Methods, devices, equipment, and media for generating risk management solutions for contract texts

By combining semantic parsing and a multi-level classifier model with a multi-source knowledge base, the problem of cost processing deviations caused by unstructured language in cross-border service contracts is solved, and efficient and accurate risk management solutions are generated.

CN123117084APending Publication Date: 2026-11-17SHENZHEN MINGXIN DIGITAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611628500.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The issues include the qualitative bias in manual interpretation of cost processing caused by unstructured language in cross-border service contract texts, and the low clause hit rate of existing automated solutions.

Method used

Key contract elements are extracted through semantic parsing, dynamic clause matching is performed using a multi-source knowledge base, the nature of the contract is determined using a multi-level classifier model, and risk management solutions are generated by combining the risk assessment of the beneficial owner.

Benefits of technology

It improves the accuracy of the application of clauses, reduces the instability and risk of oversight in human judgment, ensures that the qualitative results are consistent with the actual contractual agreement, and provides a traceable and complete chain.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117084A_ABST
    Figure CN123117084A_ABST
Patent Text Reader

Abstract

This application relates to the field of risk management scheme generation technology for contract texts, and discloses a method, apparatus, equipment, and medium for generating risk management schemes for contract texts. The method includes: acquiring a cross-border service contract text and parsing to extract key elements; performing dynamic clause matching in a multi-source knowledge base to determine applicable fee agreement clauses and conflict resolution identifiers; using a multi-level classifier model to obtain the obtained nature determination result; combining the beneficial owner's risk assessment result to determine the applicable fee path; matching a target declaration document template based on the path, nature determination, and risk assessment result; extracting legal basis information to fill the target declaration document template to generate a declaration document and a legal basis summary, thus obtaining a risk management scheme. This achieves dynamic mapping of complex fee agreement clauses, automatically generates risk management schemes that meet strict regulatory requirements, and improves the automation level, consistency of judgment, and compliance reliability of declaration materials in cross-border fee cost processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of risk management scheme generation technology for contract texts, and in particular to a method, apparatus, equipment and medium for generating risk management schemes for contract texts. Background Technology

[0002] The processing of fees and costs in cross-border service contracts has long relied on manual review of bilateral fee agreements and relevant legal provisions, with personnel determining the nature of the income and selecting applicable fees and costs based on the contract descriptions. However, cross-border service contracts are often written in unstructured legal language, and the wording of the same service content varies significantly across different contracts. Manual interpretation is prone to qualitative biases due to nested clause references (such as the blurred boundaries between applicable royalties and operating profits). Existing automated solutions generally use static rule bases for keyword matching, failing to dynamically analyze the deep connections between contract semantics and agreement clauses, resulting in a low hit rate for applicable clauses. Summary of the Invention

[0003] Therefore, it is necessary to propose a method, device, equipment, and medium for generating risk management solutions for contract texts to address the problem of low hit rate of existing applicable clauses.

[0004] A method for generating risk management solutions for contract texts, the method comprising: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; Based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label and clause conflict resolution identifier; The matching results of the terms and the key elements of the contract are input into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; Obtain the beneficial owner risk assessment results, and determine the applicable cost path based on the nature determination results and the beneficial owner risk assessment results; Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in the preset document template database; The corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is filled to generate the declaration document and legal basis summary, thus obtaining the risk disposal plan.

[0005] Furthermore, the step of dynamically matching clauses in a pre-defined multi-source knowledge base based on the key elements of the contract to determine the applicable fee agreement clauses and output the clause matching results includes: Extract the service type tags from the key elements of the contract and calculate their similarity with the clause type tags in the multi-source knowledge base; Extract the tags of target clauses with similarity values ​​greater than a preset similarity value to form a set of candidate clauses; When the candidate clause set contains two or more target clause tags, conflict resolution is carried out in sequence according to the special law priority rule, the purpose interpretation rule, and the mutual negotiation procedure initiation rule to determine the final fee agreement clause; Output the terms matching results based on the aforementioned fee agreement terms.

[0006] Furthermore, the step of obtaining the risk assessment results for the beneficial owner includes: Obtain substantive operational assessment data of the transaction entity; wherein, the substantive operational assessment data includes at least one or more of the following: registered capital amount, personnel employment records, and information on the location of the actual management institution; Input the substantive operational assessment data into a preset quantitative scoring model to calculate the benefit owner's score; Based on the comparison between the beneficiary's score and a preset threshold, a risk level label for the beneficiary is output, which includes at least three levels: low risk, medium risk, and high risk.

[0007] Furthermore, before the step of inputting the clause matching result and the key elements of the contract into a preset multi-level classifier model to obtain the obtained property determination result, the method further includes: Obtain historical cost case samples and the corresponding attribute annotation results as positive samples; Generative adversarial networks are used to generate adversarial samples with blurred boundaries as negative samples; The positive and negative samples are input into the initial classifier model for iterative training until the confidence level of the initial classifier model's output for the obtained property determination reaches a preset threshold, thereby obtaining the multi-layer classifier model.

[0008] Furthermore, the step of determining the applicable cost path based on the nature determination result and the beneficial owner risk assessment result includes: When the result of the determination of the nature of the obtained product is operating profit, the first expense path shall be applied. When the nature of the income is determined to be a royalty or technical service fee, and the risk assessment result of the beneficial owner is low risk, the second fee path is determined to apply. When the nature determination result is a royalty or technical service fee, and the beneficial owner risk assessment result is medium or high risk, the first fee path is determined to apply, the current process is suspended, and a correction notice is output; wherein, the correction notice is used to instruct the user to supplement the supporting materials corresponding to the beneficial owner risk assessment; until the correction completion instruction is received, the suspension is lifted and the beneficial owner risk assessment is re-executed.

[0009] Furthermore, after the step of determining the application of the first fee path, suspending the current process, and outputting a correction notice when the obtained nature determination result is a royalty or technical service fee, and the beneficial owner's risk assessment result is medium or high risk, the process further includes: When the risk assessment result of the beneficial owner changes to low risk after reassessment, the applicable cost path will be switched to the second cost path; If the reassessed risk assessment result for the beneficial owner is still medium or high risk, maintain the first cost path and issue a correction notice again.

[0010] Further, the step of matching the target declaration document template in a preset document template database based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result includes: The applicable cost path, the result of the determination of the nature of the income, and the risk assessment result of the beneficial owner are combined to generate a joint state; Using the joint state as the matching key, a document template that has a preset mapping relationship with the joint state is retrieved from the document template database to obtain the target declaration document template.

