Demand information feedback method and device, electronic equipment and storage medium

Abstract

The invention relates to a demand information feedback method and device, electronic equipment and a storage medium. The method comprises the following steps: in response to received input information of a user, determining a user intention according to the input information; under the condition that it is determined that the input information represents knowledge questions and answers related to the field, field-related knowledge information is obtained from the structured field knowledge base according to the user intention and the input information; based on the domain-related knowledge information, processing the input information into cue word data; and based on the prompt word data, generating demand feedback information for the user through the large model. According to the scheme provided by the invention, the user intention can be identified, the existing knowledge reuse can be realized, the demand feedback information for the user can be quickly obtained, and the fluency of communication with the user and the user experience can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a demand information feedback method and device, electronic equipment and storage medium. BACKGROUND

[0002] In related technologies, with the development of artificial intelligence technology, deep learning models have been widely used in natural language processing tasks.

[0003] A representative scheme of current demand automatic generation is as follows: after receiving an original demand, a pre-configured demand retriever is used to identify the top K most relevant original demands; each original demand-derivative demand is input into a pre-configured demand production prompt template, and according to whether the original demand and the derivative demand match, a large model is selectively guided to generate a derivative demand. However, this scheme excessively relies on the static matching of preset rules and historical data. SUMMARY

[0004] To solve or partially solve the problems in related technologies, the present application provides a demand information feedback method, device, electronic equipment and storage medium, which determines the user's intention according to the input information in response to receiving the user's input information; obtains the domain-related knowledge information from the structured domain knowledge base according to the user's intention and the input information in the case that the input information represents a knowledge question and answer related to the domain; processes the input information into prompt word data based on the domain-related knowledge information; and generates demand feedback information for the user through a large model based on the prompt word data, which can identify the user's intention, realize existing knowledge reuse, quickly obtain demand feedback information for the user, and help improve the fluency of communication with the user and the user experience.

[0005] The first aspect of the present application provides a demand information feedback method, comprising: determining the user's intention according to the input information in response to receiving the user's input information; obtaining domain-related knowledge information from the structured domain knowledge base according to the user's intention and the input information in the case that the input information represents a knowledge question and answer related to the domain; processing the input information into prompt word data based on the domain-related knowledge information; and generating demand feedback information for the user through a large model based on the prompt word data.

[0006] In some embodiments, the method further comprises: determining a target path according to the deployment instruction in response to receiving the user's deployment instruction, and writing the demand feedback information into the system to generate a demand entry according to the target path.

[0007] In some embodiments, the target path includes at least one of a requirement folder path, a requirement specification path, and a requirement item path; in response to receiving the deployment instruction of the user, the target path is determined according to the deployment instruction, and the requirement feedback information is written into the system according to the target path to generate a requirement item, including at least one of the following: in response to receiving the deployment instruction of the user, in a case where the target path determined according to the deployment instruction includes the requirement folder path, obtaining a requirement specification name defined by the user, and in generating the requirement item, generating a requirement specification under the folder corresponding to the requirement specification name, and writing the requirement item as the content of the requirement specification; in response to receiving the deployment instruction of the user, in a case where the target path determined according to the deployment instruction includes the requirement specification path, in generating the requirement item, writing the requirement item as the content into the requirement specification above the corresponding requirement specification; in response to receiving the deployment instruction of the user, in a case where the target path determined according to the deployment instruction includes the requirement item path, obtaining a sub-level or a same-level of the corresponding item defined by the user, and in generating the requirement item, writing the requirement feedback information into the sub-level or the same-level of the corresponding item.

[0008] In some embodiments, the method further includes: constructing a structured domain knowledge base; wherein the construction includes: data preparation and structured processing to obtain target data; constructing a vector index according to the target data; integrating retrieval enhancement to generate a RAG model for multi-dimensional enhancement processing on query results obtained through the vector index.

[0009] In some embodiments, the input information includes current round input information and historical dialogue information.

[0010] In some embodiments, in response to receiving the input information of the user, the user intent is determined according to the input information, including: in response to receiving the input information of the user, the intent label is determined according to the current round input information and the historical dialogue information.

[0011] The second aspect of the present application provides a requirement information feedback device, which includes: a determination module for determining a user intent according to input information in response to receiving the input information of the user; an obtaining module for obtaining domain-related knowledge information from a structured domain knowledge base according to the user intent and the input information in a case where the input information represents a knowledge question and answer related to a domain; a processing module for processing the input information into prompt word data based on the domain-related knowledge information; and a first generation module for generating requirement feedback information for the user based on the prompt word data through a large model.

