Methods, devices, equipment, and media for implementing intelligent quality valves in the demand submission process.

Abstract

This invention discloses a method, apparatus, device, and medium for implementing an intelligent quality valve in the requirement submission process, relating to the field of requirement process quality review in fintech. The method includes: acquiring a requirement description and uploading it to a created large-scale intelligent platform; expanding the document based on comparison with existing similar requirements and performing standardization processing to fill in missing elements, generating a requirement document; acquiring voice data from requirement review meetings and uploading it to the large-scale intelligent platform; analyzing the voice data to generate a requirement review meeting summary and submitting the summary to a DevOps platform; creating a requirement review model, which uses retrieval enhancement technology to acquire private domain data in real time to review the requirement document currently awaiting review. This application enables automated review operations, significantly reducing the burden of manual review and improving the accuracy and efficiency of the review process.

Description

Technical Field

[0001] This application relates to the field of quality review of the demand process in financial technology, specifically to a method, device, equipment and medium for implementing an intelligent quality valve in the demand submission process. Background Technology

[0002] At present, most software development teams rely on the following steps for quality control in the requirement submission stage: (1) Requirement collection and analysis: collect user requirements through meetings, interviews, etc., and have product managers or analysts conduct preliminary analysis; (2) Requirement document writing: transform the analyzed requirements into detailed documents, including functional requirements, non-functional requirements, user stories, etc.; (3) Requirement review: organize relevant personnel (such as developers, testers, stakeholders) to review the requirement documents to ensure the accuracy and feasibility of the requirements.

[0003] However, the existing methods of submitting requirements suffer from problems such as low review efficiency, frequent requirement changes, and inconsistent understanding of requirements, which seriously affect the progress and quality of related projects. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and medium for implementing an intelligent quality valve in the requirements submission process, which can realize automated review operations, greatly reduce the burden of manual review, and improve the accuracy and efficiency of review.

[0005] In a first aspect, embodiments of this application provide a method for implementing an intelligent quality valve in a demand submission process, the method comprising:

[0006] Obtain the requirement description in text or image format and upload it to the created large model intelligent platform. Expand the document based on comparison with existing similar requirements, and perform standardization to fill in missing elements to generate a requirement document.

[0007] Acquire the audio data of the requirements review meeting and upload it to the big model intelligent platform. The big model intelligent platform analyzes the audio data of the requirements review meeting to generate a summary of the requirements review meeting and submits the summary of the requirements review meeting to the DevOps platform.

[0008] A requirement review model is created. The requirement review model obtains private domain data in real time through the retrieval enhancement technology of the retrieval enhancement generation model in order to review the requirement documents to be reviewed. The private domain data includes existing requirement documents, voice data of requirement review meetings, and requirement review meeting summaries.

[0009] In conjunction with the first aspect, in one implementation, the large-scale model intelligent platform is specifically used for:

[0010] Based on the agile access mechanism, various large network models are accessed. The advantages and characteristics of various large network models are dynamically obtained by using a multi-dimensional model evaluation system. The appropriate large network models are flexibly assembled and pushed to the current business scenario through an intelligent matching mechanism.

[0011] In conjunction with the first aspect, in one implementation, the process of obtaining a requirement description in text or image form and uploading it to the created large-scale model intelligent platform, expanding the document based on comparison with existing similar requirements, and performing standardization processing to fill in missing elements, thereby generating a requirement document, specifically includes:

[0012] Obtain a requirement description in text or image format and upload it to the Big Model Intelligent Platform. When the requirement description is in image format, the Big Model Intelligent Platform uses image recognition technology to identify the image.

[0013] The big model intelligent platform identifies key information in the requirement description and matches it with similar requirements in the requirement database. Based on the structure, content and expression specifications of the requirements, and combined with the contextual semantics of the requirement description, the big model intelligent platform classifies, performs structured analysis and document expansion of the requirement description, and fills in missing elements to generate a standardized requirement document.

[0014] The elements of the requirements document include the background and objectives of the requirements, the scope of the requirements, the details of the functional requirements, the non-functional requirements, the change management mechanism, the requirements traceability matrix, risk management, and related impacts.

[0015] In conjunction with the first aspect, in one implementation, the acquisition of requirements review meeting audio data and its uploading to the large model intelligent platform, the analysis of the requirements review meeting audio data by the large model intelligent platform to generate a requirements review meeting summary, and the submission of the requirements review meeting summary to the DevOps platform, specifically includes:

[0016] Record the requirement review meeting to obtain the audio data, convert the audio data into text, and upload the audio data to the big model intelligent platform.

[0017] The large-scale intelligent platform analyzes the voice data of the requirements review meeting and generates a summary of the requirements review meeting according to preset standards. The summary of the requirements review meeting includes key discussion points and decision results.

[0018] Upload the generated requirements review meeting summary to the DevOps platform and archive it.

[0019] In conjunction with the first aspect, in one implementation, the creation of a requirements review model involves the requirements review model acquiring private domain data in real time through retrieval enhancement techniques of the retrieval enhancement generation model to review the current requirements document to be reviewed. The private domain data includes existing requirements documents, requirements review meeting audio data, and requirements review meeting summaries, specifically including:

[0020] A requirement review model is created. The requirement review model obtains private domain data from the private domain retrieval database in real time through the retrieval enhancement technology of the retrieval enhancement generation model. The private domain data includes existing requirement documents, voice data of requirement review meetings, and summary of requirement review meetings.