[0011] A risk management solution generation device for contract text, the device comprising: The extraction module is used to acquire the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements. The determination module is used to perform dynamic clause matching in a preset multi-source knowledge base based on the key elements of the contract, determine the applicable fee agreement clauses, and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label, and clause conflict resolution identifier; The input module is used to input the matching results of the terms and the key elements of the contract into a preset multi-level classifier model to obtain the obtained nature determination result; wherein, the obtained nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; The acquisition module is used to acquire the risk assessment results of the beneficial owner and determine the applicable cost path based on the nature determination results and the risk assessment results of the beneficial owner; The matching module is used to match the target declaration document template in a preset document template database based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result. The filling module is used to extract corresponding legal basis information from the multi-source knowledge base, fill the target declaration document template to generate the declaration document and legal basis summary, and obtain the risk disposal plan.

[0012] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; Based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label and clause conflict resolution identifier; The matching results of the terms and the key elements of the contract are input into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; Obtain the beneficial owner risk assessment results, and determine the applicable cost path based on the nature determination results and the beneficial owner risk assessment results; Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in the preset document template database; The corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is filled to generate the declaration document and legal basis summary, thus obtaining the risk disposal plan.

[0013] A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; Based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label and clause conflict resolution identifier; The matching results of the terms and the key elements of the contract are input into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; Obtain the beneficial owner risk assessment results, and determine the applicable cost path based on the nature determination results and the beneficial owner risk assessment results; Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in the preset document template database; The corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is filled to generate the declaration document and legal basis summary, thus obtaining the risk disposal plan.

[0014] The beneficial effects of this invention are as follows: By transforming unstructured contract text into computable structured elements through semantic parsing, subsequent clause matching can be accurately retrieved based on clear element tags, avoiding qualitative biases caused by semantic ambiguity during manual reading, thereby improving the accuracy of clause application. Furthermore, the clause matching results and key contract elements are input into a multi-layer classifier model, ensuring that the resulting nature determination not only relies on static clause tags but also incorporates the semantic context of the contract, guaranteeing consistency between the qualitative results and the actual contractual agreement. Simultaneously, the substantive operational assessment data of the transaction entities is independently introduced into the risk assessment, allowing the determination of cost paths to consider both the type of income and the entity's risk. Finally, based on the determined path, nature determination, and risk assessment results, a document template is jointly matched and the corresponding legal basis is extracted for filling, creating a traceable and complete chain from contract input to risk disposal output, effectively reducing the instability and oversight risks associated with manual item-by-item judgment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0016] in: Figure 1 This is a diagram illustrating the application environment of a method for generating risk management solutions for contract text in one embodiment. Figure 2 A flowchart of a method for generating a risk management plan for contract text in one embodiment; Figure 3This is a structural block diagram of a risk management scheme generation device for contract text in one embodiment; Figure 4 This is a structural block diagram of an electronic device in one embodiment. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Figure 1 Generate an application environment diagram for the risk management scheme of the contract text in one embodiment. (Refer to...) Figure 1 The method for generating risk mitigation plans for contract texts is applied to a risk mitigation plan generation system for contract texts. This system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal; a mobile terminal can be at least one of a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers.

[0019] like Figure 2 As shown, in one embodiment, a method for generating a risk management plan for a contract text is provided. This method can be applied to both terminals and servers; this embodiment illustrates its application to a terminal. The method for generating a risk management plan for a contract text specifically includes the following steps: S1: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; S2: Based on the key elements of the contract, perform dynamic clause matching in a preset multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label and clause conflict resolution identifier; S3: Input the matching results of the terms and the key elements of the contract into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; S4: Obtain the risk assessment results of the beneficial owner, and determine the applicable cost path based on the nature determination results and the risk assessment results of the beneficial owner; S5: Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, match the target declaration document template in the preset document template database; S6: Extract the corresponding legal basis information from the multi-source knowledge base, fill the target declaration document template to generate the declaration document and legal basis summary, and obtain the risk disposal plan.

[0020] As described in step S1 above, the cross-border service contract text to be processed is obtained, and semantic parsing is performed on the contract text to extract key contract elements. This involves structurally extracting information related to cost determination from unstructured text data. The received cross-border service contract text is typically in the form of PDF (Portable Document Format), Word (Microsoft Word document format), or scanned images. Semantic parsing of the contract text can refer to using natural language processing technology to perform word segmentation, syntactic analysis, and entity recognition to understand the legal meaning of the contract terms. Extracting key contract elements specifically includes identifying fields such as service type (e.g., software license, technical consulting), payment method (e.g., one-time payment, percentage of sales), service geographical restrictions, confidentiality clauses, and intellectual property ownership. For example, numerical information such as contract amount and date can be extracted using rule-based regular expressions; alternatively, business terms such as franchise and technical services can be extracted using a pre-trained named entity recognition model.

[0021] As described in step S2 above, based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results. The clause matching results include at least the matched fee agreement clause number, clause type label, and clause conflict resolution identifier. The pre-set multi-source knowledge base is the core data support for legal reasoning in this method, and its construction is completed during the system initialization phase. The construction of this knowledge base includes: collecting original legal texts such as bilateral fee agreement texts, relevant regulations, and contracting party reservations published on relevant websites of various countries; using natural language processing technology to extract entities and relationships from the above texts, constructing a legal clause knowledge graph containing nodes such as clause number, scope of application, cost limits, and conditional constraints. Its internal structure or format can, for example, use a resource description framework triplet storage, describing the reference and conflict relationships between legal provisions in the form of clause-relationship-clause, and supporting retrieval via graph query language. In actual operation, this knowledge base supports incremental maintenance based on regulatory updates to ensure the timeliness of the legal basis, with an update frequency of no less than once per quarter. The updated knowledge base version number and effective date are recorded together. Dynamic clause matching within a pre-defined multi-source knowledge base can refer to comparing the semantic similarity of extracted key contract elements with legal clauses in the knowledge base. This process could involve, for example, calculating the cosine similarity between the text vectors of the key contract elements and the text vectors of each clause; or it could be based on a predefined rule engine, filtering potential clauses by matching keywords (such as royalties, trade secrets). Determining the applicable fee agreement clause involves selecting the legal clause that best fits the current contract from the comparison results. The output clause matching result includes the matched fee agreement clause number, clause type label, and clause conflict resolution identifier. The clause number uniquely identifies the specific legal provision; the clause type label identifies the nature of the clause (such as royalties, operating profits); and the clause conflict resolution identifier records whether the matching process resolved applicable conflicts between different legal provisions (e.g., special law prevails over common law), and the specific resolution strategy. By outputting results containing the above information, the user is clearly informed of the legal clauses upon which the current contract is based and their applicable logic.

