Supply chain contract intelligent review system and method based on large language model

Abstract

The invention discloses a supply chain contract intelligent review system and method based on a large language model, and relates to the technical field of contract review. Aiming at the defect that the existing contract review generally depends on fixed template and keyword matching, the adopted scheme comprises the following steps: receiving a contract text through a text acquisition module; preprocessing the text through a text preprocessing module; the large language model analysis module adopts a pre-trained large language model to carry out deep semantic understanding on a text and extract key information; a supply chain management domain knowledge graph is constructed through a graph construction module, and compliance verification is assisted; the intelligent analysis module performs multi-dimensional risk identification and compliance evaluation on contract content in combination with a big language model analysis result and a knowledge graph; and the visualization module provides an interactive user interface for a user to upload a contract text, and displays an examination result, a risk prompt and a compliance suggestion of the contract text. According to the invention, the supply chain contract text can be automatically examined and risk early warning can be carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of contract review, and in particular to a supply chain contract intelligent review system and method based on a large language model. Background Art

[0002] Traditional supply chain contract review relies primarily on manual item-by-item verification, which has the following significant technical limitations:

[0003] 1. Inefficiency: Manual word-by-word review is time-consuming and tedious. When faced with large-scale processing scenarios involving hundreds or thousands of contracts, it is difficult to meet the demand for rapid response and easily leads to process backlogs.

[0004] 2. Lack of accuracy: Due to the influence of the professional background, experience differences and subjective understanding bias of the reviewers, it is easy to omit the judgment of ambiguous clauses and potential risk points in the contract, or make inconsistent interpretations of similar clauses, leading to the risk of misjudgment.

[0005] 3. High cost: It is necessary to hire legal advisors or senior legal personnel with expertise in the supply chain field on a long-term basis. The high labor cost significantly increases the operating burden of the enterprise.

[0006] 4. Lagging knowledge iteration: Laws and regulations (such as supply chain-related trade regulations and breach of contract liability clauses) and industry-wide contract templates are frequently updated, making it difficult for humans to track the latest revisions in real time, which may lead to outdated review basis.

[0007] Existing technologies still have shortcomings in terms of improvement: although some rule-based natural language processing systems have been applied to contract review, their core relies on preset fixed templates and keyword precise matching mechanisms. They are not adaptable to contract texts with flexible sentence structures and complex expressions (such as customized clauses and cross-domain mixed clauses), and cannot achieve semantic understanding of the deep logical relationships and implicit risks of the clauses, making it difficult to cover the review needs in complex scenarios. Summary of the Invention

[0008] In response to the needs and shortcomings of current technological development, the present invention provides a supply chain contract intelligent review system and method based on a large language model to automatically review supply chain contract texts, identify potential risk clauses, and improve contract review efficiency and accuracy.

[0009] In a first aspect, the present invention provides a supply chain contract intelligent review system based on a large language model. The technical solutions adopted to solve the above technical problems are as follows:

[0010] An intelligent review system for supply chain contracts based on a large language model, the structure of which includes:

[0011] The text preprocessing module is used to preprocess the contract text uploaded by the user to purify the text data and lay the foundation for subsequent text analysis;

[0012] The large language model parsing module uses a pre-trained large language model to perform deep semantic understanding of pre-processed contract text and extract key information;

[0013] A graph construction module is used to build a knowledge graph in the supply chain management field. The knowledge graph includes laws and regulations, industry standards, and common risk clauses to assist in compliance verification.

[0014] An intelligent analysis module, which combines the analysis results of a large language model with the knowledge graph to conduct multi-dimensional risk identification and compliance assessment of contract content;

[0015] The visualization module is used to provide an intuitive and interactive user interface for users to upload contract texts and display risk reports of the contract texts, including review results, risk warnings and compliance recommendations.

