Supply chain trade background verification system based on block chain and AI

Abstract

The invention provides a supply chain trade background verification system based on a block chain and AI, which relates to the field of electric digital data processing and comprises an on-chain trusted data evidence storage module, a multi-source data AI identification and analysis module, an AI driving verification control module and an asynchronous cooperative monitoring module. The on-chain trusted data evidence storage module is used for carrying out on-chain evidence storage on whole-process data, the multi-source data AI identification analysis module is used for carrying out extraction and semantic understanding on original data, and the AI driving verification control module is used for carrying out consistency and rationality verification on on-chain information. The asynchronous collaborative monitoring module is used for managing a data time asynchronization problem; the system greatly improves the intelligence, accuracy and traceability of trade authenticity verification, and is suitable for various scenes such as trade finance, supply chain finance and cross-border settlement.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital data processing, in particular to a supply chain trade background verification system based on blockchain and AI. BACKGROUND

[0002] In the supply chain trade business, contracts, logistics, invoices and fund flow information are key data supporting the verification of trade authenticity. However, the current verification method has the following problems: first, the data is scattered among multiple systems and participants, forming an information island, which is difficult to verify uniformly; second, the data has low credibility and there is a risk of forgery and tampering, especially in paper documents and scanned copies; third, the verification process is highly dependent on manual review, which is low in processing efficiency, high in cost, and easily affected by human subjective factors; fourth, different information sources are time-displaced and have large format differences, making it difficult to automatically match and align. Although some trade information has been applied to some links through blockchain or OCR recognition technology, a complete mechanism from data credibility evidence, semantic understanding, intelligent verification to automatic monitoring has not yet been formed, and there is a lack of joint verification capability for multi-source heterogeneous data. AI is not fully utilized for confidence judgment and process control.

[0003] Many supply chain verification systems have been developed. After extensive search and reference, it is found that the existing supply chain systems, such as the system disclosed in CN115330545A, generally include the following steps: obtaining the data integrity result of the financing party according to the financing party's various cross-border supply chain data, the data item weight proportion of the various cross-border supply chain data, and the preset data integrity calculation method; obtaining the data credibility result of the financing party according to the various cross-border supply chain data and the preset data credibility calculation method; obtaining the data validity coefficient of the financing party according to the data integrity result of each financing party and the data credibility result of each financing party; corresponding matching of each cross-border supply chain data to obtain the data matching degree between each data; and obtaining the data verification result of the financing party according to the data matching degree between each data and the data validity coefficient. However, this system mainly solves the financing problem and is not suitable for handling trade background verification problems. SUMMARY

[0004] The present application aims to solve the existing problems and provides a supply chain trade background verification system based on blockchain and AI.

[0005] The present application adopts the following technical solutions:

[0006] A supply chain trade background verification system based on blockchain and AI, comprising an on-chain trusted data evidence module, a multi-source data AI identification and analysis module, an AI-driven verification control module, and an asynchronous collaborative monitoring module.

[0007] The on-chain trusted data storage module is used for storing and verifying full-process data on the chain, the multi-source data AI identification and analysis module is used for extracting and semantically understanding the original data, the AI-driven verification control module is used for verifying the consistency and rationality of the on-chain information, and the asynchronous collaborative monitoring module is used for managing the data time asynchronous problem.

[0008] The on-chain trusted data storage module includes a four-flow recording unit, an abstract generation unit and an index tracing unit, the four-flow recording unit records contract, logistics, invoice and fund flow information through blockchain technology, the abstract generation unit formats the on-chain data and generates a hash abstract, and the index tracing unit retrieves information in the blockchain.

[0009] The multi-source data AI identification and analysis module includes a text recognition unit, a semantic understanding unit and a standard output unit, the text recognition unit is used to identify text information from non-text type data, the semantic understanding unit is used to analyze the logical semantics in the text information, and the standard output unit is used to convert the parsed information into a standard format file output.

[0010] The AI-driven verification control module includes a target driving unit, a logic verification unit and a confidence control unit, the target driving unit is used to output the flow information required for verification, the logic verification unit is used to logically verify the matching degree between the flow information and the target file, and the confidence control unit is used to confidence evaluate the verification result and control the subsequent process.

