Risk assessment method and device based on trusted data space and related equipment

Abstract

The invention discloses a risk assessment method and device based on a trusted data space and related equipment. The method comprises the steps of obtaining a digital contract, participant information, transaction data, environment data and data space operation data of the trusted data space; carrying out anchor point hash calculation to determine whether tampering is carried out; obtaining a plurality of risk indexes and risk characteristic values; and based on the global weight and the risk feature value of each risk index, obtaining the size of each risk level preset by the digital contract. According to the method, whether the data information is tampered or not can be quickly determined through anchor point hash calculation, changes of multi-source heterogeneous data such as digital contracts, participant information, transaction data, environment data and data space operation data can be determined in real time by utilizing the preset large language model, and risk judgment is made in real time. In addition, when the risk is judged based on the risk characteristic value of each risk index, the importance degree of each risk index in the global data contract is considered, so that the judgment is better.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of big data, and in particular to a risk assessment method and device based on a trusted data space and related equipment. BACKGROUND

[0002] At present, for digital contracts in a trusted data space, risk assessment is mainly performed in a manner combining a traditional rule engine and manual review. The traditional rule engine performs risk judgment based on preset fixed rules. For a complex and variable trusted data space environment, the adaptability is poor, and it is unable to dynamically handle changes in multi-source heterogeneous data (such as time-series transaction data and unstructured environmental policy data) in the trusted data space, and the data tampering risk detection efficiency is low. SUMMARY

[0003] The main purpose of the present application is to propose a risk assessment method and device based on a trusted data space and related equipment, aiming to solve the problems in the prior art that the adaptability is poor based on a traditional rule engine, it is unable to dynamically handle changes in multi-source heterogeneous data (such as time-series transaction data and unstructured environmental policy data) in the trusted data space, and the data tampering risk detection efficiency is low.

[0004] To achieve the above-mentioned purpose, the present application proposes a risk assessment method based on a trusted data space, comprising: obtaining a digital contract, participant information, transaction data, environmental data, and data space running data of a trusted data space; the digital contract, the participant information, the transaction data, the environmental data, and the data space running data are multi-source heterogeneous data; respectively performing anchor point hash calculation based on the digital contract, the participant information, the transaction data, the environmental data, and the data space running data to determine whether they are tampered with; if not tampered with, performing text semantic analysis on the digital contract through a preset large language model to obtain a plurality of risk indicators corresponding to each clause in the digital contract; performing semantic recognition and associated field matching on each data or information through a preset large language independent model corresponding to the participant information, the transaction data, the environmental data, and the data space running data respectively to obtain risk characteristic values of each risk indicator; the risk characteristic value of each risk indicator is obtained based on the field content of the corresponding risk indicator; obtain sizes of each risk level preset for the digital contract based on global weights of each risk indicator and risk characteristic values of each risk indicator; wherein the global weights of each risk indicator are obtained based on types of clauses to which each risk indicator belongs through preset risk causal model analysis, and the preset risk causal model is obtained based on a set of fulfillment data, wherein the set of fulfillment data includes types of digital contract clauses and fulfillment results.

[0005] The application also provides a risk assessment device based on a trusted data space, comprising: An obtaining module is configured to obtain a digital contract, participant information, transaction data, environment data, and data space running data of the trusted data space; the digital contract, the participant information, the transaction data, the environment data, and the data space running data are multi-source heterogeneous data. A processing module is configured to perform anchor point hash calculation based on the digital contract, the participant information, the transaction data, the environment data, and the data space running data, and determine whether they are tampered with. If they are not tampered with, a preset large language model is used to perform text semantic analysis on the digital contract to obtain a plurality of risk indicators corresponding to each clause in the digital contract. A preset large language independent model corresponding to the participant information, the transaction data, the environment data, and the data space running data is used to perform semantic recognition and associated field matching on each data or information to obtain risk characteristic values of each risk indicator; the risk characteristic values of each risk indicator are obtained based on field contents of the corresponding risk indicator. obtain sizes of each risk level preset for the digital contract based on global weights of each risk indicator and risk characteristic values of each risk indicator; wherein the global weights of each risk indicator are obtained based on types of clauses to which each risk indicator belongs through preset risk causal model analysis, and the preset risk causal model is obtained based on a set of fulfillment data, wherein the set of fulfillment data includes types of digital contract clauses and fulfillment results.

[0006] The application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method of any of the above embodiments.

[0007] The application also provides a computing device comprising a processor, wherein the processor has a computer program stored thereon, and the computer program is executed to implement the method of any of the above embodiments.

[0008] In the embodiment of the present application, when risk assessment is performed based on the trusted data space, firstly, whether each data and information is tampered with can be determined based on the anchor point hash, then, the preset large language model is used to perform semantic analysis on the digital contract and construct a plurality of risk indicators of each clause in the digital contract, and the preset large language model corresponding to the participant information, transaction data, environmental data, and data space running data is used to perform semantic conversion on the participant information, transaction data, environmental data, and data space running data respectively, identify fields related to each risk indicator, and then determine the risk characteristic value of each risk indicator according to the field corresponding to each risk indicator, and make a risk judgment based on the risk characteristic value of each risk indicator. This method can determine the changes of multi-source heterogeneous data such as digital contract, participant information, transaction data, environmental data, and data space running data in real time by using each preset large language model, and make a risk judgment in real time. In addition, when judging the risk based on the risk characteristic value of each risk indicator, the importance of each risk indicator in the global data contract is also considered, so the judgment is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0010] Figure 1 The step diagram of the risk assessment method based on the trusted data space in an embodiment of the present application; Figure 2 The module diagram of the risk assessment device based on the trusted data space in an embodiment of the present application; Figure 3 The module diagram of the computer readable storage medium in an embodiment of the present application; Figure 4 The module diagram of the computing device in an embodiment of the present application.

[0011] The implementation, functional features and advantages of the present application will be further described with reference to the drawings. DETAILED DESCRIPTION

[0012] The principles and spirits of the present application will be described below with reference to a plurality of exemplary embodiments. It should be understood that these embodiments are presented only to enable those skilled in the art to better understand and implement the present application, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0013] Those skilled in the art will understand that embodiments of this application can be implemented as a system, device, method, or computer program product. Therefore, this application can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0014] According to the embodiments of this application, a risk assessment method, apparatus and related equipment based on trusted data space are proposed.

[0015] In this article, it is important to understand that any number of elements in the accompanying figures is for illustrative purposes and not for limitation, and any naming is for distinction only and has no limiting meaning.

[0016] The principles and spirit of this application will be explained in detail below with reference to several representative embodiments.

[0017] Exemplary methods like Figure 1 As shown in the embodiments of this application, the risk assessment method based on trusted data space includes the following steps S100-S600: S100: Obtain digital contracts, participant information, transaction data, environmental data, and data space operation data from a trusted data space; the digital contracts, participant information, transaction data, environmental data, and data space operation data are multi-source heterogeneous data.

[0018] In this application embodiment, the Trusted Data Space (TDS) is a multi-party data sharing and interaction environment that ensures the authenticity, integrity, and confidentiality of data during its flow through data encryption, identity authentication, and access control, while also meeting compliance requirements (such as the General Data Protection Regulation (GDPR) and data security laws). It serves as the basic carrier for the performance of digital contracts.

[0019] In this embodiment, the digital contract is a core protocol that regulates data interaction behavior in a trusted data space. It clarifies the rights and obligations of data providers, users, and service providers, and serves as the rule basis for data collection, use, and transfer. For example, in this embodiment, the digital contract may include the following: Contract terms: In the trusted data space, contract terms need to comply with laws and regulations such as data security law and personal information protection law, and at the same time, the use scenarios of data (such as only for data analysis, not for secondary resale), data processing methods (such as whether to allow sharing after desensitization, whether to need encrypted transmission), breach of contract responsibilities (such as compensation standards for out-of-range use), dispute resolution mechanisms (such as arbitration agency selection) and the like are explicitly stated. For example: In a certain manufacturing trusted data space, the contract terms between A enterprise and B enterprise stipulate that "B enterprise can use A enterprise's production energy consumption data for energy efficiency optimization analysis, and the data is valid for 3 months, and cannot be used for cooperation projects with C enterprise".

[0020] Signatory information: including the unique identification (such as digital certificate number) of all parties participating in the digital contract in the trusted data space, the type of the subject (enterprise, institution, individual), the information of the legal representative / person in charge, the contact information, etc., to ensure that the identity of the signatory can be traced and verified. For example: The signatory information records "Enterprise X (unified social credit code XXX, digital certificate number XXX), institution Y (institutional legal certificate number XXX)", and is associated with the identity authentication system of the trusted data space.

[0021] Performance period: Clearly define the time range of data provision, use, settlement and other behaviors, including data delivery start / end time, use permission effective / invalidation time, fee settlement period, etc., which is synchronized recorded by the timestamp system of the trusted data space to ensure that the time cannot be tampered with. For example: "From January 1, 2024 to December 31, 2024, A enterprise needs to provide real-time inventory data to B enterprise every day, and B enterprise needs to pay the data use fee of the previous month before May 5th of each month."

[0022] Performance amount: The fee standard and payment method of data transaction, including unit price (such as 0.5 yuan per data), total amount, payment period (such as by time, by month), payment channel (need to pass through the compliance payment interface in the trusted data space), and is bound with the performance progress of the contract (such as automatically triggering payment after data delivery meets the standard). For example: "Based on the actual logistics trajectory data volume transmitted each month, settle 0.1 yuan per data by the bank interface in the data space before the 10th of each month."

