Text content checking method and device, storage medium and equipment
By using large language models to drive text structured mapping and difference detection, the problem of low accuracy in compliance verification of financial product text content has been solved, achieving efficient automated verification, reducing manual costs and improving risk identification capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING WATERDROP TECH GRP CO LTD
- Filing Date
- 2026-05-08
- Publication Date
- 2026-06-05
AI Technical Summary
The accuracy of compliance verification of existing financial product text content is poor, and manual verification is inefficient, making it difficult to adapt to the needs of high-frequency product iteration.
The system employs text structure mapping and prompt word extraction based on a large language model, combined with a business keyword database for prompt word selection. It utilizes two major language models for structure extraction and difference detection, generates verification results, and performs automatic correction.
It improves the accuracy and effectiveness of text content verification, reduces the cost of manual review, and achieves full coverage and enhanced risk identification capabilities.
Smart Images
Figure CN122154689A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of natural language processing technology, and in particular to a method, apparatus, storage medium, and device for verifying text content. Background Technology
[0002] As the penetration of the internet and financial products continues to increase, compliance testing is necessary to ensure the compliance of financial products. This is especially true for insurance products, which involve significant customer interests and require legal compliance verification of their content.
[0003] Currently, compliance verification of financial product texts typically relies on manual comparison. For example, financial institutions often depend on internal professionals to perform clause comparison tasks, such as staff in actuarial, legal compliance, and underwriting management roles, to make human judgments. However, due to the large volume of financial product texts, manual compliance verification has low accuracy, consumes significant compliance human resources, has a low adaptability rate to multi-dimensional compliance requirements, and struggles to meet the demands of the industry's digital transformation and high-frequency product iteration. Summary of the Invention
[0004] In view of this, this application provides a method, apparatus, storage medium and device for verifying text content, the main purpose of which is to solve the problems of poor accuracy and effectiveness in the compliance verification of text content of existing financial products.
[0005] According to one aspect of this application, a method for verifying text content is provided, comprising: Obtain the text content to be inspected and the text content of the compliance standards; Based on the text content, a text structure mapping and a first prompt word and a second prompt word are determined. Based on the first large language model, the text structure mapping, the first prompt word, and the compliance standard text content are structured and extracted to obtain the text extraction result. The first prompt word is used to characterize words that indicate that the text content matches the clause statement, and the second prompt word is used to characterize words that indicate that the text content matches the consistency comparison statement. Based on the second language model, the text extraction results and the second prompt word are used to perform difference detection to obtain the verification result of the text content.
[0006] Furthermore, the step of determining the text structure mapping and the first and second prompt words based on the text content includes: Obtain a business keyword library, which includes keywords set based on knowledge information in the financial product field; Based on the business keyword library, matching keywords are extracted from the text content, and a text structure mapping is constructed based on the keywords; In response to a business verification task instruction, a first prompt word and a second prompt word are dynamically filtered based on the received business verification task information. The first prompt word includes role information, task instruction, and a first output constraint. The second prompt word includes data source information, verification logic information, and a second output constraint. The business verification task information is used to characterize the information for verifying and filtering the first prompt word and the second prompt word.
[0007] Furthermore, the text extraction results obtained by performing structured extraction on the text structured mapping, the first prompt word, and the compliant standard text content based on the first major language model include: The first input text is constructed based on the text structure mapping, the first prompt word, and the compliant standard text content, and the first large language model is invoked. The first input text is used as input to the first large language model for processing to obtain text extraction results, which include the target text and the matching status of the target text.
[0008] Furthermore, the step of performing difference detection on the text extraction results and the second prompt word based on the second language model to obtain the verification result of the text content includes: The second input text is constructed based on the second prompt word and the text extraction result, and the second large language model is invoked; The second input text is processed as input to the second large language model to obtain the verification result, which includes the compliance detection result and the verification result with the difference identifier.
[0009] Furthermore, after performing difference detection on the text extraction results and the second prompt word based on the second language model to obtain the verification result of the text content, the method further includes: If the compliance test result is an abnormal verification result, a verification warning message is generated and the verification result with a difference identifier is output. In response to the correction instruction of the verification result, at least one of the text content, the first prompt word, and the second prompt word is corrected based on the correction information carried by the correction instruction.
[0010] Furthermore, after obtaining the text content to be inspected and the compliance standard text content, the method further includes: The text content and the compliance standard text content are denoised, and the denoised text content and the compliance standard text content are logically structured according to regular expressions to obtain multiple text content and compliance standard text content with independent logical structures.