[0011] The asynchronous collaborative monitoring module includes an asynchronous buffer alignment unit, an artificial review scheduling unit and a visual report unit, the asynchronous buffer alignment unit is used to align the multi-modal data of different flows through feature extraction mapping method, the artificial review scheduling unit is used to secondarily schedule the verification tasks with low confidence, and the visual report unit is used to visually process the entire verification information and generate a report file.

[0012] Further, the text recognition unit includes a file preprocessor, a character recognition processor and a text format processor, the file preprocessor is used to format convert the file, convert the file into the format required by the character recognition processor, the character recognition processor is used to identify and extract text information, and the text format processor is used to format the extracted text information, prevent garbled code, and meet the subsequent processing requirements.

[0013] Further, the character recognition processor adopts a multi-modal large model technology, such as distilling the Llama4 Maverick model, to maintain the accuracy of character recognition while improving the inference efficiency, meet the needs of real-time processing of online services, support text input and image input, extract key business entities from natural language texts such as contracts, invoices, and emails, and identify business elements such as "payee", "currency", and "payment time" in contract clauses.

[0014] Further, the semantic understanding unit includes a semantic parsing processor, a syntax dependency processor, and an information normalization processor. The semantic parsing processor is used to extract entity information with unique and independent meanings from the text. The syntax dependency processor is used to construct dependency structures and extract semantic roles. The information normalization processor is used to map entity information into semantic roles of dependency structures.

[0015] Further, the semantic parsing processor integrates knowledge graphs with large model technology, embedding entities and relationships in the knowledge graph into semantic space, enhancing the semantic understanding and reasoning capabilities of large models, improving the accuracy and richness of semantic representation, and distinguishing between synonyms / fuzzy words such as "payee" and "payee account".

[0016] Further, the logic verification unit includes a field comparison processor, a semantic conflict detector, and an abnormal rule matcher. The field comparison processor is used to directly compare multi-source fields. The semantic conflict detector is used to determine whether there is a conflict between entity information in the target file and entity information in the flow information. The abnormal rule matcher is used to set a rule library and determine whether there is a situation that meets the set abnormal rules.

[0017] Further, the confidence control unit includes a confidence processor, a dynamic threshold adjuster, and a process control processor. The confidence processor generates confidence based on the verification result. The dynamic threshold adjuster dynamically adjusts the confidence threshold based on historical data of both parties to the transaction. The process control processor controls the subsequent process based on the mathematical relationship between the confidence and the confidence threshold.

[0018] Further, the confidence processor calculates the confidence S of the target file according to the following formula:

[0019]

[0020] Where n is the ratio of the number of inconsistent fields detected by the field comparison processor to the total number of detected fields, m1 is the number of conflicts detected by the semantic conflict detector, m2 is the number of rules matched by the abnormal rule matcher, A is the semantic conflict threshold, B is the rule conflict threshold, alpha is the semantic base, and beta is the rule base.

[0021] Further, the dynamic threshold adjuster calculates the dynamic threshold T according to the following formula:

[0022] T=T0+lambda*sigma-mu*rho;

[0023] Wherein, T0 is a basic threshold, sigma is a confidence degree standard deviation of historical transactions, rho is a complexity factor of current transactions, lambda is a historical adjustment coefficient, and mu is a current adjustment coefficient.

[0024] When the confidence degree is greater than or equal to the dynamic threshold, it indicates that the target file passes the verification and is directly recorded in the blockchain, and when the confidence degree is less than the dynamic threshold, it indicates that the target file is not trustworthy and needs human intervention for verification.

[0025] The beneficial effects obtained by the present application are:

[0026] The system stores the key trade elements such as contracts, logistics, invoices and fund flow information in the whole process by the blockchain mode, guarantees that the data cannot be tampered with, solves the problems of information silos and difficult data traceability in traditional systems, introduces a large model and performs appropriate fine-tuning or distillation to solve the text recognition problems such as handwritten body, fuzzy noise points, watermarks or seal cover, constructs a confidence degree system taking multi-modal data extraction quality, semantic matching degree and multi-dimensional data verification as the core, introduces a dynamic threshold mechanism, adjusts the verification strategy in real time according to the transaction complexity and historical behavior, intelligently determines whether to trigger the manual review process, introduces an event time axis and cross-modal buffer alignment mechanism to guarantee the consistency of multi-source data in the time dimension, and effectively improves the verification accuracy and process stability.

