Contract compliance auxiliary auditing method and system based on large language model

Abstract

The invention discloses a contract compliance auxiliary auditing method and system based on a large language model, and relates to the technical field of artificial intelligence auditing, and the method comprises the steps: carrying out the semantic analysis of an industry template and protocol entries to obtain contract semantics, determining the mapping relation between the contract semantics and legal entries, and the conflict condition of the contract semantics, identifying a right obligation subject based on an industry template to determine a contract target tendency, defining a risk value of a corresponding legal entry based on the contract target tendency to construct a static auditing unit, constructing a variable factor library, and generating a simulation scene according to the variable factor library to construct a live auditing unit; the contract details are input into a static auditing unit to be analyzed to obtain contract terms and corresponding grades, and the contract terms output in the static auditing model are input into a simulation unit to be subjected to sensitivity analysis to obtain a risk thermodynamic diagram and performance suggestions. According to the method, performance suggestions based on real-time data are provided for enterprises, and potential disputes caused by market uncertainty are effectively dealt with.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence auditing, in particular to a contract compliance auxiliary auditing method and system based on a large language model. BACKGROUND

[0002] With the continuous expansion of enterprise scale and the increasing complexity of business, contract auditing has become a key and arduous task in enterprise operation management. This process not only concerns the efficiency of business promotion, but also affects the sensitive nerves of legal compliance, financial tax, and qualification risk control, etc. Its importance is self-evident. However, in the face of a large number of contract texts with different structures, the traditional method of relying on manual review is particularly difficult. Auditors need to spend a lot of time and effort, but it is still difficult to avoid omissions and misjudgments, and low efficiency has become a bottleneck restricting the development of enterprises.

[0003] As a document with legal effect, the rigor and accuracy of the contract are directly related to the protection of the rights and interests of the enterprise and the safety of the operation. However, in actual operation, the contract text is often complex in structure and numerous in clauses, combined with specific requirements in different industries and different scenarios, making it difficult for manual processing to be in a state of being overworked and having little effect. Even experienced legal personnel also face the challenges of fast knowledge update and limited case accumulation, making it difficult to fully cover all potential risk points.

[0004] At present, although some enterprises try to introduce rule engines or single artificial intelligence technology to assist in auditing, these solutions still have obvious shortcomings. Although the rule engine can quickly match the preset conditions, it is difficult to deal with the diversity and flexibility of contract texts, and problems such as insufficient extraction accuracy and missing key fields occur frequently. The single AI model may deviate from the audit results due to limited training data or algorithm bias, and the subsequent maintenance cost is high, which is difficult to adapt to the changing business needs.

[0005] In order to solve these problems, a contract compliance auxiliary auditing method and system based on a large language model are needed. SUMMARY

[0006] To solve the above problems, the present application proposes a contract compliance auxiliary auditing method and system based on a large language model, which can deeply integrate advanced technologies such as natural language processing and machine learning, intelligently identify key information in contracts, automatically compare regulations and policies, and real-time alert potential risks, while having good scalability and ease of use, helping enterprises build an efficient and accurate contract management system, and escorting the development of business.

[0007] Among them, a contract compliance auxiliary auditing method based on a large language model includes the following steps: S1, obtain past contract data, past litigation cases, and supporting legal articles to obtain original data, preprocess the original data to obtain a distributed database, the distributed database including contract terms, litigation terms, and legal terms, the contract terms including industry templates and agreement terms; S2, perform semantic analysis on the industry templates and agreement terms to obtain contract semantics, determine the mapping relationship between the contract semantics and the legal terms, and the conflict situation of the contract semantics; Based on the industry template, the right and obligation subject is identified to determine the contract target tendency, and based on the contract target tendency, the risk value of the corresponding legal term is defined to construct a static audit unit; S3, obtain market environment fluctuation, construct a variable factor library according to the market environment fluctuation, and generate a simulation scenario according to the variable factor library to construct a live audit unit; S4, obtain the current contract details, input the contract details into the static audit unit for analysis to obtain the contract terms and their corresponding grades; The contract terms output from the static audit model are input into the simulation unit for sensitivity analysis to obtain a risk heat map and a performance suggestion; S5, the contract terms and their corresponding grades, the risk heat map, and the performance suggestion are summarized to generate a contract compliance auxiliary audit strategy.