[0011] Furthermore, the method also includes: The status indicator of the text content is detected, and the status indicator is used to represent the status of the standard interface. When the status identifier is the first identifier, the compliance standard text content in the target compliance page is retrieved through the standard interface, and the target compliance page is marked with a source tracing identifier in order to trace the source of the compliance standard text content; When the status identifier is the second identifier, an interface exception warning message is generated to reload the compliance standard text content.
[0012] According to another aspect of this application, a text content verification device is provided, comprising: The acquisition module is used to acquire the text content to be inspected and the text content of the compliance standards. The extraction module is used to determine the text structure mapping and the first prompt word and the second prompt word based on the text content, and to perform structured extraction on the text structure mapping, the first prompt word and the compliance standard text content based on the first large language model to obtain the text extraction result. The first prompt word is used to characterize words that indicate that the text content matches the clause statement, and the second prompt word is used to characterize words that indicate that the text content matches the consistency comparison statement. The detection module is used to perform difference detection on the text extraction results and the second prompt word based on the second language model to obtain the verification result of the text content.
[0013] Furthermore, the extraction module is specifically used to acquire a business keyword library, which includes keywords set based on knowledge information in the financial product domain; extract matching keywords from the text content based on the business keyword library, and construct a text structured mapping based on the keywords; respond to a business verification task instruction, dynamically filter a first prompt word and a second prompt word based on the received business verification task information, wherein the first prompt word includes role information, task instruction, and a first output constraint, and the second prompt word includes data source information, verification logic information, and a second output constraint, and the business verification task information is used to represent the information for verifying and filtering the first prompt word and the second prompt word.
[0014] Furthermore, the extraction module is specifically used to construct a first input text based on the text structure mapping, the first prompt word, and the compliant standard text content, and to call the first large language model; to process the first input text as input to the first large language model to obtain a text extraction result, wherein the text extraction result includes the target text and the matching status of the target text.
[0015] Furthermore, the detection module is specifically used to construct a second input text based on the second prompt word and the text extraction result, and to call the second large language model; the second input text is processed as input to the second large language model to obtain a verification result, the verification result including a compliance detection result and a verification result with a difference identifier.
[0016] Furthermore, the device also includes: The generation module is configured to generate a verification warning message and output the verification result with a difference identifier if the compliance detection result is an abnormal verification result; and in response to the correction instruction of the verification result, correct at least one of the text content, the first prompt word, and the second prompt word based on the correction information carried by the correction instruction.
[0017] Furthermore, the device also includes: The processing module is used to denoise the text content and the compliance standard text content, and to perform logical structure processing on the denoised text content and the compliance standard text content according to regular expressions to obtain multiple text content and compliance standard text content with independent logical structures.
[0018] Furthermore, The detection module is further configured to detect the status identifier of the text content, which is used to characterize the status of the standard interface. When the status identifier is a first identifier, the compliance standard text content in the target compliance page is retrieved through the standard interface, and the target compliance page is traced to trace the source of the compliance standard text content. When the status identifier is a second identifier, an interface anomaly warning message is generated to reload the compliance standard text content.
[0019] According to another aspect of this application, a computer-readable storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform an operation corresponding to the text content verification method described above.
[0020] According to another aspect of this application, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-mentioned text content verification method.
[0021] By employing the above technical solutions, the technical solutions provided in the embodiments of this application have at least the following advantages: This application provides a method, apparatus, storage medium, and device for verifying text content. Compared with the prior art, the embodiments of this application obtain the text content to be verified and the compliance standard text content; determine the text structure mapping and first and second prompt words based on the text content; and perform structured extraction on the text structure mapping, the first prompt words, and the compliance standard text content based on a first major language model to obtain the text extraction result. The first prompt word is used to characterize words indicating that the text content matches clause statements, and the second prompt word is used to characterize words indicating that the text content matches consistency comparison statements; perform difference detection on the text extraction result and the second prompt word based on a second major language model to obtain the verification result of the text content. By filtering and accurately focusing on contextual retrieval, noise input in the comparison process is reduced, and the two major language models are organically combined at the architectural level, which significantly enhances the risk identification capability, greatly improves interpretability, reduces the cost of manual review, and achieves full coverage of text verification, thereby improving the effectiveness of text content verification.
[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description
[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a text content verification method provided in an embodiment of this application is shown; Figure 2 This illustration shows a block diagram of a text content verification device provided in an embodiment of this application; Figure 3A schematic diagram of the structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation
[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0026] The embodiments of this invention can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.