[0021] The requirements review model reviews the current requirements document based on the existing requirements document, answers questions about the current requirements document, and generates new private domain data based on the review results.

[0022] Specifically, when reviewing the current requirement documents awaiting review, the review process is based on a quality scoring method, as follows:

[0023] ;

[0024] in, This indicates the quality score of the current requirements document awaiting review, and when If the value is not less than the set threshold, the review is considered passed; otherwise, a suggestion for modification will be provided. Indicates the first The scores for each evaluation factor, This indicates the total number of review factors. Indicates the first The weights of each review factor, and .

[0025] In conjunction with the first aspect, in one implementation method,

[0026] The overall architecture of the retrieval enhancement generation model includes a private domain retrieval database construction stage, a retrieval ranking stage, and a suggestion generation stage.

[0027] The private domain retrieval database construction phase involves building a private domain retrieval database based on private domain data, so that when the current requirement document to be reviewed is related to existing historical requirement documents and needs to be enhanced for retrieval, data can be queried.

[0028] The retrieval and ranking stage is a stage in which data is searched and ranked in the private domain retrieval database based on the current documents to be reviewed;

[0029] The prompt generation stage is the stage in which the requirements review model generates review results based on the retrieval and ranking stage.

[0030] In conjunction with the first aspect, in one implementation method,

[0031] For the private domain retrieval database construction phase, specifically:

[0032] Collect existing private domain data, and perform content decomposition and standardized labeling on the collected private domain data to achieve data cleaning, and introduce desensitization and semantic classification mechanisms in the cleaning process;

[0033] The cleaned data is segmented into data blocks. A custom private domain-aware vector generation network is used to generate multi-dimensional dynamic knowledge vectors for the data blocks by integrating context timestamps, business domains, and structural information and taking three-channel input. This enables vector embedding of the data blocks. The custom private domain-aware vector generation network is constructed by introducing a domain attention mechanism on the basis of the Transformer encoder.

[0034] The data blocks after vector embedding are uniformly stored in a high-efficiency vector database to realize the construction of a private domain retrieval database, and an incremental monitoring and update mechanism is set up to listen for the addition and modification operations of private domain data.

[0035] For the retrieval and ranking stage, specifically:

[0036] Based on keyword matching or embedding similarity recall mechanism, a multi-task intent recognition network is introduced to construct an intent-guided recall recognition network.

[0037] For the issues or requirements in the current requirement documents to be reviewed, the constructed intent-guided recall identification network determines whether it involves a specific requirement category recall intent. If so, a recall strategy is dynamically selected, and relevant data blocks are obtained from the private domain retrieval database based on the issues or requirements and in combination with the cosine similarity calculation method to achieve data block recall. Furthermore, by introducing a semantic extension mechanism, the recall capability of hyponyms and hypernyms is enhanced based on an external general knowledge graph.

[0038] The ranking score of data blocks is calculated based on the ranking factor to refine the ranking of the recalled data blocks, resulting in refined data blocks. Specifically, the calculation of the ranking score involves...

[0039] ;

[0040] in, Indicates the first The sorting score of each data block. Indicates the first The semantic similarity between each data block and the query. Indicates the first Timeliness score for each data block Indicates the first The frequency with which each data block is used or verified by experts Indicates the first The percentage of data blocks used in previous successful reviews. , , , Indicates the weighting coefficient;

[0041] Regarding the prompt generation phase, specifically:

[0042] For the refined data blocks, a unified Prompt template is constructed based on the issues in the current requirements document to be reviewed. The Prompt template includes a requirements summary, referenced review rules, relevant historical review case snippets, and review focus points.

[0043] Based on the sorted data blocks, the context information of the requirement review model is constructed. Based on the constructed context information, the requirement review model responds to and reviews the issues or requirements in the current requirement document to be reviewed according to the Prompt template, generates review results, and actively reviews and corrects the generated review results.

[0044] Secondly, embodiments of this application provide an intelligent quality valve implementation device for a demand submission process, the intelligent quality valve implementation device for a demand submission process comprising:

[0045] The requirement document writing module is used to obtain requirement descriptions in text or image form and upload them to the created large model intelligent platform. Based on the comparison with existing similar requirements, the document is expanded and standardized to fill in missing elements and generate a requirement document.

[0046] The requirement review summary module is used to acquire the voice data of the requirement review meeting and upload it to the big model intelligent platform. The big model intelligent platform analyzes the voice data of the requirement review meeting to generate a requirement review meeting summary and submits the requirement review meeting summary to the DevOps platform.

[0047] The review processing execution module is used to create a requirement review model. The requirement review model obtains private domain data in real time through the retrieval enhancement technology of the retrieval enhancement generation model to review the requirement documents to be reviewed. The private domain data includes existing requirement documents, requirement review meeting voice data, and requirement review meeting summary.

