A multimodal AI collaborative product requirement process management system, method, storage medium, and device

The multimodal AI-collaborative product requirement process management system solves the problems of low data entry efficiency and poor requirement quality in the traditional model. It enables automatic voice form filling, real-time intelligent guidance, and similar requirement recognition, thereby improving the efficiency and quality of requirement management.

CN123117748APending Publication Date: 2026-11-17JIANGSU XCMG CONSTRUCTION MACHINERY RESEARCH INSTITUTE LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611290510.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Traditional product demand management models suffer from problems such as low input efficiency, low demand quality, and delayed identification of duplicate demands. They also lack intelligent guidance through multimodal AI collaboration and pre-identification of similar demands.

Method used

The product requirement process management system, which adopts multimodal AI collaboration, includes a front-end interaction layer, an agent collaborative scheduling middleware layer, a voice structured mapping, contextual question and answer generation, and a full-weighted vector matching calculation unit, to achieve automatic voice form filling, real-time intelligent guidance, and similar requirement recognition.

Benefits of technology

It improved the efficiency of demand entry, enhanced the quality of demand, reduced reliance on staff experience, accurately identified repetitive demands, and improved market response efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117748A_ABST
    Figure CN123117748A_ABST
Patent Text Reader

Abstract

This invention discloses a multimodal AI collaborative product requirement process management system, method, storage medium, and device, belonging to the field of artificial intelligence and enterprise digital R&D management technology. It includes a front-end interaction layer, providing a visual interactive entry point for requirement submitters and requirement analysts, allowing them to manually or verbally input requirement information. The agent collaborative scheduling middleware layer includes a voice structured mapping calculation unit, a contextual question-and-answer generation calculation unit, and a full-weighted vector matching calculation unit. This invention constructs a multi-layered collaborative architecture, relying on a private large-scale model cluster to achieve coordinated voice structured form filling, contextual intelligent question answering, and weighted similarity requirement deduplication within the same product line. Multiple AI sub-algorithms are uniformly scheduled through a collaborative task scheduling unit, and a session storage unit handles temporary data transfer, enabling multimodal AI capabilities to work collaboratively.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a multimodal AI collaborative product demand process management system, method, storage medium, and device, belonging to the technical field. Background Technology

[0002] As enterprises deepen their digital transformation, product requirement management has become a crucial link between the market and R&D. Efficiently handling massive, multi-source requirements is key to enhancing product competitiveness. Traditional purely manual management models have significant shortcomings: mobile requirement forms have numerous fields, requiring manual transcription and completion of verbal requirements, resulting in low input efficiency and a high risk of omissions and errors; the requirement submission and analysis phases lack standardized intelligent guidance, leading to incomplete and repeatedly rejected requirements; relying solely on manual retrieval of historical requirements prevents the proactive interception of similar requests, wasting R&D resources on duplicate requests; and industry knowledge is difficult to reuse automatically, relying entirely on personal experience, leading to information distortion and insufficient standardization, significantly reducing market response efficiency and severely hindering the enterprise's market responsiveness and digital transformation effectiveness.

[0003] For example, the product requirement management method and system based on a large-scale intelligent agent disclosed in Chinese patent application 202511519380.4 only supports static text import and lacks mobile terminal voice auto-fill form function; the intelligent agent is only used for requirement assessment and change tracking, and the solution lacks a dynamic knowledge guidance mechanism in the front-end interaction stage of user filling in the requirement form; relying on knowledge graph to identify requirement conflicts in the post-process, it cannot match similar requirements in real time during the filling process and does not cover the pre-filling verification scenario. For example, the intelligent assistance method and device for the whole process of product requirement based on image recognition and large model disclosed in Chinese patent application 202511266236.4 takes web page screenshot image recognition as the core and does not support voice input at all; the AI ​​interaction is only used for test case and interface code generation and cannot provide content assistance for requirement filling and review; there is no real-time warning function for similar requirements, and the technology focuses on downstream R&D. Summary of the Invention

[0004] This invention aims to solve the problems of low efficiency in demand input, low demand quality due to user knowledge gaps, and delayed identification of duplicate demands. It provides a multimodal AI collaborative product demand process management method and system that supports efficient multimodal input, has real-time intelligent guidance and assistance, and can identify similar product demands in advance.