[0027] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.
[0028] Based on this, in one embodiment, the present invention provides a method for verifying text content. Taking the application of this method to computer devices such as servers as an example, the server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.
[0029] This application provides a method for verifying text content, such as... Figure 1 As shown, the method includes: 101. Obtain the text content to be inspected and the compliance standard text content.
[0030] In this embodiment, the current execution terminal, acting as the entity verifying the text content, first obtains the text content and the compliance standard text content. The text content is a descriptive text, which may include, but is not limited to, product descriptions, product transaction information, and product claims information; this embodiment does not impose specific limitations on this. Meanwhile, the compliance standard text content refers to financial product standards established within the financial industry. For example, for insurance products, the compliance standard text content may include details of coverage, application instructions, and insurance terms; this embodiment does not impose specific limitations on this either.
[0031] It should be noted that the text content can be directly entered by the user or automatically generated by the system. At the same time, the compliant standard text content can be directly entered or obtained by requesting from public financial information websites. This application embodiment does not make specific limitations.
[0032] In some embodiments, the text content is applicable in fields such as finance and healthcare. For example, in the financial field, it can be applied to compliance comparison scenarios of multiple versions of legal texts such as credit contracts and financial product prospectuses. In the healthcare field, it can be applied to version management and patient rights verification of key documents such as informed consent forms and treatment agreements. For example, the text content of financial products in the financial field may include, but is not limited to, explanatory texts corresponding to financial products such as credit products, insurance products, and futures products; this application embodiment does not impose specific limitations.
[0033] 102. Based on the text content, determine the text structure mapping and the first prompt word and the second prompt word, and perform structured extraction on the text structure mapping, the first prompt word and the compliant standard text content based on the first large language model to obtain the text extraction result.
[0034] In this embodiment, the text structure mapping refers to the mapping relationship constructed between words or data in the text content according to the structure. For example, the text structure mapping is extracted from the text content, consisting of page clause name - clause statement - number of conditions. This embodiment does not impose specific limitations. Additionally, the first prompt word is used to characterize a word indicating that the text content matches a clause statement, such as a matching instruction for a clause name. The second prompt word is used to characterize a word indicating that the text content matches a consistency comparison statement, such as a consistency comparison instruction for time a or amount b. Here, the first and second prompt words can be entered by the user according to the text content, generated by the system according to the text content, or constructed based on the business needs of financial products. This embodiment does not impose specific limitations.
[0035] It should be noted that after determining the text structure mapping, the current execution end takes the text structure mapping, the first prompt word, and the compliant standard text content as input, and feeds them into the Large Language Model (LLM) for structure extraction to obtain the text extraction result. Here, the Large Language Model (LLM) refers to an artificial intelligence model based on deep learning and trained with massive amounts of text data to achieve the understanding and generation of human language.
[0036] 103. Based on the second language model, perform difference detection on the extracted text and the second prompt word to obtain the verification result of the text content.
[0037] In this embodiment, for the text extraction result obtained by the first large language model, the current execution end combines the text extraction result with the second prompt word as the input of the second large language model to perform difference detection and obtain the final verification result. The second large language model (LLM) can be the same large language model as the first large language model, or it can be a separately pre-trained large language model; this embodiment does not impose specific limitations.
[0038] In some embodiments, after obtaining the verification results, a structured detection report is generated and pushed through the business interaction layer. The report content includes product code, clause name, difference fields, page value, file value, risk level, and model inference basis. Upon receiving the report, the business interaction layer triggers the notification submodule, sending a pending task message to the business manager corresponding to this financial product, prompting "Compliance detection result is inconsistent, please confirm." After logging into the system and viewing the highlighted differences and model inference basis, the manager selects "Confirm Risk and Correct." Furthermore, they can choose to jump to the correction submodule, update the page text to "30 days," and automatically trigger a new round of regression testing until verification is passed, achieving closed-loop management.
[0039] In another embodiment of this application, for further definition and explanation, the step of determining the text structure mapping and the first and second prompt words based on the text content includes: Obtain a business keyword database; Based on the business keyword library, matching keywords are extracted from the text content, and a text structure mapping is constructed based on the keywords; In response to the business verification task instruction, the first prompt word and the second prompt word are dynamically selected based on the received business verification task information.