[0048] Thirdly, embodiments of this application provide a device for implementing an intelligent quality valve for a demand submission process. The device includes a processor, a memory, and a program for implementing an intelligent quality valve for a demand submission process stored in the memory and executable by the processor. When the program for implementing an intelligent quality valve for a demand submission process is executed by the processor, it implements the steps of the method for implementing an intelligent quality valve for a demand submission process described above.

[0049] Fourthly, embodiments of this application provide a computer-readable storage medium storing a smart quality valve implementation program for a demand submission process, wherein when the smart quality valve implementation program for a demand submission process is executed by a processor, it implements the steps of the smart quality valve implementation method for a demand submission process described above.

[0050] The beneficial effects of the technical solutions provided in this application include:

[0051] By automatically generating requirement documents and requirement review meeting summaries, and introducing a large-scale intelligent platform based on artificial intelligence technology, and by optimizing the requirement review model through retrieval enhancement, the requirement review model simulates the expert review process, intelligently reviews requirement documents, and automates the review process. This greatly reduces the burden of manual review and improves the accuracy and efficiency of the review. Attached Figure Description

[0052] Figure 1 A flowchart illustrating the implementation method of the intelligent quality valve in the requirement submission process of this application;

[0053] Figure 2 A schematic diagram of the functional modules of the intelligent quality valve implementation device for the application submission process;

[0054] Figure 3 This is a schematic diagram of the hardware structure of the intelligent quality valve implementation device for the application submission process. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0057] In a first aspect, embodiments of this application provide a method for implementing an intelligent quality valve in a demand submission process.

[0058] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the implementation method of the intelligent quality valve in the submission process required by this application. Figure 1 As shown, the implementation method of the intelligent quality valve in the requirement submission process includes:

[0059] S1: Obtain the requirement description in text or image format and upload it to the created large model intelligent platform. Expand the document based on comparison with existing similar requirements, and perform standardization to fill in missing elements and generate a requirement document.

[0060] S2: Acquire the audio data of the requirements review meeting and upload it to the big model intelligent platform. The big model intelligent platform analyzes the audio data of the requirements review meeting to generate a summary of the requirements review meeting and submits the summary of the requirements review meeting to the DevOps platform.

[0061] S3: Create a requirement review model. The requirement review model obtains private domain data in real time through the retrieval enhancement technology of the retrieval enhancement generation model to review the requirement documents to be reviewed. The private domain data includes existing requirement documents, voice data of requirement review meetings, and summary of requirement review meetings.

[0062] This application not only enables the writing, tracking, and change management of requirements documents, but also integrates tightly with the DevOps platform. Through integration, quality gates can be set at the requirements stage to ensure that only requirements that have been fully reviewed and verified can enter the development stage.

[0063] It should be noted that the large model intelligent platform is specifically used for: accessing various large network models based on an agile access mechanism, dynamically obtaining the advantageous features of various large network models using a multi-dimensional model evaluation system, and flexibly assembling suitable large network models through an intelligent matching mechanism and pushing them to the current business scenario.

[0064] Furthermore, the large-scale model intelligence platform can be the Yangtze Lingxi Financial Large-Scale Model Intelligence Platform. Based on an agile access mechanism, it rapidly introduces various mainstream large-scale network models (such as Wenxin Yiyan, ChatGPT, DeepSeek, etc.) from the market. Utilizing a multi-dimensional model evaluation system, it dynamically acquires the advantageous features of each model and intelligently matches suitable large-scale network models, flexibly assembling them and recommending them to applicable business scenarios. Various AI capabilities are provided to users as plug-ins, offering out-of-the-box services. Combined with data and system interaction plug-ins, creative ideas can be easily realized by building intelligent agents. The connection and coupling of large-scale network models make intelligent agents more aligned with business scenario needs, forming a large-scale model application ecosystem shared and co-built by all employees.

[0065] Furthermore, in one embodiment, the requirement description in text or image form is obtained and uploaded to the created large-scale model intelligent platform. Based on comparison with existing similar requirements, the document is expanded and standardized to fill in missing elements, generating a requirement document, specifically including:

[0066] S101: Obtain the requirement description in text or image form and upload the obtained requirement description to the large model intelligent platform. When the requirement description is in image form, the large model intelligent platform will recognize the image based on image recognition technology.

[0067] S102: The large-scale intelligent platform identifies key information in the requirement description and matches it with similar requirements in the requirement database. Based on the structure, content, and expression specifications of the requirements, and combined with the contextual semantics of the requirement description, the large-scale intelligent platform categorizes, performs structured analysis, and expands the document, filling in missing elements to generate a standardized requirement document. The elements of this requirement document include the requirement background and objectives, requirement scope, detailed functional requirements, non-functional requirements, change management mechanism, requirement tracking matrix, risk management, and related impacts. It should be noted that after the generated requirement document undergoes review as a document awaiting review, it will be stored as an existing requirement document in the private domain data.

[0068] Specifically, requirement documents are written using AI. When creating a requirement, users can simply describe it or upload an image indicating the functions needed. They then submit this content to the large-scale intelligent platform, where a dedicated requirement agent performs image recognition and preliminary semantic extraction. The identified key information is matched against similar requirements in the requirement database. Based on the requirement's structure, content, and expression specifications, combined with the contextual semantics of the user's input, the large-scale intelligent platform categorizes, structures, and expands the requirement description. The expansion logic is primarily based on a mechanism of "semantic similarity comparison + structured element template filling," ensuring complete document content, clear boundaries, and well-defined acceptance criteria.

