Contract text processing method and device based on large language model, equipment and medium

Abstract

The application discloses a contract text processing method and device based on a large language model, equipment and medium, applied to the technical field of financial science and technology text processing, the current contract text preprocessing result is analyzed by the contract text analysis large language model, the extracted entity is aligned and processed by the knowledge graph alignment method, and the current contract text associated risk mining result is generated by combining the associated risk mining method; a dynamic risk analysis intelligent execution suggestion module is used for risk analysis, automatic conversion processing is carried out through a blockchain smart contract execution and evidence storage module, and target smart contract execution and evidence storage results are obtained, so that the problems that manual checking or rule matching is difficult to analyze complex contract texts, risk point identification efficiency is low, the execution process lacks intelligent support, and supervision and tracing are difficult are solved, the difficulty of contract text analysis is reduced, and the efficiency, reliability and intelligence of contract text processing are improved.

Description

Technical Field

[0001] This invention relates to the field of financial technology text processing technology, and in particular to a method, apparatus, device and medium for contract text processing based on a large language model. Background Technology

[0002] In corporate banking, the analysis of corporate contracts, credit agreements, guarantee documents, and loan contracts is a crucial foundation for risk control. Therefore, the process of analyzing contract texts is of paramount importance. Currently, traditional contract review typically relies on manual verification and keyword-based rule matching.

[0003] In the process of developing this invention, the inventors discovered the following shortcomings in existing technologies: They suffer from weak semantic understanding capabilities, difficulty in updating rules, difficulty in identifying cross-contract risks, and high workload for manual review. Furthermore, traditional contract analysis systems mostly rely on rule engines or keyword matching methods, which can only achieve superficial text extraction and classification. They cannot effectively understand the contextual semantics, logical dependencies, and conditional triggering mechanisms of contract clauses, thus exhibiting significant limitations in automated risk analysis and decision support. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for processing contract text based on a large language model, for implementation.

[0005] According to one aspect of the present invention, a method for processing contract text based on a large language model is provided, comprising: The system preprocesses the acquired contract text to be parsed in real time, and then parses the preprocessed contract text using a pre-trained contract text parsing big language model to obtain the current contract text parsing result. The entity is extracted from the parsing result of the current contract text, and the extracted entities are aligned with the knowledge graph using the knowledge graph alignment method. The association risk mining result of the current contract text is then generated by combining the association risk mining method. The dynamic risk analysis intelligent execution suggestion module is used to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text, and combined with the preset risk quantification and level classification methods, a current intelligent execution suggestion report is generated. Based on the current smart execution suggestion report, the blockchain smart contract execution and notarization module automatically converts and processes the data to obtain the target smart contract execution and notarization results, which are then fed back to the user.

[0006] According to another aspect of the present invention, a contract text processing apparatus based on a large language model is provided, comprising: The current contract text parsing result determination module is used to preprocess the acquired contract text to be parsed in real time, and to parse the preprocessed current contract text preprocessing result through a pre-trained contract text parsing big language model to obtain the current contract text parsing result. The current contract text association risk mining result generation module is used to extract entities from the current contract text parsing result, perform knowledge graph alignment processing on the extracted entities using the knowledge graph alignment method, and generate the current contract text association risk mining result by combining the association risk mining method. The current intelligent execution suggestion report generation module is used to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text using the dynamic risk analysis intelligent execution suggestion module, and generate a current intelligent execution suggestion report by combining the preset risk quantification and level classification methods. The target smart contract execution and notarization result determination and feedback module is used to automatically convert and process the current smart execution suggestion report through the blockchain smart contract execution and notarization module to obtain the target smart contract execution and notarization result, and then feed it back to the user.

[0007] According to another aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the contract text processing method based on a large language model as described in any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the contract text processing method based on a large language model as described in any embodiment of the present invention.

[0009] The technical solution of this invention preprocesses the acquired contract text to be parsed in real time, and then parses the preprocessed contract text using a pre-trained contract text parsing big data model to obtain the current contract text parsing result. Entities are extracted from the current contract text parsing result, and the extracted entities are aligned using a knowledge graph alignment method. A related risk mining method is then used to generate the current contract text related risk mining result. A dynamic risk analysis intelligent execution suggestion module is used to perform multi-dimensional risk analysis on the current contract text related risk mining result, and a current intelligent execution suggestion report is generated using a preset risk quantification and level classification method. Based on the current intelligent execution suggestion report, the blockchain smart contract execution and notarization module automatically converts and processes the data to obtain the target smart contract execution and notarization result, which is then fed back to the user. This solves the problems of complex and difficult-to-parse contract text, low efficiency in risk point identification, lack of intelligent support for execution processes, and difficulty in regulatory traceability that exist with manual verification or rule matching. It reduces the difficulty of contract text parsing and improves the efficiency, reliability, and intelligence of contract text processing.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a contract text processing method based on a large language model according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the structure of a contract text processing device based on a large language model according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided according to Embodiment 3 of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "target," "current," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] It is worth noting that the information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse; if the user chooses to refuse, the process will proceed to the expert decision-making process.

[0016] Example 1 Figure 1 The flowchart of a contract text processing method based on a large language model is provided in Embodiment 1 of the present invention. This embodiment is applicable to scenarios of parsing contract text or intelligent question-and-answer query. The method can be executed by a contract text processing device based on a large language model, which can be implemented in hardware and / or software.

