A blockchain-based full-process management and risk control system for electronic guarantee letters

Abstract

This application relates to the technical field of electronic guarantee letters. Specifically, it discloses a blockchain-based full-process management and risk control system for electronic guarantee letters. It retrieves the user's credit record and insurance record according to the user's identity information, and uses natural language analysis and processing algorithms based on deep learning for data splitting and semantic encoding. It performs causal association semantic dynamic walking on the encoded record features to generate a multi-modal joint representation of credit risk, so as to automatically judge whether the insurance application request is within the credit risk safety level. In this way, it can effectively analyze and utilize the complex semantic information in the user's credit record and insurance record, making the basis for credit risk judgment more comprehensive and accurate. At the same time, the model can be continuously updated and adjusted according to new data, no longer restricted by fixed formulas and preset coefficient factors, improving the accuracy and effectiveness of risk judgment in different scenarios.

Description

Technical Field

[0001] This application relates to the technical field of electronic guarantee letters, and more specifically, to a full-process management and risk control system for electronic guarantee letters based on blockchain. Background Art

[0002] An electronic guarantee letter is an electronic guarantee certificate issued by a guarantor to a beneficiary at the request of an applicant, and is used to fulfill the payment or compensation liability in case of the applicant's default. However, there are many problems in the existing system: lack of effective credit investigation and risk prevention measures, making it difficult to cope with the counterfeit and shoddy products and credit risks in the supply chain; the guarantor needs to invest additional efforts in verifying the authenticity of transactions, increasing the risk control cost, and the data information has low credibility and is easy to be tampered with.

[0003] Based on this, the existing patent CN114155093A proposes a full-process management and risk control system for electronic guarantee letters based on blockchain. It uses blockchain technology to build an electronic guarantee letter management platform. Users submit insurance application requests through smart contracts. The platform receives and processes the requests, verifies the user's identity and credit status, calculates the credit score (XY) to judge the risk level, then conducts risk coefficient analysis and review and approval. Finally, the issued electronic guarantee letter is encrypted and distributed to the specified storage address to effectively prevent the electronic guarantee letter from being tampered with and improve data security.

[0004] In this patent, to determine whether an insurance application request is within the credit risk safety level, the credit score is obtained by counting overdue data and default data and then compared with a preset threshold. However, this method only relies on numerical data such as the number of overdue times and amounts, and cannot analyze and utilize the complex semantic information in the text data, resulting in the possible omission of important background and context factors during judgment. Moreover, this method uses fixed formulas and preset coefficient factors, which are generally set based on historical data, and it is difficult to adapt to new market changes or special situations, lacking flexibility and dynamic adjustment ability, thus affecting the accuracy and effectiveness of judgment.

[0005] Therefore, an optimized full-process management and risk control solution for electronic guarantee letters based on blockchain is expected. Summary of the Invention

[0006] This application provides a full-process management and risk control system for electronic guarantee letters based on blockchain, which can effectively analyze and utilize the complex semantic information in the user's credit investigation record and user's insurance application record, making the basis for credit risk judgment more comprehensive and accurate. At the same time, the model can be continuously updated and adjusted according to new data, no longer restricted by fixed formulas and preset coefficient factors, and improving the accuracy and effectiveness of risk judgment in different scenarios.

[0007] According to one aspect of the present application, a blockchain-based full-process management and risk control system for electronic guarantee letters is provided, including: a platform construction module for constructing an electronic guarantee letter management platform using blockchain technology; a request module for initiating an insurance application request to the electronic guarantee letter management platform; a verification module for verifying the user identity in response to the electronic guarantee letter management platform receiving the insurance application request to determine whether the insurance application request is within the credit risk safety level; a guarantee letter signing and distribution module for approving the insurance application request in response to the insurance application request being within the credit risk safety level, and issuing and encrypting and storing an electronic guarantee letter for the approved insurance application request.

[0008] Wherein, the verification module includes: a retrieval unit for retrieving the user's credit record and the user's insurance application record according to the user identity information; a record splitting and encoding unit for performing data splitting and semantic encoding on the user's credit record and the user's insurance application record to obtain a time series distribution of the user's credit record semantic embedding encoding vector and a time series distribution of the user's insurance application record semantic embedding encoding vector; a record time series semantic walking unit for performing causally associated driven record semantic dynamic walking on the time series distribution of the user's credit record semantic embedding encoding vector and the time series distribution of the user's insurance application record semantic embedding encoding vector respectively to obtain a user's credit record time series semantic walking encoding vector and a user's insurance application record time series semantic walking encoding vector; a record multimodal combination unit for fusing the user's credit record time series semantic walking encoding vector and the user's insurance application record time series semantic walking encoding vector to obtain a user's credit risk multimodal combined encoding vector; a credit risk judgment unit for obtaining an identification result indicating whether the insurance application request is within the credit risk safety level according to the user's credit risk multimodal combined encoding vector.

[0009] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the record splitting and encoding unit includes:

[0010] A data splitting sub-unit for splitting the user's credit record and the user's insurance application record according to the time dimension to obtain a list of user's credit record items and a list of user's insurance application items;

[0011] A user's credit record semantic encoding sub-unit for performing semantic encoding on each user's credit record item in the list of user's credit record items to obtain a time series distribution of the user's credit record semantic embedding encoding vector;

[0012] A user's insurance application record semantic encoding sub-unit for performing semantic encoding on each user's insurance application item in the list of user's insurance application items to obtain a time series distribution of the user's insurance application record semantic embedding encoding vector.

[0013] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the user credit record semantic encoding subunit is used to: use a semantic encoder including an RNN to perform semantic encoding on each user credit record item in the list of user credit record items to obtain a time series distribution of the user credit record semantic embedding encoding vectors.

[0014] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the user insurance record semantic encoding subunit is used to: use the semantic encoder including an RNN to perform semantic encoding on each user insurance item in the list of user insurance items to obtain a time series distribution of the user insurance record semantic embedding encoding vectors.