[0012] In some embodiments, the device further includes: a second generation module for determining a target path according to a deployment instruction of a user in response to receiving the deployment instruction of the user, and generating a requirement item by writing the requirement feedback information into a system according to the target path.

[0013] The third aspect of the present application provides an electronic device, comprising: a processor; and a memory having stored executable codes, which, when executed by the processor, cause the processor to perform the method as described above.

[0014] The fourth aspect of the present application provides a computer-readable storage medium having stored executable codes, which, when executed by a processor of an electronic device, cause the processor to perform the method as described above.

[0015] The technical solutions provided by the present application can include the following beneficial effects: The technical solutions provided by the present application can include the following beneficial effects:

[0016] The technical solutions provided by the present application can also quickly and accurately generate demand items.

[0017] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings, in which like reference characters refer to like parts throughout the figures, and in which:

[0019] Figure 1 is a flowchart of a demand information feedback method shown in an embodiment of the present application; Figure 2 is a demand management structure diagram shown in an embodiment of the present application; Figure 3 is a structure diagram of a demand information feedback device shown in an embodiment of the present application; Figure 4 is a structure diagram of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0020] Embodiments of the present application will be described in more detail below with reference to the drawings. Although embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

[0021] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this application and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms "first," "second," "third," etc. can be employed in this application to describe various information, such information should not be limited by these terms. These terms are only used to distinguish one piece of information from another. For example, a first information can also be termed a second information, and, similarly, a second information can also be termed a first information, without departing from the scope of the present application. Therefore, the features defined with "first," "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0023] Currently, there is a demand for an automatically generated representative solution, such as receiving an original demand, identifying the top K most relevant original demands using a pre-configured demand retriever; inputting each original demand-derivative demand into a pre-configured demand generation prompt template, selectively guiding a large model to generate a derivative demand according to whether the original demand and the derivative demand match. However, this solution relies too much on the static matching of preset rules and historical data.

[0024] It can be understood that relying too much on the static matching of preset rules and historical data can cause the following problems: 1) Limitations of demand retriever: The pre-configured retrieval logic highly depends on the integrity and annotation quality of the historical demand library. In the face of emerging fields or ambiguously expressed original demands, the semantic generalization ability may not be sufficient to screen out key associated demands, and the generation result may fall into the "known trap" and lack innovation.

[0025] 2) Rigidity of template-driven generation: Demand generation prompt templates essentially constrain the creativity of large models through binary matching. When there are multiple implicit templates or complex context dependencies in the original demand, the mechanical screening of templates may fragment the systematic relevance between demands, resulting in fragmented and low-completion solutions.

[0026] To solve the above problems, the embodiment of the application provides a demand information feedback method. The method comprises the following steps: in response to receiving input information of a user, determining a user intention according to the input information; in a case where it is determined that the input information represents a knowledge question and answer related to a field, obtaining field-related knowledge information from a structured field knowledge base according to the user intention and the input information; processing the input information into prompt word data based on the field-related knowledge information; and generating demand feedback information for the user by using a large model based on the prompt word data. The method can identify the user intention, realize knowledge reuse, quickly obtain the demand feedback information for the user, and help improve the fluency of communication with the user and the user experience.

[0027] The technical solutions of the embodiments of the application are described in detail below with reference to the drawings.

[0028] Figure 1 FIG. 1 is a flowchart of a demand information feedback method according to an embodiment of the application.

[0029] Referring to FIG. 1, a demand information feedback method comprises the following steps. Figure 1 The method comprises the following steps. Step 101: in response to receiving input information of a user, determining a user intention according to the input information.

[0030] In this step, the user can be a demand engineer or other person who needs to generate a demand.

[0031] In some embodiments, the input information can be an original demand. The original demand can be related text content or a document file of a series of requirements that a certain project or a certain project in a certain field needs to achieve.

[0032] In some embodiments, the original demand can also be excluded. For example, the AI generates C919-A demand according to the C919 demand in the knowledge base. In this scenario, the knowledge in the structured field knowledge base is used as the input of demand generation according to the user's instruction.

[0033] In some embodiments, the input information comprises current input information and historical dialogue information.