[0016] Optionally, the text preprocessing modules involved include:

[0017] Word segmentation units are used to split the continuous contract text into the smallest units with independent semantics, transforming the text from an unordered sequence of characters into a structured vocabulary set;

[0018] The stop word removal unit is used to process the contract text after word segmentation, removing function words and auxiliary words that have no actual semantic contribution or words that are meaningless to the core information of the contract, thereby reducing the interference of redundant data on analysis;

[0019] The stem extraction unit is used to process the words after the stop words are removed, restore the different forms of the same word to the root or stem, unify the word expression form, reduce the complexity of the feature space, and enhance the consistency of the word semantics.

[0020] Further optional, the large language model involved is specifically a GPT series, LLaMA series or BERT derivative model with strong semantic understanding and information extraction capabilities. It is trained on large-scale text corpus to capture the contextual logic, professional terminology associations and implicit semantics in the contract text, and conducts in-depth semantic understanding of the contract text after word segmentation, stop word removal and stem extraction, and accurately extracts the three key information of the contract subject, clause content and responsibility allocation.

[0021] Further optionally, the knowledge graph constructed by the graph construction module involved is organized in the core form of a structured triple of "entity-relationship-entity", covering three key contents: supply chain-related laws and regulations, supply chain industry general standards, and risk clauses commonly included in supply chain contracts.

[0022] Optionally, the intelligent analysis modules involved specifically include:

[0023] The element extraction unit is used to automatically identify key risk factors in the contract based on the deep semantic understanding of the contract text based on the large language model and map them to corresponding entities in the knowledge graph;

[0024] The compliance verification unit is used to compare the extracted risk factors with the laws, regulations, and general industry standards in the knowledge graph to verify whether there are any violation risks in the contract terms;

[0025] The risk quantification unit is used to quantify the overall risk level of the contract text by using a predefined risk rating system based on the risk characteristics of the contract and combining it with a large language model to analyze the risk-related tendencies of the clause semantics;

[0026] The risk reasoning unit is used to leverage the graph structure of the knowledge graph to uncover hidden risks associated between clauses and identify potential performance risk chains.

[0027] The risk generation unit is used to automatically generate risk warnings that comply with legal regulations for identified risk points and provide compliance recommendations related to the knowledge graph.

[0028] In a second aspect, the present invention provides a method for intelligent review of supply chain contracts based on a large language model. The technical solutions adopted to solve the above technical problems are as follows:

[0029] A supply chain contract intelligent review method based on a large language model includes the following steps:

[0030] S1. Provide an intuitive and interactive user interface for users to upload contract texts and perform pre-processing operations on the contract texts to purify the text data and lay the foundation for subsequent text analysis;

[0031] S2. Use a pre-trained large language model to perform deep semantic understanding on the pre-processed contract text and extract key information;

[0032] S3. Build a knowledge graph in the supply chain management field. The knowledge graph includes laws and regulations, industry standards, and common risk clauses to assist in compliance verification.

[0033] S4. Combine the analysis results of the large language model with the knowledge graph to conduct multi-dimensional risk identification and compliance assessment of the contract content;

[0034] S5. Through an intuitive and interactive user interface, the review results, risk warnings and compliance recommendations of the contract text are displayed.

[0035] Optionally, execute step S1 to perform the following preprocessing operations on the contract text:

[0036] S1.1. Perform word segmentation on the contract text to break the continuous contract text into the smallest units with independent semantics, converting the text from an unordered character sequence into a structured vocabulary set;

[0037] S1.2. Process the segmented contract text to remove function words and auxiliary words that do not contribute to the actual semantics, or words that are meaningless to the core information of the contract, to reduce the interference of redundant data on the analysis;

[0038] S1.3. Process the words after stop words are removed, restore different forms of the same word to roots or stems, unify the word expression form, reduce the complexity of the feature space, and enhance the consistency of word semantics.

[0039] Further optional, the large language model involved is specifically a GPT series, LLaMA series or BERT derivative model with strong semantic understanding and information extraction capabilities. It is trained on large-scale text corpus to capture the contextual logic, professional terminology associations and implicit semantics in the contract text, and conducts in-depth semantic understanding of the contract text after word segmentation, stop word removal and stem extraction, and accurately extracts the three key information of the contract subject, clause content and responsibility allocation.