[0015] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the record time series semantic random walk unit includes:

[0016] A user credit record semantic implicit feature mining subunit, which is used to perform implicit feature mining on each user credit record semantic embedding encoding vector in the time series distribution of the user credit record semantic embedding encoding vectors to obtain a time series distribution of user credit record semantic embedding deep implicit feature encoding vectors;

[0017] A user credit record semantic causal feature construction subunit, which is used to construct a user credit record semantic causal association topological feature matrix based on the time series distribution of the user credit record semantic embedding deep implicit feature encoding vectors;

[0018] A user credit record semantic surface context dynamic random walk subunit, which is used to perform feature dynamic random walk driven by dynamic graph convolution on the user credit record semantic causal association topological feature matrix and the time series distribution of the user credit record semantic embedding encoding vectors to obtain a user credit record semantic surface context dynamic random walk semantic encoding vector;

[0019] A user credit record semantic hidden layer feature extraction subunit, which is used to perform the feature dynamic random walk driven by dynamic graph convolution on the user credit record semantic causal association topological feature matrix and the time series distribution of the user credit record semantic embedding deep implicit feature encoding vectors to obtain a user credit record semantic hidden layer context dynamic random walk semantic encoding vector;

[0020] A user credit record feature fusion subunit, which is used to fuse the user credit record semantic surface context dynamic random walk semantic encoding vector and the user credit record semantic hidden layer context dynamic random walk semantic encoding vector to obtain the user credit record time series semantic random walk encoding vector.

[0021] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the user credit record semantic causal feature construction subunit includes:

[0022] The user credit record semantic causal factor calculation secondary subunit is used to calculate the semantic causal association factor between any two user credit record semantic embedding depth implicit feature coding vectors in the time series distribution of the user credit record semantic embedding depth implicit feature coding vector to obtain a user credit record semantic causal association topology matrix composed of multiple user credit record semantic causal association factors;

[0023] The user credit record semantic causal trigger secondary subunit is used to perform causal triggering of the gated activation function on the user credit record semantic causal association topology matrix to obtain the user credit record semantic causal association topology feature matrix.

[0024] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the user credit record semantic causal factor calculation secondary subunit includes:

[0025] The semantic association matrix calculation tertiary subunit is used to calculate the association matrix between any two user credit record semantic embedding depth implicit feature coding vectors to obtain a user credit record semantic association matrix;

[0026] The causal factor calculation tertiary subunit based on statistical features is used to perform causal factor calculation based on statistical features on the user credit record semantic association matrix to obtain the user credit record semantic causal association factor corresponding to the user credit record semantic association matrix, where the user credit record semantic causal association factor is related to the mean, variance, maximum value, and causal bias value of the user credit record semantic association matrix.

[0027] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the causal factor calculation tertiary subunit based on statistical features is used for:

[0028] In response to the variance of the user credit record semantic association matrix being greater than or equal to a predetermined threshold, taking the weighted average of the distances between any two user credit record semantic embedding depth implicit feature coding vectors in the time series distribution of the user credit record semantic embedding depth implicit feature coding vector as the causal bias value;

[0029] In response to the variance of the user credit record semantic association matrix being less than the predetermined threshold, taking the weighted value of the mean of the user credit record semantic association matrix as the causal bias value.

[0030] In the above blockchain-based full-process management and risk control system for electronic guarantee letters, the credit risk judgment unit is configured to: input the multi-modal joint coding vector of the user's credit risk into a credit risk recognizer based on a classifier to obtain the recognition result, and the recognition result is used to indicate whether the insurance application request is within the credit risk safety level.

[0031] A blockchain-based full-process management and risk control system for electronic guarantee letters provided by this application retrieves the user's credit investigation record and the user's insurance application record according to the user identity information, and uses deep learning-based natural language analysis and processing algorithms to perform data splitting and semantic coding on the user's credit investigation record and the user's insurance application record. Then, causal association semantic dynamic walks are respectively performed on the semantic embeddings of the encoded user credit investigation records and the semantic embedding features of the user insurance application records, so as to automatically judge whether the insurance application request is within the credit risk safety level based on the credit risk multi-modal joint representation between the sequential semantic walk features of the user credit investigation record and the sequential semantic walk features of the user insurance application record obtained after association. This application can effectively analyze and utilize the complex semantic information in the user's credit investigation record and the user's insurance application record, making the basis for credit risk judgment more comprehensive and accurate. At the same time, the model can be continuously updated and adjusted according to new data, no longer restricted by fixed formulas and preset coefficient factors, improving the accuracy and effectiveness of risk judgment in different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings in the following description only relate to some embodiments of this application and do not limit this application.

[0033] Figure 1 It is a schematic block diagram of the blockchain-based full-process management and risk control system for electronic guarantee letters according to the embodiment of this application.

[0034] Figure 2 It is a schematic block diagram of the verification module in the blockchain-based full-process management and risk control system for electronic guarantee letters according to the embodiment of this application.

[0035] Figure 3 It is a schematic diagram of the data flow of the verification module in the blockchain-based full-process management and risk control system for electronic guarantee letters according to the embodiment of this application.

[0036] Figure 4 It is a schematic block diagram of the record splitting and coding unit in the blockchain-based full-process management and risk control system for electronic guarantee letters according to the embodiment of this application.

[0037] Figure 5Schematic block diagram of the timing semantic walk unit in the blockchain-based full-process management and risk control system for electronic guarantee letters according to an embodiment of the present application.

[0038] Figure 6 Schematic block diagram of the semantic causal feature construction subunit of the user credit record in the blockchain-based full-process management and risk control system for electronic guarantee letters according to an embodiment of the present application. Detailed implementation manners

[0039] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the scope of protection of the present application.

[0040] Based on the existing patent CN114155093A, the present application proposes a blockchain-based full-process management and risk control system for electronic guarantee letters. Figure 1 Schematic block diagram of the blockchain-based full-process management and risk control system for electronic guarantee letters according to an embodiment of the present application. As Figure 1 shown, the blockchain-based full-process management and risk control system for electronic guarantee letters includes: a platform construction module 10 for constructing an electronic guarantee letter management platform using blockchain technology; a request module 20 for initiating an insurance application request to the electronic guarantee letter management platform; a verification module 30 for verifying the user identity in response to the electronic guarantee letter management platform receiving the insurance application request to determine whether the insurance application request is within the credit risk safety level; and a guarantee letter signing and distribution module 40 for approving the insurance application request in response to the insurance application request being within the credit risk safety level, and issuing an electronic guarantee letter for the approved insurance application request and encrypting and storing it.

