AI-based demand review system and method

Abstract

The invention discloses an AI-based demand review system and method, and relates to the technical field of review data processing. The invention aims to support the demand engineering of aviation products, deeply review demand data by applying the AI technology, help to identify and record errors which do not meet the demand standard, and ensure the compliance and integrity of the demand, so as to judge whether the demand can be approved or not and enter the subsequent development and test stage. In addition, a mature base large model is selected, and checking and reviewing of demand data are carried out in the aspects of grammar formats of demand description, demand content, demand traceability, description of derivative demands and the like in the process of executing checking or reviewing of demand documents. According to the method, the efficiency of demand capture and demand review can be remarkably improved, rapid and comprehensive demand review results are provided for demand engineers and development teams, cross-department cross confirmation and compliance check are supported, and smooth proceeding of subsequent development and test activities is effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of review data processing, and in particular to an AI-based demand review system and method. Background Art

[0002] Requirements review mainly reviews the requirements scope and objectives, requirements document details, completeness and consistency, technical feasibility, business value and priority, legal risks and compliance, user experience and usability, and requirements change management. Currently, there are two typical methods for requirements review activities in requirements engineering:

[0003] 1. If there is no requirement management platform, the requirement developers will generate requirement documents based on the relevant requirements after the requirements definition and architecture design. The requirement reviewers will manually check the requirement documents and confirm the completeness, traceability and standardization of the requirements one by one. This usually includes multiple rounds of review, and the reviewers will rely on experience and standard manuals to check whether the requirements meet the regulations one by one. Since there are multiple levels of requirement items from aircraft-level requirements to system-level requirements, the requirement reviewers need to mark them on multiple documents and wait for the requirement compilers to verify or modify them.

[0004] 2. If there is a demand management platform, such as IBM Doors, the demand developers will enter the stakeholder requirements into the demand management platform after the requirements definition and architecture design are refined, and the demand allocation and link relationship will be established through the platform. The demand reviewers will review and confirm the demand attributes one by one through the demand management platform to determine whether the demand meets the demand standard requirements. The review results will be recorded in the demand management platform, waiting for the demand compiler to verify or modify.

[0005] Technical disadvantages of method #1: low work efficiency. Due to the complexity of requirement items and attributes, more manpower is needed to determine the basic rules of completeness, consistency, necessity, redundancy, etc. of the requirements; checking and marking each item in multiple documents increases the risk of omissions, and needs to rely on the experience of reviewers, which is easily affected by factors such as fatigue and negligence; manual inspection of traceability of requirements at different levels is difficult, and searching and confirming the integrity of parent-child relationships across multiple files is prone to errors and cannot be tracked in real time; there is a greater risk of leakage when confidentiality is required.

[0006] Technical disadvantages of method #2: Although method #2 improves work efficiency compared to method #1, specific grammar and content integrity checks still need to be manually confirmed one by one, which limits efficiency; the use of the demand management platform requires a certain learning cost, and the system operation process is complicated, which affects the efficiency of demand review; the demand management platform can support the establishment and tracking of demand relationships, but it still has limitations in deep-level demand analysis such as identification of grammatical errors, ambiguous expressions, and content contradictions; demand review opinions still require manpower to organize, and it is difficult to evaluate the work of reviewers.

[0007] In response to the above problems, the present invention proposes an AI-based demand review system and method. Summary of the invention

[0008] The purpose of the present invention is to propose an AI-based requirement review system and method to solve the problems raised in the background technology, aiming to support the requirement engineering of aviation products. By applying AI technology to conduct in-depth review of requirement data, it helps to identify and record errors that do not meet requirement standards, ensure the compliance and completeness of requirements, and thus determine whether they can be approved and enter subsequent development and testing stages.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] An AI-based demand review system, comprising:

[0011] The syntax format checking module is used to check whether the syntax of the requirement description conforms to the standard;

[0012] The content integrity review module is used to analyze whether the requirement description is complete and without omissions, detect potential contradictions in the requirement content, and prevent the requirement description from contradicting itself or conflicting with other requirements;

[0013] The traceability analysis module is used to identify and check the hierarchical relationship of requirements, ensure that non-derived requirements are correctly traced back to parent requirements, and provide a visual display of the requirement traceability path;

[0014] The derived requirement description checking module uses a deep learning model to check whether the description of derived requirements conforms to the "function-oriented" principle, automatically identifies potential content conflicts between derived requirements and parent requirements, and ensures design consistency;

[0015] The intelligent question-and-answer module provides users with real-time answers to their needs based on the company's private knowledge base. It supports users to ask questions in natural language in the interface and quickly obtain explanations of demand standards and specifications.