[0027] To enable a further understanding of the features and technical contents of the present application, please refer to the following detailed description and drawings of the present application. However, the drawings provided are only for reference and illustration, and are not intended to limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 It is a schematic diagram of the overall structure framework of the present application;

[0029] Figure 2 It is a schematic diagram of the trusted data storage module of the present application;

[0030] Figure 3 It is a schematic diagram of the multi-source data AI recognition and analysis module of the present application;

[0031] Figure 4 It is a schematic diagram of the AI-driven verification control module of the present application;

[0032] Figure 5 It is a schematic diagram of the asynchronous collaborative monitoring module of the present application;

[0033] Figure 6The actual effect verification schematic diagram for the trade background of the present application is shown in the figure. DETAILED DESCRIPTION

[0034] The following is a specific embodiment to illustrate the embodiments of the present application, and those skilled in the art can understand the advantages and effects of the present application from the disclosure. The present application can be implemented or applied by other different embodiments, and various modifications and changes can be made based on different views and applications without departing from the spirit of the present application. In addition, the drawings of the present application are only simple schematic illustrations, not the actual size, and the prior declaration is made. The following embodiments will further illustrate the related technical content of the present application, but the disclosed content is not used to limit the protection scope of the present application.

[0035] Example one.

[0036] The present embodiment provides a supply chain trade background verification system based on blockchain and AI, which combines Figure 1 , including on-chain trusted data storage module, multi-source data AI identification and analysis module, AI-driven verification control module and asynchronous collaborative monitoring module;

[0037] The on-chain trusted data storage module is used for storing data on chain in the whole process, the multi-source data AI identification and analysis module is used for extracting and semantic understanding of original data, the AI-driven verification control module is used for consistency and rationality verification of on-chain information, and the asynchronous collaborative monitoring module is used for managing data time asynchronous problem;

[0038] The trusted data storage module includes four-flow recording unit, abstract generation unit and index tracing unit, the four-flow recording unit records contract, logistics, invoice and fund flow information through blockchain technology, the abstract generation unit is used for formatting on-chain data and generating hash abstract, and the index tracing unit is used for searching information in the blockchain;

[0039] The multi-source data AI identification and analysis module includes text recognition unit, semantic understanding unit and standard output unit, the text recognition unit is used for identifying text information from non-text type data, the semantic understanding unit is used for analyzing logical semantics in text information, and the standard output unit is used for converting analysis information into standard format file output;

[0040] The text recognition unit includes file preprocessor, character recognition processor and text format processor, the file preprocessor is used for format conversion of file, converting the file into the format required by the character recognition processor, the character recognition processor is used for recognizing and extracting text information, and the text format processor is used for formatting the extracted text information, preventing garbled code and meeting the subsequent processing requirements;

[0041] The character recognition processor adopts a multi-modal large model technology, such as distilling the Llama4 Maverick model, to maintain the accuracy of character recognition while improving the inference efficiency, meet the needs of real-time processing of online services, support text input and image input, extract key business entities from natural language texts such as contracts, invoices, and emails, and identify business elements such as "payee", "currency", and "payment time" in contract clauses.

[0042] The semantic understanding unit includes a semantic parsing processor, a syntax dependency processor, and an information normalization processor. The semantic parsing processor is used to extract entity information with unique independent meanings from the text. The syntax dependency processor is used to construct dependency structures and extract semantic roles. The information normalization processor is used to map entity information into semantic roles of dependency structures.

[0043] The semantic parsing processor integrates knowledge graphs and large model technology, embedding entities and relationships in the knowledge graph into semantic space, enhancing the semantic understanding and reasoning capabilities of large models, improving the accuracy and richness of semantic representation, and distinguishing between synonyms / fuzzy words such as "payee" and "payee account".

[0044] The AI-driven verification control module includes a target driving unit, a logic verification unit, and a confidence control unit. The target driving unit is used to output flow information required for verification. The logic verification unit is used to perform logic verification on the matching degree between flow information and target files. The confidence control unit is used to perform confidence evaluation on the verification result and control the subsequent process.