[0041] Exemplarily, in the platform construction module 10, it is used to construct an electronic guarantee letter management platform by using blockchain technology. It should be understood that the reason for constructing an electronic guarantee letter management platform by using blockchain technology is to solve the problems of imperfect credit investigation, insufficient risk control, and easy data tampering existing in the existing electronic guarantee letter system. The traditional electronic guarantee letter system lacks an effective credit investigation system and a perfect risk prevention measure, and it is difficult to cope with the fake and shoddy products in the supply chain and the credit risks caused by them. In addition, banks, insurance companies or other guarantee institutions need to invest a large amount of manpower and material resources to verify the authenticity of transactions, increasing the additional risk control cost. At the same time, the credibility of the existing electronic guarantee letter data information is low and it is easy to be tampered with, further increasing the difficulty of risk management. By using blockchain technology to construct an electronic guarantee letter management platform, the security and reliability of the system can be effectively improved. The core advantage of blockchain technology lies in its distributed ledger feature, which ensures the immutability and transparency of data.

[0042] Exemplarily, in the request module 20, it is used to initiate an insurance application request to the electronic guarantee letter management platform. It should be understood that the reason for initiating an insurance application request to the electronic guarantee letter management platform is that such a request is a necessary step to obtain an electronic guarantee letter issued by a guarantee institution to the beneficiary. An electronic guarantee letter is an electronic guarantee certificate, which is issued by a bank, an insurance company, a guarantee company or other guarantors to the beneficiary at the request of the applicant, and is used to, when the applicant fails to fulfill his responsibilities or obligations according to the bilateral agreement, the guarantor substitutes him to fulfill a certain amount of payment or economic compensation liability within a certain time limit. Therefore, the insurance application request is not only a key action to start the whole process, but also an important link to ensure the security and reliability of the transaction. Specifically, when a user hopes to obtain an electronic guarantee letter through a guarantee institution, he needs to submit an insurance application request to the electronic guarantee letter management platform. This request contains detailed performance resource information, performance date, transferor information, guarantee institution information, beneficiary information and other contents. These information together constitute a complete insurance application request, providing the necessary data support for subsequent risk assessment and review. In order to initiate an insurance application request to the electronic guarantee letter management platform, the user first needs to interact with the platform through a blockchain smart contract. A blockchain smart contract is an automatically executed contract agreement, which is directly written on the blockchain and automatically triggers corresponding operations according to preset conditions. The user sends an insurance application request through the blockchain smart contract, which means that all relevant information will be recorded on the blockchain, ensuring the authenticity and immutability of the information.

[0043] Exemplarily, in the verification module 30, it is used to verify the user identity in response to the receipt of the insurance application request by the electronic guarantee letter management platform to determine whether the insurance application request is within the credit risk security level. It should be understood that the reason for verifying the user identity in response to the receipt of the insurance application request by the electronic guarantee letter management platform to determine whether the insurance application request is within the credit risk security level is to ensure the security and reliability of the entire electronic guarantee letter system. Specifically, as an important financial instrument, the electronic guarantee letter is used to protect the rights and interests of both trading parties, especially to provide economic compensation when one party fails to fulfill its contractual obligations. Therefore, accurately assessing the credit risk of the applicant is crucial. First of all, by verifying the user identity, the authenticity of the insurance application request can be confirmed, preventing the occurrence of fraud. Since the amount of funds involved in the electronic guarantee letter is usually large and is related to the interests of multiple parties, it is necessary to ensure that the identities of all participating parties are true and reliable. If the user identity cannot be effectively verified, it may lead to false insurance application requests entering the system, thus bringing unnecessary risks and losses to the guarantee institution.

[0044] Exemplarily, in the guarantee letter signing and distribution module 40, it is used to approve the insurance application request in response to the insurance application request being within the credit risk security level, and issue an electronic guarantee letter for the approved insurance application request and encrypt and store it. It should be understood that the electronic guarantee letter, as an electronic guarantee certificate, is issued by a bank, insurance company, guarantee company or other guarantor at the request of the applicant to the beneficiary, and is used to have the guarantor perform a certain amount of payment or economic compensation liability within a certain time limit on behalf of the applicant when the applicant fails to fulfill its responsibilities or obligations as agreed by both parties. Therefore, the insurance application request is not only a key action to initiate the entire process, but also an important link to ensure the security and reliability of the transaction. By submitting the insurance application request, the user can clarify the guarantee service it needs and provide necessary information for subsequent risk assessment and review. Once the electronic guarantee letter management platform receives the insurance application request and confirms that the request is within the credit risk security level, the system will approve the insurance application request. Specifically, when the user's insurance application request is considered to be within the credit risk security level, it indicates that its credit status is good and it has a low default risk. At this time, the system will enter the next approval process. First of all, the guarantee letter issuing module will conduct a detailed risk coefficient analysis on the received insurance application request, including but not limited to multi-dimensional information such as waiting time and accessed data. These analysis results will help determine whether to approve the insurance application request. For the insurance application requests that pass the review, the system will issue corresponding electronic guarantee letters. The issuing process not only involves generating the electronic guarantee letter itself, but also needs to ensure its security. For this purpose, the system uses an encryption algorithm to encrypt the electronic guarantee letter to prevent unauthorized access and tampering. The encrypted electronic guarantee letter will be stored in a designated secure storage address to ensure that only authorized relevant parties can access and use these files.

[0045] Accordingly, in the verification module, the technical concept of the present application is to retrieve the user's credit record and the user's insurance record according to the user identity information, and use natural language analysis and processing algorithms based on deep learning to perform data splitting and semantic encoding on the user's credit record and the user's insurance record. Then, causal association semantic dynamic walks are respectively performed on the semantic embeddings of each encoded user credit record and the semantic embedding features of each user insurance record, so as to automatically determine whether the insurance application request is within the credit risk safety level based on the multi-modal joint representation of the credit risk between the sequential semantic walk features of the user credit record and the sequential semantic walk features of the user insurance record obtained after association. The present application can effectively analyze and utilize the complex semantic information in the user's credit record and the user's insurance record, making the basis for credit risk judgment more comprehensive and accurate. At the same time, the model can be continuously updated and adjusted according to new data, no longer restricted by fixed formulas and preset coefficient factors, improving the accuracy and effectiveness of risk judgment in different scenarios.