[0008] Preferably, in S2, the specific content of the semantic analysis of the industry templates and the agreement terms to obtain the contract semantics and determine the mapping relationship between the contract semantics and the legal terms includes: According to the use scene of the industry template, the scene is divided to obtain a scene category group; Perform semantic analysis on the industry templates and the agreement terms under different scene category groups to obtain a triple including subject, behavior, and object, which is defined as contract semantics; Extract the key words of the contract semantics and the legal terms to obtain contract key words and legal key words; Directly match the contract key words and the legal key words to obtain a first-level mapping; Calculate the semantic similarity of the contract key words and the legal key words, and match according to the semantic similarity to obtain a second-level mapping; Analyze and extract the contract key words and the legal key words from the past litigation cases to obtain a third-level mapping; The first-level mapping, the second-level mapping, and the third-level mapping constitute the mapping relationship between the contract semantics and the legal terms.

[0009] Preferably, the specific content of the conflict situation of the contract semantics is: Obtain the contract semantics, and calculate the similarity of the subject, the behavior, and the object in the triple respectively; The similarity substandard range is preset, one of the triples is determined as a reference, and whether the other two are in the similarity substandard range is determined; If at least one of the other two is in the similarity substandard range, the current contract semantics is defined as a contract semantics group, and is recorded as an internal conflict case; If the other two are not in the similarity substandard range, the review is continued.

[0010] Preferably, the contract target tendency is determined based on the industry template, the evaluation value of the corresponding legal term is defined based on the contract target tendency, and the specific content of the static review unit is constructed by calculating the contract risk value, which includes: The contract target tendency includes an offensive tendency and a defensive tendency, and the offensive tendency and the defensive tendency are respectively preset with an evaluation value of a set of legal terms; The evaluation value of the corresponding legal term is selected according to the contract target tendency; The legal term mapped by the current contract is obtained, and the evaluation value of the legal term is accumulated to obtain a first risk value; The internal conflict case is obtained, the contract semantics group and its similarity are extracted, the mapped legal term of the corresponding contract semantics and the evaluation value of the legal term are found; The difference value of the similarity in the contract semantics group is calculated as a similarity difference value, the similarity difference value is multiplied by the evaluation value of the corresponding legal term, and the second risk value is accumulated to obtain a second risk value; The first risk value and the second risk value are added to obtain the contract risk value.

[0011] Preferably, the market environment fluctuation is obtained, a variable factor library is constructed according to the market environment fluctuation, and the specific content of the live review unit is constructed by generating a simulation scenario according to the variable factor library, which further includes: The variable factor library includes a macro fluctuation library and a litigation factor library; The market environment fluctuation is obtained, and financial market data, macroeconomic indicators, industry dynamics are obtained by analyzing the market environment fluctuation, and the corresponding factor type and factor element are extracted, and the correlation relationship of the factor element is recorded to obtain the macro fluctuation library; The litigation term is analyzed to obtain the litigation subject, the litigation type, the litigation node, the litigation amount and the judgment result to obtain the litigation structure database; The market environment fluctuation of the window period before and after the litigation node under different litigation types is collected, and the causal relationship between the litigation term and the market environment fluctuation is verified; The long-term impact prediction is combined with the market environment fluctuation of the window period before and after the litigation node; And the industry infection effect relationship is evaluated based on the litigation subject and the litigation type, and then a legal risk transmission path to market risk is generated; The litigation intensity factor is obtained by evaluating the structure data in the litigation structure database; The market reaction factor is obtained in combination with the market environment fluctuation of the window period before and after the litigation node under different litigation types; The litigation factor library includes the litigation intensity factor, the market reaction factor, the industry infection effect relationship, and the legal risk transmission path to market risk.

[0012] Preferably, the current contract details are obtained, and the contract details are input into the static audit unit for analysis to obtain the specific content of the contract terms and the corresponding grades thereof: The contract details are input into the static audit unit for legal term processing to obtain the corresponding legal terms; The contract details are input into the static audit unit for semantic similarity analysis to obtain the internal conflict situation; The risk value is calculated according to the corresponding legal terms and the internal conflict situation; The grade range of the preset risk value is combined to output the corresponding grade.