[0040] To enable the large language model to better understand the detection requirements and thus improve the accuracy of text content verification, the current execution end first obtains a business keyword library when determining the text structure mapping and the first and second prompt words. This business keyword library includes keywords based on knowledge information from the financial product domain. For example, for insurance products, the business keyword library may include, but is not limited to, insurance payouts, insurance claims, and medical coverage; this embodiment does not impose specific limitations. Furthermore, the current execution end extracts matching keywords from the text content based on the business keyword library. At this point, extraction can be performed directly using word matching methods to construct the text structure mapping based on the extracted keywords.
[0041] In some embodiments, the business keyword library includes waiting period, outpatient (emergency) medical insurance benefits under social security, and reimbursement ratio. The text content is scanned for keywords. For example, if the third item contains the keyword "waiting period", the keyword entry content is extracted as "The waiting period for medical treatment for illness is 15 days for the first insurance or non-renewal policy of this product, and there is no waiting period for renewal or accidental medical treatment". If the fifth item contains the keyword "outpatient (emergency) medical insurance benefits under social security", this item is marked as a core liability clause. If the seventh item contains the keyword "reimbursement ratio", the complex condition content is extracted as "If the insured takes out insurance as a participant in basic medical insurance... Plan 1... 10%... Plan 2... 30%...".
[0042] Subsequently, the user or product verification personnel send a business verification task instruction to trigger the verification of the text content. At this time, the business verification task instruction carries business verification task information to dynamically filter the first prompt word and the second prompt word. The business verification task information represents the information used for verifying and filtering the first prompt word and the second prompt word, and may include, but is not limited to, prompt words entered by the user or verification personnel for different financial products. During dynamic filtering, random filtering based on time thresholds, filtering based on scenario condition sequences, and filtering based on specified condition rules can be used to determine the first prompt word and the second prompt word. At this point, the first prompt word obtained through filtering includes role information, task instructions, and a first output constraint. The second prompt word obtained through filtering includes data source information, verification logic information, and a second output constraint. Among them, role information is used to represent the verification role, such as underwriting experts, claims personnel, etc.; task instructions are used to represent instructions for the large language model to process, such as locating the original paragraph based on the clause name; data source information is used to represent the source entity to be compared, such as a responsible person, a product name, a project name, etc.; verification logic is used to represent the logic for instructing the large language model to perform comparison, such as numerical comparison, semantic comparison, or logical conflict comparison, etc.; the first output constraint and the second output constraint are used to represent the content that restricts the output results, such as result classification, output format, etc. The embodiments of this application do not make specific limitations.
[0043] In some embodiments, to ensure accuracy when extracting structured text mappings from text content, the current execution end cleans the text content, including format unification, noise removal, and key information marking. For example, format unification refers to parsing potentially string-based JSON content in the "Insurance Notice" and "Insurance Terms" into plain text. For instance, different data types such as images, PDFs, and text are uniformly converted to the text format T_text to ensure consistent input format received by the subsequent comparison engine. Noise removal refers to removing general legal statements unrelated to the current liability terms, such as dispute resolution and contact information, retaining only paragraphs directly related to the "Social Security Outpatient (Emergency) Medical Insurance Benefits" liability. Key information marking refers to filtering and marking information, such as marking fragments containing key numerical values, such as "15 days," "30 days," and "0 yuan."
[0044] In some embodiments, when determining the text structure mapping, the text structure mapping can serve as an intermediate structured object of the large language model. For example, the waiting period mapping can be to map the third text as a key-value pair {"field":"waiting period", "condition":"first-time insurance / illness", "value":"15 days", "condition_renewal":"renewal / accident", "value_renewal":"no waiting period"}. Another example is the coverage mapping, which can be to mark the fifth text as an insurance liability definition and associate it with the main insurance type "social security outpatient (emergency) medical insurance". Yet another example is the deductible and limit mapping, which can be to extract {"deductible":"0 yuan", "daily_limit":"50 yuan", "period_limit":"100 yuan"} from the sixth text.
[0045] In another embodiment of this application, for further definition and explanation, the step of extracting the structured text mapping, the first prompt word, and the compliant standard text content based on the first large language model to obtain the text extraction result includes: The first input text is constructed based on the text structure mapping, the first prompt word, and the compliant standard text content, and the first large language model is invoked. The first input text is used as input to the first large language model for processing to obtain the text extraction result.
