Automatic Review Method for Procurement and Sales Contracts Based on Artificial Intelligence Large Model

Abstract

This invention relates to the field of artificial intelligence technology and discloses an automated review method for procurement and sales contracts based on a large-scale AI model. The method includes: first, based on the large-scale AI model, performing context-aware and deep semantic deconstruction on the target contract text to generate multi-level semantic expressions; then, automatically matching these expressions with a preset three-dimensional compliance risk map to obtain contextualized risk diagnosis results. Subsequently, combining the original contract semantic context, the large-scale model reconstructs the clauses to form optional revision schemes, determines the final decision through human-computer interaction, and outputs the finalized contract through global logic self-checking. Finally, the multi-level semantic expressions, risk diagnosis results, final decision, and finalized text are used as historical cases for targeted feedback optimization of the model, forming an adaptive review strategy and knowledge system. This invention can improve the efficiency of automated review of procurement and sales contracts.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an automatic review method for procurement and sales contracts based on a large artificial intelligence model. Background Technology

[0002] Existing procurement and sales contract review technologies have significant limitations in semantic processing, making it difficult to achieve deep contextual perception and multi-level semantic deconstruction of contract texts. Traditional methods often rely on fixed keyword matching or simple rule engines, which can only extract surface information from contract clauses and cannot accurately capture the logical connections and dependencies between clauses. This results in an incomplete understanding of semantic expressions in complex transaction scenarios, which in turn affects the completeness and accuracy of risk identification.

[0003] In the risk diagnosis and clause revision stages, existing technologies lack a systematic contextual matching and dynamic calibration mechanism. Traditional risk maps are simplistic and fail to effectively integrate multi-dimensional information such as transaction objectives, performance constraints, and object parameters. Furthermore, risk level determination is mostly based on static threshold judgments, without contextual assessment combined with the original semantic context of the contract, which can easily lead to misjudgment or level deviation. At the same time, the generation of clause revision schemes lacks logical coherence and multi-strategy optimization, and the revised content may be disconnected from the overall semantics of the contract. In addition, the human-computer interaction is not sufficiently collaborative, and user decisions are difficult to integrate efficiently into the review process, resulting in low review efficiency and poor practicality of revision schemes. This fails to meet the requirements of accuracy and efficiency in the review of complex contracts. Therefore, how to improve the efficiency of automatic review of procurement and sales contracts has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides an automatic review method for procurement and sales contracts based on a large artificial intelligence model to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an automatic review method for procurement and sales contracts based on a large-scale artificial intelligence model, comprising:

[0006] S1. Based on a pre-set large-scale artificial intelligence model, the target contract text is subjected to context perception, and the perception results are subjected to deep semantic deconstruction to obtain the multi-level semantic expression of the target contract text. The large-scale artificial intelligence model includes semantic association rules, domain knowledge structure, common sense logic rules in the contract domain, and a historical dispute case library.

[0007] S2. Perform automated context matching between the multi-level semantic expression and the preset three-dimensional compliance risk map to obtain the contextualized risk diagnosis result of the target contract text;

[0008] S3. Based on the logical judgment rules of the aforementioned artificial intelligence big data model, the contextualized risk diagnosis results and the original contract semantic context in the target contract text are reconstructed using multiple clause schemes to obtain the optional revision schemes of the target contract text.

[0009] S4. Perform human-computer interaction processing on the contextualized risk diagnosis results and the optional revision schemes, and receive the user's decision instructions to obtain the user's final decision result;

[0010] S5. Perform a global logical self-check on the final decision result and calibrate the language standardization in the self-check result to obtain the final draft contract text of the target contract text.

[0011] S6. Using the multi-level semantic expression, the contextualized risk diagnosis result, the final decision result, and the finalized contract text as historical review case data of the target contract text, perform targeted feedback optimization on the artificial intelligence big model to obtain the adaptive review strategy and knowledge system of the target contract text.

[0012] In a preferred embodiment, the step of performing context awareness on the target contract text based on a preset large-scale artificial intelligence model, and performing deep semantic deconstruction on the perception results to obtain a multi-level semantic expression of the target contract text, including:

[0013] Based on the semantic association rules in the pre-set artificial intelligence big data model, the logical connections and dependencies of the target contract text are perceived in context to obtain the perception result of the target contract text.

[0014] Based on the knowledge structure in the aforementioned large-scale artificial intelligence model, the perception results are subjected to deep semantic deconstruction to obtain the semantic components of the target contract text;

[0015] The semantic components are decomposed hierarchically to obtain the multi-level semantic expression of the target contract text.

[0016] In a preferred embodiment, the step of automatically matching the multi-level semantic expression with a preset three-dimensional compliance risk map to obtain the contextualized risk diagnosis result of the target contract text includes:

[0017] Extract the core transaction objectives, performance constraints, and specific transaction object parameters from the multi-level semantic expression;

[0018] The core transaction objectives, the performance constraints, and the specific transaction object parameters are respectively mapped to the transaction type dimension, obligation fulfillment dimension, and object attribute dimension of the preset three-dimensional compliance risk map;

[0019] Under the transaction type dimension, risk pattern identification is performed on the core transaction objective to obtain the basic risk list of the first dimension in the target contract text;

[0020] Under the aforementioned obligation fulfillment dimension, the performance constraints are analyzed to obtain a list of conditions and risks in the second dimension of the target contract text;

[0021] Under the object attribute dimension, the historical dispute case library in the artificial intelligence big model is compared with the specific transaction object parameters to obtain the object risk list of the third dimension in the target contract text;

[0022] Cross-correlation analysis is performed on the basic risk list, the conditional risk list, and the object risk list to obtain the initial risk diagnosis results of the target contract text;

[0023] The initial risk diagnosis results are dynamically calibrated to obtain the contextualized risk diagnosis results of the target contract text.

[0024] In a preferred embodiment, the step of dynamically calibrating the initial risk diagnosis result to obtain the contextualized risk diagnosis result of the target contract text includes:

[0025] Identify risk items in the initial risk diagnosis results that exceed a preset risk threshold;

[0026] Based on the three-dimensional compliance risk map, the risk items are traced back in reverse to obtain the matching path of the risk items;

[0027] Based on the matching path, extract the original semantic context associated with the risk item from the multi-level semantic expression;

[0028] Based on the original semantic context, the risk item is evaluated in a scenario. When the description level of the risk item is too high, the risk level of the risk item is lowered according to the limiting information in the original semantic context to obtain the calibrated risk item.

[0029] The calibrated risk items are used as the contextualized risk diagnosis results of the target contract text.

[0030] In a preferred embodiment, the logical judgment rules based on the artificial intelligence big data model are used to reconstruct multiple clauses of the contextualized risk diagnosis results and the original contract semantic context in the target contract text, resulting in optional revision schemes for the target contract text, including:

[0031] The logical judgment rules in the aforementioned large-scale artificial intelligence model are analyzed by type to obtain the risk attribution and causation rules, multi-strategy generation rules, and semantic coherence embedding rules of the large-scale artificial intelligence model.

[0032] Based on the aforementioned risk attribution and causation rules, the contextualized risk diagnosis results are analyzed item by item to obtain the root cause clauses and impact transmission paths of the contextualized risk diagnosis results.

[0033] Based on the multi-strategy generation rules, the root cause clauses, and the impact transmission path, a revision strategy guide for the target contract text is generated;

[0034] By applying the semantic coherence embedding rules and combining them with the original contract semantic context in the target contract text, the revision strategy guidance is adjusted collaboratively to obtain the draft of the revised clauses of the target contract text.

[0035] The draft amendments are comprehensively evaluated to obtain the preferred draft amendments to the target contract text;

[0036] Modal construction is performed on the preferred draft amendment to obtain the optional amendment schemes for the target contract text.

[0037] In a preferred embodiment, the step of performing human-computer interaction processing on the contextualized risk diagnosis results and the optional revision scheme, and receiving the user's decision instructions to obtain the user's final decision result, includes:

[0038] Based on the risk diagnosis sub-items in the contextualized risk diagnosis results, the contextualized risk diagnosis results are associated and mapped with the optional revision schemes to obtain the interactive decision list of the target contract text;

[0039] Based on the interactive decision list, the risk diagnosis sub-items are displayed synchronously to obtain the display result of the target contract text;

[0040] Respond to the user's operation on the interactive decision list and receive the user's decision instructions on the risk diagnosis sub-item;

[0041] The decision instruction is logically consistent to obtain the user's verified decision instruction.