[0022] As described in step S3 above, the clause matching results and key contract elements are input into a preset multi-layer classifier model to obtain the result of the nature determination. This result includes the probability distribution of whether the service belongs to royalty, technical service fee, or operating profit. The preset multi-layer classifier model is used to automatically determine the nature of income for cross-border services. The model's training is completed offline: historical fee case samples and corresponding income nature labeling results (e.g., labeled as royalty, technical service fee, or operating profit) are collected as positive samples; simultaneously, a generative adversarial network is used to generate adversarial samples with ambiguous boundaries as negative samples to enhance the model's robustness; the positive and negative samples are input into the initial classifier model for iterative training until the initial classifier model's output confidence in determining the nature of income reaches a preset threshold, thus obtaining the multi-layer classifier model. The specific structure of the model can be, for example, a deep neural network containing an input layer, several hidden layers (e.g., fully connected layers or long short-term memory network layers), and an output layer. The output layer uses a Softmax activation function (a normalized exponential function that maps vectors to probability distributions) to output the probability of each category. During the inference phase, the matching results of the clauses (such as clause number and type label) and key elements of the contract (such as service content description and payment method) are fused into features and converted into numerical vectors or text sequences acceptable to the model. These are then input into the model, which calculates the probability values ​​of the service belonging to different income categories through forward propagation. The resulting nature determination includes a probability distribution of whether the service belongs to royalty, technical service fee, or operating profit. For example, the output format is [0.1, 0.2, 0.7], representing the probabilities of each category. By outputting probability distributions rather than single labels, the confidence level of the determination can be intuitively displayed, providing a quantitative basis for subsequent risk decisions.

[0023] As described in step S4 above, the beneficial owner risk assessment results are obtained, and the applicable cost path is determined based on the nature determination results and the beneficial owner risk assessment results. Obtaining the beneficial owner risk assessment results may refer to a quantitative risk assessment of the actual business operations of the transaction entity. This assessment process is typically based on substantive business assessment data, including registered capital amount, personnel employment records, and information on the location of the actual management institution. A preset quantitative scoring model is used to score the above data, and based on the comparison of the score with preset thresholds, a risk level label such as low risk, medium risk, or high risk is output. Determining the applicable cost path based on the nature determination results and the beneficial owner risk assessment results is a decision-making process that combines legal nature determination with anti-tax avoidance testing. The applicable cost path determines the subsequent cost processing method and the type of declaration documents. The decision-making logic could be as follows: when the nature of the income is determined to be operating profit, it is usually classified as profit related to a permanent establishment, and the first expense path (such as the regular corporate income tax filing path) applies; when the nature of the income is determined to be royalties or technical service fees, and the beneficial owner's risk assessment result is low risk, the second expense path (such as the path of enjoying preferential treatment under the fee treaty) applies; when the nature of the income is determined to be royalties or technical service fees, and the risk assessment result is medium or high risk, to avoid the risk of abusing the treaty, the first expense path applies, and a risk warning mechanism is triggered. "Expense path" refers to the method of handling cross-border service fee costs determined based on the nature of the income and the beneficial owner's risk assessment result, including the first expense path applying domestic law benchmark costs and the second expense path applying preferential treatment under the fee treaty.

[0024] As described in step S5 above, based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in a pre-set document template database. This pre-set document template database stores standardized declaration document frameworks required for different cost processing scenarios. The database can be constructed, for example, by pre-defining various document templates, such as the "XX Management Measures" and "Beneficial Owner Information Collection Form," and configuring applicable triggering conditions for each template, according to regulations such as the "XX Management Measures." The process of matching the target declaration document template involves associating the current decision state with the templates in the database. Specifically, the applicable cost path, the obtained nature determination result, and the beneficial owner risk assessment result are combined to generate a joint state (e.g., royalties + second path + low risk). Using this joint state as the matching key, document templates with a pre-set mapping relationship to this joint state are retrieved from the document template database.

[0025] As described in step S6 above, corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is populated to generate the declaration document and legal basis summary, thus obtaining the risk management plan. Extracting corresponding legal basis information from the multi-source knowledge base can refer to retrieving and extracting relevant legal provisions, official annotations, and case law support from the knowledge base based on the aforementioned determined fee agreement clauses and nature determination results. For example, if Article 12 of the AB Regional Agreement applies, the original text of the definition of royalties and the cost limit provisions of that clause are extracted. Populating the target declaration document template to generate the declaration document and legal basis summary involves automatically filling the extracted structured data and legal text into the corresponding placeholders in the target declaration document template. For example, this process can utilize a document generation engine to fill in key contract elements, clause matching results, and nature determination probabilities into the data items of the declaration form; simultaneously, the legal basis information is summarized and formatted to generate a legal basis summary paragraph. Finally, the integrated set of documents constitutes a complete risk management plan. This solution not only includes final cost handling recommendations, but also provides a complete chain of evidence and legal support, enabling users to directly use this solution to declare costs or respond to cost audits.

[0026] In the aforementioned technical process, a pre-set multi-source knowledge base provides accurate mapping of legal provisions, a multi-layer classifier model quantifies the nature of income, and the beneficial owner's risk assessment results serve as a key constraint, collectively determining the applicable cost path. These three elements work synergistically to automatically identify the optimal path in complex cross-border service scenarios that complies with cost agreements while effectively mitigating anti-tax avoidance risks, and automatically generate compliant declaration documents based on this path. This end-to-end automated processing, from semantic parsing to document generation, improves the efficiency and accuracy of cross-border cost processing.

[0027] In one embodiment, step S2, which involves dynamically matching clauses in a pre-defined multi-source knowledge base based on the key elements of the contract to determine applicable fee agreement clauses and output the clause matching results, includes: S201: Extract the service type tags from the key elements of the contract and calculate the similarity between them and the clause type tags in the multi-source knowledge base; S202: Extract the tags of target clauses with similarity values ​​greater than the preset similarity value to form a set of candidate clauses; S203: When the candidate clause set contains two or more target clause tags, conflict resolution shall be carried out in sequence according to the special law priority rule, the purpose interpretation rule, and the mutual negotiation procedure initiation rule to determine the final fee agreement clause; S204: Output the terms matching result according to the aforementioned fee agreement terms.

[0028] As described in step S201 above, service type tags are extracted from the key elements of the contract and their similarity is calculated with the clause type tags in the multi-source knowledge base. The service type tags are semantic summaries of the core transaction content in cross-border service contracts, covering textual descriptions of specific business forms such as proprietary technology licensing, customized software development, and management consulting. The process of extracting service type tags from the unstructured text of the contract relies on natural language processing technology to perform entity recognition and relation extraction, abstracting the service descriptions scattered throughout the various clauses of the contract into standardized tag text. The clause type tags in the multi-source knowledge base are standardized indexes of the various payment types (such as royalty clauses, operating profit clauses, and technology service fee clauses) in bilateral fee agreements. The similarity calculation aims to quantify the semantic closeness between the contract text description and the definition of the agreement clauses. This calculation process can, for example, employ a cosine similarity algorithm based on word vectors, or a semantic similarity measurement method based on a pre-trained language model. By mapping service type tags to a high-dimensional vector space and calculating the spatial distance between them and the vectors of each clause type tag, a numerical indicator representing the degree of matching between the two can be obtained, thus providing a quantitative basis for subsequent clause selection.