[0023] In the embodiments of the present application, the participant data refers to the basic information and credit-related data of all subjects (such as enterprises, institutions, individuals, etc.) accessing the trusted data space, which is used to assess the compliance and credibility of the subjects and is the identity credential for data space access and interaction. For example, in the embodiments of the present application, the participant data can include the following contents: Enterprise basic information: includes the business license information (registered capital, scope of business, registered address) submitted by the enterprise when registering in the trusted data space, industry classification (such as manufacturing, finance), organizational structure (such as department connection and responsible person), and data security responsible person information. For example: the enterprise basic information of an e-commerce enterprise records "the scope of business includes network sales, and the data security responsible person is Wang XX (contact information XXX)", and is associated with the enterprise qualification audit record of the trusted data space.

[0024] Credit record: the credit performance of the subject in the trusted data space and external system, including whether there are data violation behaviors (such as using data beyond the scope), administrative penalty records (such as being punished by the supervision department due to data leakage), and court credit information, which is dynamically updated by the credit evaluation system of the trusted data space. For example: Enterprise Z was marked "violated the rules in June 2023, credit level dropped to B level, and restricted from participating in high-sensitive data transactions" in the credit record due to unauthorized sharing of user data in 2023.

[0025] Historical performance: the performance of the subject in the past digital contract, including whether to deliver data on time, whether the data quality meets the standard (such as accuracy, completeness), whether to pay the fee as agreed, etc., which is used for risk assessment of subsequent contract cooperation. For example: the historical performance record of enterprise Y shows "100% timely performance in contracts with 3 enterprises in 2022, data accuracy rate 98%; 1 time delayed delivery due to system failure in 2023, has been compensated according to the contract".

[0026] In the embodiments of the present application, transaction data refers to the records generated by data transactions and related supporting behaviors in the trusted data space, which are the process credentials of data value circulation, leaving traces throughout the process and being tamper-proof. For example, in the embodiments of the present application, transaction data can include the following contents: Transaction flow: records the key information of each data transaction, including transaction number (unique identifier in the data space), transaction parties (associated with the signing party information), transaction data type (such as user portrait data, sensor data), transaction quantity (such as 1000 records), transaction time (accurate to milliseconds, associated with timestamp), transaction status (such as to be paid, completed, canceled) and the like. For example: transaction flow record "number T20240501001, A enterprise purchases 'East China region home appliance sales data' 500 from B enterprise, transaction time 2024-05-01 09:30:15, status: completed".

[0027] Payment record: The record of the fund flow associated with the transaction flow, including the payment amount, payment time, payment channel (such as the bank account connected by the data space, third-party payment platform), payment voucher (such as electronic receipt number), payee account information (which needs to be recorded in the data space), etc., to ensure that the fund flow is traceable. For example: the payment record shows “corresponding to transaction T20240501001, B enterprise paid 500 yuan through XX bank interface on 2024-05-01 09:35:20, voucher number P20240501001”.

[0028] In the embodiments of the present application, environmental data refers to external dynamic information that affects data transactions and subject decisions within a trusted data space, which is used to assist participants in assessing risks and optimizing strategies, and is an external reference system for data interaction. For example, in the embodiments of the present application, environmental data can include the following contents: Market situation: Market supply and demand and price fluctuation information related to data transactions, such as market transaction volume, average price, and popular procurement direction (such as for product research and development) of a certain type of data (such as new energy vehicle user data), which is synchronized in real time by the data space through an industry database interface. For example: the environmental data records “in Q1 2024, the market transaction volume of manufacturing industry energy consumption data increased by 15% year-on-year, the average price was 0.8 yuan per piece, and the main purchaser was an energy-saving service enterprise”.

[0029] Policy and regulation changes: Legal and regulatory updates related to data security, privacy protection, and transaction compliance, which are automatically captured and pushed by the policy monitoring module of the data space. For example: “on June 1, 2024, the ‘XX Province Data Transaction Management Method’ was formally implemented, adding the ‘sensitive data transaction needs to be recorded’ clause, and the data space has updated the contract template in synchronization”.

[0030] Industry trends: Information such as technical development and major events in the industry to which the data belongs, such as the emergence of new data security technologies (such as the application of federated learning in medical data) in a certain industry, and data cooperation cases of leading enterprises, to help participants adjust data strategies. For example: “in May 2024, the financial industry trusted data space introduced a new mode of ‘data asset pledge’, and a bank has completed the first data pledge loan through this mode”.

[0031] In the embodiments of the present application, data space running data refers to the records generated in the process of running the trusted data space itself, which is used to ensure the stability, security and auditability of the data space, and is the log archive of data space operation and maintenance. For example, in the embodiments of the present application, the data space running data can include the following contents: Data storage status: Record the storage location of data in the trusted data space (such as the number of distributed storage nodes), storage format (such as encrypted files, structured databases), storage capacity (used / remaining space), data backup status (such as whether off-site backup is completed, backup time), data life cycle (such as creation time, expiration time) and other information. For example: "Enterprise A's '2023 sales data' is stored in node S001, the format is AES encrypted JSON, and it has completed 3 off-site backups, and is expected to expire on December 31, 2026."

[0032] Data transmission record: The whole link information of data flow in space, including transmission initiator / receiver (associated participant identification), transmission data ID (associated transaction serial number), transmission time, transmission path (such as node A to node B), transmission status (such as success, failure, retry times), encryption method (such as TLS1.3) and other information, to ensure that data transmission is traceable. For example: "2024-05-01 09:32:00, enterprise A transmits data ID D1001 to B, path S001→S003, encryption method TLS1.3, transmission success, time 2.5 seconds."

[0033] Access log: Record of all subjects accessing data space and data, including access subject (associated participant information), access time, access object (such as a contract, a group of data), access operation (such as query, download, modification), access permission verification result (such as pass, reject and reason), terminal device information (such as IP address, device fingerprint) and other information. For example: "2024-05-01 10:00:15, enterprise B accesses data D1001 through IP 113.XXX.XXX, operation: download, permission verification passed, device fingerprint: XXX."

[0034] S200: Based on the digital contract, the participant information, the transaction data, the environment data, and the data space running data, respectively, anchor point hash calculation is performed to determine whether it is tampered with.

[0035] In the embodiments of the present application, whether the digital contract, the participant information, the transaction data, the environment data, and the data space running data are tampered with can be determined based on the following steps S210-240: S210: Determine the core fields of the digital contract, the participant information, the transaction data, the environment data, and the data space running data, and the anchor point clauses of each core field in the digital contract; S220: bind the core fields of the digital contract, the participant information, the transaction data, the environment data, and the data space running data with their corresponding anchor clauses to generate respective binding results; S230: perform hash operations on the binding results corresponding to the digital contract, the participant information, the transaction data, the environment data, and the data space running data respectively to obtain current contract anchor hashes corresponding to the digital contract, the participant information, the transaction data, the environment data, and the data space running data; S240: compare the current contract anchor hashes corresponding to the digital contract, the participant information, the transaction data, the environment data, and the data space running data respectively with initial contract anchor hashes corresponding to the digital contract, the participant information, the transaction data, the environment data, and the data space running data in the digital space to determine whether the digital contract, the participant information, the transaction data, the environment data, and the data space running data are tampered with.

[0036] In the embodiments of the present application, the core fields of the digital contract can be key information directly bound to the validity of the data contract, rights and obligations, such as: performance guarantee margin amount, data delivery timestamp, signatory identity, authorized use period, data use range definition field, etc. If these fields are tampered with, the legal effect of the digital contract will be directly destroyed.

[0037] In the embodiments of the present application, the anchor clause of the core field of the digital contract in the digital contract can be the unique identifier (ContractID) of the digital contract. Binding the core field of the digital contract with the unique identifier of the digital contract and then performing hash encryption can ensure the legal effect of the digital contract.

[0038] In the embodiments of the present application, the core fields of the participant information can be key information related to identity authenticity and credit validity, such as: enterprise unified social credit code, legal representative identity card number, credit rating score, number of historical default records and timestamp, etc. If these core fields are tampered with, the participant's identity may be impersonated, the credit status may be distorted, and the signing qualification of the data contract or the performance credibility may be affected.

[0039] In the embodiments of the present application, the anchor clause of the core field of the participant information in the digital contract can be the participant qualification clause. Binding the core field of the participant information with the anchor clause and then performing hash encryption can ensure the non-tamperability of the participant's identity and credit information.

[0040] In the embodiments of the present application, the core fields of the transaction data can be information related to transaction authenticity and amount accuracy, such as: transaction amount, payment timestamp, transaction party account identifier, logistics waybill number, and receipt confirmation timestamp. If these fields are tampered with, it will lead to false transactions and incorrect fund settlement, and violate the transaction authenticity guarantee clause in the contract.

[0041] In the embodiments of the present application, the anchor clause of the core fields of the transaction data in the digital contract can be a transaction subject and amount agreement related clause. The core fields in the transaction data are bound to the anchor clause, and then hashed and encrypted, which can ensure that the transaction link is traceable.

[0042] In the embodiments of the present application, the core fields of the environmental data can be key information related to data contract validity conditions and risk early warning, such as: policy and regulation validity / invalidity timestamp, industry benchmark price, market supervision level identifier, and sudden risk event record. If these fields are tampered with, it may lead to distortion of external conditions on which the contract is based, and violation of the contract environmental adaptability clause.

[0043] In the embodiments of the present application, the anchor clause of the core fields of the environmental data in the digital contract can be an external condition reference clause. The core fields of the environmental data are bound to the corresponding anchor clause, and then hashed and encrypted, which can ensure the effectiveness of the environmental data as a supplementary basis for the contract.