[0046] In one embodiment, as Figure 2 and Figure 3 shown, the verification module 30 includes: a retrieval unit 31 for retrieving the user's credit record and the user's insurance record according to the user identity information; a record splitting and encoding unit 32 for performing data splitting and semantic encoding on the user's credit record and the user's insurance record to obtain the time series distribution of the semantic embedding encoding vectors of the user's credit record and the time series distribution of the semantic embedding encoding vectors of the user's insurance record; a record sequential semantic walk unit 33 for performing causal association-driven record semantic dynamic walks on the time series distribution of the semantic embedding encoding vectors of the user's credit record and the time series distribution of the semantic embedding encoding vectors of the user's insurance record respectively to obtain the sequential semantic walk encoding vectors of the user's credit record and the sequential semantic walk encoding vectors of the user's insurance record; a record multi-modal joint unit 34 for fusing the sequential semantic walk encoding vectors of the user's credit record and the sequential semantic walk encoding vectors of the user's insurance record to obtain a user credit risk multi-modal joint encoding vector; and a credit risk judgment unit 35 for obtaining an identification result indicating whether the insurance application request is within the credit risk safety level according to the user credit risk multi-modal joint encoding vector.

[0047] Exemplarily, in the retrieval unit 31, the user's credit record and insurance record are retrieved according to the user identity information. It should be understood that when the user submits an insurance request to the electronic guarantee letter management platform through the blockchain smart contract, the system will automatically obtain and verify the user's identity information. This process usually includes basic information such as the user's name, ID number, and contact information, and may further confirm the user's identity by combining biometric technologies such as fingerprint or facial recognition. Once the user's identity is confirmed, the system will interact with external credit agencies and internal databases based on this identity information to retrieve the user's credit record and insurance record. Specifically, the system will establish a connection with a credit assessment agency (such as the central bank credit reference center or other third-party credit service providers) through a secure data interface and send a query request containing the user's identity information. These credit agencies have detailed user credit history data, including repayment records, default situations, credit scores, etc. Through a standardized data exchange protocol (such as API), the system can quickly obtain the latest credit record. At the same time, the system will also access the user's insurance record stored internally. These records usually contain all the user's past insurance behaviors, such as the time of insurance, amount, guarantee agency, and whether there has been a claim, etc. To ensure data consistency and integrity, the system may use a distributed database or blockchain technology to store these records, which not only improves data security but also ensures data immutability.

[0048] In one embodiment, as Figure 4 shown, the record splitting and encoding unit 32 includes: a data splitting sub-unit 321 for splitting the user's credit record and the user's insurance record according to the time dimension to obtain a list of user credit record items and a list of user insurance items; a user credit record semantic encoding sub-unit 322 for performing semantic encoding on each user credit record item in the list of user credit record items to obtain a time series distribution of the user credit record semantic embedding encoding vectors; a user insurance record semantic encoding sub-unit 323 for performing semantic encoding on each user insurance item in the list of user insurance items to obtain a time series distribution of the user insurance record semantic embedding encoding vectors.

[0049] Exemplarily, in the data splitting sub-unit 321, considering that the credit investigation and insurance application behaviors of users are essentially processes that change dynamically over time. For example, the repayment behavior in the credit investigation record, the occurrence time of credit operations, as well as the insurance application time and claim settlement time in the insurance application record, etc., all have an obvious chronological order. Therefore, in order to be able to understand and analyze the inherent characteristics between data more accurately and clearly, in the technical solution of this application, the user credit investigation record and the user insurance application record are split according to the time dimension to obtain a list of user credit investigation record items and a list of user insurance application items. In this way, the changes in the user's credit behavior at different times can be clearly observed. For example, by checking the fluctuations in the user's repayment overdue situation over time, it is possible to judge whether their credit status is gradually improving or deteriorating, which helps to predict the future credit risk trend.

[0050] Exemplarily, in the user credit investigation record semantic encoding sub-unit 322, considering that each user credit investigation record item contains its own semantic information and meaning, therefore, in the technical solution of this application, semantic encoding is performed on each user credit investigation record item in the list of user credit investigation record items to capture the deep features in the text, and a time series distribution of user credit investigation record semantic embedding encoding vectors is obtained. Specifically, in a specific example of this application, the user credit investigation record semantic encoding sub-unit is used to: use a semantic encoder containing an RNN to perform semantic encoding on each user credit investigation record item in the list of user credit investigation record items to obtain the time series distribution of the user credit investigation record semantic embedding encoding vectors. By using a semantic encoder containing an RNN to perform semantic encoding on each user credit investigation record item in the list of user credit investigation record items to obtain the time series distribution of the user credit investigation record semantic embedding encoding vectors. It should be understood that the meaning of each credit investigation record item often depends on its preceding and following record items, that is, the context information. When processing sequences, the RNN can transfer the information from the previous time step to the current time step, enabling the model to understand the semantics of each record item based on the context and retaining the characteristics of the time series, thereby more accurately representing the semantic information of each record item.

[0051] Exemplarily, in the user insurance record semantic encoding subunit 323, considering that different time points of insurance information are expressed in each user insurance item, and each piece of information contains the semantic meaning between contexts. Based on this, in the technical solution of this application, semantic encoding is performed on each user insurance item in the list of user insurance items to obtain the time series distribution of the user insurance record semantic embedding encoding vector. Specifically, in an example of this application, the user insurance record semantic encoding subunit is configured to: use the semantic encoder including RNN to perform semantic encoding on each user insurance item in the list of user insurance items to obtain the time series distribution of the user insurance record semantic embedding encoding vector. It is also possible to accurately capture the context semantic information of each insurance item in the entire sequence by using the semantic encoder including RNN to perform semantic encoding on each user insurance item in the list of user insurance items, and obtain the time series distribution of the user insurance record semantic embedding encoding vector.