[0013] Preferably, the contract terms output from the static audit model are input into the simulation unit for sensitivity analysis to obtain the specific content of the risk heat map and the performance suggestion: The contract terms output from the static audit model are parsed into quantifiable parameters, and the quantifiable parameters are mapped to the element factors in the macro fluctuation library to obtain dynamic simulation driving factors; The isolated risk value is obtained by testing the influence of the variation of a single dynamic simulation driving factor on the performance risk; Based on the correlation relationship of the factor elements, the orthogonal test method is used to combine the dynamic simulation driving factor variation containing the correlation relationship, and the influence of the combined dynamic simulation driving factor variation containing the correlation relationship on the performance risk is obtained to obtain the synergistic risk value; The dynamic simulation driving factors are assigned probability distributions and subjected to random simulation dynamic simulation driving factor variation, and the influence of the random simulation dynamic simulation driving factor variation on the performance risk is obtained to obtain the random risk value; The litigation terms in the simulation process of the isolated risk value, the synergistic risk value, and the random risk value are counted; The dynamic simulation driving factors are taken as the X-axis, and the isolated risk value, the synergistic risk value, and the random risk value are taken as the Y-axis, and the different risk values are color rendered to establish a risk heat map; The litigation term occurrence is marked as a scatter point on the risk heat map, and the performance suggestion is generated in combination with the risk heat map.

[0014] A contract compliance auxiliary audit system based on a large language model, comprising: The data acquisition unit: obtain the original data by obtaining the past contract data, past litigation cases and supporting legal articles, preprocessing the original data to obtain a distribution database, the distribution database includes contract terms, litigation terms and legal terms, the contract terms include industry templates and agreement terms; The model construction unit: the semantic analysis of the industry templates and the agreement terms obtains the contract semantics, the mapping relationship between the contract semantics and the legal terms is determined, and the conflict of the contract semantics is determined, the contract target tendency is determined based on the industry template identification of the right and obligation subject, the risk value of the corresponding legal terms is defined based on the contract target tendency to construct a static audit unit, the market environment fluctuation is obtained, the variable factor library is constructed based on the market environment fluctuation, and the simulation scene is generated according to the variable factor library to construct a live audit unit; The strategy generation unit: the current contract details are obtained, the contract details are input into the static audit unit for analysis to obtain contract terms and corresponding grades, the contract terms output from the static audit model are input into the simulation unit for sensitivity analysis to obtain a risk heat map and a performance suggestion, and the contract terms and corresponding grades, the risk heat map and the performance suggestion are summarized to generate a contract compliance auxiliary audit strategy.

[0015] An electronic device includes a memory and a processor, the memory stores a computer program, and the processor calls the computer program in the memory to realize the content of the contract compliance auxiliary audit method based on the large language model.

[0016] A storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by a processor to realize the content of the contract compliance auxiliary audit method based on the large language model.

[0017] In summary, compared with the traditional technology, the contract compliance auxiliary audit method and system based on the large language model have some advantages: 1、The application integrates multi-source heterogeneous data (contracts, litigation, legal articles) to construct a structured knowledge base, accurately matches contract terms and legal specifications through semantic analysis, realizes the identification of right and obligation subjects and the quantification of risk value, and the data-driven static audit unit can quickly locate clause defects, significantly improves the contract review efficiency and compliance, and reduces the risk of manual omission; 2、The application introduces a market environment fluctuation factor library and a simulation deduction mechanism, and integrates external variables (such as price fluctuation and policy adjustment) into the risk assessment model. Through dynamic generation of simulation scenes and sensitivity testing, the risk heat map can be intuitively presented, the performance suggestion based on real-time data is provided for enterprises, and potential disputes caused by market uncertainty can be effectively dealt with; 3, The application adopts a "static audit + dynamic simulation" dual-engine collaborative architecture, which not only guarantees the legality verification of the basic clauses, but also reveals the implicit risks through quantitative analysis. The final output of the hierarchical evaluation results and the visual report form a complete closed loop from risk identification to decision optimization, helping enterprises to establish a standardized and intelligent contract management system and comprehensively improve the compliance management capability.

[0018] The technical method of the present application will be further described in detail below by means of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A step diagram of a contract compliance auxiliary auditing method based on a large language model according to the present application; Figure 2 A unit schematic diagram of a contract compliance auxiliary auditing system based on a large language model according to the present application. DETAILED DESCRIPTION

[0020] The technical method of the present application will be further described in detail below by means of the accompanying drawings and examples.

[0021] The following description of at least one example embodiment is merely exemplary in nature and is in no way intended to limit the scope of the application or its application or uses.

[0022] Techniques, systems, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered as part of the specification, where appropriate.

[0023] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Thus, other examples of example embodiments can have different values.

[0024] Unless otherwise defined, technical or scientific terms used in the present application should be interpreted as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0025] The present application provides a contract compliance auxiliary auditing method and system based on a large language model.