[0034] It can be understood that the user intention is accurately understood through multiple rounds of interaction, and the demand feedback information is obtained through the large model according to the intention, so as to generate the demand item.

[0035] It can be understood that the demand feedback information can be corrected through multiple rounds of interaction, so as to correct the generated demand item, thereby iteratively optimizing the demand feedback information, and waiting for the demand engineer to confirm the target demand feedback information.

[0036] The output result is matched with the real intention of the user, and the efficiency of demand writing is improved. The demand refers to the demand being written at the time. The AI generated demand item can generate a demand specification, or it can generate a demand item in the demand specification being written.

[0037] In some embodiments, in response to receiving the input information of the user, the user intention is determined according to the input information, comprising: in response to receiving the input information of the user, determining the intention label according to the current input information and the historical dialogue information.

[0038] For example, the input information of the user is identified, and in a case where it is determined that the input information belongs to direct input, the intention is judged by using a large model, and then the intention label is output.

[0039] For example, the input information of the user is identified, and in a case where it is determined that the input information belongs to input through a shortcut function, the intention label is carried in the request.

[0040] Specifically, for example, the input information of the user is parsed, and the user intention is determined in the following two ways.

[0041] a) Based on the large model, the current user message and the historical user message are automatically identified, the range of user intention is limited in order to provide accuracy, and the few-shot prompt word is used to help the large model to accurately identify.

[0042] b) If the user starts a chat through the preset shortcut function on the interactive interface, the preset intention in the shortcut function is captured. The shortcut function provides a prompt word template of different types of intentions or tasks, and the user only needs to edit based on the preset model, modify some parts and click to send. Through the way of shortcut function, the user intention can be enhanced and accurately sent to the background application for processing, which improves the intention recognition degree and accuracy.

[0043] Step 102, in a case where it is determined that the input information represents a knowledge question and answer related to a field, obtaining field related knowledge information from a structured field knowledge base according to the user intention and the input information.

[0044] For example, first, it is determined whether the input information represents a knowledge question and answer related to a field, so that the large model can judge whether it is necessary to perform knowledge retrieval. If not, it is not necessary to retrieve and directly reply. If yes, the knowledge base retrieval is performed, that is, the knowledge base retrieval tool is called, and the field related knowledge information is obtained from the structured field knowledge base according to the user intention and the input information.

[0045] It can be understood that when the knowledge base is called, if the knowledge base does not retrieve the corresponding content, it will reply that the corresponding content cannot be retrieved in the knowledge base, thereby giving the user the right to judge whether the generated requirement item is based on the knowledge base or the large model, so as to avoid generating requirements that are chaotic data and mislead users.

[0046] In some embodiments, the structured domain knowledge base can include knowledge content uploaded by the requirement engineer, which can be a successful example of some previously generated requirement items. The reuse of existing knowledge (knowledge content uploaded by the requirement engineer) is achieved.

[0047] Step 103, based on the domain-related knowledge information, the input information is processed into prompt word data.

[0048] It can be understood that the retrieved domain-related knowledge information is used as context information, and the input information is processed into prompt word data through the domain-related knowledge information.

[0049] It can be understood that prompt word data is key information used to guide large models to generate specific types of answers or content. They are like a key that can open the "door of thought" of the large model, allowing it to think and answer questions according to the user's wishes. Through carefully designed prompt words, the accuracy and efficiency of the large model in practical applications can be improved.

[0050] Step 104, based on the prompt word data, generate requirement feedback information for the user through the large model.

[0051] For example, obtaining context content such as prompt word data can include domain-related knowledge information + input information, calling a large model, and thus outputting the above-mentioned requirement feedback information.

[0052] Large model refers to a large language model (Large Language Model), which is a machine learning model that can understand and generate human language.

[0053] It can be understood that this step realizes the standardized output of requirement feedback information through the JSON generation capability of the large language model, combined with the pre-defined data structure and verification mechanism.

[0054] Requirement feedback information refers to the requirement feedback answer generated based on the prompt word data.

[0055] It should be noted that this step can include the following three points, specifically: a) Structure constraint management Define a general business data template based on JSON Schema (including field type / must-have item / nested structure); dual-channel generation strategy: LLM native JSON output and rule engine fallback guarantee; dynamic Schema adaptation: support online updating of data structure configuration (such as field addition / deletion / format change).