[0069] Then, standardization processing is performed, and missing elements are filled in, ultimately generating a standardized requirements document that conforms to the DOMM3 level standard. The missing element filling operation refers to the large-scale intelligent model platform filling in missing elements, which are specified by a standard-certified requirements document template. These elements include: requirements background and objectives, requirements scope, detailed functional requirements, non-functional requirements, change management mechanism, requirements traceability matrix (establishing the correspondence between requirements and development and testing phases), risk management, and related impacts. DOMM3 level refers to the "System and Software Engineering Development and Operation Integration Capability Maturity Model" standard (GB / T 42560—2023, abbreviated as "DevOps National Standard") published by the China Electronics Technology Standardization Institute.

[0070] The large model intelligent platform of this application can assist in writing requirements documents, improve the efficiency and standardization of requirements document writing. Similarly, the large model intelligent platform is also applicable to the verification and assistance in writing other project documents such as design documents, and comprehensively optimizes the document management process.

[0071] Furthermore, in one embodiment, the audio data of the requirements review meeting is acquired and uploaded to the large model intelligent platform. The large model intelligent platform analyzes the audio data of the requirements review meeting to generate a requirements review meeting summary, and submits the requirements review meeting summary to the DevOps platform, specifically including:

[0072] S201: Record the requirement review meeting to obtain the audio data of the requirement review meeting, convert the audio data of the requirement review meeting into text content, and upload the audio data of the requirement review meeting to the big model intelligent platform.

[0073] S202: The large-scale intelligent platform analyzes the voice data of the requirements review meeting and generates a summary of the requirements review meeting according to preset standards. The summary of the requirements review meeting includes key discussion points and decision results.

[0074] S203: Upload the generated requirements review meeting summary to the DevOps platform and archive it.

[0075] Specifically, AI is used to summarize requirement review meetings. Recordings are made during these meetings to obtain audio data, which is then converted into text and uploaded to a large-scale intelligent platform. This platform analyzes the audio data and automatically generates a summary of the requirement review meeting according to preset standards, facilitating the organization and archiving of meeting minutes. Finally, the summary is uploaded to the DevOps platform as a requirement review node document for future reference.

[0076] In this application, the DevOps platform refers to an integrated development and operations management platform that has passed Level 3 certification of the "System and Software Engineering Development and Operations Integration Capability Maturity Model" (GB / T 42560—2023, hereinafter referred to as the "DevOps National Standard") formulated by the China Electronics Technology Standardization Institute. The DevOps platform, centered around the entire lifecycle of the R&D process, constructs an integrated architecture encompassing process management, development processes, quality control, continuous delivery, measurement and analysis, outsourcing management, and multi-system integration. The DevOps platform incorporates requirements, tasks, and versions into unified management, realizing the online and visual nature of the development process, and platform-integrating service systems, code projects, and human resource outsourcing resources, effectively improving resource scheduling efficiency.

[0077] Furthermore, in one embodiment, a requirements review model is created. This model acquires private domain data in real time using retrieval enhancement techniques from a retrieval enhancement generation model to review the current requirements document to be reviewed. The private domain data includes existing requirements documents, requirements review meeting audio data, and requirements review meeting summaries, specifically including:

[0078] S301: Create a requirement review model. The requirement review model obtains private domain data from the private domain retrieval database in real time through the retrieval enhancement technology of the retrieval enhancement generation model. The private domain data includes existing requirement documents, voice data of requirement review meetings, and summary of requirement review meetings.

[0079] S302: The requirements review model reviews the current requirements document based on the existing requirements documents, answers questions about the current requirements document, and generates new private domain data based on the review results.

[0080] Specifically, when reviewing the current requirement documents awaiting review, the review process is based on a quality scoring method, as follows:

[0081] ;

[0082] in, This indicates the quality score of the current requirements document awaiting review, and when If the value is not less than the set threshold, the review is considered passed; otherwise, a suggestion for modification will be provided. Indicates the first The scores for each evaluation factor, This indicates the total number of review factors. Indicates the first The weights of each review factor, and The evaluation factors include the completeness of the background information, the clarity of the requirements, and the explicitness of the acceptance criteria.

[0083] It should be noted that in requirement review scenarios, existing requirement documents, requirement review meeting audio data, and requirement review meeting summaries—i.e., historical data such as requirements and requirement reviews—are considered private domain data. The requirement review model needs to be trained and adapted to this type of private domain data before it can be used. However, traditional model training methods require expensive GPU machine clusters, resulting in high training costs and the inability to take effect in real time, failing to meet the need for rapid knowledge updates. To address this issue, this application uses the retrieval enhancement technology of the Retrieval Augmentation Generative Model (RAG) to optimize the requirement review model. Specifically, the requirement review model uses the retrieval enhancement technology of the RAG to obtain private domain data from the private domain retrieval database in real time. Based on existing requirement documents and the requirement document standards required by DOMM Level 3, the requirement review model reviews the requirement document to be reviewed and answers questions related to it. Simultaneously, it can update or apply the private domain data in real time. Furthermore, the requirements review model improves the recall rate of knowledge data through multi-dimensional recall and refined knowledge segmentation techniques, and further enhances the practicality and accuracy of the review by selecting the most relevant knowledge into the generation / question-answering context through fine ranking and relevance screening strategies.