[0016] The requirement consistency and uniqueness verification module is used to check whether the requirement content is repeated or similar to other requirements, ensure the uniqueness of each requirement, automatically identify repeated or similar requirement descriptions, and avoid overlap between different requirements.

[0017] An AI-based demand review method includes the following contents:

[0018] S1. Data preparation: Collect demand confirmation data, technical standard documents, and domain knowledge base, pre-process the above data and integrate them into a unified format;

[0019] S2. Model selection and training: Select a large model as the base, and based on its natural language processing capabilities, use the enterprise's demand confirmation data to fine-tune and train the base model;

[0020] S3. System development:

[0021] S3.1. Develop a user interface to support requirement document uploading, editing and result display;

[0022] S3.2. Develop core functional modules to check, review and improve requirement descriptions;

[0023] S3.3, integrate models and functional modules, and debug system functions to optimize model calculation speed and front-end loading speed;

[0024] S4. System deployment and application: Deploy the developed system in the enterprise LAN environment to ensure data security; provide API interface to achieve seamless integration of front-end with model and functional modules to ensure real-time response of user operation and model calculation;

[0025] S5. Continuous optimization: Based on user feedback and performance monitoring results, continuously improve model accuracy and system response speed; regularly update data sets, expand training data, and maintain the model's adaptability to new demand formats and industry standards.

[0026] Preferably, the data preprocessing in S1 specifically refers to:

[0027] Detailed annotation of required data, cleaning of duplicate or invalid information, and ensuring the quality of training data;

[0028] Mark incorrect examples and ambiguous expressions in the requirements as negative samples for training;

[0029] Generate diverse data sets, including simulated ambiguous statements and erroneous demand samples, to enrich training data and enhance the recognition ability of the model.

[0030] Preferably, the fine-tuning and training of the base model in S2 specifically refers to:

[0031] Identify common grammatical formats for aviation domain requirements;

[0032] Identify and point out common errors in requirement descriptions through labeled negative sample training models;

[0033] The model is trained to understand the hierarchical relationship between requirements to achieve traceability checking of requirements.

[0034] Compared with the prior art, the present invention provides an AI-based demand review system and method, which has the following beneficial effects:

[0035] (1) Strong targeting: The present invention uses the enterprise's internal data set for model training to ensure that the model can accurately identify and process specific needs in the product development process, thereby improving the accuracy and efficiency of demand confirmation.

[0036] (2) High customizability: The present invention supports customized development to meet the specific needs of enterprises, solve unique problems, ensure that the tool is closely integrated with the business processes and specifications of the enterprise, and improve the practicality and adaptability of the tool.

[0037] (3) High security: The present invention adopts LAN deployment to ensure the security of enterprise data, avoid data leakage, and meet the enterprise's strict requirements for data security.

[0038] (4) Stable performance: The present invention supports the replaceability of the LLM model, and the base model can be replaced according to technological development and enterprise needs, thereby maintaining the advancement and stability of the model and adapting to the ever-changing technological trends.

