Fund product parameter auditing management system and method

Abstract

The invention provides a fund product parameter auditing management system and method, and the system comprises an announcement collection module which collects fund product files, obtains standard files according to a preset structured analysis method, and uploads the standard files to an auditing database; obtaining audit file data based on the currently input audit file and a preset structured analysis method; the auditing configuration module configures auditing rules based on the auditing file data and the auditing database; and the intelligent auditing module obtains related paragraph information based on a preset auditing rule, an auditing database and auditing file data so as to obtain an auditing result according to a user question mode based on the related paragraph information. The invention aims to provide the fund product parameter auditing management system and method, so as to solve the problems that the prior art is difficult to flexibly adapt to an actual complex application scene, cannot meet business requirements and is high in data maintenance cost, remarkably improve the timeliness and accuracy of data, and provide a more efficient, accurate and intelligent auditing scheme.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fund operation, in particular to a fund product parameter auditing management system and method. BACKGROUND

[0002] In the fund product operation business, the operation and management work has the characteristics of strong professionalism, complex process, high precision, etc. The fund operation support personnel need to extract fund operation parameters, rate information, transaction rules and other key data from various legal documents, regular announcements, temporary announcements and other unstructured data sources, and complete system parameter configuration within a strict time window. This operation process runs through the whole life cycle of the fund product, including product issuance, establishment, existence, change, termination and other stages. Due to the diversity and dispersion of information sources, as well as the strict timeliness requirement, the traditional manual processing method has the problems of high operation risk, low efficiency, etc.

[0003] The existing technology mainly adopts the following technical scheme for data collection and extraction: for structured data, data is usually collected directly from the source system; for unstructured data, specific regular expression rules are defined based on product parameter characteristics to extract data.

[0004] However, in actual operation scenarios, a large amount of data is difficult to be structured and stored in a short period of time, especially the key data in the business process often comes from non-standardized scenarios, such as email notifications, unstructured documents, etc. The existing technology cannot effectively process such unstructured data, resulting in low data collection efficiency. In addition, the existing technology has high requirements for business processes and system construction. When the business process, especially the cross-department business process, cannot be improved, and the data cannot be completely structured and stored, business personnel still have to use the offline manual collection and arrangement scheme. Business personnel extract parameters from unstructured documents such as product legal documents one by one and manually enter the system. However, this process is prone to errors such as misfilling and missing, which makes it difficult to guarantee data accuracy. Moreover, since these parameters are related to fund valuation calculation, compliance risk control, information disclosure and other downstream business systems, any error may trigger a chain reaction and cause serious business risks. Therefore, in the case of diverse and non-standardized data sources, the existing technology lacks flexibility and adaptability, and cannot effectively meet business needs. SUMMARY

[0005] The present application aims to provide a fund product parameter auditing management system and method to solve the problems of the existing technology in the actual complex application scenarios, such as lack of flexibility and adaptability, inability to meet business needs, high data maintenance cost, and insufficient data consistency and accuracy, significantly improve the timeliness and accuracy of data, reduce manual intervention, reduce data maintenance cost, and provide a more efficient, accurate and intelligent solution for business scenarios.

[0006] To solve the above technical problems, the application provides a fund product parameter auditing management system, comprising an announcement collection module, an auditing configuration module and an intelligent auditing module, wherein:

[0007] The announcement collection module is used to collect fund product files, obtain standard files based on the fund product files according to a preset structured parsing method, and upload the standard files to an auditing database;

[0008] The announcement collection module is also used to obtain auditing file data based on a currently input auditing file and a preset structured parsing method;

[0009] The auditing configuration module is used to configure auditing rules based on the auditing file data and the auditing database;

[0010] The intelligent auditing module is used to obtain relevant paragraph information based on preset auditing rules, the auditing database and the auditing file data, so as to obtain an auditing result according to a user questioning mode based on the relevant paragraph information.

[0011] The fund product parameter auditing management system provided by the application integrates the announcement collection module, the auditing configuration module and the intelligent auditing module, integrates file standardization preprocessing and intelligent auditing, significantly improves the reuse efficiency and knowledge sedimentation capability of product announcement files, extracts and structures unstructured data through the information extraction and structured parsing in the announcement collection module, integrates and processes different types of announcement files into standardized structured files that can be intelligently processed, greatly improves the auditing service efficiency, avoids the risk of human error, avoids the resource waste problem existing in traditional manual auditing, and automatically configures and intelligently audits based on the relevant information of the auditing file through the auditing configuration module and the intelligent auditing module, solves the problems that the existing technology is difficult to flexibly adapt to actual complex application scenarios, cannot meet business requirements, has high data maintenance cost, and has insufficient data consistency and accuracy, significantly improves the timeliness and accuracy of data, reduces manual intervention, reduces data maintenance cost, and provides a more efficient, accurate and intelligent solution for business scenarios through the method.

[0012] Further, the announcement collection module is configured to collect the fund product file, obtain the standard file based on the fund product file according to the preset structured parsing method, and upload the standard file to the audit database, and the announcement collection module comprises a file information unit, a paragraph information unit and a standard file unit, wherein: the file information unit is configured to collect the fund product file, extract announcement related information based on the preset association field and the fund product file, and obtain an announcement information table based on the announcement related information; the paragraph information unit is configured to obtain a paragraph information table according to the preset structured parsing method and the fund product file; and the standard file unit is configured to obtain a standard file based on the paragraph information table and the announcement information table, and upload the standard file to the audit database.

[0013] In the above scheme, the announcement collection module collects, cleanses and structures the specified fund product announcement from the data source, stores and uploads the structured data to the audit database, so as to ensure the fine storage and management of the file content and provide structured data support for subsequent auditing.