[0046] To achieve intelligent verification based on a large language model, the current execution end first constructs the first input text by combining the text structure mapping, the first prompt word, and the compliant standard text content during structured extraction, and then calls the first large language model. For example, the first input text is "Taking the 'Social Security Outpatient (Emergency) Medical Insurance Benefit' clause as an example, detail how to locate the relevant clause content of insurance product a from the above clause based on the clause name, and output the structured fields containing matching_status (matching status) and clause_content (clause content)". Here, the "Social Security Outpatient (Emergency) Medical Insurance Benefit" clause is the compliant standard text content, the original paragraph located based on the clause name is the first prompt word, and the structured fields containing matching_status (matching status) and clause_content (clause content) are the text structure mapping. Then, based on the first large language model trained with a large number of text samples, the first input text is processed to obtain the text extraction result. At this time, the text extraction result includes the target text and the target text's matching status.
[0047] In some embodiments, when the compliance standard text content consists of "Special Provisions," "Insurance Application Instructions," and "Insurance Terms" in the insurance field, the first large language model, combined with text structure mapping and first prompt words, can perform identification and processing respectively. For example, for processing the "Special Provisions," it can identify that the document contains multiple mixed provisions. In this case, structural parsing can be performed, and liability attribution can be determined based on the liability group name "Social Security Outpatient (Emergency) Medical Insurance Benefits." The large language model searches the entire text for precise matching titles, such as locating the relevant content of Articles 5 and 6, and extracting them in the original order to ensure the completeness of the liability description. At the same time, it can also record the reasoning trace of the extraction process, such as "precise anchoring" and "content harvesting," for subsequent comparison. As another example, for processing the "Insurance Application Instructions," the document structure can be scanned to locate the "9. Insurance Liability and Insurance Amount" section, and the definition, deductible, and payment ratio description of "Social Security Outpatient (Emergency) Medical Insurance Benefits" can be extracted. For example, when dealing with the "Insurance Terms and Conditions", the large language model locates "Article 6 Insurance Liability of this Insurance Contract" and extracts the definition of the insurance type and the waiting period agreement in the basic liability.
[0048] In another embodiment of this application, for further definition and explanation, the step of performing difference detection on the text extraction result and the second prompt word based on the second large language model to obtain the verification result of the text content includes: The second input text is constructed based on the second prompt word and the text extraction result, and the second large language model is invoked; The second input text is processed as input to the second large language model to obtain the verification result, which includes the compliance detection result and the verification result with the difference identifier.
[0049] To achieve intelligent verification based on a large language model, the current execution end first constructs a second input text using the second prompt word and the extracted text results during structured extraction, and then calls the second large language model. For example, the second input text is constructed using the extracted text results obtained from the first large language model and the second prompt word. For instance, the extracted text results might include: "The waiting period for medical treatment for illnesses under the first policy or non-renewal policy of this product is 15 days" (extracted from the policyholder's notice), "The waiting period is 30 days," "There is no waiting period for renewal" (extracted from the insurance terms), and "The waiting period for medical treatment for illnesses under the first policy or non-renewal policy of this product is 15 days" (extracted from the special agreement). At this point, the second input text can be constructed by combining the second prompt word, which could be: "Please compare whether the outpatient (emergency) medical insurance benefits within social security are consistent," and the risk level is output. Then, the second input text is processed as input to the second large language model to obtain the verification result. The final verification result includes both the compliance detection result and the verification result with discrepancy indicators.
[0050] In some embodiments, when the second language model processes the second input text in the above example, it can also perform semantic standardization. That is, by calling the knowledge enhancement layer through the second language model, it can confirm that "day" and "day" are semantically equivalent in the context of the insurance period, eliminating the interference caused by differences in unit expression. Then, numerical extraction and comparison are performed. The second language model extracts key numerical entities from the three texts, including the value of 15 in the "Insurance Notice", the value of 30 in the "Insurance Clause", and the value of 15 in the "Special Agreement". Finally, the second language model judges the consistency of the above values. The result is that the waiting period (30 days) stipulated in the "Insurance Clause" is inconsistent with the waiting period (15 days) stipulated in the "Insurance Notice" and the "Special Agreement". Finally, a judgment is made based on the pre-built compliance rule base. For example, the "Special Agreement" usually has higher legal effect than the "Insurance Clause", but the "Insurance Notice" is the basis for front-end display. If it is inconsistent with the main text of the "Insurance Clause", it may constitute a risk of misleading publicity or conflict of clauses. Therefore, such differences can be marked as "medium to high risk" and require manual confirmation of the priority of legal effect. This application embodiment does not make specific limitations.
