Cross-language insurance question and answer knowledge editing and updating method, system and equipment and medium

Abstract

The invention relates to the technical field of intelligent decision making, and discloses a knowledge editing and updating method, system and device for cross-language insurance questioning and answering and a medium, and the method comprises the steps that models of different languages complete unified storage of documents of different languages based on a common semantic coordinate system; when an insurance company updates a policy or a product description, a positioning module is triggered through a knowledge editing function, and content change points are identified and positioned; when the Chinese model generates a knowledge correction gradient signal, positioning a content change point corresponding to the knowledge correction gradient signal through a knowledge editing function, and finishing one-time modification and globally synchronous cross-language synchronous updating through a shared editing space; and performing cross-language consistency verification on the synchronously updated content, and completing necessary correction optimization. The method can be applied to the development of business systems such as financial science and technology, medical health, old-age care and the like, the redundant operation of modifying multiple languages one by one in the prior art is avoided, and the repeated labor cost of a multi-language team is reduced.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent decision-making, and in particular to a knowledge editing and updating method, system, device, and medium for cross-language insurance question answering. Background Technology

[0002] In current insurance question-answering systems, knowledge updates for multilingual models are primarily achieved through full retraining or independent local fine-tuning. For example, after expanding the "earthquake insurance" clause on the Chinese insurance question-answering model, it is necessary to retrain the Thai, Vietnamese, and Indonesian versions separately to ensure content synchronization. However, this approach has the following drawbacks: The update efficiency is low. Independent training in each language leads to repetitive work, and knowledge updates cannot be synchronized in real time, often requiring several weeks. There is also a serious cross-language bias. Different language models have semantic drift in their understanding of insurance clause details (such as region, compensation ratio, and scope of exclusions). In particular, the semantic alignment is insufficient after machine translation fine-tuning, which can easily lead to compliance risks. Summary of the Invention

[0003] This invention provides a knowledge editing and updating method, system, computer equipment, and medium for cross-language insurance question answering, in order to solve the technical problems of low updating efficiency and serious cross-language discrepancies in the prior art.

[0004] Firstly, a method for editing and updating knowledge in cross-language insurance question-and-answer formats is provided, including: Models for different languages ​​achieve unified storage of documents in different languages ​​based on a common semantic coordinate system; When an insurance company updates its policies or product descriptions, the knowledge editing function triggers the location module to identify and locate the points of content change. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function is used to locate the content change point corresponding to the knowledge correction gradient signal, and the cross-language synchronous update is completed through a shared editing space after one modification, and globally synchronized. Perform cross-language consistency verification on the synchronized updated content and complete necessary corrections and optimizations; Once all language versions have been updated, execute the secure output release process.

[0005] Secondly, a cross-language insurance question-and-answer knowledge editing and updating system is provided, including: The storage module is used to unify the storage of documents in different languages ​​based on a common semantic coordinate system. The location module is used to identify and locate the points of content change when the insurance company updates its policies or product descriptions, triggered by the knowledge editing function. The update module is used to locate the content change points corresponding to the knowledge correction gradient signal when the Chinese model generates the knowledge correction gradient signal, and complete the cross-language synchronous update of one modification and global synchronization through the shared editing space; The verification module is used to perform cross-language consistency verification on the synchronized updated content and to perform necessary corrections and optimizations. The output module is used to execute the secure output release process after all language versions have been updated.

[0006] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned cross-language insurance question-and-answer knowledge editing and updating method.

[0007] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned cross-language insurance question-and-answer knowledge editing and updating method.

[0008] The aforementioned cross-language insurance Q&A knowledge editing and updating method, system, computer equipment, and storage medium enables unified storage of documents in different languages ​​based on a common semantic coordinate system. When an insurance company updates its policies or product descriptions, the knowledge editing function triggers a positioning module to identify and locate content changes. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function locates the corresponding content changes and completes a one-time modification, globally synchronized cross-language update through a shared editing space. Cross-language consistency verification is performed on the updated content, and necessary corrections and optimizations are completed. After all language versions have been updated, a secure output and release process is executed. This invention avoids the redundant operations of traditional multi-language one-by-one modifications by using a unified semantic and storage base, lightweight parameter patch delivery, and a one-time modification, globally synchronized mechanism, compressing the update cycle from "days" to "minutes" and reducing repetitive labor costs for multi-language teams. Relying on a common semantic coordinate system and cross-language embedding model, it achieves semantic aggregation of synonymous concepts; combined with semantic consistency verification, it ensures that the core meaning of multi-language versions is unbiased, avoiding semantic fragmentation issues in cross-language synchronization from the source. A dual compliance verification system combining automated regulatory retrieval and manual verification is established to accurately match regulatory requirements in various countries. This system promptly triggers legal intervention for any questionable content, significantly reducing compliance risks in global operations. Update records and automated reports with unique version identifiers are generated, leaving a complete trail of key information such as changes and their impact, providing clear evidence for audits and business reviews, and improving the standardization of knowledge management. The system ensures the timeliness and accuracy of multilingual knowledge throughout the entire process. After synchronization with RAG Q&A systems in various countries, it can quickly support customer inquiries and internal business queries in different regions, providing stable knowledge support for the insurance company's global expansion. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention 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.