[0026] A contract compliance auxiliary auditing method based on a large language model, as shown in Figure 1 includes the following steps: S1, obtain past contract information, past litigation cases, and supporting legal articles to obtain raw data, preprocess the raw data to obtain a distributed database, the distributed database includes contract terms, litigation terms, and legal terms, the contract terms include industry templates and agreement terms; Massive text data is collected from enterprise historical contract library, industry template library, legal regulations database, and judicial documents network, covering various types of contract templates, supplementary agreements, litigation cases, and supporting legal articles. The original data is denoised (eliminating irrelevant format symbols), structured (extracting key elements such as parties, objects, amounts, and time limits), and standardized (labeling clause types and risk levels).

[0027] The underlying criteria for removing irrelevant format symbols from all data ensure that the denoising process does not damage the core value of the data. First, the accuracy principle avoids deleting valid symbols, such as distinguishing between commas as semantic separators and meaningless special symbols by preserving key symbols through rules; second, the consistency principle applies uniform denoising rules to the same type of data, such as preserving English punctuation or converting to Chinese punctuation for crawled data, to avoid format chaos; third, the moderation principle does not excessively remove symbols, such as < and > in code data, which are key parts of the syntax and cannot be deleted as irrelevant format symbols; fourth, the traceability principle backs up the original data before denoising, records denoising parameters, and removes symbol types for subsequent backtracking and checking.

[0028] The cleaned data is stored in a distributed database to provide high-quality corpus for subsequent model training. The core goal of this stage is to build a contract knowledge warehouse covering all business scenarios to address the bottleneck of traditional audits relying on limited experience samples.

[0029] S2, semantic analysis of industry templates and agreement terms to obtain contract semantics, determine the mapping relationship between contract semantics and legal terms, and the conflict situation of contract semantics; Based on the industry template, the rights and obligations of the subject are identified to determine the contract target tendency, and based on the contract target tendency, the risk value of the corresponding legal term is defined to build a static audit unit; Further, the specific content of S2, semantic analysis of industry templates and agreement terms to obtain contract semantics, and determining the mapping relationship between contract semantics and legal terms includes: According to the use scenarios of industry templates, scenario classification is performed to obtain a scenario category group. It can be understood that scenario classification can be divided into transaction business scenarios, service business scenarios, research and development business scenarios, and operation business scenarios according to business types.

[0030] The semantic analysis of the industry templates and protocol terms under different scene category groups obtains a triple definition containing a subject, an action, and an object, which is defined as contract semantics; It can be understood that the model selection of semantic analysis needs to consider the professionalism (legal, industry term adaptation), accuracy (structured information extraction), and scalability (multi-scene category adaptation). A general pre-training model + fine-tuning form can be used, such as a Chinese general semantic model and a lightweight text extraction model. An industry-adapted training model can also be used, such as a legal field-specific model and an industry vertical field model.

[0031] The key words of the contract semantics and legal terms are extracted to obtain contract key words and legal key words; The contract key words and legal key words are directly matched to obtain a one-layer mapping; The semantic similarity of the contract key words and legal key words is calculated, and the semantic similarity is used for semantic matching to obtain a two-layer mapping; The contract key words and legal key words are analyzed and extracted from the past litigation cases to obtain a three-layer mapping through key word matching; The one-layer mapping, two-layer mapping, and three-layer mapping constitute the mapping relationship of the contract semantics and legal terms.

[0032] Further, the specific content of the conflict of the contract semantics is; The contract semantics are obtained, and the similarity of the subject, action, and object in the triple is calculated; It can be understood that the core of semantic similarity calculation is to quantify the matching degree of two texts (such as protocol terms and industry template clauses) in the semantic level, which needs to be combined with data characteristics (structured / unstructured, short text / long text) and scene requirements (real-time, accuracy, and interpretability) to select an adaptive method.

[0033] First, the literal meaning of the text is defined: Narrow sense: similarity of the literal meaning of the text (such as “penalty for breach of contract” and “penalty for overdue”); Broad sense: similarity of the core intention / logic behind the text (such as “providing real estate as collateral” and “using real estate as collateral”).

[0034] It can be calculated by language models and cosine similarity, or it can be calculated based on semantic vectors through deep learning.

[0035] There is a preset similarity not meeting the standard range. One of the triples is determined as a reference, and the other two are determined whether they are in the similarity not meeting the standard range. The similarity not meeting the standard range can be set by the degree of refinement in the later application. In a rigorous occasion, a larger range is set, and in a less rigorous occasion, a smaller range is set.

[0036] If at least one of the other two is in the similarity substandard range, the current contract semantics is defined as a contract semantic group and is recorded as an internal conflict case; If the other two are not in the similarity substandard range, continue to review.