[0014] Further, the paragraph information unit comprises a paragraph recognition subunit and an information processing subunit, wherein: the paragraph recognition subunit is configured to determine whether there is a picture format in the fund product file, and if so, obtain picture paragraph metadata based on the preset picture recognition technology, the preset structured parsing method and the fund product file; if not, directly extract text paragraph metadata based on the preset structured parsing method and the fund product file; and the information processing subunit is configured to take the picture paragraph metadata and the text paragraph metadata as paragraph data, identify explicit identifiers and implicit association information in the paragraph data based on a preset matching method, obtain announcement paragraph information based on the paragraph information, a preset resolution mechanism, the explicit identifiers and the implicit association information, and obtain a paragraph information table based on the announcement paragraph information.

[0015] In the above scheme, the paragraph information unit directly extracts or uses the preset picture recognition technology to realize standard structured parsing and recognition of various file formats such as Word, Excel, PDF and pictures, converts unstructured announcement file information into structured text paragraphs containing rich product elements, and ensures the fine storage and management of the file content.

[0016] Further, the standard file unit comprises a paragraph supplement subunit and a file construction subunit, wherein: the paragraph supplement subunit is configured to supplement paragraph association information to the paragraph information table according to a preset supplement method, so as to obtain a complete paragraph information table; and the file construction subunit is configured to obtain an initial paragraph structure according to a preset construction method, obtain an initial standard file based on the complete paragraph information table and the initial paragraph structure, add paragraph markers to the initial standard file based on the announcement information table to obtain a standard file, and upload the standard file to the audit database.

[0017] In the above scheme, considering the homogeneity of the fund legal documents, to avoid the loss of the file basis information in the individual document paragraph and affect the effect of subsequent parameter auditing, the standard file unit is used to supplement the associated fund product information (such as fund name, fund code) of each document paragraph, and the paragraphs of the same document are added with paragraph markers and assembled into standardized and structured standard files, thereby providing structured data support for subsequent auditing.

[0018] Further, the announcement collection module further comprises a monitoring and maintenance unit, wherein: the monitoring and maintenance unit is configured to perform timing detection on the auditing database according to a preset file version and a monitoring strategy, to trigger manual intervention alarm when an abnormal detection result is obtained; and obtain a user feedback log, to optimize the auditing database based on the user feedback log.

[0019] In the above scheme, the monitoring and maintenance unit is further arranged in the announcement collection module, to perform timing detection according to a preset file version and a monitoring strategy after the document is reorganized in a structured and standardized manner, and to scan the integrity and timeliness of the knowledge base in combination with the alarm mechanism at regular intervals, and to optimize the tag weight in combination with the user feedback log, thereby realizing self-optimization of the auditing database and ensuring its continuous availability and high matching degree with the business scenario.

[0020] Further, the intelligent auditing module is configured to obtain relevant paragraph information based on the preset auditing rules, the auditing database and the auditing file data, to obtain an auditing result based on the relevant paragraph information according to the user questioning manner, comprising: the intelligent auditing module comprises a knowledge recall unit, a prompt word assembly unit and a result analysis unit, wherein: the knowledge recall unit is configured to obtain original prompt words of a target paragraph based on the auditing file data and a preset retrieval rule, and to obtain optimized prompt words based on the preset auditing rules, the auditing database and the original prompt words, to obtain relevant paragraph information based on the optimized prompt words, semantic matching retrieval, text full-text retrieval and the auditing file data; the prompt word assembly unit is configured to obtain a target questioning sentence based on the user questioning manner, the relevant paragraph information and a preset sentence model library; and the result analysis unit is configured to obtain the auditing result based on the target questioning sentence and the user questioning manner.

[0021] In the above scheme, the intelligent auditing module realizes multi-modal analysis and semantic optimization technology by knowledge recall, prompt word assembly and result analysis in combination with the preset auditing rules, efficiently extracts and audits product parameters from the product announcement paragraphs, to realize intelligent auditing, thereby ensuring the consistency and accuracy of product elements, effectively solving the auditing problems caused by data complexity or ambiguity in traditional methods, realizing efficient processing of massive documents, and further improving the coverage and accuracy of the auditing.

[0022] Further, the prompt assembling unit comprises a type determining subunit and a sentence generating subunit, wherein: the type determining subunit is configured to acquire the question parameter content when the user question mode is determined as the preset type; and acquire the comparison parameter content based on the related paragraph information when the user question mode is determined as not the preset type; and the sentence generating subunit is configured to acquire the target question sentence based on the large model reasoning prompt word, the preset sentence model library, and the question parameter content or the comparison parameter content when the target paragraph has the large model reasoning prompt word.

[0023] Further, the sentence generating subunit is further configured to acquire the target question sentence based on the preset default value, the preset sentence model library, and the question parameter content when the target paragraph has no large model reasoning prompt word.

[0024] In the above scheme, the parameter content is processed in different ways according to the user question type to generate the most suitable question sentence, which provides accurate input for subsequent model reasoning, improves the auditing accuracy, supports different types of question and answer processing of "extraction" and "comparison", and can flexibly adjust the auditing strategy according to user needs.

[0025] Further, the result analyzing unit is configured to acquire the auditing result based on the target question sentence and the user question mode, comprising: acquiring the initial answer based on the target question sentence and the user question mode; acquiring the question auditing result based on the preset format and the initial answer when the user question mode is determined as the preset type; acquiring the comparison auditing result based on the preset format and the initial answer when the user question mode is determined as not the preset type and the initial answer is determined as correct based on the regular expression; and acquiring the reference auditing data based on the target question sentence and the user question mode, and acquiring the checking auditing result based on the preset format, the initial answer, and the reference auditing data when the user question mode is determined as not the preset type and the initial answer is determined as incorrect based on the regular expression.

[0026] In the above scheme, the result analyzing unit analyzes the target question sentence generated in the previous step to acquire the auditing result, and adopts different processing methods according to the different question modes, so as to perform review and other operations, realize efficient processing of model replies, facilitate quick extraction and review of product parameter information, and meet diversified business needs.