[0044] In the embodiments of the present application, the core fields of the data space running data can be key information related to data security and permission control, such as: data accesser identity identifier, access timestamp, access operation type (read / write / process), data transmission node IP, and storage location hash value. If these fields are tampered with, it may cover up unauthorized access and data leakage, and violate the data security protection clause in the contract.

[0045] In the embodiments of the present application, the anchor clause of the data space running data in the digital contract can be a clause related to access permission level agreement. The core fields of the digital space running data are bound to the anchor clause, and then hashed and encrypted, which can ensure that the data flow complies with the permission rules.

[0046] For example, for a digital contract, assuming that the core field is F, and the unique identifier of the digital contract is ContractID, the core field of the digital contract is bound to the digital contract identifier, such as structurally splicing the core field value F with ContractID to obtain a binding word result S.

[0047] In addition, a separator can also be added during splicing to avoid confusion between the field value and the ID, such as: S = F + "|" + ContractID; Wherein, “|” is a separator.

[0048] After obtaining the binding result, the binding result can be hashed and encrypted, as follows: Hanchor = SHA-256(S) = SHA - 256( F + "∣" + ContractID ) Wherein, SHA-256 is a hash algorithm, Hanchor is the current contract anchor hash value obtained by encrypting the binding result of the core field of the current digital contract and the anchor clause, which includes both the core field of the digital contract and the anchor clause.

[0049] In addition, when the digital contract is initially generated, the initial contract anchor hash value of the digital contract is also stored, wherein the initial contract anchor hash value of the digital contract is obtained by hashing and encrypting the binding result of the core field of the initial digital contract and the unique identifier of the initial digital contract. Then, the current contract anchor hash value of the current digital contract is compared with the initial contract anchor hash value. If they are consistent, it means that the core field and the anchor clause in the current digital contract have not been tampered with. If they are not consistent, it means that the information field or the anchor clause in the current digital contract has been tampered with, which is risky.

[0050] Similarly, when the initial participant information, transaction data, environmental data, and data space running data are generated, the core field and the anchor clause of each are bound, and the binding result is hashed and encrypted to obtain the initial contract hash value. Then, when the participant information, transaction data, environmental data, and data space running data are obtained, the current core field and the current anchor clause of each are bound, and the binding result is hashed and encrypted to obtain the current contract anchor hash value of each. Then, the current contract anchor hash value of each is compared with the initial contract anchor hash value of each to determine whether the core field and the anchor clause of the participant information, transaction data, environmental data, and data space running data have been tampered with.

[0051] In the embodiments of the present application, when the current contract anchor hash value of the digital contract, participant information, transaction data, environmental data, and data space running data is different from the original contract anchor hash value, the anchor clause bound can be used to accurately locate the clause that violates the contract due to data tampering, and a breach of contract reminder can be sent.

[0052] S300: performing text semantic analysis on the digital contract by a preset large language model to obtain a plurality of risk indicators corresponding to each clause in the digital contract.

[0053] In the embodiments of the present application, each risk indicator can be obtained based on the following steps S310-S320: S310: determining a clause type of each clause in the digital contract based on the preset large language model, the clause type including a core class clause, an auxiliary class clause, and an external environment class clause.

[0054] S320: constructing each risk indicator under each type of clause based on the core class clause, the auxiliary class clause, and the external environment class clause, respectively.

[0055] In an embodiment of the present application, the preset large language model can include a contract parsing model. Based on the contract parsing model, the data contract is parsed. According to the semantics of each clause in the data contract, the priority of each clause is determined, and then each clause is divided into a core class clause, an auxiliary class clause, and an external environment class clause.

[0056] In an embodiment of the present application, the contract parsing model can be pre-trained and fine-tuned based on a BERT (Bidirectional Encoder Representations from Transformers) model. Specifically, it can include a pre-training stage and a fine-tuning stage, as follows: Pre-training stage The training sample of the pre-training stage can use a large-scale unlabeled text corpus, such as the multilingual, structured encyclopedia knowledge contained in Wikipedia, and the text content of various books contained in Books Corpus.

[0057] The input format of each training sample during training can be in the form of including two sentences for each sample. The input sequence of each training sample can be constructed in the following way: 1) Add special mark [CLS] (used for subsequent classification task aggregation representation) at the beginning; 2) Separate the two sentences with [SEP] mark; 3) The vector of each token (vocabulary) is composed of three parts: Token Embeddings, the embedding vector of the vocabulary itself; Segment Embeddings, the embedding that distinguishes the two sentences (such as 0 for sentence A and 1 for sentence B); Position Embeddings, the position of the token in the sequence.

[0058] Two self-supervised tasks can be used to train the model during pre-training, such as a masked language model (MLM) and a next sentence prediction model. Based on the masked language model, 15% of the tokens in the input text are randomly masked, for example, 80% of the tokens in the input text are replaced with a special mark [MASK], 10% of the tokens are replaced with a random word in the vocabulary, and 10% of the tokens remain unchanged. The model needs to predict the masked token, thereby forcing the learning of the context semantic association. In addition, 50% of the two sentences input in each training sample are continuous real sentence pairs, and 50% are randomly spliced sentence pairs. Based on the next sentence prediction model, the output of [CLS] is used to predict whether the two sentences are continuous, and the semantic logic at the sentence level is learned.

[0059] In addition, in the embodiments of the present application, the loss function in the pre-training optimization process can be the sum of the MLM (Masked Language Modeling) loss function and the NSP (Next Sentence Prediction) loss function. The model parameters are iteratively optimized by the gradient descent algorithm (such as Adam) until the loss converges.

[0060] Fine-tuning stage During fine-tuning, the existing BERT pre-training model can be fine-tuned based on the text data in the field of digital contracts and digital rights (such as supplementary clauses and dispute records).

[0061] During fine-tuning, the parameters of the pre-trained model can be frozen, and only the top network is updated, with the goal of training in the field of classification, matching, and other tasks (such as contract clause compliance judgment, priority judgment, core class clause judgment, auxiliary class clause judgment, external environment class clause judgment, and risk indicator judgment) to make the model more accurately understand the terminology in the field of data contracts.

[0062] For example, in the embodiments of the present application, through the fine-tuning stage, the contract analysis model can judge that the step-by-step performance clause in the digital contract is a core clause, and the corresponding risk indicators are the step-by-step clause default risk indicator, the threshold calculation error risk indicator, etc. For another example, through the fine-tuning stage, the contract analysis model can judge that the performance milestone clause in the digital contract is a core clause, and the corresponding risk indicators are the milestone progress lag risk indicator, the milestone data quality risk indicator, etc. For example, through the fine-tuning stage, the contract analysis model can determine that the data availability clause in the digital contract is the core clause, and the corresponding risk indicators are the availability timeout affecting business risk indicator and the availability stability insufficient risk indicator.

[0063] For example, through the fine-tuning stage, the contract analysis model can determine that the data availability clause in the digital contract is the core clause, and the corresponding risk indicators are the availability timeout affecting business risk indicator and the availability stability insufficient risk indicator.

[0064] For example, through the fine-tuning stage, the contract analysis model can determine that the auxiliary clause in the digital contract is the core clause, and the corresponding risk indicators are the report format error risk indicator and the contact information update delay risk indicator.

[0065] For example, through the fine-tuning stage, the contract analysis model can determine that the external condition reference clause and the contract environment adaptability clause in the digital contract are external environment clauses, and the corresponding risk indicators are the policy change leading to limited data use range risk indicator and the industry fluctuation affecting performance risk indicator.

[0066] It should be noted that the above core clauses, auxiliary clauses, external environment clauses, and risk indicators corresponding to each clause are exemplary. The specific core clauses, auxiliary clauses, and external environment clauses contained in the clauses and risk indicators are based on specific digital contracts.

[0067] S400: Through the preset large language independent model corresponding to the participant information, the transaction data, the environment data, and the data space running data, semantic recognition and associated field matching are performed on each data or information, respectively, to obtain risk characteristic values of each risk indicator. The risk characteristic value of each risk indicator is obtained based on the field content of the corresponding risk indicator.

[0068] In the embodiments of the present application, when the participant information, the transaction data, the environment data, and the data space running data are semantically converted and the fields corresponding to each risk indicator in the participant information, the transaction data, the environment data, and the data space are identified, different large language model models can be used respectively.

[0069] In the embodiments of the present application, for the participant information, the participant credit model can be used to perform semantic conversion on the obtained participant information to determine the fields in the participant information corresponding to each risk indicator, and then the fields in the participant information corresponding to each risk indicator can be mapped to the corresponding risk indicator to obtain the risk characteristic value of each risk indicator.

[0070] In the embodiments of the present application, the participant credit model can also be pre-trained and fine-tuned based on the BERT (Bidirectional Encoder Representations from Transformers) model. The training process in the pre-training stage is the same as that of the contract analysis model, and is not described here. In the fine-tuning stage, the parameters of the pre-trained model can be frozen, and only the top network is updated to fine-tune the training of the fields corresponding to the risk indicators related to the participant information as the target, so that the participant credit model can more accurately understand the fields corresponding to each risk indicator in the participant information, such as the credit record field, the historical performance field, etc., and make the participant credit model more accurately understand the participant in the risk features corresponding to each risk indicator.

[0071] For example, for the report format error risk indicator and the contact information update delay risk indicator, the participant credit model can read the participant information and determine the related fields in the participant information corresponding to the auxiliary clauses in the digital contract, such as the report format field and the contact information update field. Then, based on the report format field, the risk feature value of the report format error risk indicator is determined, and based on the contact information update field, the risk feature value of the contact information update delay risk indicator is determined.