[0029] As described in step S202 above, target clause tags with similarity values ​​greater than a preset similarity value are extracted to form a candidate clause set. The preset similarity value is a threshold used to define whether service content and specific clauses have a potential connection; its specific value is based on statistical analysis or empirical setting of historical matching cases. All similarity values ​​calculated in the above steps are compared against a threshold, and clause tags with values ​​exceeding the preset similarity value are marked as target clause tags. All target clause tags that meet the threshold conditions are aggregated to form a candidate clause set. This candidate clause set may contain a single element or multiple elements, corresponding to a single matching path or multiple potential matching paths, respectively. When the candidate clause set contains only one target clause tag, it is usually directly identified as an applicable fee agreement clause; however, when it contains two or more target clause tags, it indicates ambiguity or overlap in the application of the clauses, requiring the introduction of a deeper conflict resolution mechanism.

[0030] As described in step S203 above, when the candidate clause set contains two or more target clause tags, conflict resolution is carried out sequentially according to the special law priority rule, the purpose interpretation rule, and the mutual consultation procedure initiation rule to determine the final fee agreement clause. The situation where the candidate clause set contains two or more target clause tags triggers the clause conflict resolution mechanism. The special law priority rule refers to the principle that, in the application of law, when both general provisions and special provisions address the same matter but their contents are inconsistent, the special provisions take precedence. In the context of international fee agreements, this is reflected in the fact that specific clauses concerning specific types of income or specific industries (such as special clauses concerning royalties) take precedence over general clauses (such as general clauses concerning operating profits). If the special law priority rule can clearly distinguish that one clause has the legal status of priority application, then that clause is determined as the final fee agreement clause.

[0031] When the special law priority rule fails to effectively resolve conflicts, the purposive interpretation rule is adopted as a secondary judgment logic. This rule requires exploring the legislative purpose and contractual intent of the fee agreement, and analyzing the economic substance and theoretical objectives behind the clauses. For example, for certain service activities in borderline areas, by comparing different clauses, the type of clause that most accurately reflects the economic substance of the transaction can be selected. This process can, for example, utilize a constructed fee agreement knowledge graph, trace the legislative background annotations and OECD model annotations of the clauses, and conduct semantic correlation analysis on the applicable boundaries of conflicting clauses to determine the only clause that best meets the purpose of the agreement. When the aforementioned two levels of rules cannot resolve clause conflicts, or when there are serious ambiguities in the interpretation of the clauses, the mutual consultation procedure initiation provision is implemented as the final handling strategy. It is determined that a definitive decision cannot be made through automated logic, and human intervention or an official adjudication mechanism needs to be introduced. The result of conflict resolution will output a mutual consultation procedure initiation indicator, prompting the user that the case is in a highly complex or ambiguous area, and suggesting that a consultation application be filed with the relevant department in accordance with the mutual consultation procedure clause in the fee agreement.

[0032] As described in step S204 above, the clause matching results are output based on the fee agreement terms. Specifically, after determining the final fee agreement terms, a structured clause matching result is generated as output. This result provides a definitive mapping of information by associating ambiguous contract text with complex legal text. The clause matching result includes at least the matched fee agreement clause number (e.g., Clause 12 of the AB Region Fee Agreement), clause type label (e.g., royalties), and clause conflict resolution indicator (indicating whether a conflict resolution process and path occurred). This information collectively constitutes the core input data for subsequent determination of the nature of the fee cost and document generation. By outputting clause matching results containing rich metadata, not only is the applicable legal basis of the current contract clarified, but a complete decision-making logic chain is also provided to cost auditors, facilitating subsequent risk review and compliance verification.

[0033] In one embodiment, step S4 of obtaining the risk assessment results for the beneficial owner includes: S401: Obtain substantive operational assessment data of the transaction entity; wherein, the substantive operational assessment data includes at least one or more of the following: registered capital amount, personnel employment records, and information on the location of the actual management institution; S402: Input the substantive business assessment data into a preset quantitative scoring model to calculate the benefit owner score; S403: Based on the comparison result between the beneficiary's score and the preset threshold, output the beneficiary's risk level label, which includes at least three levels: low risk, medium risk, and high risk.

[0034] As described in step S401 above, substantive operational assessment data of the transaction entity is obtained. This substantive operational assessment data includes at least one or more of the following: registered capital amount, personnel employment records, and information on the location of the actual management institution. The substantive operational assessment data of the transaction entity (i.e., the recipient of the cross-border service fee) is collected through multiple channels to verify its qualification to independently conduct substantive business activities. This substantive operational assessment data is not a single-dimensional financial indicator, but rather a structured data sequence encompassing legal entity attributes, human resource allocation, and physical existence, aiming to reflect the commercial substance of the enterprise from different perspectives.

[0035] Specifically, the registered capital amount is derived from the business registration documents or annual audit report, representing the company's equity capital base and used to measure its financial capacity to bear operational risks; personnel employment records are obtained from social security payment lists or human resource management systems, including the number of registered employees, job distribution, and salary payment details, used to verify whether the company has the human resources support required to carry out relevant business; information on the location of the actual management institution is obtained through office space lease contracts, utility bill payment records, or GPS coordinate verification, used to confirm the location for making substantive business decisions and daily management.

[0036] As described in step S402 above, the substantive business assessment data is input into a preset quantitative scoring model to calculate the beneficial owner score. The preset quantitative scoring model is an assessment framework built upon rule-based algorithms and statistical analysis, used to quantitatively represent the substantive business status of the transaction entity. The construction process of this quantitative scoring model can, for example, during the system initialization phase, use a large amount of enterprise sample data accumulated in a historical anti-tax avoidance case database to determine the weight coefficients of each assessment indicator (such as registered capital, number of employees, and consistency of management location) through expert scoring or machine learning algorithms. Alternatively, it can be calibrated using the analytic hierarchy process combined with confirmed conduit company samples from the historical anti-tax avoidance case database. The calibrated weight coefficients have an AUC value of no less than 0.85 on the validation set, thereby establishing a multi-dimensional weighted calculation model. During the calculation process, the acquired substantive business assessment data is first preprocessed, for example, by calculating the ratio of registered capital to contract transaction amount, converting personnel employment records to full-time equivalents, and matching and verifying the location information of the actual management institution with the geographical coordinates of the contracting parties to the cost agreement. Subsequently, each indicator is scored according to the preset scoring logic. For example, this dimension will be assigned a higher negative weight when the actual management location is confirmed to be in a cost-advantaged area and lacks a reasonable business purpose; a deduction mechanism will also be triggered when the registered capital is significantly lower than the conventional operating capital level of similar enterprises. This calculation process can be formally represented as the following decision function: .