[0037] Further, based on the industry template, the subject of rights and obligations is identified to determine the contract target tendency, and based on the contract target tendency, the evaluation value of the corresponding legal term is defined and the contract risk value is calculated to construct the specific content of the static review unit, which includes: The contract target tendency includes offensive tendency and defensive tendency, and the offensive tendency and the defensive tendency are respectively preset with a set of evaluation values of legal terms; It can be understood that the offensive tendency refers to that when the contract subject sets the contract target, the core orientation is to actively obtain excess rights and interests, expand business boundaries, seize market advantages or achieve breakthrough results, the target setting has clear "aggressiveness" and "expansiveness", and usually accompanied by a certain willingness to bear risks, the core is to create or obtain "incremental value" through contract performance.

[0038] The defensive tendency refers to that when the contract subject sets the contract target, the core orientation is to avoid potential risks, protect core rights and interests from being damaged, maintain the stability of existing business or control the loss range, the target setting has clear "safety" and "stability", and usually gives priority to risk prevention and control, the core is to preserve "stock value" through contract performance.

[0039] Based on the industry template, the subject of rights and obligations can be identified to determine the contract target tendency set by the contract subject.

[0040] The evaluation value of the corresponding legal term is selected according to the contract target tendency; The legal term mapped by the current contract is obtained, and the evaluation value of the legal term is accumulated to obtain a first risk value; An internal conflict case is obtained, the contract semantic group and its similarity are extracted, the mapping legal term of the corresponding contract semantics and the evaluation value of the legal term are found; The difference of similarity in the contract semantic group is calculated as a similarity difference, the similarity difference is multiplied by the evaluation value of the corresponding legal term and accumulated to obtain a second risk value; The first risk value and the second risk value are added to obtain the contract risk value.

[0041] S3, according to the market environment fluctuation, a variable factor library is constructed, and a simulation scenario is generated according to the variable factor library to construct a live review unit; Further, the market environment fluctuation is obtained to construct a variable factor library according to the market environment fluctuation, and the specific content of the live review unit constructed according to the variable factor library generating simulation scenarios further includes: The variable factor library includes a macro fluctuation library and a litigation factor library, and the macro fluctuation library and the litigation factor library exist side by side.

[0042] The market environment fluctuation is obtained, and financial market data, macroeconomic indicators, industry dynamics are obtained by analyzing the market environment fluctuation, and corresponding factor types and factor elements are extracted, and the correlation of the factor elements is recorded to obtain a macro fluctuation library; It can be understood that the macro fluctuation library is a structured database formed by systematically collecting, analyzing market environment fluctuation data, refining core influencing factors and combing factor correlation, which is used to quantify the influence of market fluctuations on business (such as contract performance, risk assessment, decision making).

[0043] The market environment fluctuation is obtained, and multi-dimensional original data affecting market stability is comprehensively collected to ensure that the data source is authoritative, comprehensive and real-time. The data source can be financial market data: real-time market information of stock exchanges (such as Shanghai and Shenzhen stock exchanges, Nasdaq), interbank market transaction data, third-party financial data platforms (Wind, Bloomberg, and Tonghuashun), macroeconomic indicators: official data (GDP growth, CPI, PPI, monetary policy interest rate, etc.) released by the National Bureau of Statistics, the People's Bank of China and the Ministry of Finance, cross-border economic data of international organizations (IMF, World Bank), industry dynamic data: reports of industry associations (such as China Banking Association and Automobile Industry Association), announcements of leading enterprises, policy supervision documents (such as industry control policies and new rules), and mainstream financial media industry analysis.

[0044] The market environment fluctuation is analyzed and the factors are extracted, and the "factor type" and "factor element" with clear influence logic are disassembled from the original data to realize the structured transformation of data.

[0045] The cause-and-effect, correlation and transmission relationship between factors is combed to form a structured database that can be traced and reused.

[0046] The litigation terms are analyzed to obtain the litigation subject, litigation type, litigation node, litigation amount and judgment result to obtain a litigation structure database; It can be understood that the litigation structure database is formed by structurally analyzing public / non-public litigation terms (such as judgment documents, litigation announcements and case files), extracting core case elements and standardizing storage, which is used to quickly locate key information of the case, analyze litigation risks and trace the progress of the case.

[0047] Collect the market environment fluctuation of the window period before and after the litigation node under different litigation types and verify the causal relationship between the litigation term and the market environment fluctuation; Make long-term impact prediction combined with the market environment fluctuation of the window period before and after the litigation node; And evaluate the industry contagion effect relationship based on the litigation subject and litigation type to generate the legal risk to market risk transmission path; Evaluate the structure data in the litigation structure database to obtain the litigation intensity factor; Obtain the market reaction factor combined with the market environment fluctuation of the window period before and after the litigation node under different litigation types; The litigation factor library includes litigation intensity factor, market reaction factor, industry contagion effect relationship, and legal risk to market risk transmission path.