[0072] In the embodiments of the present application, for the transaction data, the transaction contract model can be used to perform semantic conversion on the obtained transaction data to determine the fields in the transaction data corresponding to each risk indicator, and then map the fields in the transaction data corresponding to each risk indicator to each risk indicator to obtain the risk feature value of each risk indicator.

[0073] In the embodiments of the present application, for the transaction data, the transaction contract model can be used to perform semantic conversion on the obtained transaction data to determine the fields in the transaction data corresponding to each risk indicator, and then map the fields in the transaction data corresponding to each risk indicator to each risk indicator to obtain the risk feature value of each risk indicator.

[0074] For example, for the milestone progress lag risk indicator and the milestone data quality substandard risk indicator, the transaction data can be semantically converted based on the transaction contract schema, and fields corresponding to the milestone performance clause in the digital contract are determined, such as a performance progress field, a time node field, a quality field, and a key field missing rate. Then, the progress gap feature and the time node deviation feature are determined based on the performance progress field and the time node field, the risk feature value of the milestone progress lag risk indicator is determined based on the progress gap feature and the time node deviation feature, the quality deviation feature is determined based on the quality field, and the risk feature value of the milestone data quality substandard risk indicator is determined based on the quality deviation feature and the key field missing rate.

[0075] For example, for the ladder clause breach risk indicator and the threshold calculation error risk indicator, the transaction data can be semantically converted based on the transaction contract schema, and fields corresponding to the ladder performance clause in the digital contract are determined, such as a condition threshold field, an action execution field, a clause indicator field, a statistical indicator field, and a threshold trigger state field. Then, the risk feature value of the ladder clause breach risk indicator is determined based on the condition threshold field and the action execution field, and the risk feature value of the threshold calculation error risk indicator is determined based on the clause indicator field, the statistical indicator field, and the threshold trigger state field.

[0076] In the embodiments of the present application, for the data space running data, the running monitoring model can be used to semantically convert the obtained data space running data, determine the fields in the data space running data corresponding to each risk indicator, and map the fields in the data space running data corresponding to each risk indicator to each risk indicator to obtain the risk feature value of each risk indicator.

[0077] The running monitoring model can also be pre-trained and fine-tuned based on the BERT (Bidirectional Encoder Representations from Transformers) model. The training process in the pre-training stage is the same as that of the contract analysis model, and is not repeated here. In the fine-tuning stage, the parameters of the pre-trained model are frozen, and only the top network is updated. The fields corresponding to the risk indicators related to the data space running data are taken as the target for fine-tuning training, so that the running monitoring model can more accurately understand the fields in the data space running data corresponding to each risk indicator, such as the access log field, the data access timeout field, and the storage state field.

[0078] For example, for the overuse of unauthorized risk indicator and the transfer authorization out-of-bound risk indicator, the data space running data can be semantically converted based on the running monitoring model, and fields corresponding to the data rights and use range clauses in the digital contract are determined therefrom, such as the timestamp field, the permission state field, the domain identification field, and the transfer authorization level field. Then, based on the timestamp field and the permission state field, the risk characteristic value of the overuse of unauthorized risk indicator is determined, and based on the domain identification field and the transfer authorization level field, the risk characteristic value of the transfer authorization out-of-bound risk indicator is determined.

[0079] For example, for the availability timeout affects business risk indicator and the availability stability insufficient risk indicator, the data space running data can be semantically converted based on the running monitoring model, and fields corresponding to the availability clauses in the digital contract are determined therefrom, such as the response time deviation field, the continuous timeout duration field, the success rate deviation field, and the failure frequency field. Then, based on the response time deviation field and the continuous timeout duration field, the risk characteristic value of the availability timeout affects business risk indicator is determined, and based on the success rate deviation field and the failure frequency field, the risk characteristic value of the availability stability insufficient risk indicator is determined.

[0080] In the embodiments of the present application, for the environment data, the environment adaptation model can be used to semantically convert the environment data, determine the fields in the environment data corresponding to each risk indicator, and then map the fields in the environment data corresponding to each risk indicator to each risk indicator to obtain the risk characteristic value of each risk indicator.

[0081] The environment adaptation model can also be pre-trained and fine-tuned based on the BERT (Bidirectional Encoder Representations from Transformers) model. The training process in the pre-training stage is the same as that of the contract analysis model, and is not described here. In the fine-tuning stage, the parameters of the pre-training model are frozen, and only the top network is updated. The field corresponding to the risk indicator related to the environment data is taken as the target for fine-tuning training, so that the environment adaptation model can more accurately understand the fields in the environment data corresponding to each risk indicator, such as the policy field and the external reference clause field. For example, for the policy change leading to limited data use range risk indicator, the environment data can be semantically converted based on the environment adaptation model, and the field corresponding to the external condition reference clause in the digital contract is determined therefrom, such as the policy field, and the risk characteristic value of the policy change leading to limited data use range risk indicator is determined based on the policy field.

[0082] For example, for the industry fluctuation impact performance progress risk indicator, the environmental data can be semantically converted based on the environmental adaptation model, and the corresponding fields corresponding to the external condition reference clauses in the digital contract, such as the policy field and the industry dynamic field, can be determined therefrom, and the risk characteristic value of the industry fluctuation impact performance progress risk indicator can be determined based on the industry dynamic field.

[0083] In the embodiments of the present application, based on the contract analysis model, the participant credit model, the transaction contract model, the operation monitoring model, and the environmental adaptation model, the semantic conversion is performed on the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data respectively, so as to identify the relevant fields corresponding to each risk indicator in the multi-modal data, and then determine the risk characteristic value of each defense indicator. No matter how the multi-modal data changes, the risk characteristics can be identified in real time, which provides a theoretical basis for real-time judgment of the risk size.

[0084] S500: Based on the global weight and the risk characteristic value of each risk indicator, the size of each risk level preset by the digital contract is obtained; wherein the global weight of each risk indicator is obtained by analyzing a preset risk causal model based on the type of the clause to which each risk indicator belongs, and the preset risk causal model is obtained by pre-training based on a performance data set, and the performance data set includes the type of the digital contract clause and the performance result.

[0085] In the embodiments of the present application, the global weight of each risk indicator can be determined based on the following steps S510-S550: S510: Based on the relative importance of the core class clause, the auxiliary class clause, and the external environment class clause, a first judgment matrix is constructed by the causal model.

[0086] In the embodiments of the present application, the causal model can be pre-trained based on a performance data set, and the performance data set includes the type of the digital contract clause and the performance result.

[0087] In the embodiments of the present application, the causal model can select a model suitable for causal inference, such as a structural equation model, a causal forest model, and a double machine learning model. The training of the causal model can generally include the following stages: Data collection and preprocessing stage: In this stage, the historical performance data set is mainly collected, including the text content (clause details) of the digital contract and the performance result (such as success / failure, performance rate, default loss, etc.).

[0088] After data collection, in the preprocessing stage, the terms of the digital contract in the force compliance data set can be classified, which can be classified into three categories based on expert experience: core terms, auxiliary terms, and external environment terms.

[0089] Training stage: During training, different types of terms can be converted into features available to the model, and interaction features can also be added between terms, such as the combination of core terms and auxiliary terms, and the compliance result as the dependent variable, and the term features as the independent variable, so that the causal model can estimate the relative importance of the three types of terms.

[0090] In the embodiments of the present application, the relative importance of the core terms, auxiliary terms, and external environment terms in the digital contract can be judged based on the causal model. For example, the causal model judges that the core terms are more important than the auxiliary terms, and assigns an importance scale value of 5, where the importance scale value 5 represents that the importance of the core terms is 5 times that of the auxiliary terms; the core terms are more important than the external environment terms, and assign an importance scale value of 3, where the importance scale value 3 represents that the importance of the core terms is 3 times that of the external environment terms; the external environment terms and the auxiliary terms are equally important, and assign an importance scale value of 1, and thus the following first judgment matrix can be obtained: C1 core terms C2 auxiliary terms C3 external environment terms C1 core terms 1 5 3 C2 auxiliary terms 1 / 5 1 1 C3 external environment terms 1 / 3 1 1 S520: Based on the first judgment matrix, determine the reference weight of each of the core terms, the auxiliary terms, and the external environment terms.

[0091] In the embodiments of the present application, when calculating the reference weights of the core terms, auxiliary terms, and external environment terms based on the first judgment matrix, the eigenvector of the first judgment matrix can be calculated, as follows: First, add each column of the first judgment matrix: Column 1 sum: 1 + 1 / 5 + 1 / 3 ≈ 1 + 0.2 + 0.333 ≈ 1.533 Column 2 sum: 5 + 1 + 1 = 7 Column 3 sum: 3 + 1 + 1 = 5 Then, divide each element of the first judgment matrix by the sum of its column to obtain the normalized matrix, as follows: C1 core terms C2 auxiliary terms C3 external environment terms C1 core terms 0.652 0.714 0.6 C2 auxiliary clause 0.130.143 0.2 C3 external environment clause 0.217 0.143 0.2 Then, the average value of each row of the standard matrix is calculated to obtain the feature vector corresponding to the core clause, the auxiliary clause and the external environment clause, i.e., the benchmark weight, as follows: C1 core clause benchmark weight = (0.652 + 0.714 + 0.6) / 3 = 0.655 C2 auxiliary clause benchmark weight = (0.130 + 0.143 + 0.2) / 3 = 0.158 C3 external environment benchmark weight = (0.217 + 0.143 + 0.2) / 3 = 0.187 In the embodiment of the present application, the core clause in the digital contract is the most important principle, which is converted into a quantitative calculation basis that the core clause weight accounts for 65.5% in risk assessment. In the subsequent fuzzy comprehensive evaluation, the influence of the risk indicators corresponding to the core clause on the final result must be greater than that of the risk indicators corresponding to the auxiliary clause and the external environment clause, so that the final risk level determination can accurately reflect the rigid requirements of the digital contract.