[0010] Figure 1 This is a schematic diagram of an application environment for a cross-language insurance question-and-answer knowledge editing and updating method according to an embodiment of the present invention.

[0011] Figure 2 This is a flowchart illustrating a knowledge editing and updating method for cross-language insurance question answering in one embodiment of the present invention.

[0012] Figure 3 This is a schematic diagram of the knowledge editing and updating system for cross-language insurance question answering in one embodiment of the present invention.

[0013] Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention.

[0014] Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0015] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] The knowledge editing and updating method for cross-language insurance question answering provided in this invention can be applied to, for example... Figure 1In this application environment, the client uses models of different languages ​​to uniformly store documents in different languages ​​based on a common semantic coordinate system. When an insurance company updates its policies or product descriptions, the knowledge editing function triggers the positioning module to identify and locate content changes. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function locates the content changes corresponding to the knowledge correction gradient signal, and a one-time modification and global synchronization update is completed through a shared editing space. The updated content undergoes cross-language consistency verification and necessary corrections and optimizations. After all language versions have been updated, a secure output and release process is executed. This invention avoids the redundant operations of traditional multi-language one-by-one modifications by using a unified semantic and storage base, lightweight parameter patch delivery, and a one-time modification and global synchronization mechanism. This reduces the update cycle from "days" to "minutes," minimizing repetitive labor costs for multi-language teams. Relying on a common semantic coordinate system and a cross-language embedding model, it achieves semantic aggregation of synonymous concepts. Combined with semantic consistency verification, it ensures that the core meaning of multi-language versions is unbiased, avoiding semantic fragmentation issues in cross-language synchronization from the source. A dual compliance verification system combining automated regulatory retrieval and manual verification is constructed to accurately match the regulatory requirements of various countries. This system promptly triggers legal intervention for questionable content, significantly reducing compliance risks in global operations. Update records and automated reports with unique version identifiers are generated, leaving a complete trail of key information such as modified content and scope of impact. This provides clear evidence for audits and business reviews, improving the standardization of knowledge management. The system ensures the timeliness and accuracy of multilingual knowledge throughout the entire process. After synchronization with RAG Q&A systems in various countries, it can quickly support customer inquiries and internal business queries in different regions, providing stable knowledge support for the insurance company's global expansion. Clients can be, but are not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a dedicated server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0017] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the knowledge editing and updating method for cross-language insurance question answering provided in this embodiment of the invention includes the following steps: S10: Models for different languages ​​achieve unified storage of documents in different languages ​​based on a common semantic coordinate system; S20: When an insurance company updates its policies or product descriptions, the knowledge editing function triggers the positioning module to identify and locate the points of content change. S30: When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function is used to locate the content change point corresponding to the knowledge correction gradient signal, and the cross-language synchronous update is completed through a shared editing space after one modification and global synchronization. S40: Perform cross-language consistency verification on the synchronized updated content and complete necessary corrections and optimizations; S50: After all language versions have been updated, execute the secure output release process.

[0018] S10 establishes a "semantic interoperability foundation" for multilingual models, breaks down storage barriers between documents in different languages, enables centralized and standardized management of various language knowledge, and provides a unified data base for subsequent cross-language synchronous updates.

[0019] S10 constructs a "Common Semantic Coordinate System," a unified semantic benchmark built upon cross-language semantic alignment technology. This system maps concepts, terms, and documents expressing the same or similar meanings in different languages ​​(such as Chinese, English, Japanese, and Korean) to the same semantic dimension. Specifically, it trains cross-language embedding models (such as LaBSE and mT5) to encode document text in different languages ​​into high-dimensional vectors. Based on the rules of the "Common Semantic Coordinate System," these vectors are standardized to ensure that vectors of synonymous content are in close proximity in the high-dimensional space. Finally, the standardized vectors and their corresponding original documents are uniformly stored in a distributed knowledge database, forming a unified knowledge reserve pool covering multiple languages. This avoids redundant maintenance of knowledge documents by different language teams, reducing management costs. Furthermore, it provides a semantic-level matching basis for subsequent modifications and global synchronization, ensuring rapid location of corresponding content in different languages ​​during updates.