[0037] In the formula, This indicates the rating of all beneficiaries. This indicates the total number of dimensions of the evaluation indicators. Indicates the first Input of substantive operational assessment data in multiple dimensions Indicates that for the first The normalized scoring function pre-defined for the dimensional data. Indicates the first The weighting coefficients corresponding to the dimensional indicators. Through the above weighted summation calculation, a scalar score that comprehensively reflects the eligibility of the beneficial owners of the transaction entities is generated. The current results provide a quantitative basis for subsequent risk level determination, enabling the assessment process to shift from experience-based to data-driven.

[0038] As described in step S403 above, based on the comparison between the beneficial owner score and the preset threshold, a beneficial owner risk level label is output. The risk level label includes at least three levels: low risk, medium risk, and high risk. Specifically, the calculated beneficial owner score is matched with a preset grading threshold range to determine the final risk level label. The preset threshold is not set using a single fixed value, but rather is a boundary criterion determined by combining cost compliance risk preferences and historical case statistical distribution. For example, it can be determined by collecting score samples from historical cost audit cases involving companies identified as conduit companies or lacking beneficial owner qualifications, statistically analyzing their score distribution percentiles (e.g., the 25th and 75th percentiles), and dynamically adjusting these percentiles based on the current stringency of anti-tax avoidance regulations, thereby forming a score range division corresponding to low risk, medium risk, and high risk. When the beneficial owner's score falls into the high range, it indicates that the transaction entity has sufficient registered capital, reasonable staffing, and a clearly defined location of its management institution, thus being identified as a beneficial owner with substantial business operations, and a low-risk label is issued. When the score is in the middle range, it indicates that some indicators have flaws but do not completely constitute obvious tax avoidance characteristics, and a medium-risk label is issued, prompting manual review or supplementary supporting materials. When the score is below the preset minimum safety threshold, it indicates that the enterprise has issues such as weakened capital, hollowing out, or abnormal management, and a high-risk label is issued, triggering further due diligence or a warning mechanism to deny treaty benefits. Through this tiered judgment mechanism, continuous quantitative scores can be transformed into discrete decision instructions, thereby achieving automated tiered management of cross-border transaction cost risks and providing a direct risk assessment basis for subsequent selection of applicable cost paths.

[0039] In one embodiment, before step S3, which involves inputting the clause matching result and the key elements of the contract into a preset multi-level classifier model to obtain the resulting property determination result, the method further includes: S201: Obtain historical cost case samples and the corresponding attribute annotation results as positive samples; S202: Generative adversarial networks are used to generate adversarial samples with blurred boundaries as negative samples; S203: Input the positive samples and the negative samples into the initial classifier model for iterative training until the confidence level of the initial classifier model's output for the obtained property determination reaches a preset threshold, thereby obtaining the multi-layer classifier model.

[0040] As described in step S201 above, historical fee case samples and their corresponding attribute annotation results are obtained as positive samples. Specifically, historical fee case samples are extracted from the enterprise's past cross-border service fee transaction records that have been reviewed or ultimately declared by relevant departments. These samples contain complete contract text information, service type descriptions, and the final determined cost processing results. The attribute annotation results clearly define the service type in the aforementioned historical fee case samples, such as labeling it as royalties, technical service fees, or operating profit. By collecting this real data with manual review markings, a data foundation reflecting the distribution of real business scenarios is constructed, providing high-quality input signals for subsequent model training. Positive samples covering multiple clearly defined business forms ensure that the model can learn standard classification feature mapping relationships. The historical fee case samples are derived from cross-border service fee declaration cases that have passed the review of relevant institutions, with a sample size of no less than 10,000. The annotation results are independently annotated by at least two experts with practical experience in fee agreements; in case of inconsistencies, a third expert makes the final decision.

[0041] As described in step S202 above, a generative adversarial network (GAN) is used to generate adversarial samples with ambiguous boundaries as negative samples. A GAN is a deep learning architecture that includes a generator and a discriminator. Leveraging the adversarial nature of this network, negative samples are constructed targeting the ambiguous areas in the obtained property determination. The generator is configured to receive a random noise vector and output simulated contract text description features, aiming to generate feature vectors near the boundary between royalty and technical service fees or operating profit classifications. The discriminator is used to distinguish the generated samples from real historical fee case samples. During adversarial training, the generator gradually masters the ability to generate ambiguous samples that can confuse the discriminator, such as generating text descriptions containing mixed features of technical consulting and software licensing. These adversarial samples, as negative samples, supplement edge scenarios that are difficult to cover in real-world cases, enrich the diversity of training data, and force the model to learn more refined semantic boundary features.

[0042] As described in step S203 above, positive and negative samples are input into the initial classifier model for iterative training until the output confidence of the initial classifier model in determining the obtained properties reaches a preset threshold, thus obtaining a multi-layer classifier model. The initial classifier model uses a pre-built basic network architecture, such as a classification model based on a deep neural network or support vector machine. During training, positive samples provide correct classification guidance, while adversarial samples provide negative samples, increasing the difficulty of the classification task and prompting the model to fine-tune the decision boundary of the feature space. The model's internal weight parameters are continuously adjusted through backpropagation to perform iterative training. Each iteration calculates the loss error between the predicted result and the true label, and updates the weights accordingly. The output confidence reflects the model's degree of certainty about the predicted result, usually presented as a probability value. A preset threshold (e.g., 0.85 or 0.9) is set, requiring the model's confidence in determining the obtained properties to reach a certain proportion or higher. When the model's output confidence stabilizes above the preset threshold and the loss function converges on the validation set, the training is considered complete, and the model parameters are fixed, forming the final multi-layer classifier model. This mixed-sample training method enables multi-layer classifier models to output more robust, accurate, and high-confidence property determination results when faced with service contracts with ambiguous boundaries or mixed features.

[0043] In one embodiment, step S4, which determines the applicable cost path based on the nature determination result and the beneficial owner's risk assessment result, includes: S411: When the obtained property determination result is operating profit, the first expense path is determined to apply; S412: When the nature of the income is determined to be a royalty or a technical service fee, and the risk assessment result of the beneficial owner is low risk, the second fee path shall be determined to apply. S413: When the obtained nature determination result is a royalty or technical service fee, and the beneficial owner risk assessment result is medium risk or high risk, determine the application of the first fee path, suspend the current process, and output a correction notice; wherein, the correction notice is used to instruct the user to supplement the supporting materials corresponding to the beneficial owner risk assessment; until the correction completion instruction is received, the suspension is lifted and the beneficial owner risk assessment is re-executed.

[0044] As described in step S411 above, when the nature determination result of the income is operating profit, the first cost path is determined to apply. The first cost path corresponds to the routine cost processing flow for expenses related to a permanent establishment or operating profit. In response to the nature determination result output by the multi-level classifier model indicating that the income from cross-border services belongs to operating profit, the execution logic of this path is triggered. The determination of this path means that the current transaction generally applies the operating profit clause in the cost agreement treatment, i.e., fees are levied only when the country of income source constitutes a permanent establishment, without involving the application of more complex restricted cost determination logic.