[0056] b) Output quality control Three-layer verification system: syntax verification (JSON format) + structure verification (field matching) + semantic verification (value range specification); exception handling mechanism: trigger retry strategy when generation fails (simplify context / example guide / disassemble generation); progressive generation: generate complex structures in stages and real-time verification.

[0057] c) Fault-tolerant processing, field hierarchical processing Mandatory fields: strict verification, error if failed; important fields: trigger retry mechanism; extended fields: allow null or default value; basic self-repair: automatically correct minor format issues (such as type conversion / field completion).

[0058] The demand information feedback method of the embodiment of the application, by responding to receiving input information of a user, determines the user's intention according to the input information; in the case of determining that the input information represents a knowledge question and answer related to a field, obtains field-related knowledge information from a structured domain knowledge base according to the user's intention and the input information; based on the field-related knowledge information, processes the input information into prompt word data; based on the prompt word data, generates demand feedback information for the user through a large model, which can identify the user's intention, realize existing knowledge reuse, quickly obtain demand feedback information for the user, and help improve the fluency of communication with the user and user experience.

[0059] It can be understood that the demand information feedback method of the embodiment of the application also needs to pre-construct a structured domain knowledge base.

[0060] In some embodiments, the construction method of the structured domain knowledge base can include: data preparation and structured processing to obtain target data; constructing a vector index according to the target data; integrating a retrieval enhancement generation RAG model to perform multi-dimensional enhancement processing on the query results obtained through the vector index.

[0061] It can be understood that the RAG technology of the large language model is used to construct the structured domain knowledge base, which can significantly improve the retrieval efficiency of domain knowledge and the accuracy of generated demand, while supporting incremental updating of the knowledge base to maintain the timeliness of the content.

[0062] In order to better understand the present application, the content of the present application will be further described below in combination with embodiments, but the present application is not limited to the following embodiments.

[0063] (1) Data Preparation and Structured Processing This step includes data parsing, structured processing of relevant field documents, and ensuring data quality and availability through various technical means.

[0064] a) Data Format Parsing When parsing data in different formats, a multi-modal preprocessing framework is used.

[0065] Text Extraction: Supports 12 mainstream formats such as Word, PDF, Markdown, Excel, etc. Utilizes Apache Tika and PDFMiner for deep parsing, and can automatically identify title levels, list structures, and annotation areas when processing complex layout documents.

[0066] OCR Recognition: For scanned / image documents, integrates Tesseract-OCR and Google Cloud Vision, and improves recognition accuracy through image binarization, perspective correction, and font smoothing algorithms.

[0067] Document Layout Optimization: Uses LaTeX formula parsing engine to process mathematical symbols, DocBank model to identify table / chart areas, and NLP-driven intelligent segmentation algorithm to solve paragraph breaking problems.

[0068] Multi-modal Content Extraction: Combines CLIP model to realize image-text association analysis and uses visual understanding model to extract image content.

[0069] b) Structured Slice Processing Intelligent chunking of extracted content: Fixed character number slicing: uses sliding window mechanism, sets adjustable fixed character number such as 2048 characters basic block (overlapping 256 characters).

[0070] Specific character mode segmentation: based on regular expression to identify chapter markers (such as "1.1 Research Background"), uses XML / HTML tag structure to intelligently segment academic papers, and preserves original format metadata.

[0071] Paragraph Recursive Segmentation: Uses adaptive chunking algorithm, sets maximum length 512 tokens and minimum overlap 128 tokens, maintains causal relationship through syntax analysis, and maintains logical association between clauses when processing legal documents.

[0072] Paragraph Semantic Segmentation: Integrates BERT-wwm model for semantic boundary detection, maintains context continuity while achieving knowledge unit granularity control.

[0073] c) Text Enhancement Processing Intelligent annotation of sliced text blocks: keyword extraction, keyword extraction based on large models, and the number of extracted keywords can be specified. Automatic question extraction based on large models and combined with knowledge base semantics.

[0074] (2) Constructing a vector index This step designs a technical framework for multi-modal data vectorization and storage, including three core links.

[0075] a) Unified representation of multi-modal data using a hybrid architecture: Text encoding, generate 4096-dimensional text vectors through the Transformer architecture. Image encoding, build a visual encoder based on the CLIP model, use the ViT-B / 32 architecture to process image data in the slice.