[0020] S10 includes: Construct a common semantic coordinate system for different language models; By training cross-linguistic embedding models (such as LaBSE or mT5), texts from different languages ​​are encoded into vector representations, enabling synonymous concepts to exhibit semantic aggregation characteristics in a high-dimensional space. The knowledge retrieval layer stores vectorized knowledge documents in all languages ​​through a unified vector index library.

[0021] As a core benchmark for semantic interoperability in multilingual models, this coordinate system provides a unified semantic reference dimension for models in different languages ​​(such as Chinese, English, Japanese, and Korean). By defining universal semantic mapping rules, it ensures accurate semantic alignment of concepts, terms, and clauses expressing the same meaning in different languages, laying the foundation for subsequent cross-language text encoding and content matching. Cross-language embedding models (such as LaBSE and mT5) are trained to transform text content in various languages ​​into standardized vector representations. The core value of this process lies in enabling vectors of synonymous concepts to exhibit significant aggregation characteristics in a high-dimensional semantic space—that is, vectors corresponding to texts with similar meanings are located in adjacent positions in the high-dimensional space, thereby breaking down semantic barriers between different languages ​​and enhancing the comparability of multilingual content. Led by the knowledge retrieval layer, all vectorized knowledge documents after encoding in all languages ​​are uniformly stored in a pre-defined vector index library. The vector index library adopts standardized storage specifications, which not only realizes centralized management of multilingual vectorized knowledge, but also provides efficient data call support for subsequent knowledge retrieval, content change comparison and cross-language synchronous updates, ensuring the smooth flow of data throughout the entire process.

[0022] The S20 system accurately captures updates to core knowledge documents such as insurance company policies and product descriptions, quickly identifying and locating specific changes to provide clear "targets" for subsequent knowledge correction and synchronization. First, the system has a built-in document monitoring mechanism that monitors core data sources such as the insurance company's internal policy document library and product description publishing platform in real time. When document additions, modifications, or deletions are detected, the knowledge editing function is automatically triggered, generating an "update warning signal" and simultaneously recording basic information such as update time, operator, and document name. Second, the knowledge editing function, in conjunction with the location module, employs a dual technology of "semantic difference detection + text comparison" for precise location: on one hand, by calculating the semantic vector distance between the documents before and after the update, it quickly identifies the approximate paragraphs where semantics have changed; on the other hand, it performs sentence-by-sentence text comparison of these paragraphs to identify specific modifications (such as terminology adjustments, clause additions or deletions, and wording optimizations), generating a "list of changed content" that clearly defines the type of change, its location, the original content, and the updated content.

[0023] Improve the targeting of knowledge correction: By automating monitoring and location, reduce the workload and error rate of manual inspection, ensure that the updates and changes of core documents can be captured in the first instance, buy time for subsequent cross-language synchronization, and avoid inconsistencies in multilingual knowledge caused by missed updates and incorrect location.

[0024] S20 includes: When insurance companies update policies or product descriptions, the document monitoring mechanism and semantic differential detection technology are combined to achieve accurate perception of content changes. Calculate the vector distance between the documents before and after the update. If the change in vector distance exceeds a preset threshold, the knowledge node is updated and a structured editing task is generated.

[0025] Specifically, when insurance companies update policies or product descriptions, they can accurately perceive content changes by combining newly uploaded terms and conditions documents in the RAG knowledge base with the content changes between the old and new documents.

[0026] When insurance companies update their policies or product descriptions, the system employs a collaborative model combining a document monitoring mechanism and semantic differential detection technology. This dual-dimensional approach enables precise perception of changes: firstly, it directly connects to the RAG knowledge base to obtain newly uploaded terms and conditions documents as the core update data source; secondly, it meticulously compares the specific content of the old and new documents using text comparison algorithms to accurately pinpoint changes in core terms, key terminology, and wording. Cross-validation of the results from both methods ensures that the perception of content changes is thorough and accurate.

[0027] Based on the detected content changes, the system further conducts quantitative analysis: It calculates the vector distance between the documents before and after the update, using the vectorized documents generated in step S10. The system presets a vector distance threshold (this threshold is set according to business needs and is used to determine whether the magnitude of the change warrants a knowledge update). If the calculated vector distance change exceeds the preset threshold, it indicates that the content change has a significant impact on the knowledge system. The system will then automatically trigger a knowledge node update and generate a structured editing task. This task clearly includes key information such as the update object, the scope of change, and the core correction requirements, providing clear targeted guidance for subsequent knowledge correction steps.