[0005] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution.

[0006] In a first aspect, the present invention discloses a multimodal AI collaborative product demand process management system, comprising: a front-end interaction layer and an Agent collaborative scheduling middleware layer;

[0007] The front-end interaction layer is used to provide a visual interaction entry for requirement submitters and requirement analysts, allowing requirement submitters or requirement analysts to manually input requirement information or input requirement information by voice.

[0008] The Agent-based collaborative scheduling middleware layer includes:

[0009] The speech structured mapping calculation unit is used to automatically fill in the text requirement form with the speech input requirement information collected from the visual interactive entry point;

[0010] The contextual question-answering generation calculation unit is used to automatically capture the manually input requirement information and / or the text requirement form, and generate contextual data for the corresponding business scenario. The business scenario includes the requirement submission scenario of the corresponding requirement submitter and the requirement analysis scenario of the corresponding requirement analyst. For the corresponding business scenario, the unit integrates the contextual information of the corresponding business scenario and the internal knowledge base to generate target prompt words, so that the large model can output the corresponding scenario display content for display in the front-end interaction layer, and fill the display content back into the corresponding input box of the requirement form. The internal knowledge base is a knowledge base pre-constructed based on the enterprise's product requirement management system documents, product requirement management specifications, established business process documents, and structured analysis of historical requirement cases.

[0011] The full-weighted vector matching calculation unit is used to perform full-weighted similarity identification on the original requirements in both the requirement submission and requirement analysis scenarios to determine whether there are any historical similar requirements.

[0012] Furthermore, it also includes: a private large model service layer and a data persistence layer;

[0013] The private large model service layer is used to deploy speech recognition models, industry NLP entity recognition models for R&D business, contextual dialogue large models, and text vectorization models for the Agent collaborative scheduling middleware layer to use for scheduling.

[0014] The data persistence layer is used to persistently store all static business data, historical calculation results, and system configuration information of the system.

[0015] Furthermore, the Agent-based collaborative scheduling intermediate layer also includes:

[0016] The collaborative task scheduling unit is used to uniformly schedule the speech structure mapping calculation unit, the context question-and-answer generation calculation unit, and the full weighted vector matching calculation unit. It automatically matches the execution sequence of the corresponding unit according to the business node, and issues the corresponding calculation tasks according to different business scenarios such as submission for verification and secondary review of market analysis, and summarizes the output results of each unit and feeds them back to the upper-level page.

[0017] The session storage unit is used to temporarily cache data within a single business process, including the complete form filling context, human-computer dialogue records, and real-time feature vectors of the requirements to be matched; temporary data is only stored within the lifecycle of a single business process, and the temporary cached data is automatically cleared after the business process is completed.

[0018] Furthermore, the processing procedure of the speech structure mapping calculation unit includes:

[0019] The front-end interaction layer collects the user's original audio, performs noise reduction, frame segmentation, and 16kHz resampling processing in a unified manner, and outputs standardized audio.

[0020] Standardized audio is pushed to a speech recognition model and converted into the original free text T0 of the requirements;

[0021] The industry NLP entity recognition model is invoked to perform semantic recognition on the original free text T0 based on the enterprise product entity dictionary, and a standardized business entity set E is extracted.

[0022] Semantic cleaning and standard alignment are performed on the entity values ​​in the standardized business entity set E, mapping colloquial and non-standard expressions to standard industry terms, eliminating ambiguity and redundant information in the mapped entity values, and outputting a semantically standardized entity set.

[0023] Read the system's preset form field mapping matrix M, where each row of the matrix corresponds to a different entity set in the semantically standardized entity set, and each column of the matrix corresponds to an independent input box in the required form; based on the row and column matching relationship, complete the precise binding of each entity set to the corresponding input box in the form, and establish a unique mapping relationship from entity to form field;

[0024] The entity values ​​of the semantically standardized entity set that has been matched are automatically filled into the corresponding input boxes in the form. Form fields that do not have a corresponding entity are left empty. The complete form is pushed to the front end for the submitter to verify and complete.

[0025] Furthermore, the processing procedure of the context question-answering generation computing unit includes:

[0026] In the requirement submission scenario, when the requirement submitter opens the form to edit the original requirement form, the session storage unit automatically captures all the filled text on the page and generates requirement submission scenario context data.