[0092] S530: Based on the relative importance between the risk indicators under any type of clause, a second judgment matrix corresponding to each type of clause is constructed.

[0093] In the embodiment of the present application, there are three types of clauses: core clause, auxiliary clause and external environment clause, and each type of clause corresponds to multiple risk indicators. Therefore, in step S530, a second judgment matrix needs to be constructed based on the risk indicators corresponding to each type of clause. There are three types of clauses, so there are three second judgment matrices.

[0094] When constructing the judgment matrix corresponding to each type of clause, the relative importance of multiple risk indicators of each type can also be based on the method of assigning importance degree scale value by experts or experience, which is not described here.

[0095] S540: Based on the second judgment matrix corresponding to each type of clause, the local weight of each risk indicator under each type of clause is determined.

[0096] In the embodiment of the present application, when calculating the weight (local weight) of each risk indicator under the type corresponding to the second judgment matrix of each type of clause, the feature vector of the second judgment matrix can also be calculated, and the specific calculation method is referred to step S520, which is not described here.

[0097] S550: Determine the global weight of each risk indicator based on the local weight of each risk indicator and the benchmark weight corresponding to the type clause where the risk indicator is located.

[0098] The benchmark weight corresponding to each type clause is calculated in step S520, and the local weight of each risk indicator in the type clause where the risk indicator is located is calculated in S540. In the embodiment of the present application, the product of the local weight of each risk indicator in the type clause where the risk indicator is located and the benchmark weight corresponding to the type clause is taken as the global weight corresponding to the risk indicator.

[0099] In the embodiment of the present application, the weight calculation method through two levels (clause type and risk indicator corresponding to each type clause) not only considers the importance of different types of clauses, but also considers the importance of different risk indicators under the same type. The global weight of each risk indicator obtained finally can avoid the distortion caused by simple average allocation of weight, so that when risk assessment is performed based on the global weight of each risk indicator, the real intention of the contract clause corresponding to each risk indicator can be more accurately reflected.

[0100] After determining the global weight of each risk indicator, the membership vector of each risk indicator under each preset risk level needs to be determined based on the risk characteristic value of each risk indicator.

[0101] In the embodiment of the present application, the risk levels preset by the digital contract can include three risk levels of low risk, medium risk and high risk. In other embodiments, more or fewer risk levels can also be included, and the number of risk levels is not specifically limited in the present application.

[0102] In step S400, each preset large language model can be used to determine the risk characteristic value corresponding to each risk indicator based on the participant information, transaction data, environment data and data space running data. The risk characteristic value is used to evaluate the risk size of the corresponding risk indicator.

[0103] For example, in the embodiment of the present application, based on the comprehensive identification and determination of each preset large language model, there are two risk indicators: Risk indicator A: payment delay risk (core clause); Risk indicator B: report format error risk (auxiliary clause).

[0104] Among them, the payment delay risk corresponds to the payment delay risk indicator, and the report format error risk corresponds to the report format error risk indicator. It is assumed that the risk characteristic value of the payment delay risk indicator is 30 (belonging to slight delay), and the risk characteristic value of the report format error risk is 85 (the higher the score, the higher the risk).

[0105] In the embodiments of the present application, the membership function is determined based on each risk characteristic value at which the risk indicator is at a degree of each risk level.

[0106] In the embodiments of the present application, the membership function is as follows: Low risk function: completely belongs to low risk when the risk characteristic value is less than 50; the membership degree linearly decreases to 0 when the risk characteristic value is between 50 and 70; Medium risk function: the membership degree increases from 0 to 1 when the risk characteristic value is between 50 and 70; decreases from 1 to 0 when the risk characteristic value is between 70 and 90; High risk function: completely belongs to high risk (1) when the risk characteristic value is greater than 90; the membership degree increases from 0 to 1 when the risk characteristic value is between 70 and 90.

[0107] For the risk indicator A (30), the corresponding membership degrees are as follows: Low risk membership degree: μ_L(30) ≈ 1.0 (far less than 50) Medium risk membership degree: 0 High risk membership degree: 0 Therefore, the membership vector of the risk indicator A is [1, 0, 0], which represents the degrees of low risk, medium risk and high risk respectively as 1, 0 and 0 at the current risk characteristic value of the risk indicator A.

[0108] For the risk indicator B (85), the corresponding membership degrees are as follows: Low risk membership degree: 0 (85 is far higher than 50) Medium risk membership degree: 85 is in the decreasing interval of 70-90, therefore, μ_M(85) = (90-85) / (90-70) = 0.25 High risk membership degree: 85 is in the increasing interval of 70-90, μ_H(85) = (85-70) / (90-70) = 0.75 Therefore, the membership vector of the risk indicator B is [0, 0.25, 0.75], which represents the degrees of low risk, medium risk and high risk respectively as 0, 0.25 and 0.75 at the current risk characteristic value of the risk indicator B.

[0109] In the embodiments of the present application, after the global weight of each risk indicator and the membership vector of each risk indicator characteristic value at each risk level are determined, the risk size of each risk level can be determined based on the membership vector of each risk indicator at each risk level and the global weight of each risk indicator.

[0110] In the embodiments of the present application, the risk size of each risk level can be determined based on the following steps S560-S570: S560: Construct a fuzzy relation matrix based on the membership vectors of each risk indicator under each risk level.

[0111] Still taking the risk indicator A and the risk indicator B in step S500 as an example, the membership vector of the risk indicator A is [1, 0, 0], and the membership vector of the risk indicator B is [0, 0.25, 0.75], therefore, the fuzzy relation matrix can be constructed as follows: Low risk Medium risk High risk Risk indicator A 1 0 0 Risk indicator B 0 0.25 0.75 S570: Determine the risk size of each risk level by taking the global weight of each risk indicator as the weight vector of the fuzzy relation matrix.

[0112] Suppose that the global weight Wa of the risk indicator A is 0.7 and the global weight Wb of the risk indicator B is 0.3 based on the calculation method of the global weight of each risk indicator in step S500, then the global weight vector of the risk indicators A and B is [0.7, 0.3], and the risk size of each risk level can be determined by performing a synthetic operation (such as a weighted average type synthesis) on the global weight vector of the risk indicators and the fuzzy relation matrix.

[0113] In the embodiments of the present application, the comprehensive membership vector can be calculated based on the following formula: B(j) = (Wa * Raj) + (Wb * Rbj) wherein B(j) is the membership vector of the risk level of the level j, Wa is the global weight of the risk indicator A, Raj is the membership degree of the risk indicator A under the risk level j, Wb is the global weight of the risk indicator B, and Rbj is the membership degree of the risk indicator B under the risk level j.

[0114] Therefore, B(low risk) = (0.7 * 0) + (0.3 * 1.0) = 0.3 B(medium risk) = (0.7 * 0.25) + (0.3 * 0) = 0.175 B(high risk) = (0.7 * 0.75) + (0.3 * 0) = 0.525 The comprehensive membership vector B = [0.3, 0.175, 0.525] represents that, under the current risk characteristic values of the risk indicators A and B, the degrees of being in the low risk, the medium risk, and the high risk are 0.3, 0.175, and 0.525 respectively.

[0115] In the embodiments of the present application, when risk assessment is performed based on the trusted data space, firstly, whether each data and information is tampered with can be determined based on the anchor point hash, and then a preset large language model is used to perform semantic analysis on the digital contract and construct a plurality of risk indicators of each clause in the digital contract, and a preset large language model corresponding to the participant information, transaction data, environmental data, and data space running data is used to perform semantic conversion on the participant information, transaction data, environmental data, and data space running data respectively, identify fields related to each risk indicator, and then determine the risk characteristic value of each risk indicator according to the field corresponding to each risk indicator, and make a risk judgment based on the risk characteristic value of each risk indicator. This method can determine the changes of multi-source heterogeneous data such as digital contracts, participant information, transaction data, environmental data, and data space running data in real time by using each preset large language model, and make a risk judgment in real time. In addition, when judging the risk based on the risk characteristic value of each risk indicator, the importance of each risk indicator in the global data contract is also considered, so the judgment is more accurate. In the embodiments of the present application, after obtaining the comprehensive membership degree vector, the final risk can be determined based on the size of the membership degree vector of each risk level in the comprehensive membership degree vector. For example, the final risk can be determined as the risk level with the maximum membership degree vector. For example, in the comprehensive membership degree vector B = [0.3, 0.175, 0.525], the membership degree vector of the high risk level is 0.525, which is greater than the membership degree vector 0.3 of the low risk level and the membership degree vector 0.175 of the medium risk level. Therefore, the final risk can be determined as the high risk level.

[0116] In the embodiments of the present application, after obtaining the digital contract, participant information, transaction data, environmental data, and data space running data of the trusted data space, the method further comprises: determining a dynamic performance clause in the digital contract; obtaining time series data corresponding to the dynamic performance clause based on the transaction data and the data space running data; constructing a sliding window based on the time series data; determining the performance capability of the participant based on the sliding window.

[0117] In the embodiments of the present application, the dynamic performance clause in the digital contract is, for example, a clause related to the quarterly growth rate, monthly transaction volume, and quarterly transaction volume. The dynamic performance clause can be determined by a contract analysis model.