[0027] The application provides a fund product parameter auditing management system, which converts unstructured announcement file information into structured text segments containing rich product elements through OCR and other technologies, and assembles precise questioning prompts suitable for generating artificial intelligence models by combining specific product parameter auditing rules, and then obtains reply results based on the generating artificial intelligence models, and efficiently extracts and audits product parameter information with the highest confidence by cooperating with answer analysis knowledge base. Specifically, the file preprocessing and intelligent auditing function are integrated into an integrated platform, which significantly improves the reuse efficiency and knowledge sedimentation ability of product announcement files. By automatically processing unstructured data, the service efficiency is greatly improved, the problem of resource waste in traditional manual auditing is avoided, and the risk of human error is reduced; the generating artificial intelligence model is used to compare and extract product parameters, which can timely discover and correct inconsistent data, and through semantic understanding and intelligent reasoning, the consistency and accuracy of product elements are ensured, effectively solving the auditing problems caused by data complexity or ambiguity in traditional methods; through the knowledge fragment recall module and intelligent prompt word optimization technology, the key information related to the product parameters is quickly located and extracted, combined with vector library retrieval and full-text retrieval technology, efficient processing of massive documents is realized, and the coverage and accuracy of the audit are further improved; supporting different types of question and answer processing such as "extraction" and "comparison", the auditing strategy can be flexibly adjusted according to user needs, and then the structured result analysis is used to realize efficient processing of model replies, which is convenient for quickly extracting and reviewing product parameter information and meets the diversified business needs.

[0028] The application also provides a fund product parameter auditing management method, which is realized by using the fund product parameter auditing management system.

[0029] Collect fund product files, obtain standard files based on the fund product files according to a preset structured analysis method, and upload the standard files to an auditing database;

[0030] Knowledge collection and preprocessing are performed on the currently input auditing files to obtain auditing file data;

[0031] The auditing rules are configured based on the auditing file data and the auditing database;

[0032] Related paragraph information is obtained based on the preset auditing rules, the auditing database and the auditing file data, and the auditing results are obtained based on the related paragraph information according to the user questioning mode.

[0033] The fund product parameter auditing management method provided by the application integrates the announcement collection module, the auditing configuration module and the intelligent auditing module, integrates the file standardization preprocessing and the intelligent auditing, significantly improves the reuse efficiency and knowledge sedimentation capability of the product announcement file, extracts and structurally analyzes the unstructured data through the information extraction and the structural analysis in the announcement collection module, integrates and processes different types of announcement files into the standardized structural files that can be intelligently processed, greatly improves the auditing service efficiency, avoids the risk of human error, avoids the resource waste problem existing in the traditional manual auditing, and automatically configures and intelligently audits the related information of the auditing file through the auditing configuration module and the intelligent auditing module, solves the problems that the prior art is difficult to flexibly adapt to the actual complex application scene, cannot meet the business requirements, and has high data maintenance cost and insufficient data consistency and accuracy, significantly improves the timeliness and accuracy of the data, reduces the manual intervention, reduces the data maintenance cost, and provides a more efficient, accurate and intelligent solution for the business scene through the method. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 A fund product parameter auditing management system schematic diagram provided by an embodiment of the application;

[0035] Figure 2 A fund product parameter auditing management method schematic diagram provided by an embodiment of the application;

[0036] Figure 3 A fund product announcement collection module processing flow schematic diagram provided by an embodiment of the application;

[0037] Figure 4 A fund product intelligent auditing module processing flow schematic diagram provided by an embodiment of the application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0039] Embodiment one:

[0040] The embodiment provides a fund product parameter auditing management system, which comprises an announcement collection module, an auditing configuration module and an intelligent auditing module, and wherein:

[0041] The announcement collection module is configured to collect fund product files, obtain standard files based on the fund product files according to a preset structured analysis method, and upload the standard files to the audit database.

[0042] The announcement collection module is further configured to obtain audit file data based on the currently input audit file and the preset structured analysis method.

[0043] The audit configuration module is configured to configure audit rules based on the audit file data and the audit database.

[0044] The intelligent audit module is configured to obtain relevant paragraph information based on the preset audit rules, the audit database and the audit file data, and obtain an audit result based on the relevant paragraph information according to a user questioning mode.

[0045] The fund product parameter audit management system provided in the embodiment integrates the announcement collection module, the audit configuration module and the intelligent audit module, integrates file standardization preprocessing and intelligent audit, significantly improves the reuse efficiency and knowledge sedimentation capability of product announcement files, extracts and structures unstructured data through the information extraction and structured analysis of the announcement collection module, integrates different types of announcement files into standardized structured files that can be intelligently processed, greatly improves the audit service efficiency, avoids the risk of human error, avoids the waste of resources in traditional manual audit, and automatically configures and intelligently audits based on the relevant information of the audit file through the audit configuration module and the intelligent audit module, solves the problems of the prior art, such as difficulty in flexible adaptation to actual complex application scenarios, inability to meet business requirements, high data maintenance cost, and insufficient data consistency and accuracy, significantly improves the timeliness and accuracy of data, reduces manual intervention, and reduces data maintenance cost. Through the method, a more efficient, accurate and intelligent solution can be provided for business scenarios.

[0046] Optionally, the announcement collection module is configured to collect fund product files, obtain standard files based on the fund product files according to a preset structured analysis method, and upload the standard files to the audit database, and the announcement collection module includes a file information unit, a paragraph information unit and a standard file unit, wherein: the file information unit is configured to collect fund product files, extract announcement related information based on a preset association field and the fund product files, and obtain an announcement information table based on the announcement related information; the paragraph information unit is configured to obtain a paragraph information table according to a preset structured analysis method and the fund product files; and the standard file unit is configured to obtain a standard file based on the paragraph information table and the announcement information table, and upload the standard file to the audit database.