[0028] Based on the knowledge correction of the Chinese model, S30 achieves global synchronization with a single modification through a shared editing space, quickly synchronizing the updated content of the Chinese document to all other language versions, ensuring real-time consistency of multilingual knowledge.

[0029] The core of cross-language synchronization consists of four key steps: "benchmark correction signal generation - correction signal corresponding change point location - shared space mapping - multilingual transmission". The entire process relies on "knowledge editing function", "shared editing space" and "language mapping matrix" to achieve accurate semantic transmission and accurate matching of change points. First: Since Chinese is the core business language of insurance companies, the system defaults to using the Chinese model as the benchmark for knowledge correction: The Chinese model generates a knowledge correction gradient signal for the changes located in S20. This signal is a semantic quantification of the modified content, containing key information such as correction direction, semantic weight, and scope of influence. At the same time, the system converts the model update corresponding to the correction into a "lightweight parameter patch" in the form of a low-rank matrix (small in size, high in transmission efficiency, and avoids the redundancy of retraining the entire model). Secondly, the knowledge editing function proactively associates with the knowledge correction gradient signal generated by the Chinese model. Based on the common semantic coordinate system constructed by S10, it analyzes the semantic features in the gradient signal to accurately locate the specific content change point corresponding to the correction signal (forming a double verification with the change point located by S20 to ensure the uniqueness and accuracy of the correction target). Simultaneously, it generates a "correction-change point" mapping list, clarifying the correspondence between the gradient signal and the specific modified content, providing a precise "targeting basis" for subsequent cross-language transmission. Furthermore, the generated "parameter patch" knowledge correction gradient signal and the "correction-change point" mapping list are synchronously mapped to the "shared editing space." This space is a unified editing hub connecting multilingual models, compatible with the semantic rules of different languages. It standardizes the correction signal and mapping list, eliminating semantic deviations caused by language differences and ensuring the universality of correction information and change point location. Next, based on the "common semantic coordinate system" built by S10, the standardized correction signals, parameter patches, and "correction-change point" mapping list in the shared editing space are accurately transmitted to all other target language models (such as English, Japanese, etc.) through the preset "language mapping matrix" (which records the semantic correspondence between different languages ​​and Chinese). After receiving the information, each target language model quickly locates its corresponding content change point according to the mapping list, automatically calls the parameter patch to complete the knowledge system update, and finally achieves the effect of one modification and global synchronization.

[0030] Significantly improves the efficiency and accuracy of multilingual updates and reduces operating costs: Traditional multilingual updates require different language teams to modify and review each one, which is time-consuming, costly, and prone to errors; This step uses the knowledge editing function to accurately locate the change points corresponding to the correction signals, combined with automated semantic transmission and synchronous updates, which not only achieves one-time modification and global synchronization (compressing the update cycle from "days" to "minutes"), but also ensures the consistency of the modification goals of each language version through the dual verification of "correction-change point", completely avoiding human error.

[0031] S30 includes: After generating the knowledge correction gradient signal in the Chinese model, the corresponding model update is presented in the mathematical form of a low-rank matrix, which is a lightweight "parameter patch". The parameter patches are mapped to a shared editing space and passed to each target language model through a language mapping matrix. At the same time, by constraining the semantic consistency of the output of the multilingual models, the meaning of the answers in different languages ​​is ensured to remain highly consistent.

[0032] Using the Chinese model as the benchmark for multilingual updates, after the Chinese model generates the knowledge correction gradient signal for the structured editing task generated by S22, the corresponding model update is presented in the mathematical form of a low-rank matrix. This low-rank matrix is ​​a lightweight "parameter patch" with advantages of small size and high transmission efficiency, which can effectively avoid the redundant operation of retraining the entire model and greatly improve update efficiency.

[0033] First, the generated parameter patch is mapped to a preset shared editing space—a unified editing hub connecting various language models, compatible with the semantic rules of different languages, and the parameter patch is standardized to eliminate transmission bias caused by language differences. Then, through a preset language mapping matrix (which pre-records the semantic correspondence between different target languages ​​and Chinese), the standardized parameter patch is accurately transmitted to each target language model (such as English, Japanese, Korean, etc.), realizing cross-language transmission of update commands.

[0034] Throughout the parameter patch delivery and target language model update process, a semantic consistency constraint mechanism is initiated simultaneously. By monitoring the output results of each language model in real time, the semantics of the multilingual model output are constrained to ensure that the meaning of the answers generated by different language models based on the updated knowledge remains highly consistent, thus avoiding semantic deviations that may occur during cross-language synchronization from the source.

