A contract text intelligent comparison method and system based on multimodal feature fusion

Abstract

The present invention belongs to the field of contract text comparison and provides a contract text intelligent comparison method and system based on multimodal feature fusion. The method obtains two contract texts to be compared, preprocesses each contract text, and generates a semantic vector; calculates the surface similarity, logical consistency, and legal effect equivalence between the semantic vectors of the two contract texts; extracts the text mode, digital mode, and logical mode of the contract text respectively, determines the digital mode or logical mode corresponding to the text mode, and compares the modes between the contract texts. During the comparison process, according to the requirements of the comparison task and based on the attention mechanism, the weight of each mode is dynamically allocated, the conflicting position and conflict type of the two contract texts are determined, and the modifications existing in the two and the categories to which they belong are determined and stored. The present invention can improve the accuracy of contract comparison.

Description

Technical Field

[0001] The present invention belongs to the field of contract text comparison, and specifically relates to a contract text intelligent comparison method and system based on multimodal feature fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Contract text comparison technology is a crucial component of legal technology and smart contract review, and is widely used in areas such as corporate legal affairs and financial compliance. Currently, it is typically implemented using computer methods (such as diff algorithms) and document management software.

[0004] However, existing methods can only identify literal differences and cannot determine the semantic equivalence of clauses, resulting in inaccurate recognition of semantically equivalent clauses with different wording. Furthermore, contracts generally contain text, structured data (such as amounts and dates), and clause references. However, existing methods can only process plain text data, which can easily lead to missed detection of numerical errors and clause reference conflicts. Furthermore, existing methods cannot distinguish between formatting adjustments and substantive changes, nor can they trace the intent of revisions.

[0005] In summary, existing technologies are still mainly at the level of text difference comparison and fail to truly achieve semantic-level intelligent comparison, resulting in low contract review efficiency and limited risk identification capabilities. Summary of the Invention

[0006] In order to solve the above problems, the present invention proposes a method and system for intelligent contract text comparison based on multimodal feature fusion. The present invention can accurately identify changes in the legal effect of clauses through semantic-level difference detection, rather than just textual differences; and jointly analyze text, numerical values, and logical dependencies to achieve multimodal risk prevention and control, comprehensively cover risk points, and avoid missed detections; record revision intentions through blockchain, associate changes in the legal environment, and achieve true traceability.

[0007] According to some embodiments, the present invention adopts the following technical solutions:

[0008] A method for intelligent contract text comparison based on multimodal feature fusion, comprising the following steps:

[0009] Obtain the two contract texts that need to be compared and pre-process each contract text;

[0010] Determine the text language and legal system based on the pre-processed contract text, select the first model that has been fine-tuned for the legal system, and generate a semantic vector;

[0011] Calculate the surface similarity, logical consistency, and legal equivalence between the semantic vectors of two contract texts;

[0012] Extract the textual modality, numerical modality, and logical modality of the contract text respectively, determine the numerical modality or logical modality corresponding to the textual modality, and compare the modalities between the contract texts. During the comparison process, based on the requirements of the comparison task and the attention mechanism, dynamically assign weights to each modality to determine the location and type of conflict between the two contract texts;

[0013] Based on the surface similarity, logical consistency, and legal equivalence between the semantic vectors of the two contract texts, as well as the comparison results, the modifications that exist in the two and their categories are determined, and whether the modifications pose legal risks are determined. Blockchain technology is used to store the substantive modifications, the semantic vectors of the two contract texts corresponding to the substantive modifications, and the legal risks.

[0014] As an optional implementation, the first model is pre-trained using a legal corpus containing laws, regulations, judicial interpretations and typical cases. During the training process, the parameters of the first model are fine-tuned based on the differences in expressions of different legal systems.

[0015] As an optional implementation method, the process of determining the text language and legal system based on the preprocessed contract text includes extracting the language of the contract text and the information of the two parties to the contract text, the place where the contract is signed and the place where the contract is performed, and combining the above information to determine the legal system.