[0048] S4, obtain the current contract details, input the contract details into the static audit unit for analysis to obtain the contract terms and corresponding grades; Input the contract terms output by the static audit model into the simulation unit for sensitivity analysis to obtain the risk heat map and performance suggestion; It can be understood that by inputting the structured parameters of the static audit terms into the simulation unit, combined with sensitivity analysis and risk heat map, the upgrade from "text compliance check" to "business risk quantification" can be realized. This process not only identifies potential risks, but also provides implementable performance optimization solutions to help enterprises balance contract rigor and business flexibility.

[0049] Further, the specific content of obtaining the current contract details, inputting the contract details into the static audit unit for analysis to obtain the contract terms and corresponding grades is: Input the contract details into the static audit unit for legal term processing to obtain the corresponding legal terms, such as extracting legal information in a large language model to obtain the corresponding legal terms.

[0050] Input the contract details into the static audit unit for semantic similarity analysis to obtain the internal conflict situation; Calculate the risk value according to the corresponding legal terms and internal conflict situation; There is a pre-set grade range of risk value, and the corresponding grade is output combined with the grade range.

[0051] Further, the specific content of inputting the contract terms output by the static audit model into the simulation unit for sensitivity analysis to obtain the risk heat map and performance suggestion is: Parse the contract terms output by the static audit model into quantifiable parameters, and map the quantifiable parameters to the element factors in the macro fluctuation library to obtain dynamic simulation driving factors; It can be understood that the structured contract terms output by the static audit model, such as the rights and obligations, the performance time node, the breach of contract responsibility, etc., are the core input parameters of the simulation unit. For example, if the contract stipulates that "the buyer needs to pay the balance within 30 days after the goods arrive", "30 days" is extracted as the payment period variable, and is bound with the associated parameters such as "supply chain time efficiency" and "funds turnover rate" in the variable factor library to form a set of simulation parameters that can be dynamically adjusted.

[0052] The isolated test single dynamic simulation driving factor variation influences the performance risk to obtain an isolated risk value; Based on the correlation between the factor elements, the orthogonal test method is used to combine the dynamic simulation driving factor variation containing the correlation, and then the combined dynamic simulation driving factor variation containing the correlation influences the performance risk to obtain a synergistic risk value; The dynamic simulation driving factor is assigned a probability distribution and a random simulation dynamic simulation driving factor variation is performed, and then the random simulation dynamic simulation driving factor variation influences the performance risk to obtain a random risk value; To realize the dynamic simulation of the target system, first, based on historical data and mechanism analysis, each driving factor that influences the system state is assigned a probability distribution that conforms to its actual fluctuation characteristics; then, through a random simulation method (such as Monte Carlo simulation, Latin hypercube sampling), the dynamic variation trajectory of the driving factor is generated, providing input data that conforms to the real scene for subsequent system simulation.

[0053] Statistical isolated risk value, synergistic risk value, litigation item in the simulation process; Taking the dynamic simulation driving factor as the X-axis and the isolated risk value, the synergistic risk value, and the random risk value as the Y-axis, the different risk values are color rendered to establish a risk heat map; The litigation item occurrence is marked as a scatter point on the risk heat map, and the performance suggestion is generated in combination with the risk heat map.

[0054] S5, the contract terms and their corresponding grades, the risk heat map, and the performance suggestion are summarized to generate a contract compliance auxiliary audit strategy.

[0055] A contract compliance auxiliary audit system based on a large language model, as shown in Figure 2 , includes: A data acquisition unit: obtain past contract data, past litigation cases, and supporting legal articles to obtain original data, and preprocess the original data to obtain a distribution database, the distribution database including contract items, litigation items, and legal items, the contract items including industry templates and agreement items; The model construction unit: semantic analysis is performed on the industry template and the protocol term to obtain contract semantics, a mapping relationship between the contract semantics and the legal term is determined, and a conflict of the contract semantics is determined, a contract target tendency is determined based on the industry template and the identification of the right and obligation subject, a risk value of the corresponding legal term is defined based on the contract target tendency to construct a static audit unit, market environment fluctuation is obtained, a variable factor library is constructed based on the market environment fluctuation, a simulation scenario is generated according to the variable factor library to construct a live audit unit; The strategy generation unit: obtain the contract details, input the contract details into the static audit unit to obtain the contract clauses and the corresponding grades, input the contract clauses output from the static audit model into the simulation unit to obtain the risk heat map and the performance suggestion, and summarize the contract clauses, the corresponding grades, the risk heat map and the performance suggestion to generate the contract compliance auxiliary audit strategy.