[0084] In this application, the overall architecture of the retrieval enhancement generation model includes a private domain retrieval database construction stage, a retrieval ranking stage, and a prompt generation stage. The private domain retrieval database construction stage involves building a private domain retrieval database based on private domain data, so that data can be queried when the current requirement document to be reviewed is related to existing historical requirement documents and retrieval enhancement is required. The retrieval ranking stage involves searching and ranking the data in the private domain retrieval database based on the current requirement document to be reviewed. The prompt generation stage is the stage where the requirement review model generates the review results based on the retrieval ranking stage.

[0085] In this application, regarding the private domain retrieval database construction phase, specifically:

[0086] a1: Collect existing private domain data, and perform content decomposition and standardized annotation (based on DOMM3 level requirements) on the collected private domain data to achieve data cleaning, so as to improve the subsequent embedding accuracy. In the cleaning process, desensitization and semantic classification mechanisms are introduced to ensure data privacy and security and facilitate knowledge unitization processing.

[0087] a2: The cleaned data is segmented into data blocks. A custom private domain-aware vector generation network is used to integrate context timestamps, business domain (such as financial and government domain features), and structural information (title, body text, conclusion segments, etc.) for three-channel input to generate multi-dimensional dynamic knowledge vectors for the data blocks, thereby realizing vector embedding of the data blocks. The custom private domain-aware vector generation network is built by introducing a domain attention mechanism on the basis of the Transformer encoder. The private domain-aware vector generation network can focus on industry terms and review semantics, making the embedding more in line with business scenarios.

[0088] a3: The vector-embedded data blocks are uniformly stored in a high-efficiency vector database (such as FAISS, Weaviate, etc.) to build a private domain retrieval database. An incremental monitoring and update mechanism is set up to listen for additions and modifications to private domain data. By listening for these additions and modifications, data re-embedding and index updates are automatically triggered to ensure that the retrieval database is synchronized with the latest enterprise knowledge in real time.

[0089] Specifically, in the private domain retrieval database construction phase, the first step is data collection and cleaning. This involves collecting existing private domain data such as requirement documents, voice data from requirement review meetings, and meeting summaries, and cleaning the data to ensure quality. Next, knowledge segmentation and indexing are performed. A multi-dimensional recall strategy, such as based on keywords, topics, and time, is used to segment the private domain data into data blocks. Vector embeddings of these data blocks are then generated and stored in a high-efficiency vector database, thus completing the construction of the private domain retrieval database. Furthermore, a real-time update mechanism can be introduced, and an automated process can be designed to update the private domain data periodically or in real-time, ensuring that the knowledge in the private domain retrieval database is always up-to-date.

[0090] In this application, the specific details of the search and ranking stage are as follows:

[0091] b1: Based on keyword matching or embedding similarity recall mechanism, a multi-task intent recognition network is introduced to construct an intent-guided recall recognition network (PAEN).

[0092] b2: For the issues or requirements in the current requirements document to be reviewed, the constructed intent-guided recall identification network determines whether it involves a specific requirement-type recall intent. If so, a recall strategy is dynamically selected, and relevant data blocks are obtained from the private domain retrieval database based on the issues or requirements and in combination with the cosine similarity calculation method to achieve data block recall. Furthermore, by introducing a semantic extension mechanism, the recall capability of hyponyms and hypernyms is enhanced based on an external general knowledge graph.

[0093] Specifically, by constructing an intent-guided recall identification network, it is determined whether the problems or requirements in the current requirement document to be reviewed involve specific requirement-type recall intents, such as whether they are standardized or meet the acceptance criteria. When a clear recall intent is identified, a recall strategy is dynamically selected, such as using a combination of strategies like "requirement number" retrieval and recalling related meeting content. Then, relevant data blocks are obtained based on similarity calculation to achieve data block recall. At the same time, a semantic extension mechanism is introduced to enhance the user's ability to query hyponyms and hypernyms based on external general knowledge graphs (such as a requirement terminology list) to avoid missing key information.

[0094] b3: Calculate the ranking score of the data blocks based on the ranking factor to refine the ranking of the recalled data blocks, resulting in refined data blocks. Specifically, the calculation of the ranking score involves...

[0095] ;

[0096] in, Indicates the first The sorting score of each data block. Indicates the first The semantic similarity between each data block and the query. Indicates the first Timeliness score for each data block Indicates the first The frequency with which each data block is used or verified by experts Indicates the first The percentage of data blocks used in previous successful reviews. , , , This represents the weighting coefficient; ranking factors include text similarity, document timeliness, expert review weight, and historical review success rate. Based on the calculated ranking scores, the data blocks are sorted in descending order to achieve fine-grained ranking.

[0097] When sorting data blocks based on ranking scores, a relevant ranking model can be used. This model employs an improved weighted boosting model (introducing a time-sensitive attention factor on top of XGBoost) and combines it with a lightweight graph neural network (GNN) to model the relationships between data blocks, thereby improving the ranking hierarchy and interpretability. Furthermore, data blocks with similarity below a set threshold can be filtered out to ensure the context quality of subsequent generation stages.