[0042] The verified decision instruction is coupled with the contextualized risk diagnosis result to obtain the user's final decision result.

[0043] In a preferred embodiment, the step of performing a global logical self-check on the final decision result and calibrating the language standardization in the self-check result to obtain the final draft contract text of the target contract text includes:

[0044] By performing cross-logical relationship analysis on the clause revision records in the final decision result, potential logical conflicts in the target contract text can be obtained.

[0045] Based on the common-sense logical rules of the contract domain in the aforementioned large-scale artificial intelligence model, the potential logical conflicts are corrected to obtain the logically revised text of the target contract text;

[0046] The logically revised text is subjected to language standardization processing to obtain the standard text of the target contract text;

[0047] The format of the standard text is integrated to obtain the final draft of the target contract text.

[0048] In a preferred embodiment, the step of using the multi-level semantic expression, the contextualized risk diagnosis result, the final decision result, and the finalized contract text as historical review case data for the target contract text includes:

[0049] By associating and integrating the multi-level semantic expressions, the contextualized risk diagnosis results, the final decision results, and the finalized contract text, a structured case record of the target contract text is obtained.

[0050] The contract type identifier, risk type set, revision strategy type, and the difference identifier between the user's final decision and the initial diagnosis result in the structured case record are used as the key feature data of the structured case record;

[0051] Tensor synthesis is performed on the key feature data to obtain the feature vector of the structured case record, and a timestamp and weight identifier are added to the feature vector to obtain the weight time-series vector of the structured case record;

[0052] The weighted time-series vector is stored in the historical review case library of the target contract text, and an index relationship is established between the stored historical review case library and the original contract text in the target contract text to construct the historical review case data of the target contract text.

[0053] In a preferred embodiment, the step of performing targeted feedback optimization on the large-scale artificial intelligence model to obtain the adaptive review strategy and knowledge system for the target contract text includes:

[0054] Obtain the case vector set from the historical review case library in the target contract text;

[0055] Based on the case vector set, calculate the consistency feedback coefficient between the user's final decision and the initial diagnostic results in the large artificial intelligence model;

[0056] Based on the consistency feedback coefficient, the parameters of the risk judgment and strategy generation logic of the large artificial intelligence model are updated to obtain the optimized model of the large artificial intelligence model.

[0057] The logical judgment rules of the optimization model are integrated and encapsulated with the historical review case library to obtain the adaptive review strategy and knowledge system for the target contract text.

[0058] In a preferred embodiment, the formula for calculating the consistency feedback coefficient is:

[0059] ;

[0060] in, This represents the consistency feedback coefficient. This represents the total number of cases in the case vector set. This represents the total number of risk types in the contextualized risk diagnosis results. Indicates the first The weighting coefficients of each case. This represents the preset time decay constant. Indicates the first The time decay amount in each case Indicates the first Risk dimension weights for each risk type Indicates the first In the case of the first Consistency in user decisions across different risk types Indicates the first In the case of the first Identifiers for the diagnostic accuracy of each risk type This represents an exponential function.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. This invention uses a pre-set artificial intelligence model to perform context awareness and deep semantic deconstruction on the target contract text, generating multi-level semantic expressions. Then, it combines a three-dimensional compliance risk map for automated context matching to complete contextual risk diagnosis, effectively improving the comprehensiveness and accuracy of contract risk identification and ensuring that risk items accurately match the actual needs of the contract scenario.

[0063] 2. This invention reconstructs multiple contract clauses using logical judgment rules from a large-scale artificial intelligence model, ensuring the consistency between the revised clauses and the semantic context of the original contract. Through global logical self-checking and language standardization calibration, it enhances the logical rigor and standardized expression of the final contract text. Simultaneously, targeted feedback optimization from historical review case data drives adaptive iteration of review strategies and knowledge systems, continuously improving the efficiency and reliability of automated contract review. Attached Figure Description

[0064] Figure 1 A flowchart illustrating an automatic review method for procurement and sales contracts based on a large artificial intelligence model, provided in an embodiment of the present invention;

[0065] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0066] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0067] This application provides an automatic review method for procurement and sales contracts based on an artificial intelligence large-scale model. The executing entity of this automatic review method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the automatic review method for procurement and sales contracts based on an artificial intelligence large-scale model can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0068] Reference Figure 1 The diagram shown is a flowchart illustrating an automatic review method for procurement and sales contracts based on a large-scale artificial intelligence model, according to an embodiment of the present invention. In this embodiment, the automatic review method for procurement and sales contracts based on a large-scale artificial intelligence model includes:

[0069] S1. Based on a pre-set large-scale artificial intelligence model, the target contract text is subjected to context perception, and the perception results are subjected to deep semantic deconstruction to obtain the multi-level semantic expression of the target contract text. The large-scale artificial intelligence model includes semantic association rules, domain knowledge structure, common sense logic rules in the contract domain, and a historical dispute case library.

[0070] In this embodiment of the invention, the step of performing context awareness on the target contract text based on a preset large-scale artificial intelligence model, and performing deep semantic deconstruction on the awareness results to obtain a multi-level semantic expression of the target contract text, includes:

[0071] Based on the semantic association rules in the pre-set artificial intelligence big data model, the logical connections and dependencies of the target contract text are perceived in context to obtain the perception result of the target contract text.

[0072] Based on the knowledge structure in the aforementioned large-scale artificial intelligence model, the perception results are subjected to deep semantic deconstruction to obtain the semantic components of the target contract text;

[0073] The semantic components are decomposed hierarchically to obtain the multi-level semantic expression of the target contract text.

[0074] The AI ​​big data model is built upon massive amounts of textual data, industry practice information, and historical contract cases in the field of procurement and sales contracts. It is constructed through processes such as text annotation, feature extraction, and model training. The model contains core components such as semantic association rules, domain knowledge structure, common-sense logic rules in the contract domain, and a historical dispute case database. Semantic association rules are used to identify the logical connections and dependencies in contract texts, the domain knowledge structure provides professional knowledge support related to contracts, common-sense logic rules in the contract domain ensure the logical rationality of contract clauses, and the historical dispute case database provides a reference for risk identification.

[0075] The pre-defined AI-powered large-scale model is a professional contract review model built through a specific process based on relevant data resources in the field of procurement and sales contracts. The data resources upon which it is built include massive amounts of textual data, industry practice materials, and historical contract cases in the procurement and sales contract field. The construction process involves key steps such as text annotation, feature extraction, and model training, ultimately forming a complete model containing core components such as semantic association rules, domain knowledge structure, common-sense logic rules in the contract domain, and a historical dispute case database. Semantic association rules are used to identify the logical connections and dependencies in contract texts, providing a basis for contextual awareness and helping to generate the perceived results of the target contract text. The domain knowledge structure covers core contract elements, common clause types, and industry practices, supporting deep semantic deconstruction of the perceived results to obtain semantic components. Common-sense logic rules in the contract domain ensure the logical rationality of contract clauses, correcting potential logical conflicts and forming logically revised texts. The historical dispute case database contains a large number of dispute cases arising from the transaction object, providing a reference for generating an object risk list.

[0076] The training data specifically includes over 1 million procurement and sales contract texts, covering various transaction types such as raw material procurement, finished product sales, and technical service procurement, and encompassing contract scenarios of enterprises of different industries and sizes; over 500,000 pieces of industry practice data, including common practices and normative requirements in various industries for contract signing, performance, and dispute resolution; and over 300,000 historical contract dispute cases, each with detailed records of transaction object parameters, causes of disputes, risk manifestations, and handling results, providing rich samples for the model to learn risk identification and revision strategies.

[0077] The verification effect is mainly reflected in the various outputs generated by the model. The model can accurately perform context awareness and deep semantic deconstruction of the target contract text, generating complete and clearly hierarchical multi-level semantic expressions, ensuring that no semantic components are omitted and the hierarchical division is reasonable. After automatically matching the multi-level semantic expressions with the three-dimensional compliance risk map, the generated contextualized risk diagnosis results can comprehensively cover various risks in the dimensions of transaction type, obligation performance, and object attribute. After dynamic calibration, the risk level judgment is accurate, with no risk misjudgment or level deviation. The optional revision schemes generated based on logical judgment rules can accurately address risk issues in the root cause clauses and are consistent with the original contract. Maintaining high semantic coherence and possessing strong practical operability, the multi-modal solutions can meet the needs of different review scenarios. The finalized contract text, generated after human-computer interaction processing, global logic self-checking, and language standardization calibration, is free of logical conflicts and linguistic ambiguities, and complies with legal regulations and industry practices. At the same time, through targeted feedback optimization of historical review case data, the model's risk assessment and strategy generation logic are continuously improved, and the adaptive review strategy and knowledge system are constantly enriched, resulting in a continuous improvement in the accuracy and efficiency of subsequent contract reviews. The risk identification accuracy rate reaches over 95%, the contract review efficiency is improved by over 80% compared to traditional methods, and the user adoption rate of the revised solution reaches over 90%.