[0047] In the implementation process, the announcement collection module interfaces with multiple upstream data sources, collects and cleans data of the specified product announcement, and stores and uploads the structured data to the knowledge base. The file information unit first performs an announcement file information extraction action: records the announcement-related information of the announcement file in the product announcement information table, and the specific fields include: associated fund product, file number, file name, file type, file size, etc., to realize the structured storage of the basic information of the announcement file.

[0048] Optionally, the paragraph information unit includes a paragraph identification subunit and an information processing subunit. The paragraph identification subunit is configured to determine whether there is a picture format in the fund product file, and if so, obtain picture paragraph metadata based on a preset picture recognition technology, a preset structured analysis method, and the fund product file; and if not, directly extract text paragraph metadata based on a preset structured analysis method and the fund product file. The information processing subunit is configured to take the picture paragraph metadata and the text paragraph metadata as paragraph data, identify explicit identifiers and implicit associated information in the paragraph data based on a preset matching method, obtain announcement paragraph information based on the paragraph information, a preset resolution mechanism, the explicit identifiers, and the implicit associated information, and obtain a paragraph information table based on the announcement paragraph information.

[0049] In the implementation process, the paragraph information unit performs an announcement file paragraph information extraction action: for various file formats such as Word, Excel, PDF, and pictures, adopts technical means such as direct extraction or OCR recognition, and performs structured analysis according to chapter titles and natural paragraphs to generate paragraph metadata including chapter numbers and chapter titles; then, through regular expression matching, keyword pattern matching, and benchmark index matching, explicit fund identifiers (i.e., display identifiers) and implicit associated information in the paragraphs are identified, and a preset anaphora resolution mechanism (such as “the fund” corresponding to the current document fund name) is established to process context association and record in the file paragraph information table, to ensure fine storage and management of the file content. The paragraph information unit adopts direct extraction or a preset picture recognition technology to realize standard structured analysis and recognition of various file formats such as Word, Excel, PDF, and pictures, to convert unstructured announcement file information into structured text paragraphs containing rich product elements, and to ensure fine storage and management of the file content.

[0050] Optionally, the standard file unit comprises a paragraph supplement subunit and a file construction subunit, wherein: the paragraph supplement subunit is configured to supplement the paragraph information table with paragraph-related information according to a preset supplement method to obtain a complete paragraph information table; and the file construction subunit is configured to obtain an initial paragraph structure according to a preset construction method, to obtain an initial standard file based on the complete paragraph information table and the initial paragraph structure, to add paragraph markers to the initial standard file based on the announcement information table to obtain a standard file, and to upload the standard file to the audit database.

[0051] In the implementation process, the standard file unit performs a standard file assembly action: considering the homogenization characteristics of fund legal documents, to avoid the loss of file basic information of individual file paragraphs and affect the effect of subsequent parameter auditing, the standard file unit supplements each file paragraph with paragraph-related information, i.e., fund product information (such as fund name and fund code) associated therewith, and reassembles the paragraphs of the same file into a standardized and structured file. First, the system constructs a document tree with the fund code as the primary key, reorganizes the original paragraphs in a three-level nested structure of “file-chapter-paragraph”, second, injects file basic information (such as release date and document unique ID), and at the same time, adds fund identification and version stamp (i.e., paragraph markers) to each paragraph, and finally uploads the file assembled according to the specification to the knowledge base, providing structured data support for subsequent auditing.

[0052] Optionally, the announcement collection module further comprises a monitoring and maintenance unit, wherein: the monitoring and maintenance unit is configured to perform periodic detection on the audit database according to a preset file version and a monitoring strategy, to trigger a manual intervention alarm when an abnormal detection result is obtained, and to obtain user feedback logs to optimize the audit database based on the user feedback logs.

[0053] In the implementation process, the monitoring and maintenance unit in the announcement collection module performs a periodic monitoring and maintenance action: the system periodically scans the integrity (checks missing fields) and timeliness (such as invalid file fragments) of the knowledge base through a preset file version change detection monitoring strategy and an automatic alarm mechanism (such as triggering manual intervention for knowledge fragments); at the same time, the system optimizes the tag weight in combination with the user feedback logs, realizes a self-optimization closed loop of the knowledge base, and ensures its continuous availability and high matching degree with business scenarios.

[0054] In the implementation process, the verification index unit is also arranged in the announcement collection module to perform knowledge verification and indexing: after document restructuring, the system performs knowledge verification through an automated rule engine to ensure content integrity (such as checking mandatory fields and dependency association) and compliance (combined with a regulatory rule library and NLP sensitive word detection); at the same time, the system builds a multi-level intelligent index to achieve efficient retrieval through paragraph-level semantic tags and deep analysis capabilities. In addition, the system supports user feedback learning to dynamically optimize the index to significantly improve accuracy and analysis efficiency, and ultimately forms a verifiable and highly available knowledge management system.

[0055] Optionally, the intelligent auditing module is configured to obtain relevant paragraph information based on preset auditing rules, an auditing database, and auditing file data, and obtain an auditing result based on the relevant paragraph information according to a user questioning manner, including: the intelligent auditing module includes a knowledge recall unit, a prompt word assembly unit, and a result analysis unit, wherein: the knowledge recall unit is configured to obtain original prompt words of a target paragraph based on the auditing file data and the preset retrieval rules, and obtain optimized prompt words based on the preset auditing rules, the auditing database, and the original prompt words, to obtain the relevant paragraph information based on the optimized prompt words, semantic matching retrieval, text full-text retrieval, and the auditing file data; the prompt word assembly unit is configured to obtain a target questioning sentence based on the user questioning manner, the relevant paragraph information, and a preset sentence model library; and the result analysis unit is configured to obtain the auditing result based on the target questioning sentence and the user questioning manner.