[0016] As an optional implementation method, the process of calculating the surface similarity, logical consistency and legal equivalence between the semantic vectors of two contract texts includes: based on the semantic similarity matrix, calculating the literal similarity between the semantic vectors of the two contract texts, preliminarily determining that the part with the literal similarity less than the set value is the suspected modification area, using the pre-trained second model to perform dependency syntax analysis on the suspected modification area, verifying whether the modification of the contract terms affects the overall logic of the contract, marking the suspected modification areas where there are logical conflicts, using the pre-trained third model, combined with current legal rules, to determine whether the suspected modification area changes the rights and obligations relationship, and marking the suspected modification area that changes the rights and obligations relationship.

[0017] As an optional implementation method, the process of extracting the text modality, digital modality and logical modality of the contract text respectively includes: extracting the text modality containing the clause type and legal entity through named entity recognition; using the conditional random field algorithm to identify the digital modality containing the amount, date and associated legal threshold; and marking the clause reference relationship and trigger conditions through a pre-built clause dependency graph to form a logical modality.

[0018] As an optional implementation method, the digital mode or logical mode corresponding to the text mode is determined, and the modes are compared between the contract texts. During the comparison process, according to the comparison task requirements and based on the attention mechanism, the weights of each mode are dynamically allocated. The process includes: after associating the text mode with its corresponding digital mode or logical mode, a detection is performed. If the detection requirement is numerical conflict detection, the weight of the digital mode is increased to a preset first threshold; if the detection requirement is clause equivalence or logical consistency judgment, the weight of the logical mode is increased to a preset second threshold.

[0019] As an optional implementation method, the process of determining the location and type of conflict between two contract texts includes: through a cross-modal constraint propagation mechanism, reversely mapping the rule constraints of the logical modality to the semantic analysis of the text modality and the data verification process of the digital modality, forming a closed-loop verification chain, locating the location where the two contract texts conflict, and the type of conflict generated.

[0020] As an optional implementation method, the process of determining the modifications that exist in the two contracts and their categories, and determining whether the modifications pose legal risks includes: based on the surface similarity, logical consistency, and legal effect equivalence between the semantic vectors of the two contract texts, as well as the comparison results, if the modification of the two contract texts is font format or typesetting adjustment, it is considered not to be a substantive modification; if the text expression is different but the legal effect is the same, it is considered not to be a substantive modification; otherwise, it is considered to be a substantive modification. If the substantive modification does not comply with legal provisions, there is a legal risk.

[0021] An intelligent contract text comparison system based on multimodal feature fusion, including:

[0022] A preprocessing module is configured to obtain two contract texts to be compared and preprocess each contract text;

[0023] a semantic vector generation module configured to determine the text language and legal system based on the preprocessed contract text, select a first model that has been pre-tuned for the legal system, and generate a semantic vector;

[0024] a three-dimensional calculation module configured to calculate surface similarity, logical consistency, and legal effect equivalence between semantic vectors of two contract texts;

[0025] The multimodal comparison module is configured to extract the textual modality, numerical modality, and logical modality of the contract text, determine the numerical modality or logical modality corresponding to the textual modality, and compare the modalities between the contract texts. During the comparison process, based on the requirements of the comparison task and using the attention mechanism, the weights of the modalities are dynamically assigned to determine the locations and types of conflicts between the two contract texts.

[0026] The modification confirmation and storage module is configured to determine the modifications that exist between the two contract texts and their categories based on the surface similarity, logical consistency and legal effect equivalence between the semantic vectors of the two contract texts, as well as the comparison results, and to determine whether the modifications pose legal risks. The module uses blockchain technology to store the substantive modifications, the semantic vectors of the two contract texts corresponding to the substantive modifications, and the legal risks.

[0027] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above method are completed.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention addresses the common problems of semantic ambiguity, reference ambiguity and expression equivalence in contract texts. It calculates from three dimensions: surface similarity, logical consistency and legal effect equivalence, and quantifies the impact of differences in clause wording on the effectiveness of the contract, so as to eliminate false detections caused by different literal expressions and determine whether the modification is within the scope of legal provisions. It breaks through the surface matching limitations of traditional semantic analysis and improves the accuracy and efficiency of contract risk prevention and control.