[0045] The asynchronous collaborative monitoring module includes an asynchronous buffering alignment unit, an artificial review scheduling unit, and a visual reporting unit. The asynchronous buffering alignment unit is used to align multi-modal data by feature extraction mapping alignment method. The artificial review scheduling unit is used to reschedule verification tasks with low confidence. The visual reporting unit is used to visualize the entire verification information and generate a report file.

[0046] The logic verification unit includes a field comparison processor, a semantic conflict detector, and an abnormal rule matcher. The field comparison processor is used to directly compare multi-source fields. The semantic conflict detector is used to determine whether there is a conflict between entity information in the target file and entity information in the flow information. The abnormal rule matcher is used to set a rule library and determine whether there is a situation that meets the set abnormal rules.

[0047] The confidence control unit comprises a confidence processor, a dynamic threshold adjuster and a flow control processor, the confidence processor generates a confidence degree based on the verification result, the dynamic threshold adjuster dynamically adjusts a confidence threshold based on historical data of both parties of a transaction, and the flow control processor controls a subsequent flow based on a mathematical relationship between the confidence degree and the confidence threshold.

[0048] The confidence processor calculates the confidence degree S of the target file according to the following formula:

[0049]

[0050] wherein n is a ratio of the number of inconsistent fields detected by the field comparison processor to the total number of detected fields, m1 is the number of conflicts detected by the semantic conflict detector, m2 is the number of rules matched by the abnormal rule matcher, A is a semantic conflict threshold, B is a rule conflict threshold, a is a semantic base, and β is a rule base.

[0051] The dynamic threshold adjuster calculates the dynamic threshold T according to the following formula:

[0052] T = T0 + λ · σ - μ · ρ;

[0053] wherein T0 is a basic threshold, σ is a standard deviation of the confidence degree of historical transactions, ρ is a complexity factor of the current transaction, λ is a historical adjustment coefficient, and μ is a current adjustment coefficient.

[0054] When the confidence degree is greater than or equal to the dynamic threshold, it indicates that the target file passes the verification and is directly recorded in the blockchain, and when the confidence degree is less than the dynamic threshold, it indicates that the target file is not trustworthy and needs human intervention for verification.

[0055] Embodiment two.

[0056] This embodiment includes all the contents of embodiment one and provides a supply chain trade background verification system based on blockchain and AI, comprising an on-chain trusted data storage module, a multi-source data AI identification and analysis module, an AI-driven verification control module and an asynchronous collaborative monitoring module.

[0057] The on-chain trusted data storage module is used to store all-process data on the chain, the multi-source data AI identification and analysis module is used to extract and understand the semantics of the original data, the AI-driven verification control module is used to verify the consistency and rationality of the on-chain information, and the asynchronous collaborative monitoring module is used to manage the data time asynchronous problem.

[0058] In combination with Figure 2The trusted data evidence module includes a four-flow recording unit, an abstract generation unit and an index tracing unit, the four-flow recording unit records contract, logistics, invoice and fund flow information through blockchain technology, the abstract generation unit is used for formatting the on-chain data and generating a hash abstract, and the index tracing unit is used for retrieving information in the blockchain;

[0059] In combination Figure 3 The multi-source data AI identification and analysis module includes a text identification unit, a semantic understanding unit and a standard output unit, the text identification unit is used for identifying text information from non-text type data, the semantic understanding unit is used for analyzing the logical semantics in the text information, and the standard output unit is used for converting the analysis information into a standard format file output;

[0060] In combination Figure 4 The AI driven verification control module includes a target driving unit, a logic verification unit and a confidence control unit, the target driving unit is used for outputting the flow information required for verification, the logic verification unit is used for logically verifying the matching degree between the flow information and the target file, and the confidence control unit is used for confidence evaluation and control of the verification result;

[0061] In combination Figure 5 The asynchronous collaborative monitoring module includes an asynchronous buffering alignment unit, an artificial review scheduling unit and a visual report unit, the asynchronous buffering alignment unit is used for aligning multi-modal data through feature extraction mapping alignment method, the artificial review scheduling unit is used for second scheduling of the verification task with low confidence, and the visual report unit is used for visual processing of the entire verification information and generating a report file;