[0027] The system integrates contextual data of requirement submission scenarios, internal knowledge base, and questions from requirement submitters to construct requirement submission scenario prompts, which are then used by the large model to output fields such as scenario conditions, requirement value, gaps / expected solutions of the current product, other competitors and their solutions, and original requirement descriptions.

[0028] The front-end pop-up displays all generated content and is configured to automatically populate all generated content back into the corresponding input boxes of the form with one click;

[0029] In a requirements analysis scenario, when a requirements analyst opens a form to edit market analysis, the session storage unit automatically captures all the filled-in text on the page and generates context data for the requirements analysis scenario.

[0030] By integrating contextual data from the requirements analysis scenario, internal knowledge base, and questions from requirements analysts, prompt words for the requirements analysis scenario are constructed to provide multi-dimensional evaluation templates for the large model to output.

[0031] The front-end pop-up displays all generated content and is configured to automatically populate all generated content back into the corresponding input boxes of the form with one click.

[0032] Furthermore, the processing procedure of the full-weighted vector matching calculation unit includes:

[0033] Generate a title feature vector V for each complete original title of the original requirement. t Generate a descriptive feature vector V from the complete original requirement description. d ;

[0034] Weighted fusion generates a comprehensive demand vector V a The fusion formula is: V a = α×V t +β× V d Where α is the fixed weight of the original requirement title, β is the fixed weight of the original requirement description, and α+β=1;

[0035] Retrieve the weighted composite vector set V corresponding to all historical demands within the same product line from the weighted historical demand vector library. s V s = {V s1 V s2 ,...V sn}, where V si This represents the i-th weighted composite vector in the set of weighted composite vectors, where i = 1, 2, ..., n;

[0036] Batch execution of cosine similarity calculation yields the similarity Sim, where Sim = (V a ·V si ) / (||V a ||×||V si In the formula, · represents the dot product symbol, ||V a ||and||V si || indicates V a and V si The modulus length; retain the weighted composite vector set V sSimilar requirement entries with a similarity greater than or equal to a preset threshold are selected for confirmation by the requirement submitter or requirement analyst to determine whether they are similar.

[0037] Secondly, this invention also discloses a multimodal AI collaborative product demand process management method, including:

[0038] Get manual input or voice input requests initiated by the requester;

[0039] Generate original requirements based on the manually input service request or the voice input service request, and enter the requirement form;

[0040] Extract the feature vectors of the original requirement title and requirement description. Perform a first full-weighted similarity identification on the feature vectors of the requirement title and requirement description. Obtain the confirmation result of the requirement submitter. If there are historical similar requirements, abandon the submission of the original requirement. If there are no historical similar requirements, submit the original requirement.

[0041] Obtain confirmation results from the requirements analyst regarding the second full-weighted similarity identification of the submitted original requirements.

[0042] If there are similar historical requirements, the original requirement will be merged and tracked with the corresponding similar requirement;

[0043] If there are no similar historical requirements, the system automatically captures the original requirements and the input information of the requirements analyst to generate contextual data for the requirements analysis scenario. It then integrates the contextual data of the requirements analysis scenario with the internal knowledge base to generate prompts for the requirements analysis scenario, which are then used by the large model to analyze and output detailed requirements descriptions for display in the front-end interaction layer. The displayed content is then filled back into the corresponding input boxes of the requirements form.

[0044] Furthermore, when the requester manually enters the business request on the page, all the text already filled in on the page is captured to generate context data of the request filling scenario. The context data of the request filling scenario, the internal knowledge base, and the requester's question are integrated to generate prompt words for the request filling scenario, so that the large model can analyze and output detailed request filling information for display in the front-end interaction layer, and the displayed content is filled back into the corresponding input box of the request form.

[0045] Thirdly, the present invention also discloses a computer-readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method described in the second aspect.

[0046] Fourthly, the present invention also discloses a computer device, comprising,

[0047] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing the method described in the second aspect.

[0048] The beneficial effects achieved by this invention are as follows:

[0049] (1) This invention constructs a multi-layered collaborative architecture, relying on a private large model cluster to realize the linkage of voice structured form filling, context intelligent question answering, and weighted similarity requirement deduplication function of the same product line; through the collaborative task scheduling unit to uniformly schedule multiple AI sub-algorithms, and the session storage unit to complete the temporary data transfer, the multi-modal AI capabilities can work collaboratively.