[0078] The semantic association rules in the large-scale AI model are pre-constructed by combining professional knowledge in the field of procurement and sales contracts, common logical relationships of clauses, and language expression norms. They include association patterns such as causal relationships, conditional relationships, and parallel relationships between different clauses in the contract. When processing the target contract text, the model scans the contract text sentence by sentence, identifies key terms, clause identifiers, and conjunctions in the text, and clarifies the logical connections between clauses through semantic association rule comparison. For example, the sequential dependency relationship between payment clauses and delivery clauses, and the causal relationship between breach of contract clauses and contract performance clauses. At the same time, it sorts out the dependency relationships between different chapters and different clauses, and integrates these identified and sorted logical association information to form the perceived result of the target contract text.

[0079] The knowledge structure of the AI ​​big data model is composed of domain knowledge related to procurement and sales contracts, covering the core elements of the contract, common clause types, industry practices, etc., organized according to a certain logical framework. After obtaining the perception results, the model will, based on its own knowledge structure, decompose and analyze the logically related information in the perception results, and extract the key parts that constitute the semantics of the contract one by one, including the contract subject information, the description of the transaction object, the agreement on rights and obligations, the performance period, the settlement method, the liability for breach of contract, etc. These extracted key parts are the semantic components of the target contract text.

[0080] After obtaining the semantic components, the contract text is decomposed according to its logical hierarchy and importance. First, a core semantic layer and an auxiliary semantic layer are distinguished. The core semantic layer contains the core transaction information of the contract, such as the subject matter of the transaction, the transaction amount, and the key semantic components of fulfilling core obligations. The auxiliary semantic layer contains content that provides supplementary explanations for the core semantics, such as details of the performance method and the specific procedures for dispute resolution. Then, the core semantic layer and the auxiliary semantic layer are further subdivided. The core semantic layer is divided into the transaction agreement layer, the performance requirement layer, and the liability division layer according to the transaction process sequence. The auxiliary semantic layer is divided into the explanation layer and the supplementary agreement layer according to function. Through this hierarchical division, each semantic component is mapped to a different level, ultimately forming a multi-level semantic expression of the target contract text.

[0081] The beneficial effects are that by clarifying the foundation and core components of the large-scale artificial intelligence model, the target contract text can be accurately and step-by-step completed with context perception, deep semantic deconstruction and hierarchical decomposition. The logical connections and semantic composition of the contract text are clearly presented, and multi-level semantic expressions are accurately obtained. This provides a comprehensive and accurate semantic foundation for subsequent contract risk diagnosis, clause revision and other processes, and effectively improves the accuracy and efficiency of automatic review of procurement and sales contracts.

[0082] S2. Perform automated context matching between the multi-level semantic expression and the preset three-dimensional compliance risk map to obtain the contextualized risk diagnosis result of the target contract text;

[0083] In this embodiment of the invention, the step of automatically matching the multi-level semantic expression with a preset three-dimensional compliance risk map to obtain the contextualized risk diagnosis result of the target contract text includes:

[0084] Extract the core transaction objectives, performance constraints, and specific transaction object parameters from the multi-level semantic expression;

[0085] The core transaction objectives, the performance constraints, and the specific transaction object parameters are respectively mapped to the transaction type dimension, obligation fulfillment dimension, and object attribute dimension of the preset three-dimensional compliance risk map;

[0086] Under the transaction type dimension, risk pattern identification is performed on the core transaction objective to obtain the basic risk list of the first dimension in the target contract text;

[0087] Under the aforementioned obligation fulfillment dimension, the performance constraints are analyzed to obtain a list of conditions and risks in the second dimension of the target contract text;

[0088] Under the object attribute dimension, the historical dispute case library in the artificial intelligence big model is compared with the specific transaction object parameters to obtain the object risk list of the third dimension in the target contract text;

[0089] Cross-correlation analysis is performed on the basic risk list, the conditional risk list, and the object risk list to obtain the initial risk diagnosis results of the target contract text;

[0090] The initial risk diagnosis results are dynamically calibrated to obtain the contextualized risk diagnosis results of the target contract text.

[0091] The dynamic calibration of the initial risk diagnosis results to obtain the contextualized risk diagnosis results of the target contract text includes:

[0092] Identify risk items in the initial risk diagnosis results that exceed a preset risk threshold;

[0093] Based on the three-dimensional compliance risk map, the risk items are traced back in reverse to obtain the matching path of the risk items;

[0094] Based on the matching path, extract the original semantic context associated with the risk item from the multi-level semantic expression;

[0095] Based on the original semantic context, the risk item is evaluated in a scenario. When the description level of the risk item is too high, the risk level of the risk item is lowered according to the limiting information in the original semantic context to obtain the calibrated risk item.

[0096] The calibrated risk items are used as the contextualized risk diagnosis results of the target contract text.

[0097] The core transaction objectives, which embody the core transaction intent of the contract, are selected from multi-level semantic expressions. These objectives clearly define the main purpose and core demands of both parties in conducting the transaction. At the same time, the performance constraints that restrict and regulate the performance behavior in the contract are extracted. These constraints cover key requirements such as the performance period, performance location, performance method, and quality standards. Furthermore, the specific transaction object parameters related to the transaction are accurately extracted, including detailed information such as the name, specifications, quantity, performance indicators, and technical parameters of the transaction object. This ensures that the core transaction objectives, performance constraints, and specific transaction object parameters are all complete and accurately correspond to the actual content of the contract.

[0098] The pre-defined three-dimensional compliance risk map includes transaction type dimension, obligation fulfillment dimension, and object attribute dimension. The transaction type dimension covers various procurement and sales-related transaction models and corresponding risk characteristics. The obligation fulfillment dimension focuses on various compliance requirements and risk points in the process of fulfilling the obligations of both parties to the contract. The object attribute dimension classifies the inherent attributes and potential risks of different transaction objects. The extracted core transaction objectives are mapped to the transaction type dimension, the performance constraints are mapped to the obligation fulfillment dimension, and the specific transaction object parameters are mapped to the object attribute dimension, ensuring that each extracted piece of information can be accurately matched to the corresponding dimension of the three-dimensional compliance risk map.

[0099] Under the dimension of transaction type, based on the risk feature library corresponding to various transaction modes preset in this dimension, a comprehensive comparative analysis is conducted on the core transaction target to identify the transaction mode to which the core transaction target belongs, and then the common risk types, risk manifestations and risk impact ranges under the transaction mode are determined. These identified risk information are systematically organized to form the basic risk list of the first dimension in the target contract text. The list clearly defines the name, risk description and transaction mode to which the risk belongs for each risk.

[0100] Under the obligation performance dimension, the performance constraints mapped to this dimension are broken down and analyzed item by item. Each performance constraint is checked to see if it conforms to relevant laws and regulations, industry practices, and the reasonable expectations of both parties to the contract. It is determined whether there are problems such as unclear constraints, unreasonable constraint standards, or incomplete scope of constraints. For each problem found, the corresponding risk point is identified. These risk points are organized in a unified format to form the condition risk list of the second dimension in the target contract text. The list records in detail the performance constraint item corresponding to the risk, the specific manifestation of the risk, and the cause of the risk.

[0101] In terms of object attributes, the historical dispute case library in the AI ​​big data model is retrieved. This library contains a large number of historical dispute cases related to procurement and sales contracts, and each case includes detailed transaction object parameters and corresponding dispute details. The extracted specific transaction object parameters are compared one by one with the transaction object parameters of each case in the library. The overlap and similarity of the two in various parameters are analyzed, and historical dispute cases with high similarity to the current specific transaction object parameters are screened out. Based on the dispute types, risk manifestations and impact consequences in these cases, the risks that the current contract may face in terms of object attributes are determined, forming a third-dimensional object risk list in the target contract text. The list clearly indicates the historical case references corresponding to the risks, the specific content of the risks, and the possible consequences of the risks.