[0062] The four-flow recording unit includes a contract evidence blockchain, a logistics information blockchain, an invoice information blockchain and a fund flow blockchain, the contract evidence blockchain is used for storing supply chain contract information, the logistics information blockchain is used for storing logistics information of transaction goods, the invoice information blockchain is used for storing key field information of transaction invoices, and the fund flow blockchain is used for storing fund payment information of transaction parties;

[0063] The abstract generation unit includes a field extraction processor, a field arrangement processor and a hash calculation processor, the field extraction processor is used for extracting key field information in the text, the field arrangement processor sorts the fields according to a preset rule, and the hash calculation processor performs hash calculation on the sorted field information to obtain abstract information;

[0064] The index tracing unit comprises an index construction processor, an on-chain retrieval processor and a tracing query processor, the index construction processor is configured to generate a multi-dimensional retrieval label for the on-chain data, the on-chain retrieval processor is configured to provide a retrieval-based fast information positioning service, and the tracing query processor is configured to query a historical transaction link and a corresponding file digest;

[0065] The text recognition unit comprises a file preprocessor, a character recognition processor and a text format processor, the file preprocessor is configured to perform format conversion on the file, convert the file into a format required by the character recognition processor, the character recognition processor is configured to recognize and extract character information, and the text format processor is configured to perform format processing on the extracted character information, prevent garbled code and meet subsequent processing requirements;

[0066] The character recognition processor adopts a multi-modal large model technology, for example, distills the Llama4 Maverick model, maintains the accuracy of character recognition while improving the inference efficiency, meets the demand of real-time processing of online business, supports text input and image input, extracts key business entities from natural language texts such as contracts, invoices and emails, and recognizes business elements such as “payee”, “currency” and “payment time” in contract clauses;

[0067] The semantic understanding unit comprises a semantic parsing processor, a syntax dependency processor and an information normalization processor, the semantic parsing processor is configured to extract entity information with unique and independent meanings from the text, the syntax dependency processor is configured to construct a dependency structure and extract semantic roles, and the information normalization processor is configured to map the entity information into the semantic roles of the dependency structure;

[0068] The semantic parsing processor fuses a knowledge graph and a large model technology, embeds entities and relationships in the knowledge graph into a semantic space, enhances the semantic understanding and reasoning ability of the large model, improves the accuracy and richness of semantic representation, and distinguishes near-synonymous / fuzzy words such as “payee” and “payee account”;

[0069] The standard output unit comprises a sample database, a file generation processor and a verification output processor, the sample database is configured to store file samples of all types, the file generation processor is configured to match a corresponding file sample and fill the semantic roles into the file sample to generate a target file, and the verification output processor is configured to verify the matching degree of the target file and output to an AI-driven verification control module;

[0070] The target driving unit comprises a target judgment processor, a verification element parser and a flow information extractor, the target judgment processor is used to judge whether verification is needed based on a target file type, the verification element parser is used to parse a target file to obtain a search element, and the flow information extractor is used to obtain corresponding flow information from a block chain through the search element;

[0071] The logic verification unit comprises a field comparison processor, a semantic conflict detector and an abnormal rule matcher, the field comparison processor is used to directly compare multi-source fields, the semantic conflict detector is used to judge whether entity information in a target file and entity information in flow information exist conflicts, and the abnormal rule matcher is used to set a rule library and judge whether a condition meeting an abnormal rule exists;

[0072] The confidence control unit comprises a confidence processor, a dynamic threshold adjuster and a flow control processor, the confidence processor is used to generate a confidence degree based on a verification result, the dynamic threshold adjuster is used to dynamically adjust a confidence threshold value based on historical data of transaction parties, and the flow control processor is used to control a subsequent flow based on a mathematical relationship between the confidence degree and the confidence threshold value;

[0073] The confidence processor calculates a confidence degree S of a target file according to the following formula:

[0074]

[0075] Wherein, n is a ratio of a number of inconsistent fields detected by the field comparison processor to a total number of detected fields, m1 is a number of conflicts detected by the semantic conflict detector, m2 is a number of matched rules of the abnormal rule matcher, A is a semantic conflict threshold value, B is a rule conflict threshold value, a is a semantic base, and β is a rule base;