[0035] S40 performs quality checks on all language versions of the content after synchronization and updates to ensure complete semantic consistency across different languages. It also corrects semantic discrepancies and non-standard expressions that may arise during synchronization, guaranteeing the accuracy of multilingual knowledge. A combination of automated verification and necessary manual correction is employed, focusing on two core dimensions: semantic consistency and standardization of expression. First, the semantic consistency check system is activated, performing two core checks on each language version after synchronization and updates: One is calculating the semantic vector similarity of the corresponding content in different language versions (based on the common semantic coordinate system of S10). If the similarity is lower than a preset threshold (e.g., 95%), it is judged as "semantic inconsistency." The second is performing standardization checks on the terminology and expressions in each language version, checking whether they conform to the industry standards and company standard terminology databases for the corresponding language (e.g., the English version must conform to the international terminology standards for the insurance industry, and the Japanese version must conform to the expression requirements of the Japanese insurance regulatory agency). Secondly, issues discovered during automated verification are handled in two categories: First, minor semantic deviations (similarity between 90% and 95%) are handled by the system automatically using a semantic correction model, fine-tuning based on a common semantic coordinate system to ensure semantic consistency. Second, serious semantic deviations (similarity below 90%) or non-standard expressions are handled by the system generating a "correction prompt list," clearly identifying the problem location, error type, and suggested correction direction, which is then transferred to the human review team for targeted correction. After correction, the system performs consistency verification again until all language versions meet the quality requirements.

[0036] To avoid knowledge bias caused by semantic loss and language differences during the synchronization process, we ensure that the content of each language version not only conforms to the core meaning of the Chinese benchmark, but also adapts to the expression habits and industry standards of the corresponding language, so as to provide accurate knowledge support for subsequent business operations in different countries / regions (such as customer service and compliance filing).

[0037] S40 includes: The semantic consistency check system calculates the similarity of multilingual outputs to verify the semantic correctness of the content after synchronization and updating. Retrieve relevant provisions from national legal databases and verify their consistency with local laws; If there are uncertainties during the compliance verification process, a "compliance discrepancy alert" will be generated and transferred to legal personnel for manual confirmation.

[0038] A semantic consistency check system is launched, based on a common semantic coordinate system constructed using S10. It verifies the semantic correctness of the synchronized updated content by calculating the vector similarity of the multilingual output content. The core objective is to ensure that the core meaning expressed in different language versions is completely consistent, without any semantic deviations, omissions, or misinterpretations.

[0039] The system automatically searches national legal knowledge bases, extracts relevant legal provisions related to the updated content, and verifies the consistency between the synchronized updated language versions and the corresponding local laws of the respective countries / regions. The verification focuses on core compliance points such as data privacy protection, standardized wording of clauses, and clear definition of responsibilities, ensuring that the content complies with the regulatory requirements of different regions.

[0040] If uncertainties arise during the compliance verification process (such as blurred boundaries between content and local legal provisions, multiple possible interpretations, or inability to directly determine compliance), the system will not make subjective judgments but will automatically generate a "compliance discrepancy alert." The alert will detail the questionable content segment, the corresponding original legal provisions, and the core points of contention, and will be transferred to legal personnel for manual confirmation to ensure the accuracy and rigor of the compliance assessment.

[0041] S50 completes the final implementation of multilingual knowledge updates, ensuring the compliant and secure output of updated content to various business systems, while retaining update records for subsequent traceability and auditing. Centered on "compliance review + secure release + record retention," it consists of three key steps: First, the system automatically searches the insurance industry regulatory knowledge base of various countries / regions to verify whether the updated content in each language version complies with local regulatory requirements (such as data privacy, clause wording, liability definition, etc.). If compliance risks exist, a "compliance difference alert" is generated and transferred to the legal team for confirmation until the risk is eliminated. Second, after compliance verification, the secure release process is executed: all updated language versions of the knowledge are synchronously output to the insurance company's core business systems, including the RAG Q&A system, customer service knowledge base, internal employee training platform, regulatory filing document system, etc. Encrypted transmission technology is used during synchronization to ensure the security of knowledge data and prevent information leakage or tampering during transmission. Furthermore, the system automatically generates a "knowledge update record" with a version identifier, which records the entire update process in detail, including the update trigger time, list of changed content, synchronized language range, consistency verification results, correction records, compliance review opinions, release time, operator, etc. This record will be archived to the company's knowledge management platform for subsequent update tracing, audit verification, and problem review.