[0030] The present invention constructs a joint representation space of text, logic, and numbers, adaptively adjusts the weight of each modality according to task requirements, and reversely maps the rule constraints of the logical modality to the text semantic analysis and digital parameter verification process through a cross-modal constraint propagation mechanism, forming a closed-loop verification chain of semantics-logic-parameters, which helps to accurately locate the conflict location and generate an analysis report containing conflict type and legal consequence analysis, helping to improve the accuracy of contract risk identification.

[0031] The present invention uses blockchain technology to store contract modification operations, encapsulating each contract modification operation as an independent blockchain transaction unit. This improves storage efficiency while ensuring that metadata such as modification time, modifier, and content change summary cannot be tampered with.

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0034] Figure 1 The present invention is a flowchart of an intelligent contract text comparison method based on multimodal feature fusion according to an embodiment. DETAILED DESCRIPTION

[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0036] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0038] In the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other.

[0039] Example 1

[0040] An intelligent contract text comparison method based on multimodal feature fusion, such as Figure 1 As shown, the following steps are included:

[0041] Obtain the two contract texts that need to be compared and pre-process each contract text;

[0042] Determine the text language and legal system based on the pre-processed contract text, select the first model that has been fine-tuned for the legal system, and generate a semantic vector;

[0043] Calculate the surface similarity, logical consistency, and legal equivalence between the semantic vectors of two contract texts;

[0044] Extract the textual modality, numerical modality, and logical modality of the contract text respectively, determine the numerical modality or logical modality corresponding to the textual modality, and compare the modalities between the contract texts. During the comparison process, based on the requirements of the comparison task and the attention mechanism, dynamically assign weights to each modality to determine the location and type of conflict between the two contract texts;

[0045] Based on the surface similarity, logical consistency, and legal equivalence between the semantic vectors of the two contract texts, as well as the comparison results, the modifications that exist in the two and their categories are determined, and whether the modifications pose legal risks are determined. Blockchain technology is used to store the substantive modifications, the semantic vectors of the two contract texts corresponding to the substantive modifications, and the legal risks.

[0046] In this embodiment, the first model is pre-trained using a legal corpus containing laws, regulations, judicial interpretations, and typical precedents. This allows the model to grasp the specialized semantics of legal terminology, such as the difference in legal effect between liquidated damages and late payment fees. During the training process, the parameters of the first model are fine-tuned to account for differences in expression across different legal systems.

[0047] Specifically, the parameters of the first model are fine-tuned to address the differences in expression between the civil law system (such as Chinese contracts) and the common law system (such as Anglo-American contracts). The civil law system focuses on strict matching of the literal expression of the clauses, while the common law system focuses on the correlation analysis between the clauses and precedents, generating semantic vectors suitable for different legal systems.

[0048] Among them, the first model can select a semantic encoder model.

[0049] Among them, the process of extracting the legal system to which the contract belongs extracts the language of the contract text and the information of the two parties to the contract text, the place where the contract is signed and the place where the contract is performed, and combines the above information to determine the legal system to which it belongs.

[0050] For example, if the contract text is in Chinese, both parties are Chinese companies, and the place where the contract is signed and the place where the contract is performed are both a province or city in China, then the legal system is determined to be the civil law system.

[0051] If the contract text is in English, the contracting parties are a British company and an American company respectively, the contract is signed in London, UK, and the contract is performed in Leeds, UK, then the legal system is determined to be the common law system.

[0052] In this embodiment, the impact of clause differences on legal effectiveness is calculated using a three-dimensional model. Specifically, the surface similarity, logical consistency, and legal effectiveness equivalence between the semantic vectors of the two contract texts are calculated. Specifically, the following steps are performed:

[0053] Based on the semantic similarity matrix, the literal similarity between the semantic vectors of the two contract texts is calculated, and the parts with literal similarity less than the set value are preliminarily determined to be suspected modification areas. This step is mainly used to eliminate differences caused by inconsistent textual expressions, but they are not actually inconsistent. For example, the expression of a payment period of 30 days is inconsistent with that of a payment period of one month, but their literal similarity is high, so the suspected modification area can be excluded.