[0076] The semantic conflict threshold value and the rule conflict threshold value are artificially set after actual test, for example, batch trade background real data and trade background non-real data are used to perform semantic conflict detection and rule conflict detection, a mean value of a number of conflict detections of the trade background real data is denoted as a, a mean value of a number of conflict detections of the trade background non-real data is denoted as b, and A and B are set as follows: The rule conflict threshold value is set in the same way as the semantic conflict threshold value;

[0077] In particular, the value range of a and β is (0, 1), which is used to reflect the importance of semantic conflicts and rule conflicts to the confidence degree;

[0078] The dynamic threshold adjuster calculates a dynamic threshold value T according to the following formula:

[0079] T=T0+λ·σ-μ·ρ;

[0080] Wherein, T0 is a basic threshold value, σ is a confidence degree standard deviation of historical transactions, ρ is a complexity factor of the current transaction, λ is a historical adjustment coefficient, and μ is a current adjustment coefficient;

[0081] When the confidence degree is greater than or equal to the dynamic threshold value, it indicates that the target file passes the verification and is directly recorded in the blockchain, and when the confidence degree is less than the dynamic threshold value, it indicates that the target file is not trustworthy and needs human intervention for verification.

[0082] The asynchronous buffer alignment unit includes an event timeline manager, a cross-modal alignment processor, and a time tolerance setter, the event timeline manager is used to construct a unified business event timeline, the cross-modal alignment processor is used to process the time offset of the source data, and the time tolerance setter is used to set the tolerable time difference within an acceptable range.

[0083] When the flow information extractor cannot extract the required flow information, the cross-modal alignment processor determines that the source data has a time offset, and searches in the source data not stored in the blockchain until the corresponding source data is searched and sent to the logical verification unit;

[0084] The artificial review scheduling unit includes a review task generator, an intelligent assignment processor, and a review log recorder, the review task generator is used to create a review task when the confidence degree is insufficient, the intelligent assignment processor is used to accurately assign the review task to the corresponding reviewer, and the review log recorder is used to record the process and result information of each artificial review;

[0085] The visualization report unit includes a data visualization processor, a report generation processor, and an abnormality alert processor, the data visualization processor is used to convert the verification process and result into a chart form and display, the report generation processor is used to output a standard format verification report file, and the abnormality alert processor is used to specially process the abnormal content in the verification information;

[0086] Part of the code information of the system is as follows:

[0087]

[0088]

[0089]

[0090] Two samples are prepared, each containing 40 pieces of to-be-tested data, wherein sample one contains 10 pieces of trade background real data and 30 pieces of trade background non-real data, and sample two contains 30 pieces of trade background real data and 10 pieces of trade background non-real data, the verification is performed through the system, and the verification result is as shown in Figure 6 , which has high verification accuracy.

[0091] The above disclosed is only the preferred embodiment of the present application, and is not intended to limit the protection scope of the present application, so any equivalent technical change made according to the content of the present application and the drawings is included in the protection scope of the present application, and in addition, the elements can be updated as the technology develops.

Claims

1. A blockchain and AI based supply chain trade background verification system, characterized in that, The system comprises an on-chain trusted data storage module, a multi-source data AI identification and analysis module, an AI-driven verification control module, and an asynchronous collaborative monitoring module. The on-chain trusted data storage module is used for storing all-process data on the chain, the multi-source data AI identification and analysis module is used for extracting and semantically understanding the original data, the AI-driven verification control module is used for verifying the consistency and rationality of the on-chain information, and the asynchronous collaborative monitoring module is used for managing the data time asynchronous problem. The on-chain trusted data storage module comprises a four-flow recording unit, an abstract generation unit, and an index tracing unit, the four-flow recording unit records contract, logistics, invoice, and fund flow information through blockchain technology, the abstract generation unit formats the on-chain data and generates a hash abstract, and the index tracing unit retrieves information in the blockchain. The multi-source data AI identification and analysis module comprises a text recognition unit, a semantic understanding unit, and a standard output unit, the text recognition unit is used for identifying text information from non-text type data, the semantic understanding unit is used for analyzing the logical semantics in the text information, and the standard output unit is used for converting the parsed information into a standard format file output. The AI-driven verification control module comprises a target driving unit, a logic verification unit, and a confidence control unit, the target driving unit is used for outputting the flow information required for verification, the logic verification unit is used for logically verifying the matching degree between the flow information and the target file, and the confidence control unit is used for confidence evaluation and control of the verification result. The asynchronous collaborative monitoring module comprises an asynchronous buffer alignment unit, an artificial review scheduling unit, and a visual report unit, the asynchronous buffer alignment unit is used for aligning multi-modal data through feature extraction mapping alignment method, the artificial review scheduling unit is used for second scheduling of verification tasks with low confidence, and the visual report unit is used for visualizing the entire verification information and generating a report file.