[0056] In the implementation process, the intelligent auditing module is based on generative artificial intelligence technology and combines multi-modal analysis capabilities to support manual uploading by the user or automatic screening by the system, and efficiently extracts and audits product parameters from product announcement paragraphs. At the same time, to adapt to different formats, different types of product announcements, and different types of product parameters, and to improve the auditing effect, a set of customizable auditing rule configuration schemes are provided, specifically, the preset auditing rules are configured according to the following dimensions according to the differences in product type, announcement file type, product parameter type, and other attributes:

[0057] (1) Product type: including but not limited to stock type, pure debt type, mixed type, ETF, ETF connection, etc. Different product types correspond to different announcement templates, and the sections and paragraph positions of their product parameter information are significantly different. By configuring the product type, the corresponding auditing rules can be accurately matched.

[0058] (2) Announcement file type: including but not limited to fund contract, prospectus, trust agreement, product data summary, etc. Different announcement files cover different product parameter information, and by configuring the file type, relevant parameters can be extracted and audited.

[0059] (3) Product parameter type: According to the different parameter formats, product parameters are divided into basic information, redemption information, and rate information, etc. For example, redemption information needs to be distinguished between "personal" and "institutional", and rate information may be segmented by time and amount, and the audit rules need to be configured for each interval to ensure the comprehensiveness and accuracy of the audit.

[0060] (4) Positioner: According to the different positioning methods of the text section, the positioner is divided into absolute positioning and relative positioning. Absolute positioning directly positions the text section content through the chapter number, and relative positioning positions based on the semantic similarity of the chapter title. If the positioner is not configured, the relevant fragments are recalled from the knowledge base by default.

[0061] (5) Prompt word configuration: Prompt words are divided into knowledge base prompt words and large model reasoning prompt words. Knowledge base prompt words are used to improve the accuracy of recalling document content from the knowledge base, and are recommended to be highly related to the document content. If not configured, the field name is used for recall by default. Large model reasoning prompt words are used to guide the large model to analyze the content and improve the identification accuracy. If not configured, the field name is used for judgment by default.

[0062] For any product parameter that needs to be audited, the classification rules can be configured according to the above dimensions. The unit provides a visual configuration interface, and business personnel can flexibly customize the configuration according to actual needs to meet the audit requirements in different scenarios.

[0063] In the specific implementation process, when the knowledge recall unit recalls the knowledge fragments, it first obtains the input knowledge document name through the audit file data, and performs document knowledge retrieval based on the preset retrieval rules according to the knowledge document name. Combined with the preset audit rules configured in the above, the original prompt words of the specified field are optimized through the special prompt words of the audit database. Subsequently, based on the optimized prompt words, content fragments are retrieved from the document knowledge. Usually, two retrieval methods are used: semantic matching retrieval based on vector library: retrieve fragments with high semantic and content matching through vector library; text full-text retrieval based on text content: retrieve fragments related to product parameters from the document. Finally, the product announcement paragraph containing product parameter information (i.e. relevant paragraph information) is selected, providing a data basis for subsequent audit.

[0064] Optionally, the prompt word assembly unit includes a type determination sub-unit and a sentence generation sub-unit, wherein: the type determination sub-unit is used to obtain the question parameter content when the user question mode is determined as a preset type; when the user question mode is not determined as a preset type, the comparison parameter content is obtained based on the relevant paragraph information; the sentence generation sub-unit is used to obtain the target question sentence based on the large model reasoning prompt word, the preset sentence model library, and the question parameter content or the comparison parameter content when the target paragraph has a large model reasoning prompt word.

[0065] Optionally, the sentence generation subunit is further configured to determine that the target passage does not have a large model inference prompt word, and then obtain the target questioning sentence based on a preset default value, a preset sentence model library, and a questioning parameter content.

[0066] In the specific implementation process, the answer content is divided into two types of “extraction” and “comparison” according to the questioning mode selected by the user. For the extraction type: the expected model extracts the parameter content that meets the requirements based on the semantic information of the input text segment and the product parameters; for the comparison type: the pre-existing parameter content needs to be reviewed according to the text segment information to ensure the accuracy of the parameters. If there is a large model inference prompt word in the specified field, the specified field is optimized according to the prompt word first. Then, the prompt word is analyzed by lexical analysis and syntax analysis, and each model in the sentence model library is matched to generate the most suitable target questioning sentence, providing accurate input for model inference, and realizing large model prompt word assembly.

[0067] Optionally, the result analysis unit is configured to obtain an audit result based on the target questioning sentence and the user questioning mode, including: obtaining an initial answer based on the target questioning sentence and the user questioning mode; when the user questioning mode is determined to be a preset type, obtaining a questioning audit result based on a preset format and the initial answer; when the user questioning mode is determined to be not the preset type, and the initial answer is determined to be correct based on a regular expression, obtaining a comparison audit result based on the preset format and the initial answer; when the user questioning mode is determined to be not the preset type, and the initial answer is determined to be incorrect based on the regular expression, obtaining reference audit data based on the target questioning sentence and the user questioning mode, and obtaining a checking audit result based on the preset format, the initial answer, and the reference audit data.

[0068] In the specific implementation process, the answer is obtained by inputting the prompt word information and the target questioning sentence generated in the above into the model. The results of the large model answer are also processed differently according to the different questioning types. For the answer of the extraction type: the answer is returned in the form of a JSON string, for example: [{settlement date: 20231114}, {fund code: 000059}, {sales organization code: 211}], the product parameter content in the answer can be conveniently extracted through the key-value pair method. For the answer of the comparison type: first, it is determined whether the answer is correct through a regular expression. If the answer is incorrect, the model is required to return in the form of a JSON string, and the parameter content considered by the model is extracted for further review.