[0118] For time series data in transaction data, it contains a large number of records that change frequently over time, which can directly reflect the dynamic fulfillment process of the dynamic fulfillment clause in the digital contract. Therefore, the dynamic fulfillment clause related to the transaction in the data contract can be evaluated based on the time series data in the transaction data.

[0119] For example, in the embodiment of the present application, the contract analysis model determines that the dynamic fulfillment clause in the digital contract is: the fulfillment ability of agreeing to the monthly minimum quantity of 500, the quarterly growth rate of >10%, and the payment growth of no less than 5% per quarter, and the quarterly delivery quantity needs to increase by 5% per week.

[0120] In the embodiment of the present application, the transaction flow time series in the transaction data includes the occurrence time, amount, and transaction status of each transaction, such as “2024-01-01 09:00 transaction 500”, “2024-01-02 10:30 transaction 480”, at this time, the transaction contract model can be used to obtain the transaction flow time series of each window with a sliding window of one week or one month, and then determine the transaction volume of each week or each month, so as to judge the fulfillment ability of the participant for the dynamic fulfillment clause “monthly minimum quantity of 500”; in addition, the weekly growth rate, monthly growth rate, and quarterly growth rate can also be calculated based on the transaction flow time series of each window, so as to judge the fulfillment ability of the participant for the dynamic fulfillment clause “quarterly growth rate >10%”.

[0121] In the embodiment of the present application, the transaction data also includes a payment record time series, which includes a payment amount and a payment timestamp, such as “2024-01 to 2024-03 monthly payment amounts are 100,000, 120,000, and 110,000 respectively”, therefore, the transaction contract model can also be used to obtain the payment record time series of each window with a sliding window of one week or one month, and analyze and calculate the payment amount growth rate, and then judge the fulfillment ability of the participant for the dynamic fulfillment clause “payment growth of no less than 5% per quarter”.

[0122] In the embodiment of the present application, the transaction data also includes a logistics and delivery time series, which includes the time and quantity of data delivery, such as “100 data are delivered every Monday for 12 weeks”, therefore, the transaction contract model can also be used to obtain the logistics and delivery time series of each window with a sliding window of one week or one month, and analyze to obtain the cumulative delivery quantity, and then judge the fulfillment ability of the participant for the dynamic fulfillment clause “quarterly delivery quantity needs to increase by 5% per week”.

[0123] In the embodiments of the present application, for the time series data in the data space running data, which records the state of the system level changing over time, therefore, the dynamic compliance clauses related to "data availability, access compliance" in the data contract can be evaluated based on the time series data in the data space running data.

[0124] For example, in the embodiments of the present application, it is assumed that the contract analysis model determines that the dynamic clause in the digital contract includes "data access for 5 consecutive minutes exceeds 5 seconds, quarterly data transmission volume growth rate > 10%", and the daily access during the contract term shall not be less than 5 times.

[0125] In the embodiments of the present application, the data space running data includes an access response time sequence, and the access response time sequence includes: the average response time per hour / minute, such as "2024-06-01 08:00 response time 3 seconds, 09:00 response time 4.5 seconds", then based on the running monitoring model, the access response time sequence of each window can be obtained with a one-minute sliding window, and then the index of the dynamic compliance clause "data access for 5 consecutive minutes exceeds 5 seconds" can be judged.

[0126] In the embodiments of the present application, the data space running data further includes a data transmission volume time sequence, and the data transmission volume time sequence includes: the cross-domain data transmission volume per hour / day, such as "2024-Q1 monthly transmission volume is 100GB, 105GB, 108GB", then based on the running monitoring model, the data transmission volume time sequence of each window can be obtained with a one-month sliding window, and then whether the dynamic compliance clause "quarterly data transmission volume growth rate > 10%" meets the standard can be judged.

[0127] In the embodiments of the present application, the data space running data further includes an access log time sequence, and the access log time sequence includes the time, frequency and operation type of user access data, such as "user A accesses 10 times per day from 2024-05-01 to 05-31", then based on the running monitoring model, the access log time sequence of each window can be obtained with a one-day sliding window, and then the index of the dynamic compliance clause "the daily access during the contract term shall not be less than 5 times" can be judged.

[0128] In the embodiments of the present application, in addition to obtaining time series data from transaction data and the data space running data, time data can also be obtained from participant information to evaluate some dynamic indexes in the digital contract.

[0129] For example, in the embodiments of the present application, the participant's monthly performance completion rate sequence can be obtained from the participant's information based on the participant's credit model, such as the performance completion rates of a certain enterprise from January to April 2024, which are 95%, 90%, 85%, and 80% respectively. Based on this, the subsequent trend is predicted through the LSTM (Long Short-Term Memory) model. If the prediction is lower than the "minimum completion rate of 80%" required by the digital contract for three consecutive months, the "insufficient performance ability warning" is triggered.

[0130] For another example, in the embodiments of the present application, the credit score time sequence can be obtained from the participant's information based on the participant's credit model, such as "the credit score decreased from 90 points to 75 points in Q1 2024". Based on this, the analysis of whether the credit deterioration will lead to future default can be combined with the credit maintenance clause in the digital contract, such as "the credit score cannot be lower than 80 points".

[0131] In the embodiments of the present application, the contract terms (such as the guaranteed quantity of 500 units and the growth rate of 10%) in the digital contract are static evaluation benchmarks and do not change over time, so they do not generate time series data and are only used as target thresholds for time series analysis.

[0132] In the embodiments of the present application, although the market conditions, policies and regulations and other environmental data in the environmental data have a time dimension, the change frequency is low (such as policies are updated monthly / quarterly), and they are more used for "trend judgment" rather than "high-frequency dynamic performance evaluation indicators", so they are only used as auxiliary references in special scenarios (such as the impact of policy changes on quarterly performance) and do not constitute core time series data.

[0133] In the embodiments of the present application, after obtaining the digital contract in the trusted data space, the method further comprises: determining the life cycle type of the digital contract, the life cycle type including a trial period contract, a long-term contract, and a short-term contract; if the digital contract is a trial period contract, increasing the weight of the risk indicator corresponding to the initial performance clause; if the digital contract is a long-term contract, increasing the weight of the risk indicator corresponding to the long-term impact clause; if the digital contract is a short-term contract, increasing the weight of the risk indicator corresponding to the short-term target completion degree clause.

[0134] In the embodiments of the present application, the contract analysis model, the participant credit model, the transaction contract model and the operation monitoring model can identify the general characteristics of five types of data, such as transaction amount, response time, credit score, etc. after pre-training and fine-tuning. According to the general characteristics of the five types of data, the general rules of performance behavior and default risk can be learned, but at this time, the default risk assessment does not consider the life cycle of the digital contract. For example, the digital contract can be divided into trial period contract, long-term contract and short-term contract according to the life cycle, and the core risks of digital contracts with different life cycles are different.

[0135] For example, for the trial period contract (the first 3 months), the trial period is the key period for both parties to establish trust, and the core risk characteristics are initial performance behavior, such as whether the first delivery meets the standard and whether the first month is paid on time. The initial performance behavior is highly predictive of long-term cooperation risk, such as contracts that default during the trial period, the long-term default probability is 3 times that of ordinary contracts.

[0136] For example, for long-term contracts (≥1 year), they are greatly affected by external environmental fluctuations, such as policy changes and market cycle fluctuations, and often contain force majeure clauses in long-term contracts, such as “exempt from liability due to industry policy mutations that cause performance difficulties”. Therefore, the core risk of long-term contracts needs to focus on long-term trend risk.

[0137] For example, for short-term contracts (3-12 months), the performance nodes are relatively dense, such as delivering data once a month, and the risk mainly comes from node completion rather than long-term trends, and the contract rarely contains complex force majeure clauses. Therefore, the core risk of short-term contracts needs to focus on the density of performance nodes and short-term goals.

[0138] Therefore, in the embodiments of the present application, on the basis of the pre-trained and fine-tuned contract analysis model, participant credit model, transaction contract model and operation monitoring model, an adaptation layer can be added to each model to adjust each model according to the life cycle of the digital contract, so that the core risks of each life cycle contract can be considered when assessing risks.

[0139] For example, for trial period contracts, the adaptation layer is used to strengthen the prediction weight of initial behavior. For example, the weight of the performance behavior characteristics of the first 3 months (such as the first month transaction volume and the first data quality compliance rate) is increased by 20% on the basis of each basic model. For example, the weight of the first month behavior in the basic model is adjusted from 0.1 to 0.12 after the adaptation layer.

[0140] In addition to increasing the prediction weight of initial behavior, the input of long-term behavior characteristics (such as changes in industry policy) can also be temporarily weakened, and the input of initial behavior characteristics such as initial key node completion degree (such as whether to submit qualification documents in the first week of the trial period or whether to reach the minimum transaction volume in the first month) is emphasized.

[0141] For example, in the embodiment of the present application, if the trial period contract stipulates that the minimum volume in the first month is 500, the basic model predicts a default probability of 30%, and after the adaptation layer increases the prediction weight of the first month behavior characteristics, if the actual transaction volume in the first week is only 100 (much lower than the average of 125 per week), the predicted default probability is adjusted to 50%, so that the initial risk can be captured more sensitively.

[0142] For long-term contracts, the adaptation layer is used to increase the weight of external environmental characteristics, while maintaining a high weight for core clauses such as payment and data availability. For example, the weight of the policy factor in the basic model is increased from 0.05 to 0.0575. In addition, it is also used for feature set expansion, such as adding long-term impact features such as macroeconomic indexes in the past 6 months, industry regulatory policy update frequency, and participant annual revenue fluctuation range, etc. to the basic model input.