[0050] (2) In the voice input scenario, through entity recognition, entity standardization and two-dimensional field matrix mapping, the user's spoken requirements are automatically mapped to the standardized requirements form fields, reducing the threshold for filling in requirements.

[0051] (3) Differentiate AI-assisted analysis for two business scenarios: demand reporting and market analysis. Reuse the enterprise's private knowledge base, reduce reliance on personnel experience, and improve the quality of demand documents.

[0052] (4) Based on the business characteristics of product demand, feature vectors are generated for demand titles and demand descriptions respectively, and weighted fusion is used to obtain a comprehensive vector; search constraints are added, similarity matching is only performed within the same product line, and a two-level deduplication mechanism is set up: pre-submission verification of demand and secondary review of market analysis, effectively filtering out irrelevant similar results across product lines and accurately identifying duplicate project requirements. Attached Figure Description

[0053] Figure 1 This is the overall system architecture diagram;

[0054] Figure 2 This is a schematic diagram of the voice-based structured automatic form filling process;

[0055] Figure 3 This is a schematic diagram of the contextual question-and-answer generation process;

[0056] Figure 4 This is a schematic diagram of the full-weighted vector similarity requirement matching algorithm.

[0057] Figure 5 This is a schematic diagram of the collaborative task scheduling and distribution process. Detailed Implementation

[0058] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0059] Example 1: This example introduces a multimodal AI collaborative product requirement process management system, such as... Figure 1 As shown, the system is equipped with a four-layer hierarchical architecture, which consists of four layers: front-end interaction layer, agent collaborative scheduling middleware layer, private large model service layer, and data persistence layer.

[0060] The front-end interaction layer refers to providing a visual interactive entry point for both requirement submitters and requirement analysts, including mobile and PC interaction components. Mobile interaction components include an audio capture component, a standardized requirement form filler, a floating AI intelligent assistant, and a similar requirement prompt pop-up. PC interaction components include a requirement details display component, an AI analysis assistant, and a similar requirement review pop-up.

[0061] The Agent is a collaborative scheduling middleware layer that uniformly handles front-end requests, schedules computing power for various large models, executes three core sub-algorithms, and manages sessions and data flow. It is divided into five major functional units.

[0062] (1) Speech structure mapping calculation unit, with built-in speech structure mapping sub-algorithm, can automatically fill in the form with speech information by executing the algorithm independently. For example Figure 2As shown, the specific steps are as follows: ① Audio standardization preprocessing: The mobile terminal collects the user's original audio, performs noise reduction, frame segmentation, and 16kHz resampling processing, and outputs standardized audio segments; ② Speech-to-free text: The standardized audio is pushed to the speech recognition model and completely converted into the original free text T0 of the requirements; ③ Industry named entity recognition: The industry NLP entity recognition model is called, and semantic recognition is performed on the T0 text based on the enterprise product entity dictionary to extract the standardized business entity set E = {e1 Original requirement title, e2 Region, e3 Country / Province, e4 Requirement source, e5 Customer name, e6 Customer contact information, e7 Expected delivery time, e8 Company, e9 Comprehensive scenario, e10 Requirement type, e11 Requirement classification, e12 Requirement level, e13 Target sales volume, e14 Target terminal price, e15 Original requirement description, e16 Scenario condition description, e17 Requirement value, e18 Gap between current product / expected solution, e19 Other competing products and their solutions descriptions; ④ Semantic standardization of entity values: Receive the output standardized business entity set, perform semantic cleaning and standard alignment on the entity values ​​in the set, map colloquial and non-standard expressions to standard industry terms, eliminate ambiguity and redundant information in entity values, and output a semantically standardized entity set; ⑤ Two-dimensional field matrix mapping: Read the system's preset form field mapping matrix M; the matrix rows correspond one-to-one with the standardized entity set in step ③, and the matrix columns correspond one-to-one with the independent input boxes in the requirement form; based on the row and column matching relationship, complete the accurate binding of various entity sets with the corresponding input boxes in the form, and establish a unique mapping relationship from entity to form field; ⑥ Automatic form backfilling: Automatically fill the standardized entity values ​​that have completed matrix matching into the corresponding input boxes in the form, set the form fields where no corresponding entity is identified to empty, and push the complete form to the front end for the submitter to verify and complete.