[0040] Further optionally, step S3 is executed to construct a knowledge graph with the structured triple of "entity-relationship-entity" as the core organizational form, covering three key contents: supply chain-related laws and regulations, supply chain industry general standards, and risk clauses commonly included in supply chain contracts.

[0041] Optionally, step S4 specifically includes:

[0042] S4.1. Automatically identify key risk factors in contracts based on deep semantic understanding of contract texts using a large language model and map them to corresponding entities in the knowledge graph.

[0043] S4.2. Compare the extracted risk factors with the laws, regulations, and general industry standards in the knowledge graph to verify whether there are any risk factors for violations in the contract terms;

[0044] S4.3. Quantify the overall risk level of the contract text using a predefined risk rating system tailored to contract risk characteristics and a large language model that analyzes the risk-related tendencies of clause semantics.

[0045] S4.4. Leverage the graph structure of the knowledge graph to uncover hidden risk factors between clauses and identify potential performance risk chains;

[0046] S4.5. For the identified risk factors, automatically generate risk warnings that comply with legal regulations and provide relevant compliance recommendations in the knowledge graph.

[0047] The intelligent supply chain contract review system and method based on a large language model of the present invention has the following beneficial effects compared with the prior art:

[0048] 1. This invention integrates natural language processing, large language models, and supply chain management technologies to automatically review supply chain contract texts, identify potential risk clauses, improve contract review efficiency and accuracy, implement intelligent analysis and early warning of contract content, and reduce enterprise operating costs.

[0049] 2. The present invention can automate the review process to significantly improve contract processing efficiency and shorten the review cycle; through the combination of large language models and domain knowledge graphs, it can improve the accuracy of contract review, reduce the misjudgment rate, reduce reliance on manual review, and reduce enterprise operating costs; it supports custom rules and knowledge graph updates, and can adapt to changes in laws, regulations and contract terms. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Attachment Figure 1 This is a module connection block diagram of the first embodiment of the present invention;

[0051] Attachment Figure 2 This is a flowchart of the method of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0052] In order to make the technical solution, the technical problems solved and the technical effects of the present invention more clear, the technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments.

[0053] Example 1:

[0054] Combined with attachment Figure 1 This embodiment proposes a supply chain contract intelligent review system based on a large language model, the structure of which includes:

[0055] The text preprocessing module is used to preprocess the contract text uploaded by the user to purify the text data and lay the foundation for subsequent text analysis;

[0056] The large language model parsing module uses a pre-trained large language model to perform deep semantic understanding of pre-processed contract text and extract key information;

[0057] A graph construction module is used to build a knowledge graph in the supply chain management field. The knowledge graph includes laws and regulations, industry standards, and common risk clauses to assist in compliance verification.

[0058] An intelligent analysis module, which is used to combine the parsing results of the large language model and the knowledge graph to conduct multi-dimensional risk identification and compliance assessment on the contract content;

[0059] A visualization module, which is used to provide an intuitive and interactive user interface for users to upload contract texts and display the risk reports of the contract texts. The risk reports include review results, risk warnings, and compliance suggestions.

[0060] In this embodiment, the involved text preprocessing module includes:

[0061] A word segmentation unit, which is used to split the continuous contract text into the smallest units with independent semantics (such as words or phrases), so that the text is transformed from an unordered character sequence into a structured vocabulary set;

[0062] A stop word removal unit, which is used to process the contract text after word segmentation, remove the function words and auxiliary words without actual semantic contribution (such as "of", "in", "le", etc.) or the words meaningless to the core information of the contract (such as clause content, rights and obligations, liability for breach of contract, etc.), and reduce the interference of redundant data on the analysis;

[0063] A stemming unit, which is used to process the words after stop word removal, restore the same words in different forms (such as "review", "reviewing", "reviewed") to the root or stem (such as "review"), unify the vocabulary expression form, reduce the complexity of the feature space, enhance the consistency of vocabulary semantics, and improve the efficiency and accuracy of subsequent natural language processing tasks (such as contract clause matching, risk keyword identification, text classification, etc.).