[0102] All risk items in the basic risk list, conditional risk list, and object risk list are summarized, and the relationships between risk items in different lists are analyzed to determine whether there are multiple risk items caused by the same reason, whether there is a causal relationship or mutual influence between different risk items, and related risk items are classified and integrated. At the same time, the risk of duplicate records is checked, redundant information is eliminated, and the initial risk diagnosis results of the target contract text are formed. The results comprehensively cover all risks in the three dimensions and clearly present the relationship between risks.

[0103] Each risk item in the initial risk diagnosis results is judged by referring to the preset risk threshold. The preset risk threshold is a specific risk level boundary determined by combining the industry risk standards of procurement and sales contracts, legal and regulatory requirements, and the enterprise's risk tolerance. Risk items with risk levels exceeding the threshold are screened out, and the specific location and risk description of these high-risk items are clarified.

[0104] Based on the three-dimensional compliance risk map, we trace back along the mapping path of risk items in the map to restore the complete process of risk items from information extraction and dimension mapping to risk identification. We clarify the original source of risk items in multi-level semantic expression, the specific matching nodes in the three-dimensional compliance risk map, and the map rules on which risk identification is based. Finally, we determine the matching path of risk items, which fully presents the generation logic and related information of risk items.

[0105] Following the established matching path, we trace back to multi-level semantic expressions to locate the text content directly related to the risk item, including the original text of the clauses on which the risk item is based, the related statements before and after the clauses, and the specific content of the relevant semantic components. We extract this content completely to form the original semantic context associated with the risk item, ensuring that the original semantic context can fully support the subsequent scenario assessment.

[0106] Carefully study the original semantic context, and comprehensively assess the actual risk level of the risk item in conjunction with the specific expressions and limiting information in the context. If the assessment finds that the current expression level of the risk item does not match the actual risk situation reflected by the original semantic context, and the current expression level is higher than the actual risk level, then adjust the risk level of the risk item according to the limiting conditions in the original semantic context, such as supplementary agreements on relevant obligations in the contract, clear definition of the scope of risk, etc., to adjust it to a level that matches the actual risk level, forming a calibrated risk item. The calibration process strictly follows the original semantic context to ensure that the adjusted risk level is objective and accurate.

[0107] All calibrated risk items are systematically organized to form a complete contextualized risk diagnosis result for the target contract text. Each risk item in this result has been dynamically calibrated, and the risk level accurately reflects the risk situation in the actual contract scenario, providing a precise risk basis for subsequent clause revisions.

[0108] The beneficial effects are that by accurately extracting key information, mapping dimensions, identifying risks, conducting correlation analysis, and dynamically calibrating through multiple steps, the contextualized risk diagnosis results fully cover all risks in the three dimensions of transaction type, obligation performance, and object attributes. Moreover, the risk level determination is accurate and closely matches the actual contract scenario, effectively avoiding risk misjudgment or level deviation. This provides accurate and reliable risk basis for subsequent clause reconstruction and contract revision, and improves the comprehensiveness, accuracy, and pertinence of risk diagnosis for procurement and sales contracts.

[0109] S3. Based on the logical judgment rules of the aforementioned artificial intelligence big data model, the contextualized risk diagnosis results and the original contract semantic context in the target contract text are reconstructed using multiple clause schemes to obtain the optional revision schemes of the target contract text.

[0110] In this embodiment of the invention, the logical judgment rules based on the artificial intelligence big data model are used to reconstruct multiple clauses of the contextualized risk diagnosis results and the original contract semantic context in the target contract text, resulting in optional revision schemes for the target contract text, including:

[0111] The logical judgment rules in the aforementioned large-scale artificial intelligence model are analyzed by type to obtain the risk attribution and causation rules, multi-strategy generation rules, and semantic coherence embedding rules of the large-scale artificial intelligence model.

[0112] Based on the aforementioned risk attribution and causation rules, the contextualized risk diagnosis results are analyzed item by item to obtain the root cause clauses and impact transmission paths of the contextualized risk diagnosis results.

[0113] Based on the multi-strategy generation rules, the root cause clauses, and the impact transmission path, a revision strategy guide for the target contract text is generated;

[0114] By applying the semantic coherence embedding rules and combining them with the original contract semantic context in the target contract text, the revision strategy guidance is adjusted collaboratively to obtain the draft of the revised clauses of the target contract text.

[0115] The draft amendments are comprehensively evaluated to obtain the preferred draft amendments to the target contract text;

[0116] Modal construction is performed on the preferred draft amendment to obtain the optional amendment schemes for the target contract text.

[0117] The logical judgment rules in the AI ​​big model are a comprehensive set of rules built on the scenario of reviewing procurement and sales contracts. They cover various norms related to risk analysis, revision strategy generation, and semantic coherence. When analyzing the types, we first comprehensively sort out the core functions, applicable scenarios, and objects of all logical judgment rules, and classify them according to their core uses. We clearly distinguish between risk attribution and causation rules for tracing the root causes of risks, multi-strategy generation rules for generating multiple revision paths, and semantic coherence embedding rules for ensuring that the revised clauses are semantically consistent with the original text. This ensures that the boundaries of the three types of rules are clear and comprehensive, and finally we obtain the risk attribution and causation rules, multi-strategy generation rules, and semantic coherence embedding rules of the AI ​​big model.

[0118] The risk attribution and causation rules include specific analytical steps and judgment criteria for deriving the root cause from the risk outcome. For each risk item in the contextualized risk diagnosis results, the rules start from the specific manifestation of the risk, breaking down the contract terms, transaction links, and related constraints involved in the risk layer by layer. The rules identify the direct factors that may cause the risk in each link, and pinpoint the contract terms that directly cause the risk as the root cause clauses. At the same time, following the root cause clauses, the rules also sort out other contract terms, rights and obligations agreements, and transaction process nodes that the risk may spread to, clearly presenting the complete path of the risk from its generation to its spread, forming an impact transmission path. Through the analysis of each risk item, the root cause clauses and impact transmission paths of the contextualized risk diagnosis results are comprehensively obtained.

[0119] The multi-strategy generation rules pre-define various revision methods and combinations for different risk types, causal clauses, and scopes of impact. They combine the specific problem attributes of the fundamental causal clauses, such as vague wording, missing content, logical contradictions, and insufficient compliance, with the size of the contract scope covered by the impact transmission path. The rules then match corresponding revision methods from the multi-strategy generation rules to establish a clear revision direction for each fundamental causal clause. For example, for vaguely worded causal clauses, the revision direction is to accurately define the key information; for clauses with missing content, the revision direction is to supplement necessary rights and obligations; and for logically contradictory clauses, the revision direction is to adjust the clause content to eliminate the contradiction. By integrating the revision directions of all fundamental causal clauses, a revision strategy guide for the target contract text is formed, ensuring a clear and targeted approach.

[0120] The semantic coherence embedding rule clarifies the specific requirements that revised clauses must maintain consistency with the original contract in terms of grammatical structure, expression style, logical coherence, and use of professional terminology. First, it involves in-depth study of the semantic context of the original contract text within the target contract, comprehensively understanding the overall expression habits, the logical connections between clauses, the fixed usage of professional terminology, and the writing style. Then, based on the semantic coherence embedding rule, each revision direction in the revision strategy is compared sentence by sentence with the original semantic context. The expression, word choice, sentence structure, and clause placement of the revision direction are refined and adjusted, transforming the abstract revision direction into concrete, complete clauses that seamlessly connect with the original semantics. This ensures that the revised clauses integrate seamlessly into the original contract, ultimately resulting in a draft of the revised clauses for the target contract text.

[0121] The comprehensive evaluation revolves around four core dimensions: effectiveness of risk avoidance, semantic coherence, practical operability, and compliance. Quantitative evaluation standards are established. The effectiveness of risk avoidance is assessed to determine whether the revised draft clauses accurately address the root causes of the problems and thoroughly avoid the corresponding risks. Semantic coherence is assessed to determine whether the revised clauses are naturally connected to the semantic context of the original contract and whether the wording is consistent. Practical operability is assessed to determine whether the revised clauses are easy to execute and have no obstacles to implementation during the actual performance of the contract. Compliance is assessed to determine whether the revised clauses comply with relevant laws and regulations, industry practices, and the core demands of both parties to the contract. Each revised draft clause is scored according to the evaluation standards, and the draft clause with the highest overall score and all scores meeting the preset passing thresholds is selected as the preferred revised draft clause for the target contract text.