[0017] Correspondingly, such as Figure 1 As shown, the method includes: S110. The obtained contract text to be parsed is preprocessed in real time, and the preprocessed result of the current contract text is parsed by a pre-trained contract text parsing big language model to obtain the current contract text parsing result.

[0018] The contract text to be parsed can include different types, such as loan contracts, guarantee agreements, and pledge agreements.

[0019] The contract text parsing big language model can be a model for intelligently parsing and analyzing contract text. This model may also include a natural language question-and-answer interface for responding to received question-and-answer commands. The current contract text parsing result can be the parsing result obtained from parsing the contract text.

[0020] Optionally, the real-time preprocessing of the acquired contract text to be parsed includes: acquiring the contract text to be parsed and the current contract text file format in real time through the contract text data acquisition and multimodal parsing module; performing multimodal parsing on the contract text to be parsed according to the current contract text file format to obtain the current contract text multimodal parsing result; automatically identifying and classifying the first paragraph content corresponding to the contract text to be parsed and the current contract text multimodal parsing result through a pre-built automatic contract text classification model to obtain the target contract classification result; wherein, the automatic contract text classification model is trained based on standard financial contract corpus to achieve matching of corresponding contextual text based on the target contract classification result; and processing the current contract text multimodal parsing result through the text structuring reconstruction unit and semantic full masking technology enhancement unit corresponding to the preset text preprocessing and semantic enhancement module to obtain the current contract text preprocessing result.

[0021] The contract text data acquisition and multimodal parsing module is a model that collects contract texts and performs different types of parsing operations based on the file format of the contract texts. This module serves as the system's data entry point, responsible for collecting contract documents from multiple heterogeneous data sources both inside and outside the bank and performing preliminary parsing.

[0022] Specific implementation methods include: data interface with the bank's internal contract management system and electronic signature platform through a predefined application programming interface (API). This API enables standardized data exchange between the system and the contract management system and electronic signature platform, supporting automatic acquisition of contract information, status synchronization, and metadata transmission, thereby improving the efficiency and consistency of data collection. External scanned image data can be accessed through pre-configured file transfer protocols or object storage services. The module's built-in intelligent document parsing engine can adaptively process various file formats.

[0023] Specifically, for text-based PDF and Word contract text file formats, the internal document object model is directly parsed to accurately extract text and structural information; for image-based PDFs and scanned documents, an optical character recognition engine optimized for financial documents is used for processing. This engine can use a deep learning model incorporating attention mechanisms, which can effectively resist noise interference such as seal coverage, binding line shadows and image tilt, significantly improving the accuracy of text recognition.

[0024] Furthermore, a pre-built automatic contract text classification model is used to automatically identify and classify the first paragraph of the contract text to be parsed and the multimodal parsing results of the current contract text, resulting in the target contract classification result. Specifically, this classification model is a contextual semantic model pre-trained on a standard financial contract corpus. It can match the target contract classification result with the corresponding contextual textual basis, thereby deeply understanding and accurately capturing the key features in the contract text, achieving automated identification and classification of contract types, and improving the accuracy and generalization ability of classification. A concrete example is the ability to automatically identify and classify contract types (i.e., the target contract classification result) based on the contract title and first paragraph. This result can include loan contracts, guarantee agreements, or pledge agreements, providing contextual basis for subsequent differentiated analysis processes.

[0025] Accordingly, the current contract text multimodal parsing result is processed by the text structuring reconstruction unit and semantic full masking technology enhancement unit corresponding to the preset text preprocessing and semantic enhancement module to obtain the current contract text preprocessing result.

[0026] In this embodiment, the text preprocessing and semantic enhancement module is responsible for deep cleaning, structure reconstruction, and semantic enhancement of the raw text obtained from upstream parsing. Regarding text structure reconstruction, this module uses carefully designed regular expression patterns to match the chapter and clause numbers in the contract, and combines this with layout information such as paragraph indentation and font size to reconstruct the complete chapter-tree logical structure of the contract document. Simultaneously, this module can also use a deep learning model with an object detection architecture to locate and identify table areas in the document. By analyzing the row and column structure, it accurately restores the table content into machine-readable structured data (by detecting the intersections of table lines to determine row and column boundaries, establishing a two-dimensional coordinate mapping relationship, and reorganizing cells according to row and column positions into JSON or CSV format, preserving the table header hierarchy and cell merging attributes).

[0027] Furthermore, regarding semantic full-word masking enhancement, this module deploys a domain-adaptive named entity recognition pipeline. This pipeline employs a full-word masking enhancement algorithm fine-tuned on a financial legal corpus, enabling semantic understanding of Chinese financial terms. It utilizes a deep large-scale language model architecture to capture long-distance dependencies in contract text, improving the accuracy of entity boundary identification and type classification. Further, it can be combined with a bidirectional long short-term memory network and a conditional random field decoding layer for accurate identification and classification of named entities in the text, including entities such as responsible persons, organizations, amounts, and dates. The identified amount, date, and other information are standardized to provide high-quality, standardized text input for subsequent deep semantic analysis.