[0054] Next, we conduct a logical consistency check. Using the pre-trained second model, we perform dependency syntax analysis on suspected modified areas to verify whether the contract clause changes affect the overall contract logic. We also mark suspected modified areas with logical conflicts. For example, if a payment time change is discovered, we determine whether this change conflicts with the delivery time. If so, we mark it.

[0055] The second model can use the BERT or BiLSTM model to implement dependency syntax analysis and accurately parse the clause text structure.

[0056] Finally, the legal effect equivalence assessment uses a pre-trained third model, combined with current legal rules, to determine whether suspected modifications alter rights and obligations. Any suspected modifications that alter rights and obligations are marked. For example, payment within 30 days is legally equivalent to payment within one month and does not alter rights and obligations.

[0057] The third model can use a graph neural network model to model the logical dependencies between clauses.

[0058] This embodiment systematically quantifies the impact of differences in clause wording on contract effectiveness from three dimensions: surface similarity, logical consistency, and legal effectiveness equivalence. It breaks through the surface matching limitations of traditional semantic analysis, provides an explainable and quantifiable legal effectiveness assessment system for contract review, and significantly improves the accuracy and efficiency of contract risk prevention and control.

[0059] In this embodiment, the process of extracting the text modality, numerical modality, and logical modality of the contract text includes:

[0060] Named entity recognition is used to extract text modalities containing clause types (such as payment clauses, confidentiality clauses, etc.) and legal entities (such as Party A, Party B, contract amount, etc.).

[0061] A conditional random field algorithm is used to identify digital patterns containing amounts, dates, and associated legal thresholds (such as the upper limit of liquidated damages being 30% of the contract amount).

[0062] Through the pre-built clause dependency graph, the clause reference relationship (such as Article 5.1 in the contract references Article 3.2) and trigger conditions (such as the trigger dependency of the "breach of contract clause" on "payment delayed for more than 15 days") are marked to form a logical mode.

[0063] Next, after associating the text modality with its corresponding digital modality or logical modality, a test is performed. If the test requirement is numerical conflict detection, the digital modality weight is increased to the preset first threshold (for example, the digital modality weight is increased to 80% to strengthen parameter verification); if the test requirement is clause equivalence or logical consistency judgment, the logical modality weight is increased to the preset second threshold (for example, the logical modality weight is increased to 70% to strengthen dependency verification).

[0064] In this embodiment, the process of determining the location and type of conflict between two contract texts involves: using a cross-modal constraint propagation mechanism, the rule constraints of the logical modality are reverse-mapped to the semantic parsing of the text modality and the data verification process of the digital modality. This forms a closed-loop verification chain of text semantics, logical relationships, and numerical parameters, locating the location of the conflict between the two contract texts and the type of conflict. For example, if the textual clause states "payment in installments" but the specific content is "a lump-sum payment amount," this is considered a conflict, the conflict location is automatically located, and the conflict type is marked, i.e., a logical contradiction.

[0065] This embodiment can adaptively adjust the contribution / weight of each modality according to task requirements (such as conflict detection and clause equivalence determination), so that the corresponding modality dominates during verification. Through the cross-modal constraint propagation mechanism, the rule constraints of the logical modality are reversely mapped to the text semantic analysis and digital parameter verification process, forming a closed-loop verification chain of semantics-logic-parameters. It can accurately locate specific conflicting locations such as clause paragraphs and digital cells, and generate a report containing the types of contradictions (such as semantic ambiguity, logical contradictions or parameter exceeds the limit, etc.), which is helpful for the subsequent analysis of legal consequences and the generation of explainability reports.

[0066] The above solution, through dynamic inter-modal collaboration and conflict tracing mechanisms, provides the legal technology field with an intelligent solution that combines high precision and strong explainability, significantly promoting the transition of contract management from manual review to automation, precision, and traceability.

[0067] The process of determining whether any modifications exist and their categories, and whether any modifications pose legal risks, includes: Based on the superficial similarity, logical consistency, and legal equivalence of the semantic vectors of the two contract texts, as well as the comparison results, if the modification of the two contract texts is a font format or typesetting adjustment, it is considered not a substantive modification. That is, the font or typesetting change does not affect the semantics or legal effect, such as Party A changing its font from "bold" to "Songti".