[0122] Modal construction, based on different application scenarios and user needs and preferences in procurement and sales contract review, transforms the preferred draft amendment clauses into specific solutions in different modalities. For conservative review scenarios with the highest priority of risk control, a revision solution with the most rigorous clause stipulations and the most comprehensive risk coverage is constructed, clearly defining and detailing various rights, obligations, and liabilities for breach of contract. For balanced review scenarios that balance risk control and transaction flexibility, a revision solution that retains a certain degree of room for negotiation under the premise of controllable core risks is constructed, ensuring both contract security and improving transaction efficiency. For flexible review scenarios that focus on transaction efficiency, a revision solution that simplifies processes and streamlines clauses based on compliance is constructed, removing redundant provisions and optimizing the performance process. Each modal solution clearly presents the specific wording of the amendment clauses, explanations of applicable scenarios, and core advantages. The three types of solutions are systematically integrated to form optional amendment solutions for the target contract text.

[0123] The beneficial effects are that by completing the logical judgment rule type analysis, risk root cause tracing, revision strategy guidance generation, semantic collaborative adjustment, revision draft selection and multimodal solution construction in a step-by-step manner, the generated optional revision solutions can not only accurately target and solve contract risks, but also maintain a high degree of consistency with the original contract semantics, while covering the needs of different review scenarios, providing users with diversified choices, effectively improving the accuracy, adaptability and efficiency of procurement and sales contract clause revision, and providing strong support for contract review.

[0124] S4. Perform human-computer interaction processing on the contextualized risk diagnosis results and the optional revision schemes, and receive the user's decision instructions to obtain the user's final decision result;

[0125] In this embodiment of the invention, the step of performing human-computer interaction processing on the contextualized risk diagnosis results and the optional revision schemes, and receiving the user's decision instructions to obtain the user's final decision result includes:

[0126] Based on the risk diagnosis sub-items in the contextualized risk diagnosis results, the contextualized risk diagnosis results are associated and mapped with the optional revision schemes to obtain the interactive decision list of the target contract text;

[0127] Based on the interactive decision list, the risk diagnosis sub-items are displayed synchronously to obtain the display result of the target contract text;

[0128] Respond to the user's operation on the interactive decision list and receive the user's decision instructions on the risk diagnosis sub-item;

[0129] The decision instruction is logically consistent to obtain the user's verified decision instruction.

[0130] The verified decision instruction is coupled with the contextualized risk diagnosis result to obtain the user's final decision result.

[0131] First, the contextualized risk diagnosis results are broken down to identify each independent risk diagnosis sub-item. Each sub-item contains core information such as risk type, risk level, risk description, and the location of related contract clauses. For each risk diagnosis sub-item, all solutions that can specifically address the risk are selected from the available revision options. A one-to-one correspondence between the risk diagnosis sub-item and the corresponding revision solution is established. At the same time, the correspondence is supplemented with explanatory information such as the applicable scenarios of the solution, the core revision content, and the risk avoidance effect. All risk diagnosis sub-items, corresponding revision solutions, and supplementary explanations are compiled into a booklet in a unified format to form an interactive decision list for the target contract text.

[0132] Following the established structure of the interactive decision checklist, risk diagnosis sub-items are presented in a layered manner. First, all risk diagnosis sub-items are displayed in descending order of risk level. Below each risk diagnosis sub-item's display area, all corresponding optional revision solutions are listed simultaneously. Each solution clearly presents the complete revision clause content, the relationship logic with the risk sub-item, and the expected effects after implementation. At the same time, the original contract clause corresponding to the risk diagnosis sub-item is marked with a link, which allows users to easily access and trace the source at any time. This ensures that users can intuitively obtain the correspondence between risks and solutions and obtain the display results of the target contract text.

[0133] A dedicated interactive response module is built to capture user actions on the interactive decision list in real time, including clicking to select a revision plan corresponding to a certain risk diagnosis sub-item, rejecting all optional plans and manually entering custom revision opinions, and confirming to retain the original clauses without revision. When a user completes a decision operation on one or more risk diagnosis sub-items, the system immediately receives the user's operation instructions, accurately identifies and records the content of the instructions, clarifies the user's decision intention for each risk diagnosis sub-item, and fully receives the user's decision instructions for all risk diagnosis sub-items.

[0134] A rule base for verifying contract logic consistency is constructed. This rule base includes core logical requirements such as common clause logical relationships in procurement and sales contracts, the principle of equal rights and obligations, and the sequence of performance processes. All user decision instructions are substituted into this rule base for comprehensive verification. The rule base checks for logical conflicts between decision instructions of different risk diagnosis sub-items. For example, if one decision instruction stipulates shortening the delivery cycle, but another related decision instruction does not adjust the payment node accordingly, this situation is judged as a logical conflict. At the same time, it checks whether individual decision instructions contradict the overall contract framework and core transaction objectives. Conflicting or contradictory instructions are marked and prompted to the user. After the user confirms the modification or maintains the original decision, the verified decision instruction for the user is formed.

[0135] The verified decision instructions are categorized and collected according to their corresponding risk diagnosis sub-items. Each risk diagnosis sub-item and its verified decision instructions form a set of associated data. Auxiliary information such as the generation time of the decision instruction, user operation traces, and logical verification results are supplemented in each set of data. All associated data are integrated with the contextualized risk diagnosis results to ensure that the final processing method of each risk diagnosis sub-item can find the corresponding risk basis in the contextualized risk diagnosis results, forming a complete data set covering the entire process of risk identification, user decision-making, and logical verification, which is the user's final decision result.

[0136] The beneficial effects are that by establishing a precise link between risks and revision plans, providing clear information display, real-time interactive response, and rigorous logical verification, users can fully understand contract risks and make reasonable decisions. At the same time, it ensures the logical consistency and relevance of decision-making instructions, giving full play to the user's subjective judgment and improving the scientificity and accuracy of decision-making through system assistance, thus laying a solid foundation for the finalization of the contract text.

[0137] S5. Perform a global logical self-check on the final decision result and calibrate the language standardization in the self-check result to obtain the final draft contract text of the target contract text.

[0138] In this embodiment of the invention, the step of performing a global logical self-check on the final decision result and calibrating the language standardization in the self-check result to obtain the final draft contract text of the target contract text includes:

[0139] By performing cross-logical relationship analysis on the clause revision records in the final decision result, potential logical conflicts in the target contract text can be obtained.

[0140] Based on the common-sense logical rules of the contract domain in the aforementioned large-scale artificial intelligence model, the potential logical conflicts are corrected to obtain the logically revised text of the target contract text;

[0141] The logically revised text is subjected to language standardization processing to obtain the standard text of the target contract text;

[0142] The format of the standard text is integrated to obtain the final draft of the target contract text.

[0143] A comprehensive review of all clause revision records in the final decision-making process was conducted, including newly added, modified, retained, and deleted clauses. The purpose, rights and obligations, and performance milestones corresponding to each revision record were clearly defined. Cross-analysis was performed on all revision records from multiple dimensions, including causal relationships, conditional constraints, sequential performance, and equivalence of rights and obligations. For example, the process examined whether there were logical contradictions between the payment terms and delivery deadlines in the payment terms, whether the compensation standards in the breach of contract clauses matched the contract value and the extent of loss caused by the breach, and whether the performance location clause conflicted with the transportation and acceptance clauses. All logical contradictions and issues discovered through cross-analysis were summarized to identify potential logical conflicts in the target contract text.

[0144] The common-sense logic rules in the contract domain within the AI-powered large-scale model are constructed based on industry practices, legal requirements, and common performance logic of procurement and sales contracts. These rules encompass core elements such as the order of performance processes, the principle of equal rights and obligations, the matching of risks and responsibilities, and the coherence of clauses. For each potential logical conflict, an analysis is conducted against the corresponding common-sense logic rules in the contract domain to clarify the root cause of the conflict and the direction for compliant correction. For example, if the potential logical conflict is "the delivery period is agreed to be within 10 days after the contract is signed, while the payment terms stipulate payment within 30 days after delivery," it conforms to the order of performance processes and requires no correction. However, if the conflict is "the delivery period is agreed to be within 5 days after the contract is signed, while the production cycle for the purchased customized equipment is typically 15 days," then according to the principle of reasonableness in performance processes, the delivery period is adjusted to "within 15 days after the contract is signed." All potential logical conflicts are corrected one by one to form a logically revised version of the target contract text.