[0028] Optionally, before parsing the preprocessed current contract text using a pre-trained contract text parsing large language model to obtain the current contract text parsing result, the process includes: acquiring and pre-training the initially constructed contract text parsing large language model based on standard unlabeled contract joint description text to obtain an initial contract text parsing large language model; generating at least one target contract triple dataset based on the standard unlabeled contract joint description text; wherein the target contract triple dataset includes the current extraction instruction, the current input contract text, and the current extraction instruction output result; and performing supervised retraining on the initial contract text parsing large language model according to each target contract triple dataset including the current extraction instruction, the current input contract text, and the current extraction instruction output result, and confirming that the contract text parsing large language model has been trained when the training result meets the preset accuracy requirements.

[0029] In this embodiment, the training of the large language model for contract text parsing includes a first stage of unsupervised training and a second stage of supervised optimization training.

[0030] The standard unmarked joint description text of the contract may include contract texts such as financial legal documents and regulatory provisions.

[0031] Specifically, the model is pre-trained using a massive amount of unlabeled financial legal documents and regulations to master the relevant professional terminology and syntactic structures, resulting in an initial large-scale language model for contract text parsing. Then, a triplet dataset consisting of the current extraction instruction, the current input contract text, and the output of the current extraction instruction can be constructed. Supervised fine-tuning then teaches the model to extract predefined structured information from the clauses according to the instructions. Finally, when the training results meet the preset accuracy requirements, the large-scale language model for contract text parsing is confirmed to be successfully trained.

[0032] A specific example is for any contractual clause. Model execution function This function encapsulates the reasoning process of a large language model. It receives the contract terms text and predefined extraction instructions as input, automatically identifies and outputs key structured elements in the terms through semantic understanding, and the output structure can be... .in, Represents the collective of responsible parties. Representative of the set of obligations, Represents the set of conditions and exceptions. This represents a set of risk labels. The post-processing unit cross-validates the model output with the entity information extracted by the named entity recognition system in the preprocessing module to ensure entity consistency. It also initiates a review process for any identified conflicts, thereby ensuring the accuracy and reliability of the extraction results.

[0033] Optionally, the text parsing large language model also provides a natural language question-and-answer interface; after the real-time preprocessing of the acquired contract text to be parsed, and the parsing of the preprocessed current contract text preprocessing result by the pre-trained contract text parsing large language model to obtain the current contract text parsing result, the method further includes: receiving and parsing at least one current unanswered text question instruction in real time through the natural language question-and-answer interface, generating text response results for each target instruction; and providing visual feedback to the user for each of the target instruction text response results.

[0034] In this embodiment, a natural language question-and-answer interface can also be set in the text parsing large language model to parse and process one or more currently unanswered text question commands received in real time, thereby generating text response results for each target command; then, the text response results for each target command can be given to the user in a visual feedback operation.

[0035] In this embodiment, by setting a natural language question-and-answer interface for the large language model for text parsing, the understanding of contract text can be better realized, thereby improving the efficiency and accuracy of contract text parsing.

[0036] S120. Extract entities from the current contract text parsing result, perform knowledge graph alignment processing on the extracted entities using the knowledge graph alignment method, and generate the current contract text association risk mining result by combining the association risk mining method.

[0037] In this embodiment, a knowledge graph alignment and association risk mining module is set up. This module can first extract entities from the current contract text parsing results, and then perform knowledge graph alignment processing on the extracted entities using a knowledge graph alignment method. Specifically, this module first uses an entity linking unit to align the entities extracted by the semantic parsing module with the nodes in the knowledge graph. This process effectively captures the semantic similarity between names by mapping entity names and aliases to high-dimensional semantic vectors, supporting cross-entity fuzzy matching and disambiguation based on vector similarity; furthermore, it encodes names into vectors and calculates cosine similarity to achieve efficient fuzzy matching. The aligned data is stored in the native graph database to expand and enrich the knowledge graph.

[0038] Furthermore, by executing the association risk mining method in the association risk analysis submodule, a series of graph algorithms are run for in-depth mining, aiming to discover the hidden risk transmission paths and clustering phenomena in the complex network of equity, guarantee and transaction relationships between enterprises.

[0039] Specifically, firstly, loop detection algorithms (based on relationships such as shareholders, detecting circular guarantee chains formed between enterprises; automatic guarantee circles refer to closed guarantee relationship loops formed by multiple enterprises through mutual guarantees and circular guarantees, significantly amplifying the transmission of local risks and systemic risks) can be used to automatically identify guarantee circles. Then, potential risk contagion chains can be discovered by calculating the shortest paths and relationship weights between nodes; for a specific guarantor, all its external guarantee amounts are aggregated through graph traversal, thereby accurately identifying excessive guarantee behavior. This graph-based association analysis enables penetrating review of hidden risks that are difficult to detect using traditional review methods, improving the accuracy of contract text mining. In other words, the results of risk mining related to the current contract text can be obtained.

[0040] S130. Use the dynamic risk analysis intelligent execution suggestion module to perform multi-dimensional risk analysis on the current contract text associated risk mining results, and generate the current intelligent execution suggestion report by combining the preset risk quantification and level classification methods.

[0041] Among them, the dynamic risk analysis intelligent execution suggestion module is the decision-making and output center of the system. It is responsible for comprehensively judging the analysis results of the whole process, classifying risks, and generating disposal plans and reports with direct operability. Its core lies in building a complete decision-making chain from multi-dimensional risk analysis to the generation of executable suggestions.