[0063] (2) Context-based question-and-answer generation calculation unit, with built-in context-bound question-and-answer generation sub-algorithm, solves the pain points of process handlers lacking standardized filling and analysis references, and the inability to reuse enterprise knowledge in real time. For example Figure 3As shown, the specific steps are as follows: ① Real-time context caching: When the submitter / analyst opens the form to edit the original requirement form / market analysis, the session storage unit automatically captures all the filled text on the page, generates exclusive context data, and temporarily stores it; ② User input intent recognition: The AI ​​analysis pop-up receives the input information from the submitter / analyst, and distinguishes two types of business scenarios through intent recognition by the large model: requirement submission and requirement market analysis; ③ Layered Prompt construction and template generation: The exclusive input prompt is constructed by integrating context, internal knowledge base, and user questions. The large model outputs two types of content in layers: the requirement submission scenario outputs fields such as scenario conditions and requirement value, and the market analysis scenario outputs multi-dimensional evaluation templates of performance, cost, compatibility, and regulatory constraints; the internal knowledge base is a knowledge base pre-constructed based on the enterprise's product requirement management system documents, product requirement management specifications, established business process documents, and historical requirement cases after structured analysis; ④ Content backfilling interaction: The front-end pop-up displays all generated content, supports one-click backfilling of content of each dimension into the corresponding input box of the form, and retains the ability to copy, assisting the process handlers in quickly improving the original requirement and market analysis information.

[0064] (3) A full-weighted vector matching calculation unit, with a built-in full-weighted vectorized similar demand matching sub-algorithm, realizes real-time weighted deduplication during form submission and demand market analysis, and intercepts similar demands in advance. The rule is that similar demand retrieval is only performed within the same product line, and the vector fusion weight is fixed at 30% for the original demand title and 70% for the original demand description. Figure 4 As shown, the specific steps are as follows: ① Deduplication Trigger: After the submitter automatically fills in the complete requirement title and description via voice or manually fills in the complete requirement title and description, they click the submit button. The front end collects the complete original title and full text of the original requirement description for this requirement in one go, and transmits them to the Agent middle layer to trigger a full weighted similarity check before submission. Entering the market analysis page, the front end retrieves the finalized requirements already in the database, reads the original requirement title and full text of the original requirement description for this requirement from the database in one go, and transmits them to the Agent middle layer to perform a batch full weighted similarity check; ② Independent Vectorization of Full Text: Using only the full-text encoding mode, a title feature vector V is generated for the complete original title of this requirement. t Generate a descriptive feature vector V from the complete original requirement description. d ③ Weighted fusion generates a comprehensive demand vector V a Fusion formula: V a =0.3×V t + 0.7×V dWhere 0.3 is the fixed weight of the original demand title and 0.7 is the fixed weight of the original demand description. Weighted matching is only performed on historical demands within the same product line, and similarity retrieval is not triggered across product lines; ④ Same product line vector library retrieval and similarity calculation: retrieve the weighted comprehensive vector set V corresponding to all historical demands within the same product line from the weighted historical demand vector library. s = {V s1 V s2 ,...V sn}, Batch perform cosine similarity calculation: Sim = (V a V si ) = (V a ·V si ) / (||V a ||×||V si ||), with a preset similarity threshold of 0.75, only retaining similar requirement entries with a similarity of ≥0.75, where cross-product line data is not included in similarity matching; ⑤ Differentiated display of similar requirement results by scenario: In the requirement submission stage, the submission pop-up displays the number, title, and requirement status of all matching similar requirements, prompting the submitter to check and confirm whether to continue submitting; In the market analysis node, the similar requirement identification button can also display the number, title, and requirement status of each similar requirement, supporting the analysis team leader to uniformly merge and identify duplicate requirements.

[0065] (4) Collaborative task scheduling unit: It has built-in business linkage triggering logic, unified scheduling of three major algorithm modules, and automatically matches the execution sequence of corresponding algorithms according to business nodes. It realizes the orderly invocation of three sets of sub-algorithms: voice structured mapping, context intelligent question answering, and full weighted vectorized similar demand matching, and data interconnection and sharing. According to different business scenarios such as demand submission verification and market analysis secondary review, it issues corresponding calculation tasks and summarizes the output results of each module to feed back to the upper-level page.