[0145] The revised text was checked for linguistic accuracy sentence by sentence and paragraph by paragraph. Referencing standard language patterns in procurement and sales contracts, legal terminology, and industry conventions, issues such as ambiguous expressions, grammatical errors, inappropriate wording, and inconsistent terminology were corrected. For example, "delivery will be completed within approximately one month" was changed to "delivery period is within 30 days of contract signing," and "if Party B fails to deliver on time, it shall bear corresponding responsibility" was changed to "Party B shall bear liability for breach of contract if it fails to deliver the subject matter within the period stipulated in this contract." Similar expressions such as "payment method" and "settlement method" were standardized to ensure that all clauses in the text were expressed accurately, rigorously, unambiguously, and grammatically correct, meeting the linguistic requirements of contract texts and resulting in a standardized version of the target contract.

[0146] In accordance with the standard format requirements for procurement and sales contracts, the standardized text undergoes comprehensive format integration; the font, font size, line spacing, and page margins are standardized; the standardized text is formatted by chapter according to the core components of the contract, including chapters on general provisions, subject matter, quantity, quality, price or remuneration, performance period, place and method of performance, liability for breach of contract, and dispute resolution methods; the clauses under each chapter are sequentially numbered to ensure that the clause numbering logic is clear, without repetition or omission; the formatting of elements such as attachments, signature positions, and signing dates in the contract is standardized, making the entire contract text neat, uniform, well-organized, and easy to read, ultimately resulting in the final draft of the target contract text.

[0147] The beneficial effects include eliminating logical conflicts in contracts through global logical self-checking, ensuring the rigor and feasibility of contract terms, ensuring the professional accuracy and unambiguity of contract texts through language standardization calibration, and achieving standardized presentation of contract texts through format integration. The final draft contract text not only conforms to legal norms and industry practices, but also has good readability and enforceability, effectively reducing the risk of disputes caused by logical contradictions, linguistic ambiguities, or format chaos during contract performance, and improving the signing quality and performance efficiency of procurement and sales contracts.

[0148] S6. Using the multi-level semantic expression, the contextualized risk diagnosis result, the final decision result, and the finalized contract text as historical review case data of the target contract text, perform targeted feedback optimization on the artificial intelligence big model to obtain the adaptive review strategy and knowledge system of the target contract text.

[0149] In this embodiment of the invention, the step of using the multi-level semantic expression, the contextualized risk diagnosis result, the final decision result, and the finalized contract text as historical review case data for the target contract text includes:

[0150] By associating and integrating the multi-level semantic expressions, the contextualized risk diagnosis results, the final decision results, and the finalized contract text, a structured case record of the target contract text is obtained.

[0151] The contract type identifier, risk type set, revision strategy type, and the difference identifier between the user's final decision and the initial diagnosis result in the structured case record are used as the key feature data of the structured case record;

[0152] Tensor synthesis is performed on the key feature data to obtain the feature vector of the structured case record, and a timestamp and weight identifier are added to the feature vector to obtain the weight time-series vector of the structured case record;

[0153] The weighted time-series vector is stored in the historical review case library of the target contract text, and an index relationship is established between the stored historical review case library and the original contract text in the target contract text to construct the historical review case data of the target contract text.

[0154] The process of performing targeted feedback optimization on the large-scale artificial intelligence model to obtain the adaptive review strategy and knowledge system for the target contract text includes:

[0155] Obtain the case vector set from the historical review case library in the target contract text;

[0156] Based on the case vector set, calculate the consistency feedback coefficient between the user's final decision and the initial diagnostic results in the large artificial intelligence model;

[0157] Based on the consistency feedback coefficient, the parameters of the risk judgment and strategy generation logic of the large artificial intelligence model are updated to obtain the optimized model of the large artificial intelligence model.

[0158] The logical judgment rules of the optimization model are integrated and encapsulated with the historical review case library to obtain the adaptive review strategy and knowledge system for the target contract text.

[0159] The formula for calculating the consistency feedback coefficient is as follows:

[0160] ;

[0161] in, This represents the consistency feedback coefficient. This represents the total number of cases in the case vector set. This represents the total number of risk types in the contextualized risk diagnosis results. Indicates the first The weighting coefficients of each case. This represents the preset time decay constant. Indicates the first The time decay amount in each case Indicates the first Risk dimension weights for each risk type Indicates the first In the case of the first Consistency in user decisions across different risk types Indicates the first In the case of the first Identifiers for the diagnostic accuracy of each risk type This represents an exponential function.

[0162] A comprehensive review of multi-level semantic expressions, contextualized risk diagnosis results, final decision results, and finalized contract text is conducted, establishing information relationships according to the complete contract review process. Multi-level semantic expressions serve as the foundational semantic basis of the contract, corresponding to each risk item in the contextualized risk diagnosis results, clarifying the semantic foundation upon which each risk item relies. Contextualized risk diagnosis results are then linked to the final decision results, annotating the user's decision choices corresponding to each risk item. The final decision results are then linked to the finalized contract text, clearly presenting the specific manifestation of the decision instructions within the contract text. This linked information is organized in a unified structured format, including modules such as basic contract information, semantic analysis content, risk diagnosis details, user decision records, and the final contract text, forming a structured case record of the target contract text.

[0163] Key feature data is precisely extracted from structured case records. Contract type identifiers are determined based on the core transaction objectives and nature of the transactions in the structured case records, clearly indicating whether the contract belongs to a procurement contract, sales contract, or other sub-type. The risk type set summarizes all risk types in the contextualized risk diagnosis results and classifies them according to the dimensions of transaction type, obligation performance, and object attribute. The revision strategy type is extracted from the optional revision schemes corresponding to the final decision result, clearly defining the revision strategy category adopted for each risk item. The difference identifier between the user's final decision and the initial diagnosis result is generated by comparing the user's final decision and the initial risk diagnosis result, recording in detail the user's acceptance, modification, or rejection of the initial diagnosis risk items. These four types of information are integrated as key feature data of the structured case records.

[0164] Following a pre-defined feature dimension framework, key feature data is standardized, transforming non-numerical information such as contract type identifiers, risk type sets, revision strategy types, and difference identifiers into computable data in a unified format. Based on tensor synthesis logic, the processed key feature data is integrated according to a set dimensional order to construct a multi-dimensional feature vector that comprehensively represents the core information of the structured case record. The specific time when the case review was completed is recorded and added to the feature vector as a timestamp. Based on factors such as the complexity of the case, industry typicality, and risk coverage, the importance of the case is determined, and corresponding weight values ​​are assigned as weight identifiers and added to the feature vector, ultimately yielding a weighted time-series vector of the structured case record.

[0165] In accordance with the storage specifications of the historical review case database, the weighted time-series vectors are stored in the database in a unified data format to ensure the integrity and readability of the stored data. A dedicated index is established for each stored weighted time-series vector, which includes key retrieval information such as the unique identifier of the target contract text, the contract type identifier, core transaction target keywords, and risk type keywords. The index system establishes a correspondence between the stored historical review case database and the original contract text in the target contract text, enabling the corresponding original contract text and related review data to be quickly located through the index, thereby completely constructing the historical review case data of the target contract text.

[0166] By using the search function of the historical review case library, all stored weighted time-series vectors are extracted from the library, sorted and organized according to the order of contract review completion time, and categorized and labeled according to attributes such as contract type and industry field to form a complete case vector set, ensuring that the vector set contains contract review case data of different scenarios and types.

[0167] By combining the weight identifier of each case in the case vector set, the time decay corresponding to the timestamp, the dimensional weight of each risk type, the consistency of user decisions, and the diagnostic accuracy identifier, the degree of fit between the user's final decision and the initial diagnostic result of the AI ​​big data model in each case is comprehensively analyzed. The relevant information of all cases is systematically summarized, taking into account the importance, temporal correlation, and differences in risk types of different cases. Through comprehensive weighting and calculation, a consistency feedback coefficient that reflects the overall fit level is obtained, which intuitively reflects the matching between the initial diagnostic result and the user's decision.

[0168] The calculated consistency feedback coefficient is compared with a preset reasonable threshold. If the consistency feedback coefficient is higher than the threshold, it indicates that the initial diagnostic results of the AI ​​model are highly consistent with the user's decision. Only some detailed parameters in the risk assessment and strategy generation logic are fine-tuned to optimize the model's accuracy. If the consistency feedback coefficient is lower than the threshold, it indicates that the model has certain judgment bias. For the risk types with concentrated bias, the risk assessment criteria, the weight of influencing factors, and the strategy generation logic are adjusted. At the same time, the model's adaptability to different contract scenarios is optimized. An optimized AI model is formed through parameter updates.