[0042] Optionally, the step of using the dynamic risk analysis intelligent execution suggestion module to perform multi-dimensional risk analysis on the current contract text association risk mining results, and generating a current intelligent execution suggestion report by combining a preset risk quantification and level classification method, includes: using the dynamic risk analysis intelligent execution suggestion module to perform risk analysis on the current contract text association risk mining results for text compliance, business logic, and association contagion, respectively, to obtain the current text compliance risk value, the current business logic risk value, and the current association contagion risk value; obtaining preset text compliance risk weight values, business logic risk weight values, and association contagion risk weight values, and weighting and summing them with the current text compliance risk value, the current business logic risk value, and the current association contagion risk value, respectively, to obtain the current multi-dimensional risk value; processing the current multi-dimensional risk value using a preset risk quantification and level classification method to obtain the target risk value level; and matching the target risk value level with a preset strategy rule base to generate the current intelligent execution suggestion report.

[0043] In this embodiment, multi-dimensional risk analysis can include risk analysis of textual compliance, business logic, and contagion. Specifically, textual compliance risk analysis can include risks stemming from deviations of contract terms from standard templates, best practices, or mandatory laws and regulations. When the risk label set output by the semantic parsing module contains high-priority labels such as "invalid clauses," "unequal rights and obligations," and "missing or ambiguous key elements" (e.g., unclear payment terms or breach of contract liability), such risks are identified; its essence is to examine the standardization and legality of the contract text. Furthermore, for business logic risks, this dimension focuses on the inherent rationality and sustainability of the commercial arrangements involved in the contract. When the knowledge graph module calculates that the guarantor's "guarantee multiple" exceeds a preset threshold, or the semantic parsing module identifies high-risk business models such as "cross-default" or "serious maturity mismatch," such risks are identified; its core is to assess the financial and credit risks that the commercial terms themselves may cause. Finally, for contagion risks, this dimension aims to reveal potential systemic risks arising from complex relationships between entities. When a graph computing algorithm detects a guarantee circle (such as a loop where A guarantees B, B guarantees C, and C guarantees A), or when a community detection algorithm identifies an abnormally high density of risky customers within a customer's associated community, it determines that such a risk exists. Its focus is on identifying the risk contagion paths hidden behind isolated contracts. Furthermore, the current text compliance risk value, current business logic risk value, and current associated contagion risk value can be calculated.

[0044] Furthermore, preset text compliance risk weight values, business logic risk weight values, and related contagion risk weight values ​​can be obtained to weight the current text compliance risk value, current business logic risk value, and current related contagion risk value respectively, and then sum them to obtain the current multi-dimensional risk value.

[0045] Accordingly, the current multi-dimensional risk values ​​are processed using risk quantification and grading methods to obtain the target risk value level; this risk value level specifically includes high risk, medium risk, and low risk. Specifically, high risk is defined as a score exceeding a preset high-score threshold; this level typically indicates that the contract has legal validity issues that could lead to significant losses, serious over-guarantees, or complex guarantee circles, requiring immediate mandatory intervention. Medium risk is defined as a score between the low and high score thresholds; this level indicates that the contract has flaws or potential problems that require attention, but the risk is still within a controllable range and can be mitigated through supplementary conditions, negotiated amendments to clauses, etc. (e.g., requesting supplementary explanatory documents, adjusting payment plans, adding guarantor declarations, improving dispute resolution clauses, etc.). Low risk is defined as a score below the low-score threshold; this level indicates that the contract has only minor textual flaws or no critical risks, and the system only provides a warning without requiring major intervention.

[0046] In this embodiment, based on the matching of the target risk level with the strategy rule base, this module ultimately outputs a current intelligent execution suggestion report that can be directly used by business personnel. Specifically, the generator maps and generates natural language suggestions with clear operational gradients from a pre-set, large "risk level-countermeasure" strategy rule base, according to the determined risk level and specific risk factor list. Its output strategy is as follows: For high risk, suggestions begin with mandatory words such as "must," "immediately," and "reject," for example: "The borrower must be required to provide sufficient collateral before loan disbursement; otherwise, it is recommended to reject this transaction." For medium risk, suggestions use a negotiating tone such as "suggest," "should," and "negotiate," for example: "It is recommended to negotiate with the customer to clarify the specific calculation method of the penalty clause in the supplementary agreement." For low risk, the output is mostly suggestive information, such as: "Please note that this contract text is a non-standard version and has been marked for archiving. It is recommended to prioritize the use of our bank's latest standard text in subsequent business."

[0047] The technical solution of this embodiment uses a dynamic risk analysis intelligent execution suggestion module to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text. Combined with a preset risk quantification and level classification method, it generates a current intelligent execution suggestion report. This enables dynamic risk analysis processing, provides intelligent execution analysis reports, and can accurately address different types of risk outcomes, thus improving the flexibility and accuracy of intelligent analysis of contract texts.

[0048] S140. Based on the current smart execution suggestion report, the blockchain smart contract execution and notarization module automatically converts and processes the data to obtain the target smart contract execution and notarization results, which are then fed back to the user.

[0049] Among them, the blockchain smart contract execution and evidence storage module utilizes the decentralized, immutable, and programmable characteristics of blockchain technology to achieve automated execution of risk management measures and trusted evidence storage throughout the entire process.