[0066] (5) Session storage unit: used to temporarily cache data within a single business process, including the complete form filling context, human-computer dialogue records, and real-time feature vectors of the requirements to be matched; it only saves temporary data within the lifecycle of a single business process, providing support for cross-module data transfer of the collaborative task scheduling unit; it automatically cleans up the temporary cached data after the business process is completed, and does not permanently store temporary vectors and form drafts.

[0067] The private large-scale model service layer provides underlying AI computing power support and model inference services, deploying a private enterprise-dedicated model cluster. The cluster has built-in speech recognition models, industry-specific NLP (Natural Language Processing) entity extraction models for R&D business, contextual dialogue large-scale models, and text vectorization models, along with enterprise industry knowledge base retrieval capabilities. All models are connected to the internal business knowledge base, supporting the recognition and semantic parsing of professional terms in the product development field, and providing standardized AI inference capabilities for upper-layer business modules.

[0068] The data persistence layer is used to persistently store all static business data, historical calculation results, and system configuration information of the system; it includes a requirement form database, a weighted historical requirement vector library, an industry knowledge base repository, a user operation log library, and a system parameter configuration library. The system parameter configuration library pre-stores fixed weight parameters for vector fusion: title weight 0.3 and requirement description weight 0.7, for the similarity matching module to read and call.

[0069] Example 2, based on the same inventive concept as Example 1, introduces a multimodal AI collaborative product demand process management system, including: a front-end interaction layer and an Agent collaborative scheduling middleware layer;

[0070] The front-end interaction layer is used to provide a visual interaction entry for requirement submitters and requirement analysts, allowing requirement submitters or requirement analysts to manually input requirement information or input requirement information by voice.

[0071] The Agent-based collaborative scheduling middleware layer includes:

[0072] The speech structured mapping calculation unit is used to automatically fill in the text requirement form with the speech input requirement information collected from the visual interactive entry point;

[0073] The contextual question-answering generation calculation unit is used to automatically capture the manually input requirement information and / or the text requirement form, and generate contextual data for the corresponding business scenario. The business scenario includes the requirement submission scenario of the corresponding requirement submitter and the requirement analysis scenario of the corresponding requirement analyst. For the corresponding business scenario, the contextual information of the corresponding business scenario and the internal knowledge base are integrated to generate target prompt words, so that the large model can output the corresponding scenario display content for display in the front-end interaction layer, and the displayed content is backfilled into the corresponding input box of the requirement form.

[0074] The full-weighted vector matching calculation unit is used to perform full-weighted similarity identification on the original requirements in both the requirement submission and requirement analysis scenarios to determine whether there are any historical similar requirements.

[0075] Example 3 is based on the same inventive concept as other examples, such as... Figure 5 As shown in the figure, this embodiment introduces a product requirement process management method for multimodal AI collaboration, including:

[0076] Get manual input or voice input requests initiated by the requester;

[0077] Generate original requirements based on the manually input service request or the voice input service request, and enter the requirement form;

[0078] Extract the feature vectors of the original requirement title and requirement description. Perform a first full-weighted similarity identification on the feature vectors of the requirement title and requirement description. Obtain the confirmation result of the requirement submitter. If there are historical similar requirements, abandon the submission of the original requirement. If there are no historical similar requirements, submit the original requirement.

[0079] Obtain confirmation results from the requirements analyst regarding the second full-weighted similarity identification of the submitted original requirements.

[0080] If there are similar historical requirements, the original requirement will be merged and tracked with the corresponding similar requirement;

[0081] If there are no similar historical requirements, the system automatically captures the original requirements and the input information of the requirements analyst to generate contextual data for the requirements analysis scenario. It then integrates the contextual data of the requirements analysis scenario with the internal knowledge base to generate prompts for the requirements analysis scenario, which are then used by the large model to analyze and output detailed requirements descriptions for display in the front-end interaction layer. The displayed content is then filled back into the corresponding input boxes of the requirements form.

[0082] In this embodiment, when the requester manually enters the business request on the page, all the text already filled in on the page is captured to generate context data of the request filling scenario. The context data of the request filling scenario, the internal knowledge base, and the requester's question are integrated to generate prompt words for the request filling scenario, so that the large model can analyze and output detailed request filling information for display in the front-end interaction layer, and the displayed content is filled back into the corresponding input box of the request form.