[0143] For example, in the embodiment of the present application, a long-term contract stipulates that the "annual data transmission volume growth rate is ≥10%", the basic model only predicts a compliance probability of 80% based on historical transmission data, and after the adaptation layer adds the feature of "industry policy tightening leading to a 30% increase in transmission cost", the predicted compliance probability is adjusted to 50% to better judge long-term risks.

[0144] For short-term contracts, the fulfillment nodes are dense (such as delivering data once a month), the risks mainly come from node completion rather than long-term trends, and the contract contains fewer complex force majeure clauses, therefore, the adaptation layer is used to remove long-term features (such as industry data for more than 5 years), only retain short-term node features (such as the number of remaining nodes, the delay rate of completed nodes), and increase the prediction weight of node time deviation features, and the prediction weight of time deviation features in short-term contracts is higher than that in long-term contracts.

[0145] For example, in the embodiment of the present application, a 3-month short-term contract requires "completing reconciliation on the last day of each month", the basic model predicts an overall default probability of 20%, and the adaptation layer increases the weight of single-month reconciliation delay by 10% based on the node density characteristics, if the reconciliation is delayed by 2 days in the first month, the predicted default probability of the next two months is increased to 35%.

[0146] In this embodiment, the risks of digital contracts with different lifecycles are common. For example, "payment delay" is a high risk in digital contracts of any lifecycle. The basic model can capture the general pattern. On the basis of the basic model, an adaptation layer is added. Only minor adjustments are needed to cover the differences between digital contracts with different lifecycles. There is no need to train an independent model for each type of digital contract, nor is there a need for repeated training, which can reduce maintenance costs.

[0147] In addition, some digital contracts may contain clauses that allow for the conversion of a trial period to a long-term contract, such as automatic renewal for two years after the trial period is completed. In this case, the method of adding an adaptation layer to the basic model in the embodiments of this application can quickly switch from the trial period mode to the long-term mode without changing the model, and can dynamically adapt to the needs.

[0148] In this embodiment, when conducting risk assessment based on a trusted data space, the method first determines whether various data and information have been tampered with based on anchor hashing. Then, a pre-defined large language model is used to perform semantic parsing on the digital contract and construct multiple risk indicators for each clause in the digital contract. Using the pre-defined large language model corresponding to participant information, transaction data, environmental data, and data space operation data, semantic transformation is performed on these data to identify fields related to each risk indicator. Then, the risk characteristic value of each risk indicator is determined based on the fields corresponding to each risk indicator, and a risk judgment is made based on the risk characteristic value of each risk indicator. This method, using various pre-defined large language models, can determine changes in multi-source heterogeneous data such as digital contracts, participant information, transaction data, environmental data, and data space operation data in real time and make risk judgments in real time. Furthermore, when judging risk based on the risk characteristic value of each risk indicator, the importance of each risk indicator in the overall data contract is also considered, thus making the judgment more accurate.

[0149] Exemplary device This application also proposes a risk assessment device 100 for a trusted data space, such as... Figure 2 As shown in the embodiments of this application, the risk assessment setting 100 of the trusted data space includes: The acquisition module 110 is used to acquire digital contracts, participant information, transaction data, environmental data, and data space operation data of the trusted data space; the digital contracts, participant information, transaction data, environmental data, and data space operation data are multi-source heterogeneous data; Processing module 120, This is used to perform anchor hash calculations based on the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data to determine whether the data has been tampered with. If not tampered, the digital contract is textually semantically parsed by a preset large language model to obtain a plurality of risk indicators corresponding to each clause in the digital contract; Each data or information is respectively subjected to semantic recognition and associated field matching by a preset large language independent model respectively corresponding to the participant information, the transaction data, the environmental data, and the data space running data, to obtain a risk characteristic value of each risk indicator; the risk characteristic value of each risk indicator is obtained based on the field content of the corresponding risk indicator; Based on the global weight and the risk characteristic value of each risk indicator, the size of each risk level preset for the digital contract is obtained; wherein the global weight of each risk indicator is obtained by a preset risk causal model based on the type of the clause to which each risk indicator belongs, and the preset risk causal model is obtained by pre-training based on a performance data set, and the performance data set includes a digital contract clause type and a performance result.

[0150] In the embodiments of the present application, the processing module 120 is further configured to: determine the core fields of the digital contract, the participant information, the transaction data, the environmental data, and the data space running data, and the anchor point clauses of each core field in the digital contract; bind the core fields of the digital contract, the participant information, the transaction data, the environmental data, and the data space running data with their corresponding anchor point clauses to generate each binding result; perform a hash operation based on the binding result corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space running data respectively to obtain the current contract anchor point hash value corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space running data; Compare the current contract anchor point hash value corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space running data with the initial contract anchor point hash value corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space running data in the digital space to determine whether the digital contract, the participant information, the transaction data, the environmental data, and the data space running data are tampered.

[0151] In the embodiments of the present application, the processing module 120 is further configured to: identify the clause type of each clause in the digital contract based on the preset large language model, wherein the clause type includes a core class clause, an auxiliary class clause, and an external environment class clause; construct each risk indicator under each type of clause based on the core type of clause, the auxiliary type of clause, and the external environment type of clause respectively.

[0152] In the embodiment of the application, the processing module 120 is further configured to: construct a first judgment matrix based on the relative importance of the core type of clause, the auxiliary type of clause, and the external environment type of clause through the risk causal model; determine the reference weight of the core type of clause, the auxiliary type of clause, and the external environment type of clause based on the first judgment matrix; construct a second judgment matrix corresponding to each type of clause based on the relative importance between each risk indicator under any type of clause; determine the local weight of each risk indicator under each type of clause based on the second judgment matrix corresponding to each type of clause; determine the global weight of each risk indicator based on the local weight of each risk indicator and the reference weight of the type of clause where the risk indicator is located.

[0153] In the embodiment of the application, the processing module 120 is further configured to: construct a fuzzy relationship matrix based on the membership degree vector of each risk indicator under each risk level; determine the risk size of each risk level by taking the global weight of each risk indicator as the weight vector of the fuzzy relationship matrix.

[0154] In the embodiment of the application, the processing module 120 is further configured to: determine the dynamic performance clause in the digital contract; obtain time series data corresponding to the dynamic performance clause based on the transaction data and the data space operation data; construct a sliding window based on the time series data; determine the performance capability of the participant based on the sliding window.

[0155] In the embodiment of the application, the processing module 120 is further configured to: determine the life cycle type of the digital contract, the life cycle type including a trial period contract, a long-term contract, and a short-term contract; if the digital contract is a trial period contract, increase the weight of the risk indicator corresponding to the initial performance clause; if the digital contract is a long-term contract, increase the weight of the risk indicator corresponding to the long-term impact clause; if the digital contract is a short-term contract, increase the weight of the risk indicator corresponding to the short-term target completion degree clause.

[0156] In the embodiments of the present application, when risk assessment is performed based on the trusted data space, the processing module 120 first determines whether each data and information is tampered based on the anchor point hash when performing risk assessment based on the trusted data space. Then, a pre-set large language model is used to perform semantic analysis on the digital contract and construct multiple risk indicators of each clause in the digital contract. The pre-set large language model corresponding to the participant information, transaction data, environmental data, and data space running data is used to perform semantic conversion on the participant information, transaction data, environmental data, and data space running data, respectively, to identify fields related to each risk indicator. Then, the risk characteristic values of each risk indicator are determined according to the fields corresponding to each risk indicator. Finally, risk judgment is made based on the risk characteristic values of each risk indicator. This method can determine the changes of multi-source heterogeneous data such as digital contracts, participant information, transaction data, environmental data, and data space running data in real time using each pre-set large language model, and make risk judgment in real time. In addition, when judging the risk based on the risk characteristic values of each risk indicator, the importance of each risk indicator in the global data contract is also considered, so the judgment is more accurate.

[0157] Exemplary medium After introducing the method, medium, and system of the exemplary embodiments of the present application, next, with reference to Figure 3 The computer-readable storage medium of the exemplary embodiments of the present application is described with reference to Figure 3 The computer-readable storage medium shown is an optical disc 70, which stores a computer program (i.e., a program product) thereon. When the computer program is run by a processor, it will implement each step described in the above method embodiments, for example, obtain the digital contract, participant information, transaction data, environmental data, and data space running data of the trusted data space; perform text semantic analysis on the digital contract by a pre-set large language model to construct multiple risk indicators corresponding to each clause in the digital contract; perform semantic conversion on the participant information, transaction data, environmental data, and data space running data, identify each field in the participant information, transaction data, environmental data, and data space corresponding to each risk indicator, and map the fields to the corresponding risk indicators to obtain the risk characteristic values of each risk indicator; determine the global weight of each risk indicator based on the relative importance of each risk indicator; determine the membership vector of each risk indicator at each risk level based on the risk characteristic values of each risk indicator; Based on the membership vectors of each risk indicator under each risk level and the global weights of each risk indicator, the risk sizes of each risk level are determined. The specific implementation of each step is not repeated here. It should be noted that examples of the computer-readable storage medium can also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other optical, magnetic storage medium, which will not be repeated here.

[0158] Exemplary computing device After introducing the method, system and medium of the exemplary embodiments of the present application, next, with reference to Figure 4 The computing device of the exemplary embodiments of the present application.

[0159] Figure 4 A block diagram of an exemplary computing device 80 suitable for implementing exemplary embodiments of the present application is shown, which can be a computer system or a server. Figure 4 The displayed computing device 80 is only an example, and should not bring any limitation to the function and use range of the embodiments of the present application.