[0050] Optionally, the step of automatically converting the current smart execution suggestion report into a target smart contract execution and storage result through the blockchain smart contract execution and storage module to obtain the target smart contract execution and storage result includes: automatically converting the current smart execution suggestion report into code through the blockchain smart contract execution and storage module to obtain the current smart contract code; using a blockchain oracle to receive the current contract trigger judgment conditions in real time; and if it is determined that the current contract trigger judgment conditions are met, executing the current smart contract code, modifying the smart contract execution state, and generating the target smart contract execution and storage result.

[0051] In this embodiment, the current smart execution suggestion report undergoes automatic code conversion to obtain executable smart contract code. For example, a suggestion to "require additional collateral" can be encoded as: if collateral value < contract amount then trigger an alert and suspend some post-loan functions. Additionally, a consortium blockchain deployment can be set up, specifically deploying the smart contract on a consortium blockchain network built by the bank and relevant partners (such as core financing enterprises and regulatory agency nodes) to ensure the transparency and trustworthiness of the contract execution process.

[0052] Furthermore, a blockchain oracle is used to receive the current contract triggering conditions in real time. If the detection determines that the current contract triggering conditions are met, the current smart contract code is executed, the smart contract execution state is modified, and the execution and notarization results of the target smart contract are generated. Prior to this, a secure blockchain oracle reliably inputs off-chain real-world data (such as data from collateral valuation systems, account balances within the bank's core system, and external credit data) into the on-chain smart contract as the contract triggering conditions. Then, through detection, if the detection determines that the current contract triggering conditions are met, the current smart contract code is executed, the smart contract execution state is modified, and cross-chain technology ensures that the execution state of the smart contract on the blockchain (such as "triggered" or "executed") can be reliably synchronized back to the bank's internal business system in real time, guaranteeing the consistency of on-chain and off-chain data. Finally, the execution and notarization results of the target smart contract are generated.

[0053] For storing critical data on the blockchain, key data and events from the contract risk analysis process can be generated into hash fingerprints and written to the blockchain to achieve permanent and tamper-proof evidence storage. Specifically, this includes storing: the hash value of the original contract file, the core structured data summary output by the semantic parsing module, the final risk analysis report and approval process records, the smart contract code, and the full lifecycle logs of deployment, triggering, and execution.

[0054] The advantages of this setup are as follows: the blockchain smart contract execution and notarization module automatically converts the current smart execution suggestion report into code, and uses a blockchain oracle to receive the current contract trigger judgment conditions in real time to generate the target smart contract execution and notarization results; this ensures the consistency of data storage, allows for more flexible automated implementation of the report, and improves the flexibility of contract text analysis and the accuracy of the analysis results.

[0055] Optionally, after the step of automatically converting and processing the target smart contract execution and notarization result according to the current smart execution suggestion report through the blockchain smart contract execution and notarization module, and then feeding it back to the user, the method further includes: receiving multi-source feedback data results in real time, and using the feedback learning and model optimization module to perform online learning, real-time fine-tuning, and periodic incremental training of the contract text parsing large language model, so as to achieve optimization processing of the contract text parsing large language model.

[0056] The blockchain smart contract execution and storage module is the core of the system's self-evolution and continuous optimization. By constructing a complete learning loop from business feedback to model iteration, it ensures that the accuracy and adaptability of the system's risk identification and decision-making capabilities continuously improve over time.

[0057] In this embodiment, multi-source feedback data results can be received in real time, which may include manual correction feedback results, business processing feedback results, and smart contract execution feedback results.

[0058] Specifically, manual feedback can be used to record risk analysts' adoption, modification, or rejection of the system's automatically generated risk levels and implementation recommendations; the modified recommendations and their context constitute important negative samples. Business handling feedback connects to the bank's business system, collecting the actual performance of risk contracts marked by this system (such as whether repayments were made on time, whether guarantees were triggered, whether disputes arose, etc.), forming training samples with real results. Smart contract execution feedback can record the triggering and execution results of blockchain smart contracts, verifying the actual effectiveness of automated handling measures based on predictive risk.

[0059] Specifically, based on the aforementioned multi-source feedback data results, a strategy combining online learning with real-time fine-tuning and periodic incremental training can be adopted to achieve continuous evolution of model performance. Real-time online learning fine-tuning involves using an online learning algorithm based on stochastic gradient descent to fine-tune the parameters of the lightweight classifier in the semantic parsing module for high-confidence samples in the feedback data stream, enabling rapid model adaptation. Periodic incremental training involves the system periodically (e.g., weekly or monthly) initiating an incremental training process. This process uses all newly accumulated feedback data to perform knowledge distillation or efficient parameter fine-tuning of the core large language model, with the following objective function: .in, For the original training loss, For the loss based on the new feedback data, To balance hyperparameters, a reinforcement learning environment is constructed to link the generation of execution suggestions with subsequent business effects. The suggestion generation strategy is optimized using a proximal policy optimization algorithm, and corresponding reward functions can be set based on suggestion adoption rate, risk mitigation success rate, and false positive penalty.

[0060] The advantages of this setup are: by receiving multi-source feedback data results in real time, and using feedback learning and model optimization modules to perform online learning, real-time fine-tuning, and periodic incremental training of the large language model for contract text parsing, the model can be continuously optimized, thereby improving the accuracy and reliability of contract text processing and analysis. The model can also be adjusted and optimized based on feedback data during training or use.