[0069] The fund product parameter auditing management system provided by the embodiment converts unstructured announcement file information into a structured text segment containing rich product elements through technologies such as OCR, and assembles a precise question prompt suitable for a generative artificial intelligence model based on specific product parameter auditing rules, and then obtains a reply result based on the generative artificial intelligence model, efficiently extracts and audits product parameter information with the highest confidence by matching the answer to analyze the knowledge base. Specifically, the file preprocessing and intelligent auditing function are integrated into an integrated platform, significantly improving the reuse efficiency and knowledge sedimentation capability of product announcement files. By automatically processing unstructured data, service efficiency is greatly improved, avoiding the waste of resources in traditional manual auditing, while reducing the risk of human error; using a generative artificial intelligence model to compare and extract product parameters can quickly discover and correct inconsistent data, and through semantic understanding and intelligent reasoning, the consistency and accuracy of product elements are ensured, effectively solving the auditing problems caused by data complexity or ambiguity in traditional methods; through the knowledge fragment recall module and intelligent prompt word optimization technology, key information related to product parameters is quickly located and extracted, combined with vector library retrieval and full-text retrieval technology, efficient processing of massive documents is realized, further improving the coverage and accuracy of auditing; supporting different types of question and answer processing such as "extraction" and "comparison", the auditing strategy can be flexibly adjusted according to user needs, and then the structured result analysis is used to efficiently process the model reply, which is convenient for quickly extracting and reviewing product parameter information and meets the diversified business needs.

[0070] Embodiment two:

[0071] For reference Figure 1 The fund product parameter auditing management process provided by the embodiment is implemented by using the fund product parameter auditing management system described above, and the fund product files sequentially pass through the announcement collection module, the auditing configuration module and the intelligent auditing module in the process processing order, including:

[0072] Start;

[0073] The announcement collection module first uploads the fund product file, and then performs knowledge collection and preprocessing on the fund product file, including: collecting the fund product file, obtaining a standard file based on the fund product file according to a preset structured analysis method, uploading the standard file to an auditing database, and obtaining auditing file data based on the current input auditing file and the preset structured analysis method;

[0074] The auditing configuration module inputs the parameter data after knowledge collection and preprocessing into a product parameter auditing rule library to obtain an auditing rule, and then performs knowledge semantic similarity calculation based on the auditing rule, including: the auditing configuration module is configured to configure the auditing rule based on the auditing file data and the auditing database;

[0075] The intelligent auditing module processes the data processed by the auditing configuration module through the prompt statement model library to generate a target question statement, and inputs the target question statement into a question and answer processing engine to further process the extraction and auditing results: based on the preset auditing rules, the auditing database and the auditing file data, the relevant paragraph information is obtained, and the auditing results are obtained based on the relevant paragraph information according to the user's questioning method;

[0076] End.

[0077] In the specific implementation process, the embodiment aims to solve the problems of complexity of business scenarios, low efficiency, high data maintenance cost, and insufficient data consistency and accuracy in the prior art. In real business scenarios, a single department needs to improve operational efficiency through systematic means, but due to the diversity and non-standardization of data sources, the prior art is difficult to effectively cope with, especially in the product element matching process in the contract file and product management system, the cost of manual operation is high, the efficiency is low, and it is difficult to ensure the timeliness and accuracy of data. Especially in the approved-to-issue stage, the provision of accurate product element information often lags behind, affecting the business process. At the same time, due to the lack of automatic comparison and verification mechanism for product legal documents and product element information, data inconsistency is difficult to be discovered and corrected in time, resulting in the inability to guarantee the accuracy and consistency of product element information, and further affecting the normal operation of downstream business systems. To solve the above problems, the technical scheme of intelligent auditing using a large language model in the embodiment automatically parses the contract file, identifies key product element information, and compares and verifies it with existing data, thereby advancing the provision of accurate product element information to the approved-to-issue stage, significantly improving the timeliness and accuracy of data, reducing manual intervention, reducing data maintenance costs, and at the same time discovering and correcting data inconsistency in a timely manner, ensuring the consistency and accuracy of product element information, and providing reliable data support for downstream business systems. Through the implementation of the invention, the problems of high data maintenance cost, low efficiency, and insufficient accuracy in the prior art can be effectively solved, and a more efficient, accurate, and intelligent solution for business scenarios is provided.

[0078] Embodiment three:

[0079] Please refer to Figure 2 The embodiment provides a fund product parameter auditing management method, which is implemented by using the fund product parameter auditing management system, and includes the following steps:

[0080] S31, collect fund product files, obtain standard files based on the fund product files according to a preset structured parsing method, and upload the standard files to an auditing database;

[0081] S32, knowledge collection and preprocessing of the current input audit file to obtain audit file data;

[0082] S33, configuring an audit rule based on the audit file data and the audit database;

[0083] S34, obtaining relevant paragraph information based on the preset audit rule, the audit database and the audit file data, to obtain an audit result based on the relevant paragraph information according to the user's questioning mode.

[0084] The fund product parameter audit management method provided by the embodiment integrates the announcement collection module, the audit configuration module and the intelligent audit module, integrates the file standardization preprocessing and the intelligent audit, significantly improves the reuse efficiency and the knowledge sedimentation ability of the product announcement file, extracts and structures the unstructured data in the announcement collection module, and processes different types of announcement files into standard structured files that can be intelligently processed, thereby greatly improving the audit service efficiency, avoiding the risk of human error, avoiding the waste of resources in traditional manual audit, and automatically configuring and intelligently auditing based on the relevant information of the audit file through the audit configuration module and the intelligent audit module. The problems of the prior art in actual complex application scenarios, such as difficulty in flexible adaptation, inability to meet business needs, high data maintenance cost, and insufficient data consistency and accuracy are solved, the timeliness and accuracy of data are significantly improved, manual intervention is reduced, and the data maintenance cost is reduced. Through the method, a more efficient, accurate and intelligent solution can be provided for business scenarios.