[0169] The logical judgment rules, such as risk attribution and causation rules, multi-strategy generation rules, and semantic coherence embedding rules in the optimization model, are deeply integrated with case data in the historical review case library. The practical experience of typical risk handling methods, user preference decision-making directions, and effective revision strategies in the cases is incorporated into the rule system. The integrated rules and case library are systematically encapsulated to form an adaptive review strategy that can automatically adjust the review focus and optimize the revision strategy according to different contract types and risk scenarios. At the same time, industry knowledge, compliance requirements, and practical experience in the cases are integrated to improve the corresponding knowledge system, and finally, an adaptive review strategy and knowledge system for the target contract text are obtained.

[0170] The total number of cases comes from the total number of cases contained in the case vector set; the total number of risk types comes from the total number of risk types covered by the contextualized risk diagnosis results; the first The weight coefficient of the case is based on the case vector set. The importance of each case is pre-set; the preset time decay constant is determined in advance based on the actual application scenario's requirements regarding the impact of time factors on the cases; the first The time decay for each case is the difference between the current time and the [number]th time. The difference in generation time for each case; The risk dimension weights for each risk type are based on the... The importance of each risk type in the contextualized risk diagnosis results is pre-defined; In the case of the first The user decision consistency for each risk type comes from the first... In the case of the user regarding the first Statistical results on the degree of consistency in decision-making for each risk type; In the case of the first The diagnostic accuracy indicator for each risk type comes from the analysis of the first... In the case of the first The accuracy of the diagnostic results for each risk type is recorded; the calculation process of the exponential function is to use the natural constant as the base and the input value as the exponent for power operation.

[0171] The consistency feedback coefficient is calculated by first calculating the numerator component value for each case. The calculation method is to take the first case as an example. The weighting coefficients of each case are combined with a negative preset time decay constant and the first case. The product of the time decay amounts of each case is multiplied by the result of the exponential function of the exponent, and then the resulting value is multiplied by the first value corresponding to all risk types under that case. The risk dimension weights of each risk type, the first In the case of the first User decision consistency for each risk type, the first In the case of the first The sum of the products of the diagnostic accuracy indicators for each risk type is multiplied, and then the numerator values ​​for all cases are summed to obtain the total numerator value. Next, the denominator value for each case is calculated by multiplying the products of the three indicators. The weighting coefficients of each case are combined with a negative preset time decay constant and the first case. The product of the time decay of each case is multiplied by the result of the exponential function of the exponent. Then, the values ​​of the denominator items corresponding to all cases are summed to obtain the total denominator value. Finally, the total numerator value is divided by the total denominator value to obtain the consistency feedback coefficient. This value reflects the overall consistency feedback of the contextualized risk diagnosis results after all cases in the case vector set are combined with time decay characteristics, case weights, risk dimension weights, user decision consistency, and diagnostic accuracy indicators.

[0172] When the When the weight coefficient of a case increases, its influence on the consistency feedback coefficient increases. If the ratio of the numerator to denominator of the case's corresponding component is greater than the current consistency feedback coefficient, the consistency feedback coefficient will increase; conversely, it will decrease. When the preset time decay constant increases, the influence of the time decay on the exponential function result increases. The earlier the case was generated, the smaller its corresponding exponential function result, and the lower its influence on the consistency feedback coefficient. If the ratio of the numerator to denominator of an earlier case is greater than the current consistency feedback coefficient, the consistency feedback coefficient will decrease; conversely, it will increase. When the risk dimension weight of the k-th risk type increases, the influence of that risk type on the corresponding case's numerator component increases. If the weight of the k-th risk type increases, the influence of the time decay on the corresponding case's numerator component will increase. In each case, the user decision consistency for the k-th risk type and the... A larger product of the diagnostic accuracy indicators for the k-th risk type in a given case will increase the numerator value of the corresponding case, thus increasing the consistency feedback coefficient; conversely, a smaller product will decrease the consistency feedback coefficient. In a given case, when the consistency of user decisions regarding the k-th risk type increases, the contribution of the corresponding risk type to the numerator of the case increases, thereby driving up the consistency feedback coefficient; when the... When the diagnostic accuracy of the k-th risk type in a case is determined to be accurate, it will increase the contribution of the corresponding risk type to the numerical value of the case's numerator, thereby driving up the consistency feedback coefficient. When it is determined to be inaccurate, it will decrease the corresponding contribution, thereby lowering the consistency feedback coefficient.

[0173] The formula for the consistency feedback coefficient is based on the core characteristics of historical review case data and the model optimization requirements.

[0174] The weight labels of structured case records in the historical review case library determine the degree of influence of different cases on model optimization, which is the basis for the case weight related content in the formula.

[0175] The timestamps of the weight time series vectors record the completion time of the case review. Cases from older times may have less reference value for the current model optimization. Therefore, it is necessary to introduce time decay-related content to form the basis for the time decay part in the formula.

[0176] The risk types in the three-dimensional compliance risk map have different importance under the transaction type dimension, obligation fulfillment dimension, and object attribute dimension. It is necessary to distinguish their impact on consistency judgment by risk dimension weights, which is the basis for the risk dimension weights in the formula.

[0177] The difference marker between the final decision result and the initial diagnosis result can reflect the fit between the user's decision and the model's initial diagnosis. The consistency of user decisions based on the statistical results of this difference marker becomes the basis for the relevant parts of the formula.

[0178] The dynamic calibration process of contextualized risk diagnosis results determines the accuracy of each risk type's diagnosis result and forms a diagnosis accuracy indicator, which is the basis for the diagnosis accuracy indicator in the formula.

[0179] By integrating these core elements from historical review case data and the risk diagnosis process, the formula enables precise calculation of the degree of alignment between the user's final decision and the model's initial diagnostic results, providing a quantitative basis for updating model parameters and ensuring that the optimized model better meets actual review needs.

[0180] In real-world scenarios, the preset risk thresholds are set on a percentage scale: 80 for high risk, 60 for medium risk, and 40 for low risk.

[0181] The preset risk threshold is determined based on the risk impact dimensions of the procurement and sales contracts.

[0182] The economic loss dimension references the average loss amount of similar contract disputes in the industry, and combines it with 5% of the company's annual contract budget as the high-risk economic loss threshold, corresponding to a threshold of 80 points. When the economic loss that a risk item may cause exceeds this threshold, it is judged to exceed the high-risk threshold.

[0183] The performance impact dimension uses the duration of delay in fulfilling the core obligations of the contract as the standard. A delay of more than 30% in the industry's normal performance cycle is considered high risk, corresponding to a threshold of 80 points, to ensure that this threshold can screen out risk items that seriously affect contract performance.

[0184] The compliance dimension is based on the mandatory requirements of relevant laws and regulations for procurement and sales contracts. If a risk item may cause the contract terms to violate mandatory legal provisions, it will be directly judged as exceeding the high-risk threshold of 80 points.

[0185] The economic loss corresponding to a medium risk threshold of 60 points is 3% to 5% of the company's annual contract budget, and the performance delay is 15% to 30% of the industry's normal performance cycle. The risk may lead to compliance flaws in the contract terms, but these can be remedied through revision.

[0186] The low-risk threshold of 40 points corresponds to economic losses of less than 3% of the company's annual contract budget and performance delays of less than 15% of the industry's normal performance cycle. The risk items have a small impact on contract performance and compliance and do not require special calibration.

[0187] This value takes into account the three dimensions of the three-dimensional compliance risk map, ensuring that the threshold can accurately match the risk level classification under different dimensions, and is precisely linked with the dynamic calibration steps of the contextualized risk diagnosis results, effectively screening out risk items that need to be backtracked and scenario-based.

[0188] The beneficial effects are that by building historical review case data through the system, a wealth of practical evidence is provided for the large-scale artificial intelligence model. Then, by guiding the optimization of model parameters through the consistency feedback coefficient, the model can be continuously iterated and upgraded. The adaptive review strategy and knowledge system formed by integration and encapsulation can make subsequent contract reviews more in line with actual needs, greatly improve the accuracy, efficiency and adaptability of the review, and form a virtuous cycle of "case accumulation - model optimization - review quality improvement".