[0064] In this embodiment, the involved large language model is specifically the GPT series, LLaMA series or BERT-derived model with strong semantic understanding and information extraction capabilities. It is trained through a large-scale text corpus to capture the context logic, professional term associations, and implicit semantics in the contract text, conduct in-depth semantic understanding on the contract text after word segmentation, stop word removal, and stemming, and accurately extract the three key information of the contract subject, clause content, and liability allocation.

[0065] In this embodiment, the knowledge graph constructed by the involved graph construction module takes the structured triple of "entity-relationship-entity" as the core organizational form, covering three types of key contents: one is the laws and regulations related to the supply chain (such as the civil code clauses, industry supervision regulations, cross-border trade agreements, etc. corresponding to the procurement, transportation, warehousing, import and export trade, etc. links); the second is the general standards of the supply chain industry (including logistics operation specifications, quality acceptance standards, information security compliance guidelines, etc.); the third is the common risk clauses frequently appearing in supply chain contracts (such as liability for breach of contract for payment delay, agreement on compensation for damaged goods, definition clause of force majeure, clause on ownership of intellectual property rights, etc.).

[0066] By systematically sorting and associating these contents, a knowledge graph covering the entire supply chain process is formed, providing accurate knowledge support for compliance verification in contract review. When reviewing contract terms, the legal and regulatory items, industry standard requirements and risk clause characteristics in the knowledge graph can be quickly matched to assist in identifying deviations between the contract and compliance requirements, potential risk points and clause loopholes.

[0067] In this embodiment, the intelligent analysis module involved specifically includes:

[0068] The element extraction unit is used to automatically identify key risk factors in the contract (such as payment terms, delivery conditions, and liability for breach of contract) based on the deep semantic understanding of the contract text by the large language model, and map them to corresponding entities in the knowledge graph (such as "payment delay" is associated with the "breach of contract liability" entity);

[0069] The compliance verification unit is used to compare the extracted risk factors with the laws and regulations (such as relevant provisions of the Civil Code) and general industry standards (such as logistics operating specifications) in the knowledge graph to verify whether there are any violation risks in the contract terms (such as the liquidated damages ratio exceeding the statutory limit);

[0070] The risk quantification unit is used to quantify the overall risk level of the contract text (outputting a specific score or risk level label to reflect the overall risk level of the contract) by combining a predefined risk grading system for contract risk characteristics (covering dimensions such as compliance, performance feasibility, and equivalence of rights and obligations, using a high / medium / low three-level or quantitative scoring standard to clarify the risk characteristics and judgment rules corresponding to each level) with a large language model's risk-related tendency analysis of the clause semantics (such as identifying characteristics such as ambiguous expressions and imbalance of rights and obligations).

[0071] The risk reasoning unit is used to leverage the graph structure of the knowledge graph to uncover risk factors hidden between clauses (such as potential conflicts between payment terms and delivery cycles) and identify potential performance risk chains.

[0072] The risk generation unit is used to automatically generate risk warnings that comply with legal regulations for identified risk points (such as "This clause may violate Article 585 of the Civil Code") and provide compliance suggestions related to the knowledge graph (such as "It is recommended to adjust the penalty ratio to no more than 30% of the subject amount").

[0073] Example 2:

[0074] Combined with attachment Figure 2 This embodiment proposes a supply chain contract intelligent review method based on a large language model, which includes the following steps:

[0075] S1. Provide an intuitive and interactive user interface for users to upload contract texts and perform preprocessing operations on the contract texts to purify the text data and lay a foundation for subsequent text analysis.

[0076] Execute step S1 and perform the following preprocessing operations on the contract text:

[0077] S1.1. Perform word segmentation on the contract text to split the continuous contract text into the smallest units with independent semantics (such as words or phrases), converting the text from an unordered character sequence into a structured vocabulary set.