[0052] Exemplarily, in the record time series semantic random walk unit 33, causal association-driven record semantic dynamic random walks are respectively performed on the time series distribution of the user credit record semantic embedding encoding vector and the time series distribution of the user insurance record semantic embedding encoding vector to obtain the user credit record time series semantic random walk encoding vector and the user insurance record time series semantic random walk encoding vector. It should be understood that considering that the credit and insurance behaviors of users change over time, the changes in these behaviors in different time periods and their causal association effects. For example, a recent claim event may have a greater impact on the current credit status than a claim event several years ago. Therefore, the context information of the front and back time points is used to more comprehensively reflect the overall behavior pattern of the user. In this application, causal association-driven record semantic dynamic random walks are respectively performed on the time series distribution of the user credit record semantic embedding encoding vector and the time series distribution of the user insurance record semantic embedding encoding vector to obtain the user credit record time series semantic random walk encoding vector and the user insurance record time series semantic random walk encoding vector. In this way, the internal correlation relationship and key information of the complex features of the user credit and insurance records can be captured more efficiently, noise interference can be reduced, the limitation of only focusing on the surface local pattern can be broken through, the latent global dependence relationship can be deeply mined, and the evaluation result can be made more accurate and reliable.

[0053] In one embodiment, as Figure 5As shown in the figure, the record timing semantic wandering unit 33 includes: a user credit record semantic implicit feature mining subunit 331, which is used to perform implicit feature mining on each user credit record semantic embedded coding vector in the time series distribution of the user credit record semantic embedded coding vector to obtain the time series distribution of the user credit record semantic embedded deep implicit feature coding vector; a user credit record semantic causal feature construction subunit 332, which is used to construct a user credit record semantic causal association topological feature matrix based on the time series distribution of the user credit record semantic embedded deep implicit feature coding vector; a user credit record semantic surface context dynamic wandering subunit 333, which is used to perform feature dynamic wandering driven by dynamic graph convolution on the user credit record semantic causal association topological feature matrix and the time series distribution of the user credit record semantic embedded coding vector to obtain a user credit record semantic surface context dynamic wandering semantic coding vector; a user credit record semantic hidden layer feature extraction subunit 334, which is used to perform the feature dynamic wandering driven by the dynamic graph convolution on the user credit record semantic causal association topological feature matrix and the time series distribution of the user credit record semantic embedded deep implicit feature coding vector to obtain a user credit record semantic hidden layer context dynamic wandering semantic coding vector; a user credit record feature fusion subunit 335, which is used to fuse the user credit record semantic surface context dynamic wandering semantic coding vector and the user credit record semantic hidden layer context dynamic wandering semantic coding vector to obtain the user credit record timing semantic wandering coding vector.

[0054] Exemplarily, in the user credit record semantic implicit feature mining subunit 331, it should be understood that the core of implicit feature mining lies in alleviating the noise problems and limitations existing in explicit features. For example, some short-term behaviors of users may fluctuate due to external factors. However, if only relying on these short-term behavior data for risk assessment, it may lead to misjudgment. Through implicit feature mining, the system can capture the long-term behavior patterns and potential trends of users, thereby more comprehensively understanding their credit status. For instance, a certain user has had several minor overdue repayment records in the past few years, but the overall proportion of on-time repayments is relatively high. Implicit feature mining can help identify this potential good credit habit rather than simply focusing on individual overdue events. Further, implicit feature mining can efficiently compress important information while eliminating redundant information. This means that the system can still maintain high efficiency and accuracy when processing a large amount of complex data. For example, when processing the data of a user with a multi-year credit history, implicit feature mining can help the system focus on the key factors that truly affect the credit score, such as repayment frequency, default amount, and default time interval, rather than paying too much attention to some unimportant details or outliers. In addition, implicit feature mining can also reveal the internal correlations and causal relationships between data. In the credit record, various behaviors of users are often interrelated. For example, frequent small overdue may indicate greater financial pressure, thereby increasing the future default risk. By mining these implicit features, the system can better understand these internal connections and transform them into the time series distribution of deep implicit feature encoding vectors. This representation method not only retains the time order of the original data but also enhances the ability to capture the dynamic changes of user behaviors. Specifically, the processing process of the user credit record semantic implicit feature mining subunit 331 can be expressed by the formula as follows:

[0055]

[0056]

[0057]

[0058] Among them, is the time series distribution of the user credit record semantic embedding encoding vector, , , and are respectively the 1st, 2nd, th, and th user credit record semantic embedding encoding vectors in the time series distribution of the user credit record semantic embedding encoding vector, is point convolution encoding, is the convolution encoding activation function, , , , and are respectively the 1st, 2nd, th, th, th, and th user credit record semantic embedding depth implicit feature coding vectors in the time series distribution of the user credit record semantic embedding depth implicit feature coding vectors,

[0059] In one embodiment, as Figure 6 shown, the user credit record semantic causal feature construction subunit 332 includes: a user credit record semantic causal factor calculation secondary subunit 3321, configured to calculate a semantic causal association factor between any two user credit record semantic embedding depth implicit feature coding vectors in the time series distribution of the user credit record semantic embedding depth implicit feature coding vectors to obtain a user credit record semantic causal association topology matrix composed of multiple user credit record semantic causal association factors; a user credit record semantic causal trigger secondary subunit 3322, configured to perform causal triggering of a gating activation function on the user credit record semantic causal association topology matrix to obtain the user credit record semantic causal association topology feature matrix.

[0060] In one embodiment, the user credit record semantic causal factor calculation secondary subunit 3321 includes: a semantic association matrix calculation tertiary subunit, configured to calculate an association matrix between any two user credit record semantic embedding depth implicit feature coding vectors to obtain a user credit record semantic association matrix; a causal factor calculation tertiary subunit based on statistical features, configured to perform causal factor calculation based on statistical features on the user credit record semantic association matrix to obtain a user credit record semantic causal association factor corresponding to the user credit record semantic association matrix, where the user credit record semantic causal association factor is related to the mean, variance, maximum value, and causal bias value of the user credit record semantic association matrix.

[0061]

[0062]

[0063] Wherein, is matrix multiplication, is the transposed vector of is and the user credit record semantic association matrix between is The variance of is to take the maximum value in is the mean value of the causal bias value, is the semantic causal association factor corresponding to the user's credit record , , and are respectively the semantic causal association factors of the user's credit record at each position in the semantic causal association topology matrix of the user's credit record is the semantic causal association topology matrix of the user's credit record.

[0064] In one embodiment, the causal factor calculation three-level subunit based on statistical features is configured to: in response to the variance of the user's credit record semantic association matrix being greater than or equal to a predetermined threshold, use the weighted average of the distances between any two user credit record semantic embedded depth implicit feature coding vectors in the time series distribution of the user credit record semantic embedded depth implicit feature coding vector as the causal bias value; in response to the variance of the user's credit record semantic association matrix being less than the predetermined threshold, use the weighted value of the mean value of the user's credit record semantic association matrix as the causal bias value.