[0098] In this application, regarding the prompt generation stage, specifically:

[0099] c1: For the refined data blocks, a unified Prompt template is constructed based on the issues in the current requirements document to be reviewed. The Prompt template includes a requirements summary, referenced review rules (such as Article X of the DOMM3 level standard), relevant historical review case snippets, and review focus points. The review focus points are enterprise-defined focus points used to guide the generation of key points, such as acceptance criteria and the rationality of the implementation path.

[0100] c2: Based on the sorted data blocks, construct the context information of the requirement review model. Based on the constructed context information, the requirement review model responds to and reviews the issues or requirements in the current requirement document to be reviewed according to the Prompt template, generates review results, and actively reviews and corrects the generated review results.

[0101] Furthermore, custom rule templates (such as format standards and key field detectors) can be inserted as retrieval guidance tokens to enhance the accuracy of generation and consistency with the enterprise. It can also be combined with the "General Review Logic Library" so that the system generates results that support the simultaneous output of natural language and structured suggestions (e.g., "Suggest adding acceptance condition fields" + JSON structured field annotations).

[0102] The above methods can enhance the requirements review model, enabling it to acquire and apply existing private domain knowledge in real time, thereby improving review efficiency and accuracy. At the same time, through multi-dimensional recall, refined knowledge segmentation, fine ranking strategies, and relevance screening, the practicality and accuracy of the review are further enhanced.

[0103] Compared to traditional RAG methods that directly call BERT or DPR to generate vectors, this application uses a self-developed PAEN network structure and introduces private domain awareness and structural enhancement mechanisms, making it more adaptable to the dynamic and multidimensional nature of enterprise knowledge. It innovatively introduces a multi-factor fusion ranking mechanism, replacing the traditional single similarity ranking with a refined ranking model that integrates semantics, timeliness, expert authority, and historical success rate, improving accuracy and practicality. The intent-guided recall recognition network integrates sequence modeling and contextual judgment capabilities, possessing stronger task matching capabilities than conventional classifiers, especially suitable for judging complex intents in requirements review. It adopts a structural semantic collaborative embedding scheme, integrating document structure, time, and business domain semantics, far surpassing the adaptation effect of existing general vector embedding methods on private domain documents.

[0104] Furthermore, regarding the intelligent quality valve implementation method for the requirement submission process in this application, a comprehensive quality score can also be defined in practical applications, specifically:

[0105] ;

[0106] in, Indicates the overall quality score, and when When the quality threshold is not lower than the set minimum, it indicates that the requirement can proceed to the development phase. Indicates the compilation pass rate. This indicates the static code scanning score. Indicates test coverage. Indicates the number of security vulnerabilities discovered. , , , This represents the weighting parameter, which is set by the quality strategy.

[0107] The intelligent quality valve implementation method for the requirement submission process in this application embodiment automatically generates requirement documents and requirement review meeting summaries, introduces a large-scale intelligent platform based on artificial intelligence technology, and optimizes the requirement review model by retrieving and enhancing the generated model. This enables the requirement review model to simulate the expert review process, intelligently review requirement documents, and achieve automated review operations, which can greatly reduce the burden of manual review and improve the accuracy and efficiency of the review.

[0108] Secondly, embodiments of this application also provide an intelligent quality valve implementation device for the demand submission process.

[0109] In one embodiment, reference is made to Figure 2 , Figure 2 This is a functional module diagram of the intelligent quality valve implementation device for the application requirements submission process. Figure 2 As shown, the intelligent quality valve implementation device for the requirement submission process includes: a requirement document writing module, a requirement review summary module, and a review processing execution module.

[0110] The requirement document writing module is used to obtain requirement descriptions in text or image format and upload them to the created large-scale intelligent platform. Based on comparison with existing similar requirements, the document is expanded and standardized to fill in missing elements, generating a requirement document. The requirement review summary module is used to obtain the audio data of the requirement review meeting and upload it to the large-scale intelligent platform. The large-scale intelligent platform analyzes the audio data of the requirement review meeting to generate a requirement review meeting summary, which is then submitted to the DevOps platform. The review processing execution module is used to create a requirement review model. The requirement review model obtains private domain data in real time through retrieval enhancement technology to review the requirement documents to be reviewed. The private domain data includes existing requirement documents, requirement review meeting audio data, and requirement review meeting summaries.

[0111] Thirdly, embodiments of this application provide an intelligent quality valve implementation device for the demand submission process. The intelligent quality valve implementation device for the demand submission process can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0112] Reference Figure 3 , Figure 3 This is a schematic diagram of the hardware structure of the intelligent quality valve implementation device for the demand submission process involved in the embodiments of this application. In this embodiment, the intelligent quality valve implementation device for the demand submission process may include a processor, a memory, a communication interface, and a communication bus.