[0051] In another embodiment of this application, for further definition and explanation, after the step of performing difference detection on the text extraction result and the second prompt word based on the second large language model to obtain the verification result of the text content, the method further includes: If the compliance test result is an abnormal verification result, a verification warning message is generated and the verification result with a difference identifier is output. In response to the correction instruction of the verification result, at least one of the text content, the first prompt word, and the second prompt word is corrected based on the correction information carried by the correction instruction.
[0052] To meet the cyclical correction requirements for text content verification and improve the accuracy of text content correction, after the current execution terminal receives the verification result, if the compliance detection result is an abnormal verification result, it indicates that there is abnormal content in the text content. Therefore, a verification warning message is generated, and a verification result with a difference identifier is output so that technical personnel or users can view and modify it. This application embodiment does not impose specific limitations. When a user or technical personnel corrects the abnormal text content, the current execution terminal can receive a correction instruction for the verification result and correct at least one of the text content, the first prompt word, and the second prompt word based on the correction information carried by the correction instruction, to ensure the correctness of the financial product text and the effectiveness of the prompt words for the next round of verification.
[0053] In another embodiment of this application, for further definition and explanation, after obtaining the text content to be inspected and the compliance standard text content, the method further includes: The text content and the compliance standard text content are denoised, and the denoised text content and the compliance standard text content are logically structured according to regular expressions to obtain multiple text content and compliance standard text content with independent logical structures.
[0054] To ensure the compliance standard text content and the validity of the text content in the large language model-based processing, and to avoid invalid text in the validation process, the current execution end can perform noise reduction processing on the text content and the compliance standard text content, and then perform logical structure processing on the noise-reduced text content and the compliance standard text content according to regular expressions to obtain multiple text content and compliance standard text content with independent logical structures.
[0055] In some embodiments, to ensure the effectiveness of text content during large language model processing and to avoid long text strings containing multiple conventions affecting text recognition, the current execution end first performs denoising and logical segmentation. Denoising refers to removing redundant spaces, newlines, and invisible characters from the text content. For example, "Special Convention: 1," is standardized to "Special Convention: 1.". Logical segmentation uses regular expressions to identify sequence patterns in the text content, such as "1," "2.", "3," etc., to cut long text into independent lists of convention entries. For example, detailContent can be cut into 11 independent convention entries; this embodiment does not impose specific limitations.
[0056] In another embodiment of this application, for further definition and explanation, the steps and methods further include: Detect the status indicator of the text content; When the status identifier is the first identifier, the compliance standard text content in the target compliance page is retrieved through the standard interface, and the target compliance page is marked with a source tracing identifier in order to trace the source of the compliance standard text content; When the status identifier is the second identifier, an interface exception warning message is generated to reload the compliance standard text content.
[0057] To meet the pre-verification requirements of text content and filter out obvious anomalies in advance, thereby improving the speed and accuracy of verification, the current execution end first checks the status identifier of the text content. At this time, the status identifier is used to characterize the standard interface status and can be marked by a financial product publishing platform or financial transaction system, for example, the status code field `code` in the JSON object returned by the system. In some embodiments, when the status identifier is a first identifier, it indicates successful access. Therefore, the compliant standard text content in the target compliance page is retrieved through the standard interface, and the target compliance page is traced to trace the source of the compliant standard text content. When the status identifier is a second identifier, it indicates access failure. Therefore, an interface anomaly warning message is generated to reload the compliant standard text content. Furthermore, the first and second identifiers can be preset to 0, 1, etc., and this application embodiment does not impose specific limitations.
[0058] In some embodiments, the status identifier is the `code` field in the returned JSON object. If `code` is not 0, `code: 0` indicates success, which is considered an interface exception and triggers a retry mechanism. If `code` is 0, the system parses the `data` array and extracts the `clauseName` (clause name), `clauseId` (unique identifier of the clause), and `detailContent` (detailed clause content) for subsequent verification steps. Additionally, a source identification identifier can be set. For example, if `clauseName` is identified as "Special Agreement" and `clauseId` is "1178193", then `clauseUrl` is stored as a source link so that subsequent manual review can directly redirect to the original document.
[0059] This application provides a method for verifying text content. Compared with the prior art, this application obtains the text content to be verified and the compliance standard text content; determines the text structure mapping and first and second prompt words based on the text content; and performs structured extraction on the text structure mapping, the first prompt words, and the compliance standard text content based on a first major language model to obtain the text extraction result. The first prompt word is used to characterize words indicating that the text content matches clause statements, and the second prompt word is used to characterize words indicating that the text content matches consistency comparison statements; performs difference detection on the text extraction result and the second prompt word based on a second major language model to obtain the verification result of the text content. By filtering and accurately focusing on contextual retrieval, noise input in the comparison process is reduced, and the two major language models are organically combined at the architectural level, which significantly enhances the risk identification capability, greatly improves interpretability, reduces the cost of manual review, and achieves full coverage of text verification, thereby improving the effectiveness of text content verification.