[0085] Embodiment Four

[0086] Please refer to Figure 3 The fund product announcement collection module processing flow is implemented by using the fund product parameter audit management system described above, and includes the following steps:

[0087] Start;

[0088] S41, receiving fund product announcements uploaded manually by a user, synchronized by a third-party system, or crawled from an official website;

[0089] S42, data cleaning of the fund product announcement;

[0090] S43, identifying whether there is a picture format part in the fund product announcement: if yes, executing step S44, and if no, directly executing step S45;

[0091] S44, performing OCR identification on the picture format part in the fund product announcement;

[0092] S45, based on paragraph file splitting: obtaining paragraph information according to a preset structured analysis method and the fund product announcement;

[0093] S46, Assemble standard files: Assemble standard files based on the paragraph information;

[0094] S47, Upload knowledge base: Upload standard files to a database;

[0095] S48, Knowledge verification and indexing: Verify knowledge through an automated rules engine;

[0096] S49, Regular monitoring and maintenance: Regularly scan the integrity and timeliness of the knowledge base through a preset file version change detection monitoring strategy and an automated alarm mechanism, and optimize the tag weight in combination with user feedback logs to form a self-optimization closed loop for the knowledge base:

[0097] End.

[0098] In the implementation process, the product announcement collection module interfaces with multiple upstream data sources, collects and cleans the data of specified product announcements, and stores and uploads the structured data to the knowledge base. Specifically, first, the relevant information of the announcement file is recorded in the product announcement information table, including the associated fund product, file number, file name, file type, file size, etc., to realize the structured storage of the basic information of the announcement file. Then, for various file formats such as Word, Excel, PDF, and pictures, direct extraction or OCR recognition techniques are used to structure the analysis according to chapter titles and natural paragraphs, generating paragraph metadata including chapter numbers and chapter titles. Subsequently, by using regular expression matching, keyword pattern matching, and benchmark index matching, explicit fund identifiers and implicit associated information in the paragraphs are identified, and a document internal reference resolution mechanism is established (e.g., "the fund" corresponds to the current document fund name) to handle context associations and record them in the file paragraph information table, ensuring the fine storage and management of file content. Then, to avoid the loss of file basic information in individual file paragraphs affecting the effectiveness of subsequent parameter auditing, the associated fund product information (such as fund name and fund code) is supplemented for each file paragraph, and the paragraphs of the same file are reassembled into standardized and structured files: a document tree is constructed with fund code as the primary key, and the original paragraphs are reorganized according to the "file-chapter-paragraph" three-level nested structure. Secondly, file basic information (such as release date and document unique ID) is injected, and a fund identifier and version stamp are attached to each paragraph. Finally, the reassembled JSON / EXCEL files that meet the specifications are uploaded to the knowledge base, providing structured data support for subsequent auditing. After the document restructuring, knowledge verification is performed through an automated rule engine to ensure content integrity (such as checking mandatory fields and dependent item associations) and compliance (combined with the regulatory rule library and NLP sensitive word detection). At the same time, a multi-level intelligent index is constructed to achieve efficient retrieval through paragraph-level semantic tags and deep analysis capabilities. In addition, the module supports user feedback learning to dynamically optimize the index, significantly improving precision and analysis efficiency, and ultimately forming a verifiable and highly available knowledge management system. Finally, the module monitors the integrity (checks for missing fields) and timeliness (such as invalid file fragments) of the knowledge base through preset file version change detection monitoring strategies and automated alarm mechanisms (such as knowledge fragment anomalies triggering manual intervention). At the same time, the user feedback log is used to optimize the tag weight, realizing a self-optimizing closed loop of the knowledge base, ensuring its continuous availability and high matching degree with business scenarios.

[0099] Embodiment Five

[0100] For reference Figure 4 The embodiment provides a fund product intelligent auditing module processing flow, which is implemented by using the fund product parameter auditing management system.

[0101] Start;

[0102] S51, select the specified knowledge base;

[0103] S52, document knowledge retrieval: according to the input knowledge document name, based on the preset rules to carry out document knowledge retrieval;

[0104] S53, judge whether the document is retrieved: if yes, execute step S54; if no, end the process flow:

[0105] S54, knowledge base prompt word processing: combine the audit rules and the specified knowledge base to carry out prompt word optimization processing;

[0106] S55, knowledge fragment retrieval: according to the optimized prompt word to carry out semantic retrieval in the database;

[0107] S56, judge whether the input field is empty: if empty, execute step S57, if not empty, execute step S58;

[0108] S57, product parameter extraction prompt word template;

[0109] S58, product parameter audit prompt word;

[0110] S59, judge whether the large model inference prompt word is configured: if yes, execute step S510, if no, execute step S511;

[0111] S510, replace the specified large model inference prompt word;

[0112] S511, replace with default value;

[0113] S512, select the large model question;

[0114] S513, analyze the large model return result;

[0115] End.

[0116] In the specific implementation process, it is judged in step S56 whether the input field is empty, that is, the question type is judged, that is, according to the user-selected question mode, the answer content is divided into two types of "extraction" and "comparison". For the extraction type: the expected model combines the input text segment and the product parameters, and extracts the corresponding parameter content based on the semantic information; for the comparison type: the pre-existing parameter content needs to be reviewed according to the text segment information to ensure the accuracy of the parameters. If the specified field has a large model reasoning prompt word, the specified field is first optimized according to the prompt word. Subsequently, the prompt word is analyzed by lexical analysis and syntax analysis, and each model in the sentence model library is matched to generate the most suitable target question sentence, to provide accurate input for model reasoning, to realize large model prompt word assembly. Finally, according to the prompt word information and the target question sentence generated in the above, input into the large model and obtain the answer of the reply. And the result of the large model answer is processed, and different processing methods are adopted according to the difference of the question type. This embodiment supports multiple types of question and answer processing such as "extraction" and "comparison", can flexibly adjust the audit strategy according to the user's demand, realizes the efficient processing of the model reply through the structured result analysis (such as JSON format), and realizes the efficient processing of the model reply through the structured result analysis (such as JSON format). It is convenient to quickly extract and review product parameter information, and meets the diversified business needs.

[0117] The above is the preferred embodiment of the present application. It should be noted that for those skilled in the art, without departing from the principles of the present application, several improvements and refinements can be made, which are also considered within the scope of protection of the present application.