[0113] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0114] Communication interfaces include input / output (I / O) interfaces, physical interfaces, and logical interfaces. These interfaces enable interconnection of internal components within the intelligent quality valve during the demand submission process, and also facilitate interconnection between the intelligent quality valve and other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0115] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0116] The processor can be a general-purpose processor, which can call the intelligent quality valve implementation program for the demand submission process stored in memory and execute the intelligent quality valve implementation method for the demand submission process provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the intelligent quality valve implementation program for the demand submission process is called can be referred to in the various embodiments of the intelligent quality valve implementation method for the demand submission process of this application, and will not be repeated here.

[0117] Those skilled in the art will understand that Figure 3 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0118] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0119] This application stores a program for implementing a smart quality valve for a demand submission process on a computer-readable storage medium, wherein when the program is executed by a processor, it implements the steps of the method for implementing a smart quality valve for a demand submission process as described above.

[0120] The method implemented when the intelligent quality valve implementation program for the demand submission process is executed can be referred to in various embodiments of the intelligent quality valve implementation method for the demand submission process of this application, and will not be repeated here.

[0121] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0122] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0123] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0124] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0125] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0126] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for implementing an intelligent quality valve in a requirement submission process, characterized in that, The intelligent quality valve implementation method for the demand submission process includes: Obtain the requirement description in text or image format and upload it to the created large model intelligent platform. Expand the document based on comparison with existing similar requirements, and perform standardization to fill in missing elements to generate a requirement document. Acquire the audio data of the requirements review meeting and upload it to the big model intelligent platform. The big model intelligent platform analyzes the audio data of the requirements review meeting to generate a summary of the requirements review meeting and submits the summary of the requirements review meeting to the DevOps platform. A requirement review model is created. The requirement review model obtains private domain data in real time through the retrieval enhancement technology of the retrieval enhancement generation model in order to review the requirement documents to be reviewed. The private domain data includes existing requirement documents, voice data of requirement review meetings, and requirement review meeting summaries. The creation of the requirements review model involves using retrieval enhancement techniques to acquire private domain data in real time to review the requirements documents currently awaiting review. This private domain data includes existing requirements documents, audio data from requirements review meetings, and summaries of those meetings, specifically including: A requirement review model is created. The requirement review model obtains private domain data from the private domain retrieval database in real time through the retrieval enhancement technology of the retrieval enhancement generation model. The private domain data includes existing requirement documents, voice data of requirement review meetings, and summary of requirement review meetings. The requirements review model reviews the current requirements document based on the existing requirements document, answers questions about the current requirements document, and generates new private domain data based on the review results. Specifically, when reviewing the current requirement documents awaiting review, the review process is based on a quality scoring method, as follows: ； in, This indicates the quality score of the current requirements document awaiting review, and when If the value is not less than the set threshold, the review is considered passed; otherwise, a suggestion for modification will be provided. Indicates the first The scores for each evaluation factor, This indicates the total number of review factors. Indicates the first The weights of each review factor, and .

2. The method for implementing an intelligent quality valve in a demand submission process as described in claim 1, characterized in that, The large-scale model intelligent platform is specifically used for: Based on the agile access mechanism, various large network models are accessed. The advantages and characteristics of various large network models are dynamically obtained by using a multi-dimensional model evaluation system. The appropriate large network models are flexibly assembled and pushed to the current business scenario through an intelligent matching mechanism.

3. The method for implementing an intelligent quality valve in a demand submission process as described in claim 1, characterized in that, The process involves obtaining a text or image description of the requirements and uploading it to the created large-scale intelligent model platform. Based on comparison with existing similar requirements, the document is expanded and standardized to fill in missing elements, generating a requirements document. Specifically, this includes: Obtain a requirement description in text or image format and upload it to the Big Model Intelligent Platform. When the requirement description is in image format, the Big Model Intelligent Platform uses image recognition technology to identify the image. The big model intelligent platform identifies key information in the requirement description and matches it with similar requirements in the requirement database. Based on the structure, content and expression specifications of the requirements, and combined with the contextual semantics of the requirement description, the big model intelligent platform classifies, performs structured analysis and document expansion of the requirement description, and fills in missing elements to generate a standardized requirement document. The elements of the requirements document include the background and objectives of the requirements, the scope of the requirements, the details of the functional requirements, the non-functional requirements, the change management mechanism, the requirements traceability matrix, risk management, and related impacts.

4. The method for implementing an intelligent quality valve in a demand submission process as described in claim 1, characterized in that, The process of acquiring and uploading the audio data of the requirements review meeting to the large-scale intelligent platform, analyzing the audio data to generate a summary of the requirements review meeting, and submitting the summary to the DevOps platform specifically includes: Record the requirement review meeting to obtain the audio data, convert the audio data into text, and upload the audio data to the big model intelligent platform. The large-scale intelligent platform analyzes the voice data of the requirements review meeting and generates a summary of the requirements review meeting according to preset standards. The summary of the requirements review meeting includes key discussion points and decision results. Upload the generated requirements review meeting summary to the DevOps platform and archive it.

5. The method for implementing an intelligent quality valve in a demand submission process as described in claim 1, characterized in that: The overall architecture of the retrieval enhancement generation model includes a private domain retrieval database construction stage, a retrieval ranking stage, and a suggestion generation stage. The private domain retrieval database construction phase involves building a private domain retrieval database based on private domain data, so that when the current requirement document to be reviewed is related to existing historical requirement documents and needs to be enhanced for retrieval, data can be queried. The retrieval and ranking stage is a stage in which data is searched and ranked in the private domain retrieval database based on the current documents to be reviewed; The prompt generation stage is the stage in which the requirements review model generates review results based on the retrieval and ranking stage.