2. The blockchain and AI based supply chain trade background verification system as claimed in claim 1, wherein, The text recognition unit comprises a file preprocessor, a character recognition processor, and a text format processor, the file preprocessor is used for format conversion of the file, converting the file into a format required by the character recognition processor, the character recognition processor is used for recognizing and extracting text information, and the text format processor is used for formatting the extracted text information to prevent garbled code and meet the subsequent processing requirements. The character recognition processor adopts multi-modal large model technology, such as distilling Llama4 Maverick model, which improves the inference efficiency while maintaining the accuracy of character recognition, meets the demand of online business real-time processing, supports text input and image input, extracts key business entities from natural language texts such as contracts, invoices, and emails, and identifies business elements such as "paying party", "currency", and "payment time" in contract clauses.

3. The blockchain and AI based supply chain trade background verification system as claimed in claim 2, wherein, The semantic understanding unit comprises a semantic parsing processor, a syntax dependency processor and an information normalization processor, the semantic parsing processor is configured to extract entity information with unique independent meaning from the text, the syntax dependency processor is configured to construct a dependency structure and extract semantic roles, and the information normalization processor is configured to map the entity information into the semantic roles of the dependency structure. The semantic parsing processor fuses the knowledge graph and the large model technology, embeds entities and relationships in the knowledge graph into a semantic space, enhances the semantic understanding and reasoning capability of the large model, improves the accuracy and richness of semantic representation, and distinguishes between synonyms / fuzzy words such as "payee" and "payee account".

4. The blockchain and AI based supply chain trade background verification system as claimed in claim 3, wherein, The logic verification unit comprises a field comparison processor, a semantic conflict detector and an abnormal rule matcher, the field comparison processor is configured to directly compare multi-source fields, the semantic conflict detector is configured to determine whether the entity information in the target file and the entity information in the flow information conflict, and the abnormal rule matcher is configured to set a rule library and determine whether there is a condition that meets the set abnormal rule.

5. The blockchain and AI based supply chain trade background verification system as claimed in claim 4, wherein, The confidence control unit comprises a confidence processor, a dynamic threshold adjuster and a flow control processor, the confidence processor generates a confidence degree based on the verification result, the dynamic threshold adjuster dynamically adjusts the confidence threshold based on the historical data of the two parties of the transaction, and the flow control processor controls the subsequent process based on the relationship between the confidence degree and the confidence threshold.

6. The blockchain and AI based supply chain trade background verification system as claimed in claim 5 wherein, The confidence processor calculates the confidence degree S of the target file according to the following formula: Wherein, n is the ratio of the number of inconsistent fields detected by the field comparison processor to the total number of detected fields, m1 is the number of conflicts detected by the semantic conflict detector, m2 is the number of rules matched by the abnormal rule matcher, A is the semantic conflict threshold, B is the rule conflict threshold, alpha is the semantic base, and beta is the rule base.

7. The blockchain and AI based supply chain trade background verification system as claimed in claim 6 wherein, The dynamic threshold adjuster calculates the dynamic threshold T according to the following formula: T = T0 + lambda * sigma - mu * rho; Wherein, T0 is the basic threshold, sigma is the confidence degree standard deviation of historical transactions, rho is the complexity factor of the current transaction, lambda is the historical adjustment coefficient, and mu is the current adjustment coefficient; When the confidence degree is greater than or equal to the dynamic threshold, it indicates that the target file passes the verification and is directly recorded in the blockchain, and when the confidence degree is less than the dynamic threshold, it indicates that the target file is not trustworthy and needs human intervention for verification.

Images