[0042] Ensure the compliance and traceability of multilingual knowledge updates: mitigate regulatory risks through pre-release compliance verification; ensure the accuracy of knowledge in business systems through secure transmission and synchronization; and provide a basis for subsequent audits and problem investigation through complete record retention, while forming a closed-loop management of knowledge updates.

[0043] The S50 includes: Once all language versions have been updated, a knowledge update record containing version identifiers will be generated. The knowledge update records are synchronized to the RAG Q&A systems in various countries, and an automated report is generated, listing the modified content, the scope of impact, and the status of manual review.

[0044] Once all language versions have been updated and passed the dual verification of the S40 process, the system automatically generates a knowledge update record. This record contains a unique version identifier (using a dedicated encoding rule to ensure accurate traceability of each update), and also includes core information such as the update time, the range of languages ​​covered by the update, and the S40 verification result, achieving full traceability of the update process.

[0045] The generated knowledge update records are synchronized to the corresponding RAG Q&A systems in various countries to ensure that business systems in each region can obtain the latest multilingual knowledge in a timely manner, thus guaranteeing the accuracy and timeliness of knowledge in scenarios such as customer consultation responses and internal employee business inquiries.

[0046] The system synchronously generates and outputs automated reports, clearly listing the key information of this update, including specific modifications (including a comparison of the old and new content), the scope of the update's impact (business modules involved, covered regions, target user groups, etc.), and the status of manual review (mainly the review results of legal personnel in the S40 process), providing clear and intuitive evidence for business review, audit verification, and other work.

[0047] As can be seen, in the above scheme, unified storage of documents in different languages ​​is first achieved through models of different languages ​​based on a common semantic coordinate system. When the insurance company updates its policies or product descriptions, the knowledge editing function triggers the positioning module to identify and locate the content change points. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function locates the content change points corresponding to the knowledge correction gradient signal, and a cross-language synchronous update is completed through a shared editing space, achieving global synchronization after a single modification. Cross-language consistency verification is performed on the synchronized updated content, and necessary corrections and optimizations are completed. After all language versions have been updated, a secure output and release process is executed. This invention avoids the redundant operations of traditional multi-language one-by-one modifications by using a unified semantic and storage base, lightweight parameter patch delivery, and a one-time modification, global synchronization mechanism, compressing the update cycle from "days" to "minutes" and reducing the repetitive labor costs of multi-language teams. Relying on a common semantic coordinate system and a cross-language embedding model, semantic aggregation of synonymous concepts is achieved; combined with semantic consistency verification, the core meaning of multi-language versions is ensured to be unbiased, avoiding the semantic fragmentation problem of cross-language synchronization from the source. A dual compliance verification system combining automated regulatory retrieval and manual verification is established to accurately match regulatory requirements in various countries. This system promptly triggers legal intervention for any questionable content, significantly reducing compliance risks in global operations. Update records and automated reports with unique version identifiers are generated, leaving a complete trail of key information such as changes and their impact, providing clear evidence for audits and business reviews, and improving the standardization of knowledge management. The system ensures the timeliness and accuracy of multilingual knowledge throughout the entire process. After synchronization with RAG Q&A systems in various countries, it can quickly support customer inquiries and internal business queries in different regions, providing stable knowledge support for the insurance company's global expansion.

[0048] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0049] In one embodiment, a cross-language insurance question-and-answer knowledge editing and updating system is provided, which corresponds one-to-one with the cross-language insurance question-and-answer knowledge editing and updating method described in the above embodiments. For example... Figure 3 As shown, this cross-language insurance question-and-answer knowledge editing and updating system includes a storage module 101, a location module 102, an update module 103, a verification module 104, and an output module 105. Detailed descriptions of each functional module are as follows: Storage module 101 is used to unify the storage of documents in different languages ​​based on a common semantic coordinate system; The positioning module 102 is used to identify and locate the content change points when the insurance company updates its policies or product descriptions, triggered by the knowledge editing function. The update module 103 is used to locate the content change point corresponding to the knowledge correction gradient signal through the knowledge editing function when the Chinese model generates the knowledge correction gradient signal, and to complete the cross-language synchronous update of one modification and global synchronization through the shared editing space. The verification module 104 is used to perform cross-language consistency verification on the synchronized updated content and to complete the necessary correction and optimization. Output module 105 is used to execute the secure output release process after all language versions have been updated.

[0050] In one embodiment, the storage module 101 is further configured to: When insurance companies update policies or product descriptions, the document monitoring mechanism and semantic differential detection technology are combined to achieve accurate perception of content changes. Calculate the vector distance between the documents before and after the update. If the change in vector distance exceeds a preset threshold, the knowledge node is updated and a structured editing task is generated.