[0045] As described in step S412 above, when the nature of the income is determined to be royalty or technical service fee, and the beneficial owner's risk assessment result is low risk, the second cost path is determined to apply. The second cost path aims to handle low-risk situations that meet the preferential conditions of a cost agreement, typically corresponding to a withholding tax process that applies limited cost, such as applying the preferential cost of royalties (7%) or technical service fees under the AB area agreement. In this case, combining the qualitative result of the royalty or technical service fee with the low-risk beneficial owner assessment result indicates that the transaction entity has substantial business activities and the conduit arrangement risk is low. Therefore, it is determined that the preferential cost reporting process can be safely applied without implementing additional risk-blocking measures.

[0046] As described in step S413 above, when the nature of the income is determined to be a royalty or technical service fee, and the beneficial owner's risk assessment result is medium or high risk, the first fee path is applied, the current process is paused, and a correction notice is output. This correction notice instructs the user to supplement the supporting documentation corresponding to the beneficial owner's risk assessment. Specifically, when the nature of the income leans towards a royalty or technical service fee, but the beneficial owner's risk assessment result shows medium or high risk, it indicates a potential risk of fee agreement abuse. In this case, the second fee path is not directly executed; instead, it reverts to the more robust first fee path as the basic processing solution. Simultaneously, a process blocking mechanism is immediately triggered, forcibly interrupting the automated processing chain that may pose compliance risks by pausing the current process. The generated correction notice explicitly instructs the user to supplement key supporting documentation proving their beneficial owner status, such as a lease agreement for the actual management location, social security payment records for personnel, annual audit reports, or descriptions of substantive business activities. This design changes the traditional passive model of post-event remediation, achieving pre-event interception of high-risk cases and ensuring the compliance of fee declarations. After the user submits the relevant supporting documents according to the correction notice, a correction completion instruction is generated and triggered. This instruction serves as the sole credential for process resumption. Upon receipt, it automatically lifts the process suspension status set in the previous steps. After lifting the suspension, it does not directly continue with the subsequent document generation steps. Instead, it re-invokes the quantitative scoring model and recalculates and assesses the risk status of the beneficiary based on the newly supplemented supporting document data. This cyclical mechanism constructs a closed-loop verification logic of risk identification—document supplementation—reassessment confirmation. Only when the re-executed assessment result meets the risk level requirements (e.g., reduced to low risk) or confirms the nature of the risk will the process continue based on the latest assessment result, thereby ensuring that the processing of each cross-border service contract is completed under the premise of full compliance.

[0047] In one embodiment, after step S413, when the obtained nature determination result is a royalty or technical service fee, and the beneficial owner's risk assessment result is medium or high risk, determining the application of the first fee path, pausing the current process, and outputting a correction notice, the method further includes: S4141: When the risk assessment result of the beneficial owner changes to low risk after reassessment, the applicable cost path will be switched to the second cost path; S4142: If the risk assessment result of the beneficial owner after reassessment is still medium risk or high risk, maintain the first cost path and output a correction notice again.

[0048] As described in step S4141 above, when the reassessed beneficial owner risk assessment result changes to low risk, the applicable fee path is switched to the second fee path. Based on the correction notice output in the aforementioned steps, the user is guided to supplement supporting documents demonstrating the beneficial owner's substantive business activities. If the reassessed beneficial owner risk assessment result changes to low risk, it indicates that the quantitative score after supplementing the materials has met the preset low-risk threshold. For example, supplemented paid-in capital proof makes the registered capital as a percentage of the service fee exceed the preset safe harbor threshold, or supplemented personnel employment records and social security payment certificates prove that the payee has the substantive business capacity required to complete the service. In response to the confirmation of the low-risk status, the applicable fee path is switched to the second fee path. This switching process essentially adjusts the original fee cost processing logic to a process applying preferential treatment under fee agreements, such as switching from a domestic law-based fee cost path to a bilateral agreement-restricted fee cost path. By modifying the process control flags or re-calling the routing decision function, the system activates subsequent preferential fee cost calculation and document generation logic based on the updated low-risk status, ensuring that the user can obtain the expected fee benefits after completing the compliance correction.

[0049] As described in step S4142 above, if the reassessed beneficial owner risk assessment result is still medium or high risk, the first cost path is maintained, and a correction notice is issued again. Specifically, if the reassessed beneficial owner risk assessment result is still medium or high risk, it indicates that the supplementary supporting documentation provided by the user has not sufficiently eliminated suspicions of conduit arrangements or tax evasion. For example, if the supplementary documentation shows that the payee's actual management institution is still located in a tax haven, or that there is a significant mismatch between the number of employees and the scale of business, in this case, the system maintains the first cost path, i.e., maintaining a conservative approach of applying domestic law-based cost standards or not enjoying treaty benefits, to avoid the risk of cost audits due to improper application of the treaty. Simultaneously, a new round of interactive feedback is initiated, and a correction notice is issued again. This notice clearly informs the user that the current risk level has not yet met the standard and prompts further supplementation of more detailed substantive operational evidence, such as audit reports and annual financial statements. By constructing this iterative compliance closed loop, continuous interactive verification and path locking are carried out until the risk level meets the requirements or it is confirmed that preferential treatment cannot be enjoyed. In this way, while ensuring the compliance and security of costs, the scope of applicable fee regulations is explored and implemented to the greatest extent.

[0050] In one embodiment, step S5, which involves matching a target declaration document template in a pre-defined document template database based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, includes: S501: Combine the applicable cost path, the result of the determination of the nature of the income, and the result of the risk assessment of the beneficial owner to generate a joint state; S502: Using the joint state as the matching key, retrieve document templates from the document template database that have a preset mapping relationship with the joint state to obtain the target declaration document template.

[0051] As described in step S501 above, the applicable cost path, the result of the determination of the nature of income, and the result of the risk assessment of the beneficial owner are combined to generate a joint state. Specifically, based on a preset combination logic, the state information of these three dimensions—applicable cost path, result of the determination of the nature of income, and result of the risk assessment of the beneficial owner—is aggregated and encoded to generate a joint state. This joint state represents a unique state identifier for a specific cost scenario in the current cross-border service contract. Its internal structure integrates qualitative (nature of income), quantitative (cost path), and risk (beneficial owner) attributes. By constructing this joint state, the complex cost decision-making process is abstracted into a standardized state vector, providing an index basis for subsequent precise template matching. For example, when the applicable cost path is the first cost path (such as the cost of domestic law), the result of the determination of the nature of income is operating profit, and the result of the risk assessment of the beneficial owner is low risk, the generated joint state can uniquely identify this combination scenario.