6. The method for implementing an intelligent quality valve in a demand submission process as described in claim 5, characterized in that: For the private domain retrieval database construction phase, specifically: Collect existing private domain data, and perform content decomposition and standardized labeling on the collected private domain data to achieve data cleaning, and introduce desensitization and semantic classification mechanisms in the cleaning process; The cleaned data is segmented into data blocks. A custom private domain-aware vector generation network is used to generate multi-dimensional dynamic knowledge vectors for the data blocks by integrating context timestamps, business domains, and structural information and taking three-channel input. This enables vector embedding of the data blocks. The custom private domain-aware vector generation network is constructed by introducing a domain attention mechanism on the basis of the Transformer encoder. The data blocks after vector embedding are uniformly stored in a high-efficiency vector database to realize the construction of a private domain retrieval database, and an incremental monitoring and update mechanism is set up to listen for the addition and modification operations of private domain data. For the retrieval and ranking stage, specifically: Based on keyword matching or embedding similarity recall mechanism, a multi-task intent recognition network is introduced to construct an intent-guided recall recognition network. For the issues or requirements in the current requirement documents to be reviewed, the constructed intent-guided recall identification network determines whether it involves a specific requirement category recall intent. If so, a recall strategy is dynamically selected, and relevant data blocks are obtained from the private domain retrieval database based on the issues or requirements and in combination with the cosine similarity calculation method to achieve data block recall. Furthermore, by introducing a semantic extension mechanism, the recall capability of hyponyms and hypernyms is enhanced based on an external general knowledge graph. The ranking score of data blocks is calculated based on the ranking factor to refine the ranking of the recalled data blocks, resulting in refined data blocks. Specifically, the calculation of the ranking score involves... ； in, Indicates the first The sorting score of each data block. Indicates the first The semantic similarity between each data block and the query. Indicates the first Timeliness score for each data block Indicates the first The frequency with which each data block is used or verified by experts Indicates the first The percentage of data blocks used in previous successful reviews. , , , Indicates the weighting coefficient; Regarding the prompt generation phase, specifically: For the refined data blocks, a unified Prompt template is constructed based on the issues in the current requirements document to be reviewed. The Prompt template includes a requirements summary, referenced review rules, relevant historical review case snippets, and review focus points. Based on the sorted data blocks, the context information of the requirement review model is constructed. Based on the constructed context information, the requirement review model responds to and reviews the issues or requirements in the current requirement document to be reviewed according to the Prompt template, generates review results, and actively reviews and corrects the generated review results.

7. A device for implementing an intelligent quality valve in a demand submission process, characterized in that, The intelligent quality valve implementation device for the demand submission process includes: The requirement document writing module is used to obtain requirement descriptions in text or image form and upload them to the created large model intelligent platform. Based on the comparison with existing similar requirements, the document is expanded and standardized to fill in missing elements and generate a requirement document. The requirement review summary module is used to acquire the voice data of the requirement review meeting and upload it to the big model intelligent platform. The big model intelligent platform analyzes the voice data of the requirement review meeting to generate a requirement review meeting summary and submits the requirement review meeting summary to the DevOps platform. The review processing execution module is used to create a requirement review model. The requirement review model obtains private domain data in real time through the retrieval enhancement technology of the retrieval enhancement generation model in order to review the requirement documents to be reviewed. The private domain data includes existing requirement documents, requirement review meeting voice data, and requirement review meeting summary. The creation of the requirements review model involves using retrieval enhancement techniques to acquire private domain data in real time to review the requirements documents currently awaiting review. This private domain data includes existing requirements documents, audio data from requirements review meetings, and summaries of those meetings, specifically including: A requirement review model is created. The requirement review model obtains private domain data from the private domain retrieval database in real time through the retrieval enhancement technology of the retrieval enhancement generation model. The private domain data includes existing requirement documents, voice data of requirement review meetings, and summary of requirement review meetings. The requirements review model reviews the current requirements document based on the existing requirements document, answers questions about the current requirements document, and generates new private domain data based on the review results. Specifically, when reviewing the current requirement documents awaiting review, the review process is based on a quality scoring method, as follows: ； in, This indicates the quality score of the current requirements document awaiting review, and when If the value is not less than the set threshold, the review is considered passed; otherwise, a suggestion for modification will be provided. Indicates the first The scores for each evaluation factor, This indicates the total number of review factors. Indicates the first The weights of each review factor, and .

8. A device for implementing an intelligent quality valve in a demand submission process, characterized in that, The intelligent quality valve implementation device for the demand submission process includes a processor, a memory, and an intelligent quality valve implementation program for the demand submission process stored in the memory and executable by the processor, wherein when the intelligent quality valve implementation program for the demand submission process is executed by the processor, it implements the steps of the intelligent quality valve implementation method for the demand submission process as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an implementation program for a smart quality valve in the demand submission process, wherein when the implementation program is executed by a processor, it implements the steps of the method for implementing a smart quality valve in the demand submission process as described in any one of claims 1 to 6.

Images