[0078] S1.2. Process the word-segmented contract text, removing function words and auxiliary words that have no actual semantic contribution (such as "de", "zai", "le", etc.) or words that are meaningless to the core information of the contract (such as clause content, rights and obligations, liability for breach of contract, etc.), reducing the interference of redundant data on the analysis.

[0079] S1.3. Process the words after stop-word removal, restoring different morphological forms of the same word (such as "review", "reviewing", "reviewed") to the root or stem (such as "review"), unifying the vocabulary expression form, reducing the complexity of the feature space, enhancing the consistency of word semantics, and improving the efficiency and accuracy of subsequent natural language processing tasks (such as contract clause matching, risk keyword identification, text classification, etc.).

[0080] S2. Adopt a pre-trained large language model to perform deep semantic understanding on the preprocessed contract text and extract key information.

[0081] The specific large language model involved is the GPT series, LLaMA series or BERT-derived model with strong semantic understanding and information extraction capabilities. It is trained through a large-scale text corpus to capture the context logic, professional term associations and implicit semantics in the contract text, perform deep semantic understanding on the contract text after word segmentation, stop-word removal and stemming, and accurately extract the three key information of the contract subject, clause content and responsibility allocation.

[0082] S3. Construct a knowledge graph in the field of supply chain management. The knowledge graph includes laws and regulations, industry general standards and common risk clauses, which are used to assist compliance verification.

[0083] The constructed knowledge graph is organized in the core form of the structured triple of "entity-relationship-entity", covering three key categories of content: First, laws and regulations related to the supply chain (such as civil code clauses, industry regulatory regulations, cross-border trade agreements, etc. corresponding to procurement, transportation, warehousing, import and export trade, etc.); second, general standards for the supply chain industry (including logistics operation specifications, quality acceptance standards, information security compliance guidelines, etc.); third, common risk clauses that appear frequently in supply chain contracts (such as liability for breach of contract for delayed payment, compensation for damage to goods, force majeure definition clauses, intellectual property ownership clauses, etc.).

[0084] By systematically sorting and associating these contents, a knowledge graph covering the entire supply chain process is formed, providing accurate knowledge support for compliance verification in contract review. When reviewing contract terms, the legal and regulatory items, industry standard requirements and risk clause characteristics in the knowledge graph can be quickly matched to assist in identifying deviations between the contract and compliance requirements, potential risk points and clause loopholes.

[0085] S4. Combine the large language model analysis results with the knowledge graph to conduct multi-dimensional risk identification and compliance assessment of the contract content, including:

[0086] S4.1. Automatically identify key risk factors in contracts (such as payment terms, delivery conditions, and liability for breach of contract) based on a deep semantic understanding of contract texts using a large language model, and map them to corresponding entities in the knowledge graph (e.g., “payment delay” is associated with the “breach of contract” entity).

[0087] S4.2. Compare the extracted risk factors with the laws and regulations (such as relevant provisions of the Civil Code) and general industry standards (such as logistics operating specifications) in the knowledge graph to verify whether the contract terms contain risk factors for violations (such as the liquidated damages ratio exceeding the statutory limit);

[0088] S4.3. Using a predefined risk grading system tailored to contract risk characteristics (covering dimensions such as compliance, performance feasibility, and equivalence of rights and obligations, using a high / medium / low tier or quantitative scoring system, with clear risk characteristics and determination rules for each tier), combined with risk-related propensity analysis of clause semantics using a large language model (e.g., identifying ambiguous expressions and imbalanced rights and obligations), the overall risk level of the contract text is quantitatively scored (outputting a specific score or risk tier label to reflect the overall risk level of the contract);

[0089] S4.4. Leverage the graph structure of the knowledge graph to uncover hidden risk factors between clauses (e.g., potential conflicts between payment terms and delivery cycles) and identify potential performance risk chains.

[0090] S4.5. For the identified risk factors, automatically generate risk warnings that comply with legal regulations (such as "This clause may violate Article 585 of the Civil Code") and provide related compliance suggestions in the knowledge graph (such as "It is recommended to adjust the penalty ratio to no more than 30% of the subject amount").