[0051] In one embodiment, the positioning module 102 is further configured to: When insurance companies update policies or product descriptions, the document monitoring mechanism and semantic differential detection technology are combined to achieve accurate perception of content changes. Calculate the vector distance between the documents before and after the update. If the change in vector distance exceeds a preset threshold, the knowledge node is updated and a structured editing task is generated.

[0052] In one embodiment, the update module 103 is further configured to: After generating the knowledge correction gradient signal in the Chinese model, the corresponding model update is presented in the mathematical form of a low-rank matrix, which is a lightweight "parameter patch". The parameter patches are mapped to a shared editing space and passed to each target language model through a language mapping matrix. At the same time, by constraining the semantic consistency of the output of the multilingual models, the meaning of the answers in different languages ​​is ensured to remain highly consistent.

[0053] In one embodiment, the verification module 104 is further configured to: The semantic consistency check system calculates the similarity of multilingual outputs to verify the semantic correctness of the content after synchronization and updating. Retrieve relevant provisions from national legal databases and verify their consistency with local laws; If there are uncertainties during the compliance verification process, a "compliance discrepancy alert" will be generated and transferred to legal personnel for manual confirmation.

[0054] In one embodiment, the output module 105 is further configured to: Once all language versions have been updated, a knowledge update record containing version identifiers will be generated. The knowledge update records are synchronized to the RAG Q&A systems in various countries, and an automated report is generated, listing the modified content, the scope of impact, and the status of manual review.

[0055] This invention provides a cross-language insurance question-and-answer knowledge editing and updating system. It achieves unified storage of documents in different languages ​​based on a common semantic coordinate system using models of different languages. When an insurance company updates policies or product descriptions, the knowledge editing function triggers a positioning module to identify and locate content changes. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function locates the corresponding content changes and completes a one-time modification, globally synchronized cross-language update through a shared editing space. The updated content undergoes cross-language consistency verification and necessary corrections and optimizations. After all language versions have been updated, a secure output and release process is executed. This invention avoids the redundant operations of traditional multi-language one-by-one modifications by using a unified semantic and storage base, lightweight parameter patch delivery, and a one-time modification, globally synchronized mechanism, compressing the update cycle from "days" to "minutes" and reducing repetitive labor costs for multi-language teams. Relying on a common semantic coordinate system and a cross-language embedding model, it achieves semantic aggregation of synonymous concepts; combined with semantic consistency verification, it ensures that the core meaning of multi-language versions is unbiased, avoiding semantic fragmentation problems in cross-language synchronization from the source. A dual compliance verification system combining automated regulatory retrieval and manual verification is established to accurately match regulatory requirements in various countries. This system promptly triggers legal intervention for any questionable content, significantly reducing compliance risks in global operations. Update records and automated reports with unique version identifiers are generated, leaving a complete trail of key information such as changes and their impact, providing clear evidence for audits and business reviews, and improving the standardization of knowledge management. The system ensures the timeliness and accuracy of multilingual knowledge throughout the entire process. After synchronization with RAG Q&A systems in various countries, it can quickly support customer inquiries and internal business queries in different regions, providing stable knowledge support for the insurance company's global expansion.

[0056] Specific limitations regarding the knowledge editing and updating system for cross-language insurance Q&A can be found in the limitations on the knowledge editing and updating methods for cross-language insurance Q&A mentioned above, and will not be repeated here. Each module in the aforementioned knowledge editing and updating system for cross-language insurance Q&A can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0057] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a cross-language insurance question-and-answer knowledge editing and updating method on the server side.

[0058] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input system connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of a cross-language insurance question-and-answer knowledge editing and updating method. In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Models for different languages ​​achieve unified storage of documents in different languages ​​based on a common semantic coordinate system; When an insurance company updates its policies or product descriptions, the knowledge editing function triggers the location module to identify and locate the points of content change. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function is used to locate the content change point corresponding to the knowledge correction gradient signal, and the cross-language synchronous update is completed through a shared editing space after one modification, and globally synchronized. Perform cross-language consistency verification on the synchronized updated content and complete necessary corrections and optimizations; Once all language versions have been updated, execute the secure output release process.

[0059] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Models for different languages ​​achieve unified storage of documents in different languages ​​based on a common semantic coordinate system; When an insurance company updates its policies or product descriptions, the knowledge editing function triggers the location module to identify and locate the points of content change. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function is used to locate the content change point corresponding to the knowledge correction gradient signal, and the cross-language synchronous update is completed through a shared editing space after one modification, and globally synchronized. Perform cross-language consistency verification on the synchronized updated content and complete necessary corrections and optimizations; Once all language versions have been updated, execute the secure output release process.