[0052] As described in step S502 above, using the joint state as the matching key, document templates with a preset mapping relationship to the joint state are retrieved from the document template database to obtain the target declaration document template. The document template database pre-stores declaration document templates corresponding to various cost scenarios. Each template has a preset mapping relationship with a specific joint state. The generated joint state is set as the matching key, and a retrieval operation is performed in the document template database. By comparing the matching key with the preset mapping relationship, document template records that are completely consistent with the current joint state or meet preset compatibility conditions are located, and the template file corresponding to this record is extracted as the target declaration document template. This preset mapping relationship can be, for example, an index table stored in key-value pair format, where the key is the code of the joint state, and the value is the storage address or unique identifier of the document template.

[0053] Reference Figure 3 The present invention provides a device for generating risk management solutions for contract texts, the device comprising: Extraction module 902 is used to acquire the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key elements of the contract; The determination module 904 is used to perform dynamic clause matching in a preset multi-source knowledge base based on the key elements of the contract, determine the applicable fee agreement clauses, and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label, and clause conflict resolution identifier; Input module 906 is used to input the clause matching result and the key elements of the contract into a preset multi-level classifier model to obtain the obtained nature determination result; wherein, the obtained nature determination result includes the probability distribution of whether the service belongs to royalty, technical service fee or operating profit; The acquisition module 908 is used to acquire the risk assessment results of the beneficial owner and determine the applicable cost path based on the nature determination results and the risk assessment results of the beneficial owner; The matching module 910 is used to match the target declaration document template in a preset document template database based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result. The filling module 912 is used to extract the corresponding legal basis information from the multi-source knowledge base, fill the target declaration document template to generate the declaration document and legal basis summary, and obtain the risk disposal plan.

[0054] In one embodiment, determining module 904 includes: The similarity calculation submodule is used to extract service type tags from the key elements of the contract and perform similarity calculation with the clause type tags in the multi-source knowledge base; The candidate clause set formation submodule is used to extract target clause tags with similarity values ​​greater than a preset similarity value, forming a candidate clause set; The conflict resolution submodule is used to resolve conflicts in sequence according to the special law priority rule, the purpose interpretation rule, and the mutual negotiation procedure initiation rule when the candidate clause set contains two or more target clause tags, and to determine the final fee agreement clause. The clause matching result output submodule is used to output clause matching results based on the fee agreement clauses.

[0055] In one embodiment, the acquisition module 908 includes: The substantive operational assessment data acquisition submodule is used to acquire substantive operational assessment data of the transaction entity; wherein, the substantive operational assessment data includes at least one or more of the following: registered capital amount, personnel employment records, and information on the location of the actual management institution; The beneficiary owner score calculation submodule is used to input the substantive business assessment data into a preset quantitative scoring model to calculate the beneficiary owner score; The Beneficiary Owner Risk Level Label Output Submodule is used to output a Beneficiary Owner Risk Level Label based on the comparison result between the Beneficiary Owner's score and a preset threshold. The risk level label includes at least three levels: low risk, medium risk, and high risk.

[0056] In one embodiment, the risk management solution generation device for the contract text further includes: The historical cost case sample acquisition module is used to acquire historical cost case samples and the corresponding attribute annotation results as positive samples. A module is used to generate adversarial samples with blurred boundaries as negative samples using a generative adversarial network; The iterative training module is used to input the positive samples and the negative samples into the initial classifier model for iterative training until the confidence level of the output of the initial classifier model for the obtained property judgment reaches a preset threshold, thereby obtaining the multi-layer classifier model.

[0057] In one embodiment, the acquisition module 908 includes: The first cost path determination submodule is used to determine the applicable first cost path when the obtained property determination result is operating profit; The second fee path submodule is used to determine the application of the second fee path when the nature determination result of the obtained result is a royalty or technical service fee, and the risk assessment result of the beneficial owner is low risk. The pause submodule is used to determine the applicable first fee path and pause the current process when the obtained nature determination result is a royalty or technical service fee, and the beneficial owner risk assessment result is medium risk or high risk; wherein, the correction notice is used to instruct the user to supplement the supporting materials corresponding to the beneficial owner risk assessment; until the correction completion instruction is received, the pause is lifted and the beneficial owner risk assessment is re-executed.

[0058] In one embodiment, the acquisition module 908 further includes: The switching submodule is used to switch the applicable cost path to the second cost path when the risk assessment result of the beneficiary after reassessment changes to low risk. The maintenance submodule is used to maintain the first cost path and output a correction notification again when the risk assessment result of the beneficiary after reassessment is still medium or high risk.

[0059] In one embodiment, the matching module 910 includes: The joint state generation submodule is used to combine the applicable cost path, the obtained nature determination result, and the beneficial owner risk assessment result to generate a joint state; The retrieval submodule is used to retrieve document templates that have a preset mapping relationship with the joint state from the document template database, using the joint state as the matching key, to obtain the target declaration document template.

[0060] Figure 4 An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a method for generating a risk management plan for the contract text. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to execute the method for generating a risk management plan for the contract text. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0061] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; Based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label and clause conflict resolution identifier; The matching results of the terms and the key elements of the contract are input into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; Obtain the beneficial owner risk assessment results, and determine the applicable cost path based on the nature determination results and the beneficial owner risk assessment results; Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in the preset document template database; The corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is filled to generate the declaration document and legal basis summary, thus obtaining the risk disposal plan.

[0062] By transforming unstructured contract text into computable structured elements through semantic parsing, subsequent clause matching can be accurately retrieved based on clear element labels, avoiding qualitative biases caused by semantic ambiguity during manual reading and thus improving the accuracy of clause application. Building on this, the clause matching results and key contract elements are input into a multi-layer classifier model, ensuring that the resulting nature determination not only relies on static clause labels but also incorporates the semantic context of the contract, guaranteeing consistency between the qualitative results and the actual contractual agreement. Simultaneously, the substantive operational assessment data of the transaction entities is independently incorporated into risk assessment, allowing the determination of cost paths to consider both the type of income and the entity's risk. Finally, based on the determined path, nature determination, and risk assessment results, a document template is jointly matched and the corresponding legal basis is extracted and filled in, creating a traceable and complete chain from contract input to risk management output, effectively reducing the instability and oversight risks associated with manual item-by-item judgment.

[0063] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following steps: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; Based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results; wherein, the clause matching results include at least the matched fee agreement clause number, clause type label and clause conflict resolution identifier; The matching results of the terms and the key elements of the contract are input into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; Obtain the beneficial owner risk assessment results, and determine the applicable cost path based on the nature determination results and the beneficial owner risk assessment results; Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in the preset document template database; The corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is filled to generate the declaration document and legal basis summary, thus obtaining the risk disposal plan.