[0189] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0190] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automatic review method for procurement and sales contracts based on a large-scale artificial intelligence model, characterized in that: The method includes: S1. Based on a pre-set large-scale artificial intelligence model, the target contract text is subjected to context perception, and the perception results are subjected to deep semantic deconstruction to obtain the multi-level semantic expression of the target contract text. The large-scale artificial intelligence model includes semantic association rules, domain knowledge structure, common sense logic rules in the contract domain, and a historical dispute case library. S2. Perform automated context matching between the multi-level semantic expression and the preset three-dimensional compliance risk map to obtain the contextualized risk diagnosis result of the target contract text; S3. Based on the logical judgment rules of the aforementioned artificial intelligence big data model, the contextualized risk diagnosis results and the original contract semantic context in the target contract text are reconstructed using multiple clause schemes to obtain the optional revision schemes of the target contract text. S4. Perform human-computer interaction processing on the contextualized risk diagnosis results and the optional revision schemes, and receive the user's decision instructions to obtain the user's final decision result; S5. Perform a global logical self-check on the final decision result and calibrate the language standardization in the self-check result to obtain the final draft contract text of the target contract text. S6. Using the multi-level semantic expression, the contextualized risk diagnosis result, the final decision result, and the finalized contract text as historical review case data of the target contract text, perform targeted feedback optimization on the artificial intelligence big model to obtain the adaptive review strategy and knowledge system of the target contract text.

2. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The pre-defined large-scale artificial intelligence model performs context awareness on the target contract text and deep semantic deconstructs the perception results to obtain multi-level semantic expressions of the target contract text, including: Based on the semantic association rules in the pre-set artificial intelligence big data model, the logical connections and dependencies of the target contract text are perceived in context to obtain the perception result of the target contract text. Based on the knowledge structure in the aforementioned large-scale artificial intelligence model, the perception results are subjected to deep semantic deconstruction to obtain the semantic components of the target contract text; The semantic components are decomposed hierarchically to obtain the multi-level semantic expression of the target contract text.

3. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The step of automatically matching the multi-level semantic expression with a preset three-dimensional compliance risk map to obtain the contextualized risk diagnosis result of the target contract text includes: Extract the core transaction objectives, performance constraints, and specific transaction object parameters from the multi-level semantic expression; The core transaction objectives, the performance constraints, and the specific transaction object parameters are respectively mapped to the transaction type dimension, obligation fulfillment dimension, and object attribute dimension of the preset three-dimensional compliance risk map; Under the transaction type dimension, risk pattern identification is performed on the core transaction objective to obtain the basic risk list of the first dimension in the target contract text; Under the aforementioned obligation fulfillment dimension, the performance constraints are analyzed to obtain a list of conditions and risks in the second dimension of the target contract text; Under the object attribute dimension, the historical dispute case library in the artificial intelligence big model is compared with the specific transaction object parameters to obtain the object risk list of the third dimension in the target contract text; Cross-correlation analysis is performed on the basic risk list, the conditional risk list, and the object risk list to obtain the initial risk diagnosis results of the target contract text; The initial risk diagnosis results are dynamically calibrated to obtain the contextualized risk diagnosis results of the target contract text.

4. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 3, characterized in that, The dynamic calibration of the initial risk diagnosis results to obtain the contextualized risk diagnosis results of the target contract text includes: Identify risk items in the initial risk diagnosis results that exceed a preset risk threshold; Based on the three-dimensional compliance risk map, the risk items are traced back in reverse to obtain the matching path of the risk items; Based on the matching path, extract the original semantic context associated with the risk item from the multi-level semantic expression; Based on the original semantic context, the risk item is evaluated in a scenario. When the description level of the risk item is too high, the risk level of the risk item is lowered according to the limiting information in the original semantic context to obtain the calibrated risk item. The calibrated risk items are used as the contextualized risk diagnosis results of the target contract text.

5. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The logical judgment rules based on the aforementioned large-scale artificial intelligence model reconstruct multiple clauses from the contextualized risk diagnosis results and the original contract semantic context in the target contract text, resulting in optional revision schemes for the target contract text, including: The logical judgment rules in the aforementioned large-scale artificial intelligence model are analyzed by type to obtain the risk attribution and causation rules, multi-strategy generation rules, and semantic coherence embedding rules of the large-scale artificial intelligence model. Based on the aforementioned risk attribution and causation rules, the contextualized risk diagnosis results are analyzed item by item to obtain the root cause clauses and impact transmission paths of the contextualized risk diagnosis results. Based on the multi-strategy generation rules, the root cause clauses, and the impact transmission path, a revision strategy guide for the target contract text is generated; By applying the semantic coherence embedding rules and combining them with the original contract semantic context in the target contract text, the revision strategy guidance is adjusted collaboratively to obtain the draft of the revised clauses of the target contract text. The draft amendments are comprehensively evaluated to obtain the preferred draft amendments to the target contract text; Modal construction is performed on the preferred draft amendment to obtain the optional amendment schemes for the target contract text.

6. The method for automatic review of procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The process of performing human-computer interaction processing on the contextualized risk diagnosis results and the optional revision schemes, and receiving the user's decision instructions to obtain the user's final decision result, includes: Based on the risk diagnosis sub-items in the contextualized risk diagnosis results, the contextualized risk diagnosis results are associated and mapped with the optional revision schemes to obtain the interactive decision list of the target contract text; Based on the interactive decision list, the risk diagnosis sub-items are displayed synchronously to obtain the display result of the target contract text; Respond to the user's operation on the interactive decision list and receive the user's decision instructions on the risk diagnosis sub-item; The decision instruction is logically consistent to obtain the user's verified decision instruction. The verified decision instruction is coupled with the contextualized risk diagnosis result to obtain the user's final decision result.

7. The method for automatic review of procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The process of performing a global logical self-check on the final decision result and calibrating the language standardization in the self-check result to obtain the final draft contract text of the target contract text includes: By performing cross-logical relationship analysis on the clause revision records in the final decision result, potential logical conflicts in the target contract text can be obtained. Based on the common-sense logical rules of the contract domain in the aforementioned large-scale artificial intelligence model, the potential logical conflicts are corrected to obtain the logically revised text of the target contract text; The logically revised text is subjected to language standardization processing to obtain the standard text of the target contract text; The format of the standard text is integrated to obtain the final draft of the target contract text.

8. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The historical review case data for using the multi-level semantic expression, the contextualized risk diagnosis results, the final decision results, and the finalized contract text as the target contract text includes: By associating and integrating the multi-level semantic expressions, the contextualized risk diagnosis results, the final decision results, and the finalized contract text, a structured case record of the target contract text is obtained. The contract type identifier, risk type set, revision strategy type, and the difference identifier between the user's final decision and the initial diagnosis result in the structured case record are used as the key feature data of the structured case record; Tensor synthesis is performed on the key feature data to obtain the feature vector of the structured case record, and a timestamp and weight identifier are added to the feature vector to obtain the weight time-series vector of the structured case record; The weighted time-series vector is stored in the historical review case library of the target contract text, and an index relationship is established between the stored historical review case library and the original contract text in the target contract text to construct the historical review case data of the target contract text.

9. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 1, characterized in that, The process of performing targeted feedback optimization on the large-scale artificial intelligence model to obtain the adaptive review strategy and knowledge system for the target contract text includes: Obtain the case vector set from the historical review case library in the target contract text; Based on the case vector set, calculate the consistency feedback coefficient between the user's final decision and the initial diagnostic results in the large artificial intelligence model; Based on the consistency feedback coefficient, the parameters of the risk judgment and strategy generation logic of the large artificial intelligence model are updated to obtain the optimized model of the large artificial intelligence model. The logical judgment rules of the optimization model are integrated and encapsulated with the historical review case library to obtain the adaptive review strategy and knowledge system for the target contract text.

10. The automatic review method for procurement and sales contracts based on a large artificial intelligence model as described in claim 9, characterized in that, The formula for calculating the consistency feedback coefficient is as follows: ； in, This represents the consistency feedback coefficient. This represents the total number of cases in the case vector set. This represents the total number of risk types in the contextualized risk diagnosis results. Indicates the first The weighting coefficients of each case. This represents the preset time decay constant. Indicates the first The time decay amount in each case Indicates the first Risk dimension weights for each risk type Indicates the first In the case of the first Consistency in user decisions across different risk types Indicates the first In the case of the first Identifiers for the diagnostic accuracy of each risk type This represents an exponential function.

Images