[0060] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0061] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program 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 may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of 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.

[0062] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0063] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A knowledge editing and updating method for cross-language insurance question answering, characterized in that, include: Models for different languages ​​achieve unified storage of documents in different languages ​​based on a common semantic coordinate system; When an insurance company updates its policies or product descriptions, the knowledge editing function triggers the location module to identify and locate the points of content change. When the Chinese model generates a knowledge correction gradient signal, the knowledge editing function is used to locate the content change point corresponding to the knowledge correction gradient signal, and the cross-language synchronous update is completed through a shared editing space after one modification, and globally synchronized. Perform cross-language consistency verification on the synchronized updated content and complete necessary corrections and optimizations; Once all language versions have been updated, execute the secure output release process.

2. The knowledge editing and updating method for cross-language insurance question answering as described in claim 1, characterized in that, The steps for achieving unified storage of documents in different languages ​​based on a common semantic coordinate system include: Construct a common semantic coordinate system for different language models; By training a cross-linguistic embedding model, texts from different languages ​​are encoded into vector representations, enabling synonymous concepts to exhibit semantic aggregation characteristics in a high-dimensional space. The knowledge retrieval layer stores vectorized knowledge documents in all languages ​​through a unified vector index library.

3. The knowledge editing and updating method for cross-language insurance question answering as described in claim 1, characterized in that, The steps for identifying and locating content changes by triggering the positioning module through the knowledge editing function when an insurance company updates its policies or product descriptions include: When insurance companies update policies or product descriptions, the document monitoring mechanism and semantic differential detection technology are combined to achieve accurate perception of content changes. Calculate the vector distance between the documents before and after the update. If the change in vector distance exceeds a preset threshold, the knowledge node is updated and a structured editing task is generated.

4. The knowledge editing and updating method for cross-language insurance question answering as described in claim 1, characterized in that, The steps for accurately sensing content changes when an insurance company updates its policies or product descriptions, combining document monitoring mechanisms and semantic differential detection technology, include: When insurance companies update their policies or product descriptions, they can accurately perceive the changes by combining newly uploaded terms and conditions documents in the RAG knowledge base with the content changes between the old and new documents.

5. The knowledge editing and updating method for cross-language insurance question answering as described in claim 1, characterized in that, The steps of locating the content change point corresponding to the knowledge correction gradient signal through the knowledge editing function when the Chinese model generates the knowledge correction gradient signal, and completing a one-time modification and globally synchronized cross-language synchronous update through the shared editing space include: After generating the knowledge correction gradient signal in the Chinese model, the corresponding model update is presented in the mathematical form of a low-rank matrix, which is a lightweight "parameter patch". The parameter patches are mapped to a shared editing space and passed to each target language model through a language mapping matrix. At the same time, by constraining the semantic consistency of the output of the multilingual models, the meaning of the answers in different languages ​​is ensured to remain highly consistent.

6. The knowledge editing and updating method for cross-language insurance question answering as described in claim 1, characterized in that, The steps of performing cross-language consistency verification on the synchronized updated content and completing necessary corrections and optimizations include: The semantic consistency check system calculates the similarity of multilingual outputs to verify the semantic correctness of the synchronized updated content. Retrieve relevant provisions from national legal databases and verify their consistency with local laws; If there are uncertainties during the compliance verification process, a compliance discrepancy alert will be generated and transferred to legal personnel for manual confirmation.

7. The knowledge editing and updating method for cross-language insurance question answering as described in claim 1, characterized in that, The steps for executing the secure output release process after all language versions have been updated include: Once all language versions have been updated, a knowledge update record containing version identifiers will be generated. The knowledge update records are synchronized to the RAG Q&A systems in various countries, and an automated report is generated, listing the modified content, the scope of impact, and the status of manual review.

8. A cross-language insurance question-and-answer knowledge editing and updating system, characterized in that, include: The storage module is used to unify the storage of documents in different languages ​​based on a common semantic coordinate system. The location module is used to identify and locate the points of content change when the insurance company updates its policies or product descriptions, triggered by the knowledge editing function. The update module is used to locate the content change points corresponding to the knowledge correction gradient signal when the Chinese model generates the knowledge correction gradient signal, and complete the cross-language synchronous update of one modification and global synchronization through the shared editing space; The verification module is used to perform cross-language consistency verification on the synchronized updated content and to perform necessary corrections and optimizations. The output module is used to execute the secure output release process after all language versions have been updated.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the knowledge editing and updating method for cross-language insurance question answering as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the knowledge editing and updating method for cross-language insurance question answering as described in any one of claims 1 to 7.

Images