[0056] An electronic device includes a memory and a processor, the memory stores a computer program, and the processor calls the computer program in the memory to realize the content of the contract compliance auxiliary audit method based on a large language model.

[0057] A storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by a processor to realize the content of the contract compliance auxiliary audit method based on a large language model.

[0058] Finally, it should be noted that: the above embodiments are only used to illustrate the technical method of the present application, but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or replace the technical method of the present application, and these modifications or replacements cannot make the modified technical method deviate from the spirit and scope of the technical method of the present application.

Claims

1. A contract compliance auxiliary review method based on a large language model, characterized in that, Includes the following steps: S1. Obtain original data by acquiring past contract information, past litigation cases, and supporting legal provisions. Preprocess the original data to obtain a distributed database. The distributed database includes contract terms, litigation terms, and legal terms. Contract terms include industry templates and agreement terms. S2. Perform semantic analysis on industry templates and agreement terms to obtain contract semantics, determine the mapping relationship between contract semantics and legal terms, and identify any conflicts in contract semantics. Based on industry templates, identify the rights and obligations of the parties involved and determine the contractual objective tendencies. Based on the contractual objective tendencies, define the risk value of the corresponding legal terms and construct a static review unit. S3. Obtain market environment fluctuations. Construct a variable factor library based on market environment fluctuations, and generate simulated scenarios based on the variable factor library to construct a real-time audit unit. S4. Obtain the current contract details, input the contract details into the static audit unit for analysis to obtain the contract terms and their corresponding levels; The contract terms output from the static audit model are input into the simulation unit for sensitivity analysis to obtain a risk heat map and performance recommendations. S5. Summarize the contract terms and their corresponding levels, risk heatmaps, and performance recommendations to generate a contract compliance auxiliary audit strategy.

2. The contract compliance auxiliary review method based on a large language model according to claim 1, characterized in that, S2 performs semantic analysis on industry templates and agreement terms to obtain contract semantics. The specific content of determining the mapping relationship between contract semantics and legal terms includes: Based on the usage scenarios of industry templates, scenario category groups are obtained by dividing the scenarios into groups. Semantic analysis of industry templates and agreement terms under different scenario category groups yields a triplet containing subject-behavior-object, which is defined as contract semantics. Keyword extraction from contract semantics and legal terms yields contract keywords and legal keywords; A mapping layer is obtained by directly matching contract keywords and legal keywords; Calculate the semantic similarity between contract keywords and legal keywords, and obtain a two-level mapping based on the semantic similarity. Three-layer mapping was obtained by analyzing and extracting contract and legal keywords from past litigation cases; The first-level mapping, second-level mapping, and third-level mapping constitute the mapping relationship between contract semantics and legal terms.

3. The contract compliance auxiliary review method based on a large language model according to claim 2, characterized in that, The specific details of the contract semantic conflict are as follows: Obtain the contract semantics and calculate the similarity of the subject, behavior, and object in the triples respectively; There is a pre-defined range of similarity that does not meet the standard. One of the triples is selected as a reference to determine whether the other two are within the range of similarity that does not meet the standard. If at least one of the other two is within the range of similarity not meeting the standard, then the current contract semantics is defined as the contract semantic group and recorded as an internal conflict situation; If the other two are not within the scope of non-similarity, the review will continue.

4. The contract compliance auxiliary review method based on a large language model according to claim 3, characterized in that, The specific content of constructing a static review unit based on industry templates to identify the rights and obligations of the parties involved, determine the contractual objective orientation, define the evaluation value of corresponding legal terms based on the contractual objective orientation, and calculate the contract risk value includes: The contractual objective tendencies include offensive and defensive tendencies, each with a pre-defined set of legal terminology evaluation values. The evaluation value of the corresponding legal terms selected based on the contract's target orientation is obtained; Obtain the legal terms mapped to the current contract and calculate the risk value by summing the evaluation values ​​of the legal terms; Obtain information on internal conflicts, extract contract semantic groups and their similarities, and find the corresponding legal terms and their evaluation values. The similarity difference is calculated by multiplying the similarity difference by the evaluation value of the corresponding legal term and summing them up to obtain the secondary risk value. The contract risk value is obtained by adding the primary risk value and the secondary risk value.

