3) above, Matching represents the matching degree, Sim represents the course similarity, and w A,i and w B,i These represent the weights of the i-th knowledge point for School A and School B, respectively. In equation (4) above, w i G represents the weight of the node in the i-th knowledge graph. embed c1 and c2 represent the embeddings of the knowledge graph, respectively, and represent the embedding vectors of the first and second courses. In equation (5) above, Recency decays according to the course update time, and IndustryDemand is obtained in real time through the labor market API. S32. When the calculation result of the above steps satisfies the matching degree ≥ the set threshold, the smart contract conversion rule will be executed automatically.

Images