[0091] S5. Through an intuitive and interactive user interface, the review results, risk warnings and compliance recommendations of the contract text are displayed.

[0092] Specifically, a) for the review results, the user interface clearly presents the core information of the contract (such as subject information, subject details, key clause summary, etc.) through a regional panel, and uses color markings (such as green to mark compliance clauses, yellow to mark pending clauses, and red to highlight risk clauses) to distinguish different states; b) for risk warnings, the user interface combines text highlighting with pop-up instructions to accurately locate the specific clauses in the contract where risks exist, and simultaneously displays the risk type (such as compliance risk, logical contradiction risk, and ambiguous expression risk) and risk level (high / medium / low), and links the risk information in the knowledge graph. c) For the suggested modifications, a comment box is attached to the corresponding clause, providing specific modification directions (such as "It is recommended to supplement the division of responsibilities under force majeure"), modification examples (such as "Reference statement: 'If performance is impossible due to force majeure, the two parties shall negotiate an extension plan within 3 days'") and modification basis (such as "In accordance with the provisions on force majeure in Article 590 of the Civil Code"). At the same time, users are supported to directly edit and mark processed risks in the interface to achieve visual interaction and closed-loop management of review results.

[0093] In summary, the intelligent supply chain contract review system and method based on a large language model of the present invention can automatically review supply chain contract texts, identify potential risk clauses, improve contract review efficiency and accuracy, realize intelligent analysis and early warning of contract content, and reduce enterprise operating costs.

[0094] The above specific examples are used to illustrate the principles and implementation methods of the present invention in detail. These examples are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by those skilled in the art without departing from the principles of the present invention should fall within the scope of patent protection of the present invention.

Claims

1. A supply chain contract intelligent review system based on a large language model, characterized by: Its structure includes: The text preprocessing module is used to preprocess the contract text uploaded by the user to purify the text data and lay the foundation for subsequent text analysis; The large language model parsing module uses a pre-trained large language model to perform deep semantic understanding of pre-processed contract text and extract key information; A graph construction module is used to build a knowledge graph in the supply chain management field. The knowledge graph includes laws and regulations, industry standards, and common risk clauses to assist in compliance verification. An intelligent analysis module, which combines the analysis results of a large language model with the knowledge graph to conduct multi-dimensional risk identification and compliance assessment of contract content; The visualization module is used to provide an intuitive and interactive user interface for users to upload contract texts and display the review results, risk warnings and compliance recommendations of the contract texts.

2. The supply chain contract intelligent review system based on a large language model according to claim 1 is characterized in that: The text preprocessing module includes: Word segmentation units are used to split the continuous contract text into the smallest units with independent semantics, transforming the text from an unordered sequence of characters into a structured vocabulary set; The stop word removal unit is used to process the contract text after word segmentation, removing function words and auxiliary words that have no actual semantic contribution or words that are meaningless to the core information of the contract, thereby reducing the interference of redundant data on analysis; The stem extraction unit is used to process the words after the stop words are removed, restore the different forms of the same word to the root or stem, unify the word expression form, reduce the complexity of the feature space, and enhance the consistency of the word semantics.

3. The supply chain contract intelligent review system based on a large language model according to claim 2 is characterized in that: The large language model is specifically a GPT series, LLaMA series or BERT derivative model with strong semantic understanding and information extraction capabilities. It is trained on large-scale text corpus to capture the contextual logic, professional terminology associations and implicit semantics in the contract text, and conducts in-depth semantic understanding of the contract text after word segmentation, stop word removal and stem extraction, and accurately extracts the three key information of the contract subject, clause content and responsibility allocation.

4. The supply chain contract intelligent review system based on a large language model according to claim 3 is characterized in that: The knowledge graph constructed by the graph construction module is based on the structured triple of "entity-relationship-entity" as the core organizational form, covering three key contents: supply chain-related laws and regulations, supply chain industry general standards, and risk clauses commonly included in supply chain contracts.