[0060] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this application provides a text content verification device, such as... Figure 2 As shown, the device includes: The acquisition module 21 is used to acquire the text content to be inspected and the compliance standard text content; Extraction module 22 is used to determine the text structure mapping and the first prompt word and the second prompt word based on the text content, and to perform structured extraction on the text structure mapping, the first prompt word and the compliance standard text content based on the first large language model to obtain the text extraction result. The first prompt word is used to characterize the words that indicate the text content matches the clause statement, and the second prompt word is used to characterize the words that indicate the text content matches the consistency comparison statement. The detection module 23 is used to perform difference detection on the text extraction results and the second prompt word based on the second language model to obtain the verification result of the text content.
[0061] Furthermore, the extraction module is specifically used to acquire a business keyword library, which includes keywords set based on knowledge information in the financial product domain; extract matching keywords from the text content based on the business keyword library, and construct a text structured mapping based on the keywords; respond to a business verification task instruction, dynamically filter a first prompt word and a second prompt word based on the received business verification task information, wherein the first prompt word includes role information, task instruction, and a first output constraint, and the second prompt word includes data source information, verification logic information, and a second output constraint, and the business verification task information is used to represent the information for verifying and filtering the first prompt word and the second prompt word.
[0062] Furthermore, the extraction module is specifically used to construct a first input text based on the text structure mapping, the first prompt word, and the compliant standard text content, and to call the first large language model; to process the first input text as input to the first large language model to obtain a text extraction result, wherein the text extraction result includes the target text and the matching status of the target text.
[0063] Furthermore, the detection module is specifically used to construct a second input text based on the second prompt word and the text extraction result, and to call the second large language model; the second input text is processed as input to the second large language model to obtain a verification result, the verification result including a compliance detection result and a verification result with a difference identifier.
[0064] Furthermore, the device also includes: The generation module is configured to generate a verification warning message and output the verification result with a difference identifier if the compliance detection result is an abnormal verification result; and in response to the correction instruction of the verification result, correct at least one of the text content, the first prompt word, and the second prompt word based on the correction information carried by the correction instruction.
[0065] Furthermore, the device also includes: The processing module is used to denoise the text content and the compliance standard text content, and to perform logical structure processing on the denoised text content and the compliance standard text content according to regular expressions to obtain multiple text content and compliance standard text content with independent logical structures.
[0066] Furthermore, The detection module is further configured to detect the status identifier of the text content, which is used to characterize the status of the standard interface. When the status identifier is a first identifier, the compliance standard text content in the target compliance page is retrieved through the standard interface, and the target compliance page is traced to trace the source of the compliance standard text content. When the status identifier is a second identifier, an interface anomaly warning message is generated to reload the compliance standard text content.
[0067] This application provides a text content verification device. Compared with the prior art, this application obtains the text content to be verified and the compliance standard text content; determines the text structure mapping and first and second prompt words based on the text content; and performs structured extraction on the text structure mapping, the first prompt words, and the compliance standard text content based on a first major language model to obtain the text extraction result. The first prompt word is used to characterize words indicating that the text content matches the clause statement, and the second prompt word is used to characterize words indicating that the text content matches the consistency comparison statement; and performs difference detection on the text extraction result and the second prompt word based on a second major language model to obtain the text content verification result. By filtering and accurately focusing on contextual retrieval, noise input in the comparison process is reduced, and the two major language models are organically combined at the architectural level, which significantly enhances the risk identification capability, greatly improves interpretability, reduces the cost of manual review, and achieves full coverage of text verification, thereby improving the effectiveness of text content verification.
[0068] According to one embodiment of this application, a computer-readable storage medium is provided, the storage medium storing at least one executable instruction that can perform the text content verification method in any of the above method embodiments.
[0069] Figure 3 The diagram shows a structural schematic of a computer device according to one embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the computer device.
[0070] like Figure 3 As shown, the computer device may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.
[0071] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308.
[0072] Communication interface 304 is used to communicate with other network elements such as clients or other servers.
[0073] The processor 302 is used to execute program 310, specifically to execute the relevant steps in the above-described text content verification method embodiment.