[0039] In summary, the present invention uses AI tools to conduct demand review, automatically completes syntax checking, content integrity and traceability analysis, reduces the time of manual review, and significantly improves work efficiency; performs reviews based on unified standards and rules, quickly identifies and extracts key information, ensures the consistency and accuracy of requirements, and reduces human errors. At the same time, the present invention can provide high-quality review feedback based on data analysis, improve the quality of reviews, and reduce the risk of leaks through encryption and security functions, thereby comprehensively optimizing the demand confirmation process and improving the overall effectiveness of project management. In actual projects, the application of the present invention, combined with manual experience, can significantly improve the processing efficiency and accuracy of requirements, enhance the reliability of results and decision-making quality, while ensuring risk control and compliance in complex situations, and achieving comprehensive optimization of the demand management process. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only one embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 This is a flow chart of the AI-based demand review method mentioned in Example 1 of the present invention;

[0042] Figure 2 This is a module display diagram of the AI-based demand review system mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0044] This invention aims to support the requirements engineering of aviation products by applying AI technology to conduct in-depth reviews of requirement data, help identify and record errors that do not meet requirement standards, ensure the compliance and completeness of requirements, and thus determine whether they can be approved and enter subsequent development and testing stages.

[0045] The present invention selects a mature base large model to carry out inspection and review of demand data from aspects such as the grammatical format of the demand description, demand content, traceability of the demand and description of derived demands during the inspection or review of the demand document.

[0046] The present invention can significantly improve the efficiency of demand capture and demand review, provide fast and comprehensive demand review results for demand engineering personnel and development teams, support cross-departmental cross-confirmation and compliance checks, and effectively ensure the smooth progress of subsequent development and testing activities.

[0047] The following is an explanation of an AI-based demand review system and method proposed in the present invention in conjunction with relevant drawings and specific examples. The specific contents are as follows.

[0048] Embodiment 1:

[0049] See also Figure 1 The technical solution of the AI-based requirements review method includes the following key aspects:

[0050] Step 1. Prepare data: Collect demand confirmation data, technical standard documents, domain knowledge base, etc., and integrate these data into a unified format. Detailed annotation of demand data, clean up duplicate or invalid information, and ensure the quality of training data. At the same time, annotate the wrong examples and fuzzy expressions in the demand content as negative samples for training. Generate a diverse data set, including simulated fuzzy expressions and erroneous demand samples, enrich training data, and enhance the recognition ability of the model.

[0051] Step 2, select and train the model: select a mature large model (such as Llama) as the base, based on its powerful natural language processing capabilities, and use the enterprise-specific demand confirmation data to fine-tune the base model. The fine-tuning process includes: identifying the common grammatical format of aviation requirements, including the uniqueness of the word "should", the tolerance requirements of performance values, etc.; training the model through annotated negative samples to identify and point out common errors in the demand description (such as ambiguous words, inconsistent quantifiers, etc.); training the model to understand the hierarchical relationship between requirements to achieve traceability inspection of requirements.

[0052] Step 3, system development: Develop a user interface to support requirement document uploading, editing and result display; develop core functional modules to implement inspection, review and improvement of requirement descriptions; integrate models and functional modules, and debug system functions to optimize model calculation speed and front-end loading speed to ensure a smooth user experience.

[0053] Step 4: System deployment and application: Deploy in the enterprise LAN environment to ensure data security; provide API interface to achieve seamless integration of the front end with models and functional modules to ensure real-time response of user operations and model calculations.

[0054] Step 5: Continuous optimization: Based on user feedback and performance monitoring results, continuously improve model accuracy and system response speed; regularly update data sets, expand training data, and keep the model adaptable to new demand formats and industry standards.

[0055] See also Figure 2 , the AI-based demand review system mainly includes the following functional modules:

[0056] 1. Syntax format check module: Check whether the syntax of the requirement description meets the standards, such as whether the word "should" is used uniquely; identify ambiguous words (such as "quickly" and "approximately") and mark them to ensure that the description is clear and unambiguous.

[0057] 2. Content integrity review module: Analyze whether the requirement description is complete and without omissions, ensure that each requirement has clear expression and self-consistent logic; detect potential contradictions in the requirement content to prevent the requirement description from contradicting itself or conflicting with other requirements.

[0058] 3. Traceability analysis module: Identify and check the hierarchical relationship of requirements to ensure that non-derived requirements can be correctly traced back to parent requirements. Provide a visual display of the requirement traceability path to help users understand the upper and lower level relationships of requirements.