5. The supply chain contract intelligent review system based on a large language model according to claim 1 is characterized in that: The intelligent analysis module specifically includes: The element extraction unit is used to automatically identify key risk factors in the contract based on the deep semantic understanding of the contract text based on the large language model and map them to corresponding entities in the knowledge graph; The compliance verification unit is used to compare the extracted risk factors with the laws, regulations, and general industry standards in the knowledge graph to verify whether there are any violation risks in the contract terms; The risk quantification unit is used to quantify the overall risk level of the contract text by using a predefined risk rating system based on the risk characteristics of the contract and combining it with a large language model to analyze the risk-related tendencies of the clause semantics; The risk reasoning unit is used to leverage the graph structure of the knowledge graph to uncover hidden risks associated between clauses and identify potential performance risk chains. The risk generation unit is used to automatically generate risk warnings that comply with legal regulations for identified risk points and provide compliance recommendations related to the knowledge graph.

6. A supply chain contract intelligent review method based on a large language model, characterized by: It includes the following steps: S1. Provide an intuitive and interactive user interface for users to upload contract texts and perform pre-processing operations on the contract texts to purify the text data and lay the foundation for subsequent text analysis; S2. Use a pre-trained large language model to perform deep semantic understanding on the pre-processed contract text and extract key information; S3. Build a knowledge graph in the supply chain management field. The knowledge graph includes laws and regulations, industry standards, and common risk clauses to assist in compliance verification. S4. Combine the analysis results of the large language model with the knowledge graph to conduct multi-dimensional risk identification and compliance assessment of the contract content; S5. Through an intuitive and interactive user interface, the review results, risk warnings and compliance recommendations of the contract text are displayed.

7. The method for intelligent review of supply chain contracts based on a large language model according to claim 6 is characterized in that: Execute step S1 to perform the following pre-processing operations on the contract text: S1.

1. Perform word segmentation on the contract text to break the continuous contract text into the smallest units with independent semantics, converting the text from an unordered character sequence into a structured vocabulary set; S1.

2. Process the segmented contract text to remove function words and auxiliary words that do not contribute to the actual semantics, or words that are meaningless to the core information of the contract, to reduce the interference of redundant data on the analysis; S1.

3. Process the words after stop words are removed, restore different forms of the same word to roots or stems, unify the word expression form, reduce the complexity of the feature space, and enhance the consistency of word semantics.

8. The method for intelligent review of supply chain contracts based on a large language model according to claim 7 is characterized in that: The large language model is specifically a GPT series, LLaMA series or BERT derivative model with strong semantic understanding and information extraction capabilities. It is trained on large-scale text corpus to capture the contextual logic, professional terminology associations and implicit semantics in the contract text, and conducts in-depth semantic understanding of the contract text after word segmentation, stop word removal and stem extraction, and accurately extracts the three key information of the contract subject, clause content and responsibility allocation.

9. The method for intelligent review of supply chain contracts based on a large language model according to claim 8 is characterized in that: Execute step S3 to construct a knowledge graph with the structured triple of "entity-relationship-entity" as the core organizational form, covering three key contents: supply chain-related laws and regulations, supply chain industry general standards, and risk clauses commonly included in supply chain contracts.

10. The method for intelligent review of supply chain contracts based on a large language model according to claim 6, characterized in that: The step S4 specifically includes: S4.

1. Automatically identify key risk factors in contracts based on deep semantic understanding of contract texts using a large language model and map them to corresponding entities in the knowledge graph. S4.

2. Compare the extracted risk factors with the laws, regulations, and general industry standards in the knowledge graph to verify whether there are any risk factors for violations in the contract terms; S4.

3. Quantify the overall risk level of the contract text using a predefined risk rating system tailored to contract risk characteristics and a large language model that analyzes the risk-related tendencies of clause semantics. S4.

4. Leverage the graph structure of the knowledge graph to uncover hidden risk factors between clauses and identify potential performance risk chains; S4.

5. For the identified risk factors, automatically generate risk warnings that comply with legal regulations and provide relevant compliance recommendations in the knowledge graph.

Images