[0064] By transforming unstructured contract text into computable structured elements through semantic parsing, subsequent clause matching can be accurately retrieved based on clear element labels, avoiding qualitative biases caused by semantic ambiguity during manual reading and thus improving the accuracy of clause application. Building on this, the clause matching results and key contract elements are input into a multi-layer classifier model, ensuring that the resulting nature determination not only relies on static clause labels but also incorporates the semantic context of the contract, guaranteeing consistency between the qualitative results and the actual contractual agreement. Simultaneously, the substantive operational assessment data of the transaction entities is independently incorporated into risk assessment, allowing the determination of cost paths to consider both the type of income and the entity's risk. Finally, based on the determined path, nature determination, and risk assessment results, a document template is jointly matched and the corresponding legal basis is extracted and filled in, creating a traceable and complete chain from contract input to risk management output, effectively reducing the instability and oversight risks associated with manual item-by-item judgment.

[0065] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0066] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for generating risk management solutions for contract texts, characterized in that, The method includes: Obtain the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements; Based on the key elements of the contract, dynamic clause matching is performed in a pre-set multi-source knowledge base to determine the applicable fee agreement clauses and output the clause matching results. The clause matching results at least include the matched fee agreement clause number, clause type tag, and clause conflict resolution identifier. Specifically, the service type tag is extracted from the key elements of the contract and its similarity is calculated with the clause type tags in the multi-source knowledge base. Target clause tags with similarity values ​​greater than a preset similarity value are extracted to form a candidate clause set. When the candidate clause set contains two or more target clause tags, conflict resolution is performed sequentially according to the special law priority rule, the purposive interpretation rule, and the mutual negotiation procedure initiation rule to determine the final fee agreement clause. The clause matching results are output based on the fee agreement clauses. The matching results of the terms and the key elements of the contract are input into a preset multi-level classifier model to obtain the nature determination result; wherein, the nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; Obtain the beneficial owner risk assessment results, and determine the applicable cost path based on the nature determination results and the beneficial owner risk assessment results; Based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result, the target declaration document template is matched in the preset document template database; The corresponding legal basis information is extracted from the multi-source knowledge base, and the target declaration document template is filled to generate the declaration document and legal basis summary, thus obtaining the risk disposal plan.

2. The method for generating a risk management plan for a contract text according to claim 1, characterized in that, The steps for obtaining the risk assessment results for the beneficial owners include: Obtain substantive operational assessment data of the transaction entity; wherein, the substantive operational assessment data includes at least one or more of the following: registered capital amount, personnel employment records, and information on the location of the actual management institution; Input the substantive operational assessment data into a preset quantitative scoring model to calculate the benefit owner's score; Based on the comparison between the beneficiary's score and a preset threshold, a risk level label for the beneficiary is output, which includes at least three levels: low risk, medium risk, and high risk.

3. The method for generating a risk management plan for a contract text according to claim 1, characterized in that, Before the step of inputting the clause matching result and the key elements of the contract into a preset multi-level classifier model to obtain the obtained property determination result, the method further includes: Obtain historical cost case samples and the corresponding attribute annotation results as positive samples; Generative adversarial networks are used to generate adversarial samples with blurred boundaries as negative samples; The positive and negative samples are input into the initial classifier model for iterative training until the confidence level of the initial classifier model's output for the obtained property determination reaches a preset threshold, thereby obtaining the multi-layer classifier model.

4. The method for generating a risk management plan for a contract text according to claim 1, characterized in that, The step of determining the applicable cost path based on the nature determination result and the beneficial owner risk assessment result includes: When the result of the determination of the nature of the obtained product is operating profit, the first expense path shall be applied. When the nature of the income is determined to be a royalty or technical service fee, and the risk assessment result of the beneficial owner is low risk, the second fee path is determined to apply. When the nature determination result is a royalty or technical service fee, and the beneficial owner risk assessment result is medium or high risk, the first fee path is determined to apply, the current process is suspended, and a correction notice is output; wherein, the correction notice is used to instruct the user to supplement the supporting materials corresponding to the beneficial owner risk assessment; until the correction completion instruction is received, the suspension is lifted and the beneficial owner risk assessment is re-executed.

5. The method for generating a risk management plan for a contract text according to claim 4, characterized in that, After the step of determining the application of the first fee path, pausing the current process, and outputting a correction notice when the nature determination result of the obtained fee is a royalty or technical service fee, and the risk assessment result of the beneficial owner is medium or high risk, the method further includes: When the risk assessment result of the beneficial owner changes to low risk after reassessment, the applicable cost path will be switched to the second cost path; If the reassessed risk assessment result for the beneficial owner is still medium or high risk, maintain the first cost path and issue a correction notice again.

6. The method for generating a risk management plan for a contract text according to claim 1, characterized in that, The step of matching a target declaration document template in a pre-set document template database based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result includes: The applicable cost path, the result of the determination of the nature of the income, and the risk assessment result of the beneficial owner are combined to generate a joint state; Using the joint state as the matching key, a document template that has a preset mapping relationship with the joint state is retrieved from the document template database to obtain the target declaration document template.

7. A device for generating risk management solutions for contract texts, characterized in that, The device includes: The extraction module is used to acquire the cross-border service contract text to be processed, perform semantic parsing on the contract text, and extract key contract elements. The determination module is used to dynamically match clauses in a preset multi-source knowledge base based on the key elements of the contract, determine the applicable fee agreement clauses, and output the clause matching results. The clause matching results at least include the matched fee agreement clause number, clause type tag, and clause conflict resolution identifier. Specifically, it extracts the service type tag from the key elements of the contract and calculates its similarity with the clause type tags in the multi-source knowledge base; it extracts target clause tags with similarity values ​​greater than a preset similarity value to form a candidate clause set; when the candidate clause set contains two or more target clause tags, it sequentially resolves conflicts according to the special law priority rule, the purpose interpretation rule, and the mutual negotiation procedure initiation rule to determine the final fee agreement clause; and it outputs the clause matching results based on the fee agreement clauses. The input module is used to input the matching results of the terms and the key elements of the contract into a preset multi-level classifier model to obtain the obtained nature determination result; wherein, the obtained nature determination result includes the probability distribution of whether the service belongs to the royalty, technical service fee or operating profit; The acquisition module is used to acquire the risk assessment results of the beneficial owner and determine the applicable cost path based on the nature determination results and the risk assessment results of the beneficial owner; The matching module is used to match the target declaration document template in a preset document template database based on the applicable cost path, the nature determination result, and the beneficial owner risk assessment result. The filling module is used to extract corresponding legal basis information from the multi-source knowledge base, fill the target declaration document template to generate the declaration document and legal basis summary, and obtain the risk disposal plan.

8. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, causes the processor to perform the steps of the risk management scheme generation method for the contract text as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, The device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the risk management scheme generation method for the contract text as described in any one of claims 1 to 6.