[0059] 4. Derived requirement description check module: Through the deep learning model, check whether the description of the derived requirements conforms to the "function-oriented" principle, rather than directly specifying the implementation method. Automatically identify potential content conflicts between derived requirements and parent requirements to ensure design consistency.

[0060] 5. Intelligent Q&A module: Based on the company's private knowledge base, it provides users with real-time answers to their needs; it supports users to ask questions in natural language in the interface and quickly obtain explanations of demand standards and specifications.

[0061] Requirements consistency and uniqueness verification module: Checks whether the requirements are repeated or similar to other requirements to ensure the uniqueness of each requirement. Automatically identifies possible repeated or similar requirements descriptions to avoid overlap between different requirements.

[0062] In summary, the present invention is based on the AI ​​big model and combined with private data training. It ensures the clarity, consistency and unambiguity of the demand data in the development process and maintains the correct demand traceability relationship by automatically reviewing the grammatical format, content integrity, traceability and derivative description of the requirements.

[0063] The present invention can be applied to the demand review in the process of aviation product development, improve the work efficiency of demand confirmation, improve the review quality, and thus comprehensively optimize the demand confirmation process.

[0064] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An AI-based demand review system, characterized in that: include: The syntax format checking module is used to check whether the syntax of the requirement description conforms to the standard; The content integrity review module is used to analyze whether the requirement description is complete and without omissions, detect potential contradictions in the requirement content, and prevent the requirement description from contradicting itself or conflicting with other requirements; The traceability analysis module is used to identify and check the hierarchical relationship of requirements, ensure that non-derived requirements are correctly traced back to parent requirements, and provide a visual display of the requirement traceability path; The derived requirement description checking module uses a deep learning model to check whether the description of derived requirements conforms to the "function-oriented" principle, automatically identifies potential content conflicts between derived requirements and parent requirements, and ensures design consistency; The intelligent question-and-answer module provides users with real-time answers to their needs based on the company's private knowledge base. It supports users to ask questions in natural language in the interface and quickly obtain explanations of demand standards and specifications. The requirement consistency and uniqueness verification module is used to check whether the requirement content is repeated or similar to other requirements, ensure the uniqueness of each requirement, automatically identify repeated or similar requirement descriptions, and avoid overlap between different requirements.

2. An AI-based demand review method implemented by the system of claim 1, characterized in that: It includes the following: S1. Data preparation: Collect demand confirmation data, technical standard documents, and domain knowledge base, pre-process the above data and integrate them into a unified format; S2. Model selection and training: Select a large model as the base, and based on its natural language processing capabilities, use the enterprise's demand confirmation data to fine-tune and train the base model; S3. System development: S3.

1. Develop a user interface to support requirement document uploading, editing and result display; S3.

2. Develop core functional modules to check, review and improve requirement descriptions; S3.3, integrate models and functional modules, and debug system functions to optimize model calculation speed and front-end loading speed; S4. System deployment and application: Deploy the developed system in the enterprise LAN environment to ensure data security; Provide API interface to achieve seamless integration between the front end and the model and functional modules, ensuring real-time response between user operations and model calculations; S5. Continuous optimization: Based on user feedback and performance monitoring results, continuously improve model accuracy and system response speed; regularly update data sets, expand training data, and maintain the model's adaptability to new demand formats and industry standards.

3. According to the AI-based demand review method of claim 1, it is characterized in that: The data preprocessing described in S1 specifically refers to: Detailed annotation of required data, cleaning of duplicate or invalid information, and ensuring the quality of training data; Mark incorrect examples and ambiguous expressions in the requirements as negative samples for training; Generate diverse data sets, including simulated ambiguous statements and erroneous demand samples, to enrich training data and enhance the recognition ability of the model.

4. The AI-based demand review method according to claim 1, characterized in that: The fine-tuning and training of the base model described in S2 specifically refers to: Identify common grammatical formats for aviation domain requirements; Identify and point out common errors in requirement descriptions through labeled negative sample training models; The model is trained to understand the hierarchical relationship between requirements to enable traceability checking of requirements.

Images