[0083] Example 4, based on the same inventive concept as other examples, describes a computer-readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of Example 3.

[0084] Example 5, based on the same inventive concept as other examples, describes a computer device including one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing the method described in Example 3.

[0085] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0089] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multimodal AI collaborative product requirement process management system, characterized in that, include: The front-end interaction layer and the agent collaborative scheduling middleware layer; The front-end interaction layer is used to provide a visual interaction entry for requirement submitters and requirement analysts, allowing requirement submitters or requirement analysts to manually input requirement information or input requirement information by voice. The Agent-based collaborative scheduling middleware layer includes: The speech structured mapping calculation unit is used to automatically fill in the text requirement form with the speech input requirement information collected from the visual interactive entry point; The contextual question-answering generation calculation unit is used to automatically capture the manually input requirement information and / or the text requirement form, and generate contextual data for the corresponding business scenario. The business scenario includes the requirement submission scenario of the corresponding requirement submitter and the requirement analysis scenario of the corresponding requirement analyst. For the corresponding business scenario, the unit integrates the contextual information of the corresponding business scenario and the internal knowledge base to generate target prompt words, so that the large model can output the corresponding scenario display content for display in the front-end interaction layer, and fill the display content back into the corresponding input box of the requirement form. The internal knowledge base is a knowledge base pre-constructed based on the enterprise's product requirement management system documents, product requirement management specifications, established business process documents, and structured analysis of historical requirement cases. The full-weighted vector matching calculation unit is used to perform full-weighted similarity identification on the original requirements in both the requirement submission and requirement analysis scenarios to determine whether there are any historical similar requirements.

2. The multimodal AI collaborative product requirement process management system according to claim 1, characterized in that, Also includes: Private large-scale model service layer and data persistence layer; The private large model service layer is used to deploy speech recognition models, industry NLP entity recognition models for R&D business, contextual dialogue large models, and text vectorization models for the Agent collaborative scheduling middleware layer to use for scheduling. The data persistence layer is used to persistently store all static business data, historical calculation results, and system configuration information of the system.

3. The multimodal AI collaborative product requirement process management system according to claim 2, characterized in that, The Agent-based collaborative scheduling middleware layer also includes: The collaborative task scheduling unit is used to uniformly schedule the speech structure mapping calculation unit, the context question-and-answer generation calculation unit, and the full weighted vector matching calculation unit. It automatically matches the execution sequence of the corresponding unit according to the business node, and issues the corresponding calculation tasks according to different business scenarios such as submission for verification and secondary review of market analysis, and summarizes the output results of each unit and feeds them back to the upper-level page. The session storage unit is used to temporarily cache data within a single business process, including the complete form filling context, human-computer dialogue records, and real-time feature vectors of the requirements to be matched; temporary data is only stored within the lifecycle of a single business process, and the temporary cached data is automatically cleared after the business process is completed.

4. The multimodal AI collaborative product requirement process management system according to claim 3, characterized in that, The processing procedure of the speech structure mapping calculation unit includes: The front-end interaction layer collects the user's original audio, performs noise reduction, frame segmentation, and 16kHz resampling processing in a unified manner, and outputs standardized audio. Standardized audio is pushed to a speech recognition model and converted into the original free text T0 of the requirements; The industry NLP entity recognition model is invoked to perform semantic recognition on the original free text T0 based on the enterprise product entity dictionary, and a standardized business entity set E is extracted. Semantic cleaning and standard alignment are performed on the entity values ​​in the standardized business entity set E, mapping colloquial and non-standard expressions to standard industry terms, eliminating ambiguity and redundant information in the mapped entity values, and outputting a semantically standardized entity set. Read the system's preset form field mapping matrix M, where each row of the matrix corresponds to a different entity set in the semantically standardized entity set, and each column of the matrix corresponds to an independent input box in the required form; based on the row and column matching relationship, complete the precise binding of each entity set to the corresponding input box in the form, and establish a unique mapping relationship from entity to form field; The entity values ​​of the semantically standardized entity set that has been matched are automatically filled into the corresponding input boxes in the form. Form fields that do not have a corresponding entity are left empty. The complete form is pushed to the front end for the submitter to verify and complete.