[0065] Specifically, when calculating the semantic causal association factor between any two user credit record semantic embedded depth implicit feature coding vectors, the system is actually looking for the internal connections and causal relationships between these features. For example, a user's certain overdue repayment behavior may be related to previous credit limit adjustments or income changes. By constructing a global fine-grained association matrix and using a causal association energy metric function for quantization coding expression, the system can capture these complex causal relationships. Specifically, the system will analyze the degree of association between each pair of depth implicit feature coding vectors and identify which factors have significant causal relationships. This step helps the system extract the key factors that truly affect the user's credit status from the vast amount of data.

[0066] Preferably, in another specific example of the present application, the three - level subunit for calculating causal factors based on statistical features is configured to: in response to the variance of the user's credit record semantic association matrix being greater than or equal to a predetermined threshold, take the weighted average of the distances between any two user's credit record semantic embedded deep implicit feature encoding vectors in the time - series distribution of the user's credit record semantic embedded deep implicit feature encoding vectors as the causal bias value; in response to the variance of the user's credit record semantic association matrix being less than the predetermined threshold, take the weighted value of the mean of the user's credit record semantic association matrix as the causal bias value. This process is represented by the formula:

[0067]

[0068] wherein, is and the user's credit record semantic association matrix between, is the variance of, is the mean of, is and the distance between, is the number of vectors in, is the predetermined threshold, and are weighted hyperparameters, is the causal bias value.

[0069] In the above - mentioned preferred example, by regarding the low - level causal associations in a complex system as molecular - level relationships inferred based on statistical correlations, it is possible to further perform intervention prediction of causal association energy on the basis of global fine - grained statistical association representation, so as to study the fine - grained structure of causal associations and their dynamic regulation based on the high - dimensional and heterogeneous representation of causal genomics. Among them, when the aggregative distribution representation of the causal graph is greater than the predetermined threshold, bias occurs during the integration of source data based on the matrix - based graph node effect representation of the user's credit record semantic causal association factor, and when the aggregative distribution representation of the causal graph is less than the predetermined threshold, condensation - structure modeling can be directly performed through feature - pattern integration and compression. In this way, not only can the causal association energy in the system be encoded and described, but also the implicit causal intervention prediction results can be condensed, thereby obtaining a more efficient revelation of key causal associations.

[0070] Exemplarily, in the user credit record semantic causal trigger secondary subunit 3322, considering that a user credit record semantic causal association topological matrix composed of multiple user credit record semantic causal association factors is obtained, the system enters the next step - performing causal trigger on this matrix using a gated activation function. The core of this process lies in dynamically extracting and modeling causal topological relationships to enhance the expressiveness and accuracy of the model. The gating mechanism allows the model to identify key causal paths in a dynamic context, thereby strengthening important associations and weakening noise interference. For example, certain short-term behavioral fluctuations (such as a small overdue payment once) should not overly affect the overall risk assessment result. Through the gating mechanism, the system can ignore these unimportant details and focus on those long-term and significant causal relationships, such as the impact of multiple large-scale default behaviors on the credit score. On this basis, the non-linear activation function enhances the model's expressive ability by introducing non-linear elements and captures high-order regularities hidden in complex causal structures. Specifically, the processing process of the user credit record semantic causal trigger secondary subunit 3322 can be expressed by the formula:

[0071]

[0072] where is the non-linear activation function, is the normalization threshold, is for to perform gated activation processing, is the user credit record semantic causal association topological feature matrix.

[0073] Exemplarily, in the user credit record semantic surface context dynamic walk sub-unit 333, a dynamic graph convolution-driven feature dynamic walk is performed on the time series distribution of the user credit record semantic causal association topological feature matrix and the user credit record semantic embedding coding vector, aiming to generate the user credit record semantic surface context dynamic walk semantic coding vector. This process captures the deep credit behavior patterns of users by simulating the propagation of features in a complex topological structure and recursively aggregating the explicit semantics between nodes from local to global. Specifically, first, the system has obtained the time series distribution of the user credit record semantic embedding depth implicit feature coding vector through implicit feature mining and constructed the user credit record semantic causal association topological feature matrix. Next, the system combines these two key inputs - the causal association topological feature matrix and the time series distribution of the semantic embedding coding vector - and uses a dynamic graph convolutional neural network (GCN) for feature dynamic walk. In this process, the dynamic graph convolution-driven feature dynamic walk simulates the propagation process of features in a complex topological structure. Specifically, the system starts from each node (i.e., the user credit record semantic embedding coding vector at each time point) and performs a dynamic walk along the edges in the graph (i.e., the causal association path). This walk method not only considers the features of the node itself but also integrates the information of its neighbor nodes, thus forming a more comprehensive and rich context representation. For example, assume that a certain user's overdue repayment record (node A) has a causal association with several previous normal repayment records (nodes B, C, D). Through the dynamic graph convolution-driven feature dynamic walk, the system can capture the mutual influence between these records. The system starts from node A and gradually aggregates the feature information of these nodes along the connected causal paths (such as nodes B, C, D) to form a comprehensive context representation. This way enables the system to better understand the overall credit behavior pattern of the user rather than just looking at individual events in isolation. After the above dynamic walk process, the system finally generates the user credit record semantic surface context dynamic walk semantic coding vector. This vector not only contains the feature information of each node itself but also integrates the influence of its neighbor nodes, thus providing a more comprehensive and detailed context representation. Specifically, this surface context dynamic walk semantic coding vector can help the system better understand the long-term credit behavior pattern of the user and its changing trend. Specifically, the processing process of the user credit record semantic surface context dynamic walk sub-unit 333 can be expressed by the formula as follows:

[0074]

[0075] Among them, is the graph convolution processing, is the user credit record semantic surface context dynamic walk semantic coding vector.

[0076] Exemplarily, in the user credit record semantic hidden layer feature extraction sub-unit 334, for the user credit record semantic embedded depth implicit feature encoding vector, the dynamic random walk mechanism conducts a deeper semantic exploration for implicit features. The hidden layer semantics enables the potential embedding of features. Therefore, it is necessary to pay attention to the multi-hop propagation of high-order information and the distributed decoupling of deep features during the random walk process to avoid the over-smoothing phenomenon caused by the propagation of deep topological features. This hidden layer representation provides greater generalization ability for feature expression by capturing profound temporal dependencies and complex semantic patterns. Specifically, the processing process of the user credit record semantic hidden layer feature extraction sub-unit 334 can be expressed by the formula:

[0077]

[0078] where, is the user credit record semantic hidden layer context dynamic random walk semantic encoding vector.