5. The contract compliance auxiliary review method based on a large language model according to claim 4, characterized in that, Based on the market environment fluctuation data, a variable factor library is constructed. Simulated scenarios are generated from this library to construct a real-time audit unit. This also includes: The variable factor library includes a macroeconomic fluctuation library and a litigation factor library; The system acquires information on market fluctuations, analyzes these fluctuations to obtain financial market data, macroeconomic indicators, and industry dynamics, extracts corresponding factor types and factor elements, and records the relationships between factor elements to create a macroeconomic volatility database. The litigation terminology is parsed to obtain the parties involved in the litigation, the type of litigation, the litigation milestones, the amount in dispute, and the judgment results, thus creating a litigation structure database; Collect data on market environment fluctuations before and after litigation milestones for different types of litigation, and examine the causal relationship between litigation terms and market environment fluctuations. Long-term impact predictions are made by considering market environment fluctuations before and after litigation milestones. Based on the parties involved in the litigation and the type of litigation, the industry contagion effect is assessed to generate the transmission path of legal risks to market risks; The litigation intensity factor is obtained by evaluating the structural data in the litigation structure database; Market reaction factors are derived by combining market environment fluctuations before and after litigation milestones for different types of litigation. The litigation factor library includes litigation intensity factors, market reaction factors, industry contagion effect relationships, and the transmission path of legal risks to market risks.

6. The contract compliance auxiliary review method based on a large language model according to claim 5, characterized in that, Retrieve the current contract details, input the contract details into the static audit unit for analysis, and obtain the specific content of the contract terms and their corresponding levels as follows: The contract details are input into the static review unit to process legal terms and obtain the corresponding legal terms. The contract details are input into the static review unit for semantic similarity analysis to obtain information on internal conflicts. The risk value is calculated based on the corresponding legal terms and internal conflict information. It presets a risk value range and outputs the corresponding level based on the range.

7. The contract compliance auxiliary review method based on a large language model according to claim 6, characterized in that, The contract terms output from the static audit model are input into the simulation unit for sensitivity analysis to obtain a risk heatmap and specific performance recommendations, as follows: The contract terms output by the static audit model are parsed into quantifiable parameters, and the quantifiable parameters are mapped to the element factors in the macro volatility library to obtain dynamic simulation driving factors. The isolated risk value is obtained by testing the impact of a single dynamic simulation driver factor change on performance risk in isolation. Based on the correlation between factor elements, orthogonal experimental design is used to combine dynamic simulation driving factor changes containing correlation, thereby obtaining the impact of the combined dynamic simulation driving factor changes containing correlation on performance risk and thus obtaining the collaborative risk value. Assign probability distributions to dynamic simulation driving factors and perform random simulation of dynamic simulation driving factor changes to obtain the impact of random simulation of dynamic simulation driving factor changes on performance risk and thus obtain random risk value. Litigation terms used in the simulation of isolated risk values, collaborative risk values, and stochastic risk values; Using the dynamic simulation driving factor as the X-axis, and the isolated risk value, collaborative risk value, and random risk value as different Y-axis, a risk heatmap is created by color rendering for different risk values. The occurrence of litigation terms is marked as scatter points on a risk heatmap, and performance suggestions are generated based on the risk heatmap.

8. A contract compliance auxiliary review system based on a large language model, characterized in that, include: Data acquisition unit: Obtains raw data by acquiring past contract information, past litigation cases, and supporting legal provisions; preprocesses the raw data to obtain a distributed database, which includes contract terms, litigation terms, and legal terms. Contract terms include industry templates and agreement terms. Model building unit: Semantic analysis of industry templates and agreement terms to obtain contract semantics, determine the mapping relationship between contract semantics and legal terms and the conflict of contract semantics, identify the rights and obligations subjects based on industry templates to determine the contract objective tendency, define the risk value of corresponding legal terms based on the contract objective tendency to build a static review unit, obtain market environment fluctuations, build a variable factor library based on market environment fluctuations, and generate simulated scenarios based on the variable factor library to build a real-world review unit; Strategy Generation Unit: Obtain current contract details, input the contract details into the static audit unit for analysis to obtain contract terms and their corresponding levels, input the contract terms output from the static audit model into the simulation unit for sensitivity analysis to obtain risk heatmaps and performance suggestions, and summarize the contract terms and their corresponding levels, risk heatmaps and performance suggestions to generate contract compliance auxiliary audit strategies.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor, when calling the computer program in the memory, implements the content of the contract compliance auxiliary review method based on any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the content of the contract compliance auxiliary review method based on a large language model as described in any one of claims 1 to 7.

Images