5. The multimodal AI collaborative product requirement process management system according to claim 3, characterized in that, The processing procedure of the context question-answering generation computing unit includes: In the requirement submission scenario, when the requirement submitter opens the form to edit the original requirement form, the session storage unit automatically captures all the filled text on the page and generates requirement submission scenario context data. The system integrates contextual data of requirement submission scenarios, internal knowledge base, and questions from requirement submitters to construct requirement submission scenario prompts, which are then used by the large model to output fields such as scenario conditions, requirement value, gaps / expected solutions of the current product, other competitors and their solutions, and original requirement descriptions. The front-end pop-up displays all generated content and is configured to automatically populate all generated content back into the corresponding input boxes of the form with one click; In a requirements analysis scenario, when a requirements analyst opens a form to edit market analysis, the session storage unit automatically captures all the filled-in text on the page and generates context data for the requirements analysis scenario. By integrating contextual data from the requirements analysis scenario, internal knowledge base, and questions from requirements analysts, prompt words for the requirements analysis scenario are constructed to provide multi-dimensional evaluation templates for the large model to output. The front-end pop-up displays all generated content and is configured to automatically populate all generated content back into the corresponding input boxes of the form with one click.

6. The multimodal AI collaborative product requirement process management system according to claim 3, characterized in that, The processing procedure of the full-weighted vector matching calculation unit includes: Generate a title feature vector V for each complete original title of the original requirement. t Generate a descriptive feature vector V from the complete original requirement description. d ; Weighted fusion generates a comprehensive demand vector V a The fusion formula is: V a = α×V t +β× V d Where α is the fixed weight of the original requirement title, β is the fixed weight of the original requirement description, and α+β=1; Retrieve the weighted composite vector set V corresponding to all historical demands within the same product line from the weighted historical demand vector library. s V s = {V s1 V s2 ,...V sn }, where V si This represents the i-th weighted composite vector in the set of weighted composite vectors, where i = 1, 2, ..., n; Batch execution of cosine similarity calculation yields the similarity Sim, where Sim = (V a ·V si ) / (||V a ||×||V si In the formula, · represents the dot product symbol, ||V a ||and||V si || indicates V a and V si The modulus length; retain the weighted composite vector set V s Similar requirement entries with a similarity greater than or equal to a preset threshold are selected for confirmation by the requirement submitter or requirement analyst to determine whether they are similar.

7. A product requirement process management method for multimodal AI collaboration, characterized in that, include: Get manual input or voice input requests initiated by the requester; Generate original requirements based on the manually input service request or the voice input service request, and enter the requirement form; Extract the feature vectors of the original requirement title and requirement description. Perform a first full-weighted similarity identification on the feature vectors of the requirement title and requirement description. Obtain the confirmation result of the requirement submitter. If there are historical similar requirements, abandon the submission of the original requirement. If there are no historical similar requirements, submit the original requirement. Obtain confirmation results from the requirements analyst regarding the second full-weighted similarity identification of the submitted original requirements. If there are similar historical requirements, the original requirement will be merged and tracked with the corresponding similar requirement; If there are no similar historical requirements, the system automatically captures the original requirements and the input information of the requirements analyst to generate contextual data for the requirements analysis scenario. It then integrates the contextual data of the requirements analysis scenario with the internal knowledge base to generate prompts for the requirements analysis scenario, which are then used by the large model to analyze and output detailed requirements descriptions for display in the front-end interaction layer. The displayed content is then filled back into the corresponding input boxes of the requirements form.

8. The product requirement process management method for multimodal AI collaboration according to claim 7, characterized in that, When the requester manually enters the business request on the page, all the text already filled in on the page is captured to generate context data of the request filling scenario. The context data of the request filling scenario, the internal knowledge base, and the requester's questions are integrated to generate prompt words for the request filling scenario. This prompts are then used by the large model to analyze and output detailed request filling information for display in the front-end interaction layer. The displayed content is then filled back into the corresponding input box of the request form.

9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods of claims 7 to 8.

10. A computer device, characterized in that, include, One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing the method of any of claims 7 to 8.

Citation Information

Patent Citations

  • Product demand full-process intelligent auxiliary method and device based on image recognition and large model

    CN121166539A

  • Product demand management method and system based on large model agent

    CN121303746A