[0061] The technical solution of this invention preprocesses the acquired contract text to be parsed in real time, and then parses the preprocessed contract text using a pre-trained contract text parsing big data model to obtain the current contract text parsing result. Entities are extracted from the current contract text parsing result, and the extracted entities are aligned using a knowledge graph alignment method. A related risk mining method is then used to generate the current contract text related risk mining result. A dynamic risk analysis intelligent execution suggestion module is used to perform multi-dimensional risk analysis on the current contract text related risk mining result, and a current intelligent execution suggestion report is generated using a preset risk quantification and level classification method. Based on the current intelligent execution suggestion report, the blockchain smart contract execution and notarization module automatically converts and processes the data to obtain the target smart contract execution and notarization result, which is then fed back to the user. This solves the problems of complex and difficult-to-parse contract text, low efficiency in risk point identification, lack of intelligent support for execution processes, and difficulty in regulatory traceability that exist with manual verification or rule matching. It reduces the difficulty of contract text parsing and improves the efficiency, reliability, and intelligence of contract text processing.

[0062] Example 2 Figure 2 This is a schematic diagram of a contract text processing device based on a large language model, provided in Embodiment 2 of the present invention. The contract text processing device based on a large language model provided in this embodiment can be implemented through software and / or hardware, and can be configured in a terminal device or server to implement a contract text processing method based on a large language model according to the embodiments of the present invention. Figure 2 As shown, the device includes: a current contract text parsing result determination module 210, a current contract text associated risk mining result generation module 220, a current smart execution suggestion report generation module 230, and a target smart contract execution and evidence storage result determination and feedback module 240.

[0063] Among them, the current contract text parsing result determination module 210 is used to preprocess the acquired contract text to be parsed in real time, and to parse the preprocessed current contract text preprocessing result through a pre-trained contract text parsing big language model to obtain the current contract text parsing result. The current contract text association risk mining result generation module 220 is used to extract entities from the current contract text parsing result, perform knowledge graph alignment processing on the extracted entities using the knowledge graph alignment method, and generate the current contract text association risk mining result by combining the association risk mining method. The current intelligent execution suggestion report generation module 230 is used to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text using the dynamic risk analysis intelligent execution suggestion module, and generate a current intelligent execution suggestion report by combining the preset risk quantification and level classification methods. The target smart contract execution and notarization result determination and feedback module 240 is used to automatically convert and process the target smart contract execution and notarization result according to the current smart execution suggestion report through the blockchain smart contract execution and notarization module, and then feed it back to the user.

[0064] The technical solution of this invention preprocesses the acquired contract text to be parsed in real time, and then parses the preprocessed contract text using a pre-trained contract text parsing big data model to obtain the current contract text parsing result. Entities are extracted from the current contract text parsing result, and the extracted entities are aligned using a knowledge graph alignment method. A related risk mining method is then used to generate the current contract text related risk mining result. A dynamic risk analysis intelligent execution suggestion module is used to perform multi-dimensional risk analysis on the current contract text related risk mining result, and a current intelligent execution suggestion report is generated using a preset risk quantification and level classification method. Based on the current intelligent execution suggestion report, the blockchain smart contract execution and notarization module automatically converts and processes the data to obtain the target smart contract execution and notarization result, which is then fed back to the user. This solves the problems of complex and difficult-to-parse contract text, low efficiency in risk point identification, lack of intelligent support for execution processes, and difficulty in regulatory traceability that exist with manual verification or rule matching. It reduces the difficulty of contract text parsing and improves the efficiency, reliability, and intelligence of contract text processing.

[0065] Based on the above embodiments, the current contract text parsing result determination module 210 can be specifically used to: collect the contract text to be parsed and the current contract text file format in real time through the contract text data acquisition and multimodal parsing module; perform multimodal parsing on the contract text to be parsed according to the current contract text file format to obtain the current contract text multimodal parsing result; automatically identify and classify the first paragraph content corresponding to the contract text to be parsed and the current contract text multimodal parsing result through a pre-built automatic contract text classification model to obtain the target contract classification result; wherein, the automatic contract text classification model is trained based on standard financial contract corpus to achieve matching of corresponding contextual text basis according to the target contract classification result; and process the current contract text multimodal parsing result through the text structuring reconstruction unit and semantic full masking technology enhancement unit corresponding to the preset text preprocessing and semantic enhancement module to obtain the current contract text preprocessing result.

[0066] Based on the above embodiments, a contract text parsing large language model training module is further included, which can be specifically used to: before parsing the preprocessed current contract text preprocessing result using the pre-trained contract text parsing large language model to obtain the current contract text parsing result, acquire and pre-train the initially constructed contract text parsing large language model based on standard unlabeled contract joint description text to obtain an initial contract text parsing large language model; generate at least one target contract triple dataset based on the standard unlabeled contract joint description text; wherein the target contract triple dataset includes the current extraction instruction, the current input contract text, and the current extraction instruction output result; supervised retraining of the initial contract text parsing large language model is performed according to each target contract triple dataset including the current extraction instruction, the current input contract text, and the current extraction instruction output result; and confirm that the contract text parsing large language model has been trained when the training result meets the preset accuracy requirements.