[0079] Exemplarily, in the user credit record feature fusion sub-unit 335, the system has generated two types of semantic encoding vectors through the feature dynamic random walk driven by dynamic graph convolution: the user credit record semantic surface context dynamic random walk semantic encoding vector and the user credit record semantic hidden layer context dynamic random walk semantic encoding vector. The surface context dynamic random walk semantic encoding vector mainly focuses on the features of the node itself and the influence of its direct neighbors, providing a relatively intuitive and explicit credit behavior pattern; while the hidden layer context dynamic random walk semantic encoding vector deeply explores higher-level relationships and deep causal associations, revealing those complex patterns and potential risk factors hidden beneath the surface. By fusing the semantic encoding vectors of the surface layer and the hidden layer, the system can capture multi-dimensional information of the user's credit behavior. For example, when evaluating a user's credit risk, the surface context dynamic random walk semantic encoding vector can help the system identify the user's recent repayment behavior and default situation, providing short-term behavior pattern analysis. At the same time, the hidden layer context dynamic random walk semantic encoding vector can reveal the user's long-term behavior trends and potential risk factors, such as income fluctuations and the impact of emergencies on credit behavior. Specifically, the processing process of the user credit record feature fusion sub-unit 335 can be expressed by the formula:

[0080]

[0081] where, is the fusion weighting parameter, is the user credit record time-series semantic random walk encoding vector.

[0082] Exemplarily, in the recording multi-modal joint unit 34, it is used to fuse the user's credit investigation record time-series semantic walk coding vector and the user's insurance application record time-series semantic walk coding vector to obtain a user credit risk multi-modal joint coding vector. It should be understood that considering that the user's credit investigation record time-series semantic walk coding vector mainly reflects their credit performance in the credit field, such as repayment ability and willingness; the user's insurance application record time-series semantic walk coding vector focuses on showing the user's behavior and potential risk status in terms of risk protection. The two depict the user from different perspectives. Based on this, in the technical solution of this application, the user's credit investigation record time-series semantic walk coding vector and the user's insurance application record time-series semantic walk coding vector are fused to comprehensively consider the credit investigation and insurance application record information, and a user credit risk multi-modal joint coding vector is obtained, so as to more comprehensively reflect the user's overall behavior pattern. In a specific example, the user's credit investigation record time-series semantic walk coding vector and the user's insurance application record time-series semantic walk coding vector are fused by means of weighted summation or an attention mechanism to comprehensively consider the credit investigation and insurance application record information.

[0083] Exemplarily, in the credit risk judgment unit 35, it is used to obtain an identification result indicating whether the insurance application request is within the credit risk security level according to the user credit risk multi-modal joint coding vector. In one embodiment, the credit risk judgment unit is configured to: input the user credit risk multi-modal joint coding vector into a credit risk identifier based on a classifier to obtain the identification result, and the identification result is used to indicate whether the insurance application request is within the credit risk security level. That is, classification processing is performed on the user credit risk multi-modal joint coding vector obtained by multi-modal joint of the user credit investigation record time-series semantic walk coding vector and the user insurance application record time-series semantic walk coding vector, so as to realize the automatic judgment of whether the insurance application request is within the credit risk security level. It should be understood that the classifier has a powerful pattern recognition ability and can learn the mapping relationship between different credit risk characteristics and security levels after training. The credit risk identifier constructed based on the classifier can give full play to its advantages, analyze and judge the credit risk characteristics contained in the multi-modal joint coding vector, and thus give an accurate identification result. By clarifying whether the insurance application request is within the credit risk security level, financial institutions and insurance companies can effectively screen out high-risk insurance application requests, avoid providing insurance services to users with too high credit risks, thereby reducing potential default risks and economic losses, and ensuring the stable operation of their own businesses. In a specific example, the credit risk identifier based on the classifier uses a softmax classifier, and the Softmax classifier consists of the following parts: Input layer: Receives the user credit risk multi-modal joint coding vector. Fully Connected Layer: Maps the input vector to a hidden layer or directly to the Softmax layer. Usually contains a weight matrix. Softmax layer: Converts the output of the fully connected layer into a probability distribution.

[0084] In summary, the blockchain-based full-process management and risk control system for electronic guarantee letters according to the embodiments of the present application is described. It retrieves the user's credit record and insurance record based on the user's identity information, and uses natural language analysis and processing algorithms based on deep learning to perform data splitting and semantic encoding on the user's credit record and insurance record. Then, causal association semantic dynamic walks are respectively performed on the semantic embeddings of each encoded user credit record and the semantic embedding features of each user insurance record, so as to automatically determine whether the insurance request is within the credit risk safety level based on the multi-modal joint representation of credit risks between the sequential semantic walk features of the user credit record and the sequential semantic walk features of the user insurance record obtained after association. The present application can effectively analyze and utilize the complex semantic information in the user's credit record and insurance record, making the basis for credit risk judgment more comprehensive and accurate. At the same time, the model can be continuously updated and adjusted according to new data, no longer restricted by fixed formulas and preset coefficient factors, improving the accuracy and effectiveness of risk judgment in different scenarios.

[0085] Those of ordinary skill in the art will realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0086] It should be understood that the specific examples herein are only for helping those skilled in the art better understand the embodiments of the present application, rather than limiting the scope of the embodiments of the present application.

[0087] It should also be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the various processes do not mean the order of execution, and the execution order of the various processes should be determined by their functions and internal logics, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0088] It should also be understood that the various implementation manners described in this specification can be implemented alone or in combination, and the embodiments of the present application do not limit this.