[0076] b) Vector database interfacing solution, design a distributed vector storage architecture: Basic engine, use Elasticsearch database to build a vector index; Metadata association, establish a mapping label between the slice and the original document information, record timestamp and version information for retrieval filtering; Interface specification, develop a unified RESTful API to implement atomic operations such as vector writing, updating, and deleting.

[0077] c) Build a hybrid retrieval service: Retrieval routing, develop a query parser based on feature detection to automatically identify the input modality type; Distance calculation, support multiple algorithm configurations such as cosine similarity, Euclidean distance, and dot product; Hybrid retrieval, implement hybrid retrieval based on text semantics and keyword indexing, and support setting the hybrid retrieval proportion weight.

[0078] (3) Integrate RAG model (Retrieval Augmented Generation model) This step mainly enhances the knowledge retrieval process in multiple dimensions by introducing a deep learning rerank model to perform secondary relevance sorting on the initial recalled knowledge blocks. Specifically, after the first round of screening by traditional inverted index or vector semantic retrieval, the system inputs the retrieval results into a rerank model based on the Transformer architecture. This model uses a multi-head attention mechanism to deeply analyze the interaction features between the question and the document, evaluates dimensions such as keyword matching degree, context correlation strength, intent similarity, and entity overlap, and constructs a multi-layer neural network scoring system to achieve precise sorting of candidate knowledge blocks. After sorting, the system uses a dynamic sliding window mechanism to retain the Top-K relevant documents, effectively improving the recall accuracy of key information. In terms of retrieval strategy optimization, the system supports building a hybrid retrieval engine, and through a configurable weight adjustment module, it realizes the organic integration of keyword retrieval and semantic retrieval.

[0079] In some embodiments, for the structured domain knowledge base, it is noted that the domain knowledge base element relationship can include: domain knowledge base → metadata, general pattern knowledge base, and parent-child pattern knowledge base. Among them, the general pattern knowledge base can include general pattern → general pattern segment. The parent-child pattern knowledge base can include parent-child pattern document → parent-child pattern segment (parent segment, child parent segment).

[0080] By constructing enterprise knowledge through structured domain knowledge base, it can be integrated by AI, so that the original enterprise knowledge can be reused when demand is generated.

[0081] Metadata is information used to describe other data. Through the application of metadata, accurate retrieval of corresponding knowledge can be achieved when AI retrieves the knowledge base. Metadata in domain knowledge base management can be applied to knowledge base folders, general pattern documents, and parent-child pattern documents. When applying, the value of specific metadata can be defined, such as establishing a model name metadata, applying the metadata when establishing a folder, and defining the model name as C919.

[0082] Metadata application follows a parent-child inheritance relationship, making it easier for engineers to build structured domain knowledge bases. For example, when the metadata value of a folder is C919, its subfolders and documents contain this metadata. When establishing metadata at the parent node, if the child node has the metadata, it will be overwritten by the parent node. This application realizes the structured management of knowledge, ensuring that the child node is labeled with C919 and does not need to perform repetitive maintenance.

[0083] The general pattern knowledge base splits the content into independent segments. When the user inputs a question, the system automatically analyzes the keywords in the question and calculates the relevance of the keywords to each content segment in the knowledge base. According to the relevance ranking, the most relevant content segment is selected and sent to the LLM to assist in processing and more effectively answer.

[0084] Parent-child pattern knowledge base. Compared with the general pattern, the parent-child pattern uses a double-layer segment structure to balance the accuracy of retrieval and contextual information, allowing both precise matching and comprehensive contextual information. Among them, the parent block maintains a larger text unit (such as a paragraph), providing rich contextual information; the child block is a smaller text unit (such as a sentence), used for precise retrieval. The system first performs precise retrieval through the child block to ensure relevance, and then obtains the corresponding parent block to supplement the contextual information, so as to ensure accuracy and provide complete background information when generating responses.

[0085] It can be understood that the present embodiment is beneficial to improve the reusability of existing knowledge. For example, the general large language model has a relatively general ability and does not reuse the historical model data of the enterprise. When writing the demand items of C929, some requirements can follow the demand items of C919.

[0086] In some embodiments, the method can further include: in response to receiving the deployment instruction of the user, determining the target path according to the deployment instruction, and writing the requirement feedback information into the system according to the target path to generate the requirement entry.

[0087] It can be understood that the role of generating the requirement entry is to facilitate the user (such as a requirement engineer) to conveniently view the requirement entry automatically generated by the large language model.