[0068] If the text is different but the legal effect is the same, it is not considered a substantive modification; for example, the payment period is changed from "30 days" to "one month".

[0069] Others are substantive modifications, such as changes in rights and obligations. If the substantive modifications do not comply with legal provisions, such as changing "liquidated damages are 10% of the actual loss" to "liquidated damages are 80% of the actual loss", and exceed the statutory upper limit (generally speaking, the liquidated damages upper limit does not exceed 30% of the actual loss), there will be legal risks.

[0070] In this embodiment, based on the consortium chain architecture, each modification is encapsulated as a blockchain transaction unit and recorded. The recorded content includes:

[0071] Metadata: modifier, modification time, and semantic vector hash before and after modification;

[0072] Incremental content: Only substantial changes / modifications are stored (not the full text), improving storage efficiency by over 60%.

[0073] In this embodiment, a visual report is generated based on the comparison results. For example, different color codes are used to distinguish format adjustments, semantically equivalent modifications, and substantive changes, and legal risks are noted. Based on experience, compliance recommendations can also be given. For example, if "liquidated damages are 80% of the actual loss," it can be recommended to adjust to "liquidated damages are 25% of the actual loss." Modification intentions can also be traced back to any historical version through the blockchain timeline.

[0074] Example 2

[0075] An intelligent contract text comparison system based on multimodal feature fusion, including:

[0076] A preprocessing module is configured to obtain two contract texts to be compared and preprocess each contract text;

[0077] a semantic vector generation module configured to determine the text language and legal system based on the preprocessed contract text, select a first model that has been pre-tuned for the legal system, and generate a semantic vector;

[0078] a three-dimensional calculation module configured to calculate surface similarity, logical consistency, and legal effect equivalence between semantic vectors of two contract texts;

[0079] The multimodal comparison module is configured to extract the textual modality, numerical modality, and logical modality of the contract text, determine the numerical modality or logical modality corresponding to the textual modality, and compare the modalities between the contract texts. During the comparison process, based on the requirements of the comparison task and using the attention mechanism, the weights of the modalities are dynamically assigned to determine the locations and types of conflicts between the two contract texts.

[0080] The modification confirmation and storage module is configured to determine the modifications that exist between the two contract texts and their categories based on the surface similarity, logical consistency and legal effect equivalence between the semantic vectors of the two contract texts, as well as the comparison results, and to determine whether the modifications pose legal risks. The module uses blockchain technology to store the substantive modifications, the semantic vectors of the two contract texts corresponding to the substantive modifications, and the legal risks.

[0081] Example 3

[0082] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the method mentioned in the first embodiment are completed.

[0083] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of one or more computer-usable storage media (including but not limited to disk storage, CD - ROM , optical storage, etc.).

[0084] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0085] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0087] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made by those skilled in the art that fall within the spirit and principles of the present invention and do not require creative effort are intended to be within the scope of protection of the present invention.

Claims

1. A contract text intelligent comparison method based on multimodal feature fusion, characterized by: The following steps are involved: Obtain the two contract texts that need to be compared and pre-process each contract text; Determine the text language and legal system based on the pre-processed contract text, select the first model that has been fine-tuned for the legal system, and generate a semantic vector; Calculate the surface similarity, logical consistency, and legal equivalence between the semantic vectors of two contract texts; Extract the textual modality, numerical modality, and logical modality of the contract text respectively, determine the numerical modality or logical modality corresponding to the textual modality, and compare the modalities between the contract texts. During the comparison process, based on the requirements of the comparison task and the attention mechanism, dynamically assign weights to each modality to determine the location and type of conflict between the two contract texts; Based on the surface similarity, logical consistency, and legal equivalence between the semantic vectors of the two contract texts, as well as the comparison results, the modifications that exist in the two and their categories are determined, and whether the modifications pose legal risks are determined. Blockchain technology is used to store the substantive modifications, the semantic vectors of the two contract texts corresponding to the substantive modifications, and the legal risks.

2. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: The first model is pre-trained using a legal corpus containing laws, regulations, judicial interpretations and typical cases. During the training process, the parameters of the first model are fine-tuned based on the differences in expressions of different legal systems.

3. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: The process of determining the language and legal system of the contract text based on the pre-processed contract text includes extracting the language of the contract text and information about the contracting parties, the place where the contract was signed, and the place where the contract was performed, and then determining the legal system based on the above information.

4. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: The process of calculating the surface similarity, logical consistency and legal equivalence between the semantic vectors of two contract texts includes: calculating the literal similarity between the semantic vectors of the two contract texts based on the semantic similarity matrix, preliminarily determining that the part with the literal similarity less than the set value is the suspected modification area, using the pre-trained second model to perform dependency syntax analysis on the suspected modification area, verifying whether the modification of the contract terms affects the overall logic of the contract, marking the suspected modification area with logical conflicts, using the pre-trained third model, combining with current legal rules, judging whether the suspected modification area changes the rights and obligations relationship, and marking the suspected modification area that changes the rights and obligations relationship.

5. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: The process of extracting the text modality, digital modality and logical modality of the contract text includes: extracting the text modality containing the clause type and legal entity through named entity recognition; using the conditional random field algorithm to identify the digital modality containing the amount, date and associated legal threshold; and using the pre-built clause dependency graph to mark the clause reference relationship and trigger conditions to form the logical modality.

6. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: Determine the digital mode or logical mode corresponding to the text mode, and compare the modes between the contract texts. During the comparison process, according to the comparison task requirements and based on the attention mechanism, the process of dynamically allocating the weights of each mode includes: after associating the text mode with its corresponding digital mode or logical mode, perform detection. If the detection requirement is numerical conflict detection, increase the weight of the digital mode to a preset first threshold; if the detection requirement is clause equivalence or logical consistency judgment, increase the weight of the logical mode to a preset second threshold.

7. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: The process of determining the location and type of conflict between two contract texts includes: through the cross-modal constraint propagation mechanism, reversely mapping the rule constraints of the logical modality to the semantic analysis of the text modality and the data verification process of the digital modality, forming a closed-loop verification chain, locating the location of the conflict between the two contract texts, and the type of conflict generated.

8. The intelligent contract text comparison method based on multimodal feature fusion according to claim 1 is characterized in that: The process of determining the modifications that exist in the two contracts and their categories, and determining whether the modifications pose legal risks includes: based on the surface similarity, logical consistency, and legal effect equivalence between the semantic vectors of the two contract texts, as well as the comparison results, if the modifications to the two contract texts are font format or typesetting adjustments, they are considered not to be substantive modifications; if the text expressions are different but the legal effect is the same, they are considered not to be substantive modifications; other modifications are substantive modifications, and if the substantive modifications do not comply with legal provisions, there are legal risks.

9. An intelligent contract text comparison system based on multimodal feature fusion, characterized by: include: A preprocessing module is configured to obtain two contract texts to be compared and preprocess each contract text; a semantic vector generation module configured to determine the text language and legal system based on the preprocessed contract text, select a first model that has been pre-tuned for the legal system, and generate a semantic vector; a three-dimensional calculation module configured to calculate surface similarity, logical consistency, and legal effect equivalence between semantic vectors of two contract texts; The multimodal comparison module is configured to extract the textual modality, numerical modality, and logical modality of the contract text, determine the numerical modality or logical modality corresponding to the textual modality, and compare the modalities between the contract texts. During the comparison process, based on the requirements of the comparison task and using the attention mechanism, the weights of the modalities are dynamically assigned to determine the locations and types of conflicts between the two contract texts. The modification confirmation and storage module is configured to determine the modifications that exist between the two contract texts and their categories based on the surface similarity, logical consistency and legal effect equivalence between the semantic vectors of the two contract texts, as well as the comparison results, and to determine whether the modifications pose legal risks. The module uses blockchain technology to store the substantive modifications, the semantic vectors of the two contract texts corresponding to the substantive modifications, and the legal risks.

10. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the method according to any one of claims 1 to 8 are completed.

Images