[0067] Based on the above embodiments, the text parsing large language model also provides a natural language question answering interface.

[0068] Based on the above embodiments, a target instruction text response result generation and feedback module is also included, which can be specifically used to: preprocess the acquired contract text to be parsed in real time, and parse the preprocessed current contract text preprocessing result through a pre-trained contract text parsing big language model to obtain the current contract text parsing result; then, through the natural language question answering interface, receive and parse at least one current unanswered text question instruction in real time, generate each target instruction text response result; and provide visual feedback to the user on each target instruction text response result.

[0069] Based on the above embodiments, the current intelligent execution suggestion report generation module 230 can be specifically used to: use the dynamic risk analysis intelligent execution suggestion module to perform risk analysis on the current contract text association risk mining results for text compliance, business logic, and association contagion, respectively, to obtain the current text compliance risk value, the current business logic risk value, and the current association contagion risk value; obtain preset text compliance risk weight values, business logic risk weight values, and association contagion risk weight values, and respectively perform weighted summation with the current text compliance risk value, the current business logic risk value, and the current association contagion risk value to obtain the current multi-dimensional risk value; process the current multi-dimensional risk value through a preset risk quantification and level classification method to obtain the target risk value level; and match the target risk value level with a preset strategy rule base to generate the current intelligent execution suggestion report.

[0070] Based on the above embodiments, the target smart contract execution and evidence storage result determination and feedback module 240 can be specifically used to: automatically convert the current smart execution suggestion report into code through the blockchain smart contract execution and evidence storage module to obtain the current smart contract code; use a blockchain oracle to receive the current contract trigger judgment condition in real time; if the detection determines that the current contract trigger judgment condition is met, then execute the current smart contract code, modify the smart contract execution state, and generate the target smart contract execution and evidence storage result.

[0071] Based on the above embodiments, it also includes a contract text parsing large language model optimization module, which can be specifically used to: after automatically converting and processing the target smart contract execution and notarization result according to the current smart execution suggestion report through the blockchain smart contract execution and notarization module, and feeding it back to the user, receive multi-source feedback data results in real time, and use the feedback learning and model optimization module to perform online learning, real-time fine-tuning and periodic incremental training of the contract text parsing large language model, so as to achieve optimization processing of the contract text parsing large language model.

[0072] The contract text processing device based on a large language model provided in this embodiment of the invention can execute the contract text processing method based on a large language model provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0073] Example 3 Figure 3 A schematic diagram of an electronic device 10, which can be used to implement Embodiment 3 of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0074] like Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0075] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0076] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as contract text processing methods based on large language models.

[0077] In some embodiments, the large language model-based contract text processing method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the large language model-based contract text processing method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the large language model-based contract text processing method by any other suitable means (e.g., by means of firmware).

[0078] The method includes: preprocessing the acquired contract text to be parsed in real time, and parsing the preprocessed contract text using a pre-trained contract text parsing big data language model to obtain the current contract text parsing result; extracting entities from the current contract text parsing result, aligning the extracted entities using a knowledge graph alignment method, and generating the current contract text association risk mining result using an association risk mining method; using a dynamic risk analysis intelligent execution suggestion module to perform multi-dimensional risk analysis on the current contract text association risk mining result, and generating a current intelligent execution suggestion report by combining a preset risk quantification and level classification method; and automatically converting the current intelligent execution suggestion report using a blockchain smart contract execution and notarization module to obtain the target smart contract execution and notarization result, and feeding it back to the user.

[0079] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0080] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0081] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0082] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0083] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0084] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0085] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0086] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0087] Example 4 Embodiment 4 of the present invention also provides a computer-readable storage medium, wherein the computer-readable instructions, when executed by a computer processor, are used to execute a contract text processing method based on a large language model. The method includes: real-time preprocessing of the acquired contract text to be parsed, and parsing the preprocessed result of the current contract text using a pre-trained large language model for contract text parsing to obtain a current contract text parsing result; extracting entities from the current contract text parsing result, aligning the extracted entities using a knowledge graph alignment method, and generating a current contract text association risk mining result using an association risk mining method; using a dynamic risk analysis intelligent execution suggestion module to perform multi-dimensional risk analysis on the current contract text association risk mining result, and generating a current intelligent execution suggestion report using a preset risk quantification and level classification method; and automatically converting the current intelligent execution suggestion report using a blockchain smart contract execution and notarization module to obtain the target smart contract execution and notarization result, and feeding it back to the user.

[0088] Of course, the computer-executable instructions provided in the embodiments of the present invention, which include a computer-readable storage medium, are not limited to the method operations described above, but can also perform related operations in the contract text processing method based on a large language model provided in any embodiment of the present invention.

[0089] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0090] It is worth noting that in the above embodiments of the contract text processing method based on a large language model, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0091] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A contract text processing method based on a large language model, characterized in that, include: The system preprocesses the acquired contract text to be parsed in real time, and then parses the preprocessed contract text using a pre-trained contract text parsing big language model to obtain the current contract text parsing result. The entity is extracted from the parsing result of the current contract text, and the extracted entities are aligned with the knowledge graph using the knowledge graph alignment method. The association risk mining result of the current contract text is then generated by combining the association risk mining method. The dynamic risk analysis intelligent execution suggestion module is used to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text, and combined with the preset risk quantification and level classification methods, a current intelligent execution suggestion report is generated. Based on the current smart execution suggestion report, the blockchain smart contract execution and notarization module automatically converts and processes the data to obtain the target smart contract execution and notarization results, which are then fed back to the user.