[0160] As Figure 4 shown, the components of the computing device 80 can include, but are not limited to, one or more processors or processing units 801, system memory 802, a bus 803 connecting different system components including the system memory 802 and the processing unit 801.

[0161] The computing device 80 typically includes a variety of computer system readable media. Such media can be any available media that is accessible by the computing device 80 and includes both volatile and non-volatile media, removable and non-removable media.

[0162] The system memory 802 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 8021 and / or cache memory 8022. The computing device 80 can further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, ROM 8023 can be used for reading a non-removable, non-volatile magnetic media (e.g., a hard disk drive). Figure 4 not shown in the computing device 80, is typically provided as a repository for a data file used by the computing device 80, such as the operating system 804. Although not specifically shown, such computer system readable media can also include, by way of example only, a phase change memory (PRAM), static random access memory (SRAM), a Figure 4As shown in FIG. 8, disk drives, optical drives and the like, can be provided for reading from or writing to a removable nonvolatile media such as a "floppy disk" or a nonvolatile "optical disk", such as a CD-ROM, DVD-ROM or other optical media. In such cases, each drive can be connected to the bus 803 via one or more data media interfaces. The system memory 802 can include at least one program product having a set (e.g., at least one) of program modules that configure the processor 801 to perform the functions of the embodiments of the application.

[0163] Program / utility 8025 having a set (at least one) of program modules 8024 can be stored in system memory 802 and implemented or accessed by the processor 801, as shown in FIG. 8. It is appreciated that each of the program modules 8024 includes, but is not limited to, an operating system, one or more applications, other program modules, and program data, each or some combination thereof, which can include implementation of a network environment. The program modules 8024 generally carry out the functions and / or methodologies of the embodiments described herein.

[0164] The computing device 80 can also communicate with one or more external devices 804 such as a keyboard, a pointing device, a display, etc. via I / O interface 805. Further, the computing device 80 can communicate with one or more networks such as a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet via network adapter 806. As Figure 4 illustrated, the network adapter 806 is in communication with the other modules of the computing device 80 (e.g., the processing unit 801, etc.) via the bus 803. It should be appreciated that the network adapter 806 can be implemented as a separate component or integrated with other modules of the computing device 80, such as the processing unit 801. Figure 4 It is to be appreciated that other hardware and / or software modules can be used in conjunction with the computing device 80, as shown in FIG. 8.

[0165] The processing unit 801 performs various functions and data processing by running programs stored in the system memory 802, such as, obtaining a digital contract of a trusted data space, participant information, transaction data, environment data, data space running data; performing text semantic analysis on the digital contract by a preset large language model, and constructing a plurality of risk indicators corresponding to each clause in the digital contract; performing semantic conversion on the participant information, the transaction data, the environment data, and the data space running data, identifying each field in the participant information, the transaction data, the environment data, and the data space corresponding to each risk indicator, and mapping the each field to the corresponding risk indicator to obtain a risk feature value of each risk indicator; determining a global weight of each risk indicator based on a relative importance of the each risk indicator; determining a membership degree vector of each risk indicator under each preset risk level based on the risk feature value of each risk indicator; determining the risk size of each risk level based on the membership degree vector of each risk indicator under each risk level and the global weight of each risk indicator. The specific implementation of each step will not be repeated here. It should be noted that although several units / modules or sub-units / sub-modules of the risk assessment device of the trusted data space are mentioned in the foregoing detailed description, such division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0166] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present application, which are used to illustrate the technical solutions of the present application, but not to limit them. The protection scope of the present application is not limited thereto, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features thereof, within the technical scope disclosed by the present application; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0167] In addition, although the operations of the method of the present application are described in a specific order in the accompanying drawings, this does not require or imply that the operations must be performed in this specific order, or that all of the shown operations must be performed to achieve the desired result. Additionally or alternatively, some steps can be omitted, a plurality of steps can be combined into one step, and / or one step can be divided into a plurality of steps.

Claims

1. A risk assessment method based on trusted data space, comprising: Obtain digital contracts, participant information, transaction data, environmental data, and data space operation data from a trusted data space; The digital contract, the participant information, the transaction data, the environmental data, and the data space operation data are multi-source heterogeneous data; Anchor hash calculations are performed on the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data to determine whether the data has been tampered with. If the digital contract has not been tampered with, the text semantics of the digital contract are analyzed by a preset large language model to obtain multiple risk indicators corresponding to each clause in the digital contract. By using a pre-defined large language independent model corresponding to the participant information, the transaction data, the environmental data, and the data space operation data respectively, semantic recognition and related field matching are performed on each data or information to obtain the risk feature value of each risk indicator; The risk characteristic value of each risk indicator is obtained based on the field content of the corresponding risk indicator; Based on the global weights and risk characteristic values ​​of each risk indicator, the magnitude of each preset risk level of the digital contract is obtained; wherein, the global weight of each risk indicator is obtained by analyzing a preset risk causal model based on the type of clause to which each risk indicator belongs, and the preset risk causal model is pre-trained based on a performance data set, which includes the digital contract clause type and performance result.

2. The risk assessment method based on a trusted data space as described in claim 1, wherein the step of performing anchor hash calculations on the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data respectively to determine whether they have been tampered with includes: Identify the core fields of the digital contract, the participant information, the transaction data, the environmental data, the data space operation data, and the anchor clauses of each core field in the digital contract; The core fields of the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data are bound to their corresponding anchor terms to generate various binding results; Hash operations are performed on the binding results corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data respectively to obtain the current contract anchor hash value corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data. Based on the current contract anchor hash value corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data, respectively, compare it with the initial contract anchor hash value corresponding to the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data in the digital space to determine whether the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data have been tampered with.

3. The risk assessment method based on trusted data space as described in claim 1, wherein the step of performing text semantic parsing on the digital contract using a preset large language model to obtain multiple risk indicators corresponding to each clause in the digital contract includes: The clause types in the digital contract are identified based on the preset large language model. The clause types include core clauses, auxiliary clauses, and external environment clauses. Based on the core clauses, the auxiliary clauses, and the external environment clauses, risk indicators are constructed for each category of clauses.

4. The risk assessment method based on trusted data space as described in claim 3, wherein the global weights of each risk indicator are obtained based on the following method: Based on the relative importance of the core clauses, auxiliary clauses, and external environment clauses using the aforementioned risk causal model, a first judgment matrix is ​​constructed. Based on the first judgment matrix, the benchmark weights of the core clauses, the auxiliary clauses, and the external environment clauses are determined. Based on the relative importance of various risk indicators under any type of clause, a second judgment matrix is ​​constructed for each type of clause. Based on the second judgment matrix corresponding to each type of clause, the local weights of each risk indicator under each type of clause are determined; The global weight of each risk indicator is determined based on its local weight and the benchmark weight corresponding to the type clause in which it belongs.

5. The risk assessment method based on trusted data space as described in claim 1, wherein obtaining the magnitude of each preset risk level of the digital contract based on the global weight and risk characteristic value of each risk indicator includes: A fuzzy relation matrix is ​​constructed based on the membership vectors of each risk indicator at each risk level. The risk level of each risk level is determined by using the global weight of each risk indicator as the weight vector of the fuzzy relation matrix.

6. The risk assessment method based on a trusted data space as described in claim 1, further comprising, after acquiring the digital contracts, participant information, transaction data, environmental data, and data space operation data of the trusted data space: Determine the dynamic performance terms in the digital contract; Based on the transaction data and the data space operation data, obtain the time-series data corresponding to the dynamic performance terms; Based on the aforementioned time-series data, a sliding window is constructed; Based on the sliding window, the participating parties' ability to perform their obligations is determined.

7. The risk assessment method based on trusted data space as described in claim 1, wherein after obtaining the digital contract of the trusted data space, the method further includes: Determine the lifecycle type of the digital contract, which includes trial period contracts, long-term contracts, and short-term contracts; If the digital contract is a trial period contract, the risk indicator weight corresponding to the initial performance terms shall be increased; If the digital contract is a long-term contract, the weight of the risk indicator corresponding to the long-term impact clause shall be increased. If the digital contract is a short-term contract, the weight of the risk indicator corresponding to the short-term goal completion clause is increased.

8. A risk assessment device based on a trusted data space, comprising: The acquisition module is used to acquire digital contracts, participant information, transaction data, environmental data, and data space operation data from the trusted data space. The digital contract, the participant information, the transaction data, the environmental data, and the data space operation data are multi-source heterogeneous data; The processing module is used to perform anchor hash calculations based on the digital contract, the participant information, the transaction data, the environmental data, and the data space operation data to determine whether the data has been tampered with. If the digital contract has not been tampered with, the text semantics of the digital contract are analyzed by a preset large language model to obtain multiple risk indicators corresponding to each clause in the digital contract. By using a pre-defined large language independent model corresponding to the participant information, the transaction data, the environmental data, and the data space operation data respectively, semantic recognition and related field matching are performed on each data or information to obtain the risk feature value of each risk indicator; The risk characteristic value of each risk indicator is obtained based on the field content of the corresponding risk indicator; Based on the global weights and risk characteristic values ​​of each risk indicator, the magnitude of each preset risk level of the digital contract is obtained; wherein, the global weight of each risk indicator is obtained by analyzing a preset risk causal model based on the type of clause to which each risk indicator belongs, and the preset risk causal model is pre-trained based on a performance data set, which includes the digital contract clause type and performance result.

9. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-7.

10. A computing device comprising a processor having a computer program stored thereon, the computer program executing to implement the method as described in any one of claims 1-7.

Images