[0074] Specifically, program 310 may include program code that includes computer operation instructions.
[0075] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.
[0076] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.
[0077] Specifically, program 310 can be used to cause processor 302 to perform the following operations: Obtain the text content to be inspected and the text content of the compliance standards; Based on the text content, a text structure mapping and a first prompt word and a second prompt word are determined. Based on the first large language model, the text structure mapping, the first prompt word, and the compliance standard text content are structured and extracted to obtain the text extraction result. The first prompt word is used to characterize words that indicate that the text content matches the clause statement, and the second prompt word is used to characterize words that indicate that the text content matches the consistency comparison statement. Based on the second language model, the text extraction results and the second prompt word are used to perform difference detection to obtain the verification result of the text content.
[0078] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.
[0079] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for verifying text content, characterized in that, include: Obtain the text content to be inspected and the text content of the compliance standards; Based on the text content, a text structure mapping and a first prompt word and a second prompt word are determined. Based on the first large language model, the text structure mapping, the first prompt word, and the compliance standard text content are structured and extracted to obtain the text extraction result. The first prompt word is used to characterize words that indicate that the text content matches the clause statement, and the second prompt word is used to characterize words that indicate that the text content matches the consistency comparison statement. Based on the second language model, the text extraction results and the second prompt word are used to perform difference detection to obtain the verification result of the text content.
2. The method according to claim 1, characterized in that, The process of determining the text structure mapping and the first and second prompt words based on the text content includes: Obtain a business keyword library, which includes keywords set based on knowledge information in the financial product field; Based on the business keyword library, matching keywords are extracted from the text content, and a text structure mapping is constructed based on the keywords; In response to a business verification task instruction, a first prompt word and a second prompt word are dynamically filtered based on the received business verification task information. The first prompt word includes role information, task instruction, and a first output constraint. The second prompt word includes data source information, verification logic information, and a second output constraint. The business verification task information is used to characterize the information for verifying and filtering the first prompt word and the second prompt word.
3. The method according to claim 1, characterized in that, The text extraction results obtained by performing structured extraction based on the first language model on the text structured mapping, the first prompt word, and the compliant standard text content include: The first input text is constructed based on the text structure mapping, the first prompt word, and the compliant standard text content, and the first large language model is invoked. The first input text is used as input to the first large language model for processing to obtain text extraction results, which include the target text and the matching status of the target text.
4. The method according to claim 1, characterized in that, The step of performing difference detection on the text extraction results and the second prompt word based on the second language model to obtain the verification result of the text content includes: The second input text is constructed based on the second prompt word and the text extraction result, and the second large language model is invoked; The second input text is processed as input to the second large language model to obtain the verification result, which includes the compliance detection result and the verification result with difference identifier.
5. The method according to claim 4, characterized in that, After performing difference detection on the text extraction results and the second prompt word based on the second language model to obtain the verification result of the text content, the method further includes: If the compliance test result is an abnormal verification result, a verification warning message is generated and the verification result with a difference identifier is output. In response to the correction instruction of the verification result, at least one of the text content, the first prompt word, and the second prompt word is corrected based on the correction information carried by the correction instruction.
6. The method according to claim 1, characterized in that, After obtaining the text content to be inspected and the compliance standard text content, the method further includes: The text content and the compliance standard text content are denoised, and the denoised text content and the compliance standard text content are logically structured according to regular expressions to obtain multiple text content and compliance standard text content with independent logical structures.
7. The method according to any one of claims 1-6, characterized in that, The method further includes: The status indicator of the text content is detected, and the status indicator is used to represent the status of the standard interface. When the status identifier is the first identifier, the compliance standard text content in the target compliance page is retrieved through the standard interface, and the target compliance page is marked with a source tracing identifier in order to trace the source of the compliance standard text content; When the status identifier is the second identifier, an interface exception warning message is generated to reload the compliance standard text content.
8. A text content verification device, characterized in that, include: The acquisition module is used to acquire the text content to be inspected and the text content of the compliance standards. The extraction module is used to determine the text structure mapping and the first prompt word and the second prompt word based on the text content, and to perform structured extraction on the text structure mapping, the first prompt word and the compliance standard text content based on the first large language model to obtain the text extraction result. The first prompt word is used to characterize words that indicate that the text content matches the clause statement, and the second prompt word is used to characterize words that indicate that the text content matches the consistency comparison statement. The detection module is used to perform difference detection on the text extraction results and the second prompt word based on the second major language model, so as to obtain the verification result of the text content.
9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method of claim 1.