2. The method according to claim 1, characterized in that, The real-time preprocessing of the acquired contract text to be parsed includes: The contract text data acquisition and multimodal parsing module is used to collect the contract text to be parsed and the current contract text file format in real time. Based on the current contract text file format, perform multimodal parsing on the contract text to be parsed to obtain the multimodal parsing result of the current contract text; By using a pre-built automatic contract text classification model, the first paragraph of the contract text to be parsed and the multimodal parsing results of the current contract text are automatically identified and classified to obtain the target contract classification result. The automatic contract text classification model is trained based on a standard financial contract corpus to match the corresponding contextual text based on the classification result of the target contract. The current contract text multimodal parsing result is processed by the text structuring reconstruction unit and semantic full masking technology enhancement unit corresponding to the preset text preprocessing and semantic enhancement module to obtain the current contract text preprocessing result.

3. The method according to claim 2, characterized in that, Before parsing the preprocessed current contract text using a pre-trained large language model for contract text parsing to obtain the current contract text parsing result, the process includes: Obtain and pre-train the initially constructed large language model for contract text parsing based on the standard unannotated contract joint description text to obtain the initial large language model for contract text parsing. Based on the standard unlabeled contract joint description text, generate at least one target contract triple dataset; The target contract triplet dataset includes the current extraction instruction, the current input contract text, and the output result of the current extraction instruction. Based on the target contract triplet datasets, which include the current extraction instruction, the current input contract text, and the output result of the current extraction instruction, the initial contract text parsing large language model is retrained under supervision. When the training result meets the preset accuracy requirements, the contract text parsing large language model is confirmed to have been trained.

4. The method according to claim 1 or 3, characterized in that, The text parsing large language model also provides a natural language question-answering interface; After real-time preprocessing of the acquired contract text to be parsed, and parsing of the preprocessed contract text using a pre-trained large language model for contract text parsing to obtain the current contract text parsing result, the process further includes: Through the natural language question-and-answer interface, at least one text question instruction to be answered is received and parsed in real time, and text response results for each target instruction are generated. The textual responses to the target instructions will be presented to the user in a visual feedback manner.

5. The method according to claim 4, characterized in that, The dynamic risk analysis intelligent execution suggestion module is used to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text. Combined with preset risk quantification and level classification methods, a current intelligent execution suggestion report is generated, including: The dynamic risk analysis intelligent execution suggestion module is used to perform risk analysis on the current contract text association risk mining results for text compliance, business logic, and association contagion, respectively, to obtain the current text compliance risk value, the current business logic risk value, and the current association contagion risk value; Obtain the preset text compliance risk weight value, business logic risk weight value, and related contagion risk weight value, and then sum them with the current text compliance risk value, current business logic risk value, and current related contagion risk value respectively to obtain the current multi-dimensional risk value; The current multi-dimensional risk value is processed by a preset risk quantification and classification method to obtain the target risk value level; The target risk level is matched with a preset policy rule base to generate a current intelligent execution suggestion report.

6. The method according to claim 5, characterized in that, The process involves automatically converting the current smart execution suggestion report into a blockchain smart contract execution and notarization module to obtain the target smart contract execution and notarization result, including: The current smart execution suggestion report is automatically converted into code using the blockchain smart contract execution and storage module to obtain the current smart contract code. Use a blockchain oracle to receive the current contract's trigger conditions in real time; If the detection determines that the current contract trigger condition is met, the current smart contract code will be executed, the smart contract execution state will be modified, and the execution and storage results of the target smart contract will be generated.

7. The method according to claim 6, characterized in that, After the current smart execution suggestion report is automatically converted and processed by the blockchain smart contract execution and notarization module to obtain the target smart contract execution and notarization result, and then fed back to the user, the process further includes: The system receives multi-source feedback data results in real time and uses feedback learning and model optimization modules to perform online learning, real-time fine-tuning, and periodic incremental training of the large language model for contract text parsing, so as to achieve optimization of the large language model for contract text parsing.

8. A contract text processing device based on a large language model, characterized in that, include: The current contract text parsing result determination module is used to preprocess the acquired contract text to be parsed in real time, and to parse the preprocessed current contract text preprocessing result through a pre-trained contract text parsing big language model to obtain the current contract text parsing result. The current contract text association risk mining result generation module is used to extract entities from the current contract text parsing result, perform knowledge graph alignment processing on the extracted entities using the knowledge graph alignment method, and generate the current contract text association risk mining result by combining the association risk mining method. The current intelligent execution suggestion report generation module is used to perform multi-dimensional risk analysis on the risk mining results associated with the current contract text using the dynamic risk analysis intelligent execution suggestion module, and generate a current intelligent execution suggestion report by combining the preset risk quantification and level classification methods. The target smart contract execution and notarization result determination and feedback module is used to automatically convert and process the current smart execution suggestion report through the blockchain smart contract execution and notarization module to obtain the target smart contract execution and notarization result, and then feed it back to the user.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a contract text processing method based on a large language model as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute a contract text processing method based on a large language model as described in any one of claims 1-7.

Images