[0089] Unless otherwise specified, all technical and scientific terms used in the embodiments of this application have the same meanings as those commonly understood by those skilled in the technical field of this application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application. The term "and / or" used in this application includes any and all combinations of one or more of the related listed items. The singular forms "a", "above-mentioned", and "the" used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0090] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0091] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0092] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0093] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0094] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A blockchain-based electronic guarantee full-process management and risk control system, including: Platform building module, used to build an electronic letter of guarantee management platform using blockchain technology; A request module, used to initiate an insurance request to the electronic letter of guarantee management platform; A verification module, for verifying the identity of the user in response to the electronic letter of guarantee management platform receiving the insurance request to determine whether the insurance request is within the credit risk security level; a letter of guarantee issuance module, for approving the insurance request in response to the insurance request being within the credit risk security level, and issuing an electronic letter of guarantee for the approved insurance request and encrypting and storing it; The verification module is characterized in that the verification module includes: a retrieval unit, which is used to retrieve the user credit record and the user insurance record according to the user identity information; a record splitting and encoding unit, which is used to perform data splitting and semantic encoding on the user credit record and the user insurance record to obtain the time series distribution of the user credit record semantic embedding coding vector and the time series distribution of the user insurance record semantic embedding coding vector; a record time series semantic walking unit, which is used to perform causal association driven record semantic dynamic walking on the time series distribution of the user credit record semantic embedding coding vector and the time series distribution of the user insurance record semantic embedding coding vector to obtain the user credit record time series semantic walking coding vector and the user insurance record time series semantic walking coding vector; a record multimodal joint unit, which is used to fuse the user credit record time series semantic walking coding vector and the user insurance record time series semantic walking coding vector to obtain the user credit risk multimodal joint coding vector; a credit risk judgment unit, which is used to obtain an identification result indicating whether the insurance request is within the credit risk security level according to the user credit risk multimodal joint coding vector; Wherein, the record splitting encoding unit includes: A data splitting subunit, used to split the user credit record and the user insurance record according to the time dimension to obtain a list of user credit record items and a list of user insurance items; A user credit record semantic encoding subunit, configured to semantically encode each user credit record item in the list of user credit record items to obtain a time series distribution of the user credit record semantic embedding encoding vector; The user insurance record semantic coding subunit is used to semantically code each user insurance item in the list of user insurance items to obtain a time series distribution of the semantic embedded coding vector of the user insurance record.

2. The blockchain-based electronic letter of guarantee full-process management and risk control system according to claim 1 is characterized in that: The user credit record semantic encoding subunit is used to: use a semantic encoder including an RNN to semantically encode each user credit record item in the list of user credit record items to obtain a time series distribution of the user credit record semantic embedding encoding vector.

3. The blockchain-based electronic guarantee full-process management and risk control system according to claim 2 is characterized in that: The user insurance record semantic encoding subunit is used to: use the semantic encoder containing RNN to semantically encode each user insurance item in the list of user insurance items to obtain the time series distribution of the user insurance record semantic embedding encoding vector.

4. The blockchain-based electronic guarantee full-process management and risk control system according to claim 3 is characterized in that: The recording time sequence semantic walking unit includes: A user credit record semantic implicit feature mining subunit is used to perform implicit feature mining on each user credit record semantic embedding coding vector in the time series distribution of the user credit record semantic embedding coding vector to obtain a time series distribution of the user credit record semantic embedding depth implicit feature coding vector; A user credit record semantic causal feature construction subunit, for constructing a user credit record semantic causal association topological feature matrix based on the time series distribution of the user credit record semantic embedding deep implicit feature coding vector; A user credit record semantic surface context dynamic walking subunit, used to perform dynamic graph convolution-driven feature dynamic walking on the user credit record semantic causal association topological feature matrix and the time series distribution of the user credit record semantic embedding coding vector to obtain a user credit record semantic surface context dynamic walking semantic coding vector; A user credit record semantic hidden layer feature extraction subunit is used to perform the dynamic graph convolution-driven feature dynamic walk on the user credit record semantic causal association topological feature matrix and the time series distribution of the user credit record semantic embedding depth implicit feature coding vector to obtain the user credit record semantic hidden layer context dynamic walk semantic coding vector; The user credit record feature fusion subunit is used to fuse the user credit record semantic surface context dynamic wandering semantic coding vector and the user credit record semantic hidden context dynamic wandering semantic coding vector to obtain the user credit record time series semantic wandering coding vector.

5. The blockchain-based electronic guarantee full-process management and risk control system according to claim 4 is characterized in that: The user credit record semantic causal feature construction subunit includes: A user credit record semantic causal factor calculation secondary subunit is used to calculate the semantic causal association factor between any two user credit record semantic embedding depth implicit feature coding vectors in the time series distribution of the user credit record semantic embedding depth implicit feature coding vector to obtain a user credit record semantic causal association topological matrix composed of multiple user credit record semantic causal association factors; The user credit record semantic causal triggering secondary sub-unit is used to perform causal triggering of the gated activation function on the user credit record semantic causal association topology matrix to obtain the user credit record semantic causal association topology feature matrix.

6. The blockchain-based electronic guarantee full-process management and risk control system according to claim 5 is characterized in that: The user credit record semantic causal factor calculation secondary subunit includes: The semantic association matrix calculation tertiary subunit is used to calculate the association matrix between any two user credit record semantic embedding deep implicit feature encoding vectors to obtain the user credit record semantic association matrix; The third-level sub-unit for causal factor calculation based on statistical features is used to perform causal factor calculation based on statistical features on the user credit record semantic association matrix to obtain the user credit record semantic causal association factor corresponding to the user credit record semantic association matrix, wherein the user credit record semantic causal association factor is related to the mean, variance, maximum value and causal bias value of the user credit record semantic association matrix.

7. The blockchain-based electronic guarantee full-process management and risk control system according to claim 6 is characterized in that: The statistical feature-based causal factor calculation three-level sub-unit is used to: In response to the variance of the user credit record semantic association matrix being greater than or equal to a predetermined threshold, taking the weighted average of the distances between any two user credit record semantic embedding deep implicit feature encoding vectors in the time series distribution of the user credit record semantic embedding deep implicit feature encoding vectors as the causal bias value; In response to the variance of the user credit record semantic association matrix being less than the predetermined threshold, a weighted value of the mean of the user credit record semantic association matrix is ​​used as the causal bias value.

8. The blockchain-based electronic letter of guarantee full-process management and risk control system according to claim 7 is characterized in that: The credit risk judgment unit is used to: input the user credit risk multimodal joint encoding vector into a classifier-based credit risk identifier to obtain the identification result, and the identification result is used to indicate whether the insurance request is within the credit risk safety level.

Images