Claims

1. A fund product parameter audit management system, characterized by, The application discloses an intelligent auditing system and method for fund product files, and belongs to the field of fund product file auditing. The announcement collecting module is used for collecting fund product files, obtaining standard files based on the fund product files according to a preset structured parsing method, and uploading the standard files to an auditing database. The announcement collecting module is also used for obtaining auditing file data based on a currently input auditing file and a preset structured parsing method. The auditing configuration module is used for configuring auditing rules based on the auditing file data and the auditing database. The intelligent auditing module is used for obtaining relevant paragraph information based on preset auditing rules, the auditing database and the auditing file data, and obtaining an auditing result according to a user questioning mode based on the relevant paragraph information.

2. The fund product parameter audit management system according to claim 1, wherein, The announcement collecting module is used for collecting fund product files, obtaining standard files based on the fund product files according to a preset structured parsing method, and uploading the standard files to an auditing database, and comprises a file information unit, a paragraph information unit and a standard file unit. The file information unit is used for collecting fund product files, extracting announcement related information based on a preset associated field and the fund product files, and obtaining an announcement information table based on the announcement related information. The paragraph information unit is used for obtaining a paragraph information table according to a preset structured parsing method and the fund product files. The standard file unit is used for obtaining a standard file based on the paragraph information table and the announcement information table, and uploading the standard file to the auditing database.

3. The fund product parameter audit management system of claim 2, wherein, The paragraph information unit comprises a paragraph recognition subunit and an information processing subunit. The paragraph recognition subunit is used for determining whether there is a picture format in the fund product files, and obtaining picture paragraph metadata based on a preset picture recognition technology, a preset structured parsing method and the fund product files if the picture format exists; or directly extracting text paragraph metadata based on a preset structured parsing method and the fund product files if the picture format does not exist. The information processing subunit is used for taking the picture paragraph metadata and the text paragraph metadata as paragraph data, identifying explicit identifiers and implicit associated information in the paragraph data based on a preset matching mode, obtaining announcement paragraph information based on the paragraph information, a preset resolution mechanism, the explicit identifiers and the implicit associated information, and obtaining a paragraph information table based on the announcement paragraph information.

4. The fund product parameter audit management system of claim 2, wherein, The standard file unit comprises a paragraph supplement subunit and a file construction subunit. The paragraph supplement subunit is used for supplementing paragraph associated information to the paragraph information table according to a preset supplement method, so as to obtain a complete paragraph information table. The file construction subunit is used for obtaining an initial paragraph structure according to a preset construction method, obtaining an initial standard file based on the complete paragraph information table and the initial paragraph structure, adding paragraph labels to the initial standard file based on the announcement information table to obtain a standard file, and uploading the standard file to the auditing database.

5. The fund product parameter audit management system of claim 2, wherein, The announcement collecting module further comprises a monitoring and maintaining unit. The monitoring maintenance unit is configured to perform periodic detection on the audit database according to a preset file version and a monitoring strategy, to trigger manual intervention alarm when an abnormal detection result is obtained, and to obtain user feedback logs to optimize the audit database based on the user feedback logs.

6. The fund product parameter audit management system of claim 1, wherein, The intelligent audit module is configured to obtain relevant paragraph information based on preset audit rules, an audit database, and audit file data, to obtain an audit result in a user questioning manner based on the relevant paragraph information, and includes a knowledge recall unit, a prompt assembly unit, and a result analysis unit, wherein: The knowledge recall unit is configured to obtain original prompts of a target paragraph based on audit file data and preset search rules, to obtain optimized prompts based on preset audit rules, an audit database, and the original prompts, and to obtain relevant paragraph information based on the optimized prompts, semantic matching search, text full search, and audit file data; The prompt assembly unit is configured to obtain a target questioning sentence based on a user questioning manner, relevant paragraph information, and a preset sentence model library; The result analysis unit is configured to obtain an audit result based on the target questioning sentence and the user questioning manner.

7. The fund product parameter audit management system of claim 6, wherein, The prompt assembly unit includes a type determination subunit and a sentence generation subunit, wherein: The type determination subunit is configured to obtain questioning parameter content when the user questioning manner is determined to be a preset type, and to obtain comparison parameter content based on relevant paragraph information when the user questioning manner is determined not to be the preset type; The sentence generation subunit is configured to obtain a target questioning sentence based on large model reasoning prompts, a preset sentence model library, and questioning parameter content or comparison parameter content when the target paragraph contains the large model reasoning prompts.

8. The fund product parameter audit management system of claim 7, wherein: The sentence generation subunit is further configured to obtain a target questioning sentence based on a preset default value, a preset sentence model library, and questioning parameter content when the target paragraph does not contain large model reasoning prompts.

9. The fund product parameter audit management system of claim 7, wherein, The result analysis unit is configured to obtain an audit result based on the target questioning sentence and the user questioning manner, including: obtaining an initial answer based on the target questioning sentence and the user questioning manner; obtaining a questioning audit result based on a preset format and the initial answer when the user questioning manner is determined to be a preset type; obtaining a comparison audit result based on a preset format and the initial answer when the user questioning manner is determined not to be the preset type and the initial answer is determined to be correct based on a regular expression; obtaining a checking audit result based on the target questioning sentence and the user questioning manner, and based on a preset format, the initial answer, and reference audit data when the user questioning manner is determined not to be the preset type and the initial answer is determined to be incorrect based on a regular expression.

10. A method for fund product parameter audit management, characterized in that, The fund product parameter audit management system is implemented by using any one of the fund product parameter audit management systems of claims 1 to 9. Collecting a fund product file, obtaining a standard file according to a preset structured parsing method based on the fund product file, and uploading the standard file to an audit database; Knowledge collection and preprocessing are performed on a currently input audit file to obtain audit file data; An audit rule is configured based on the audit file data and the audit database; Based on a preset audit rule, an audit database and audit file data, relevant paragraph information is obtained, and an audit result is obtained based on the relevant paragraph information according to a user questioning mode.

Images