[0088] For example, the target path includes at least one of a requirement folder path, a requirement specification path, and a requirement entry path.

[0089] For example, the user (a requirement engineer) can send a deployment instruction after confirming the requirement feedback information. The deployment instruction can carry the target path and the requirement feedback information. Thus, in response to receiving the deployment instruction of the user, the target path is determined according to the deployment instruction, and the requirement feedback information is written into the system according to the target path to generate the requirement entry.

[0090] In some embodiments, in response to receiving the deployment instruction of the user, the target path is determined according to the deployment instruction, and the requirement feedback information is written into the system according to the target path to generate the requirement entry, including at least one of the following: In response to receiving the deployment instruction of the user, the target path is determined according to the deployment instruction, and the requirement feedback information is written into the system according to the target path to generate the requirement entry, including at least one of the following:

[0091] In response to receiving the deployment instruction of the user, the target path is determined according to the deployment instruction, and the requirement feedback information is written into the system according to the target path to generate the requirement entry, including at least one of the following:

[0092] In response to receiving the deployment instruction of the user, the target path is determined according to the deployment instruction, and the requirement feedback information is written into the system according to the target path to generate the requirement entry, including at least one of the following:

[0093] It can be understood that different target paths have different applicable scenarios and advantages.

[0094] It can be understood that requirement writing can be divided into early, middle and late stages.

[0095] For example, in the early stage of requirement writing, first, the requirement engineer can generate corresponding requirement feedback information by reusing the historical requirements in the knowledge base with the help of the large model. Then, when writing the requirement feedback information into the system to generate the requirement entry, the user selects the requirement folder path, automatically generates a complete requirement specification document by specifying the relevant folder path, thereby effectively solving the problems of low efficiency, repetitive labor and difficulty in knowledge reuse in traditional requirement writing, and significantly improving the writing speed of the requirement document.

[0096] For example, in the middle stage of requirement writing, first, the requirement engineer can perform intelligent review on the existing requirement specification document with the help of the large model to identify the integrity of its chapter structure. If a chapter is missing, the large model will automatically generate the corresponding requirement content as requirement feedback information. Then, when writing the requirement feedback information into the system to generate the requirement entry, the user selects the requirement specification path, thereby accurately integrating into the corresponding position of the requirement specification document through the requirement specification path specified by the user (engineer), thereby significantly improving the structural integrity and content quality of the requirement document, and effectively reducing the workload of manual checking and manual supplement.

[0097] For example, in the late stage of requirement writing, first, the requirement engineer can perform in-depth supplement and entry refinement on the existing requirement specification document with the help of the large model. The large model can generate more detailed descriptions or decomposition contents for a single requirement entry, and the large model will generate the more detailed descriptions or decomposition contents as requirement feedback information. Then, when writing the requirement feedback information into the system to generate the requirement entry, the user selects the requirement entry path, thereby accurately positioning the newly added content (i.e. more detailed descriptions or decomposition contents) as the same level (parallel items) or sub-level (lower level decomposition items) of the existing entry through the existing entry specified by the user (engineer) as a reference, thereby significantly improving the granularity, clarity and traceability of the requirement entry, effectively ensuring the integrity and logical rigor of the requirement specification, and greatly reducing the risk of omissions and logical confusion in the manual refinement process.

[0098] Figure 2 The requirement management structure shown in the embodiments of the present application is shown in the following table.

[0099] In order to better understand the present application, the content of the present application will be further described below in conjunction with the embodiments, but the present application is not limited to the following embodiments. See Figure 2 .

[0100] The present embodiment can parse the structured requirement data conforming to the JSON specification, first perform format verification and data enhancement processing, then dynamically map each data field to a visual element using a componentized front-end architecture, combine conditional styles, and finally render a set of requirement entries containing interactive functions.

[0101] For writing the demand feedback information into the system according to the target path, a demand item is generated. It should be noted that the main purpose of this step is to write the demand data (demand feedback information) generated by the large language model into the demand management system according to the path selected by the user (such as a demand engineer), so as to directly use the automatically generated data and save the manual input link of the engineer. The specific implementation steps include: (1) generating path selection; (2) system data generation.

[0102] (1) Generating path selection The main purpose of this step is to select the generation path of the automatically generated demand item as the path information written into the system.

[0103] The paths selectable by the demand engineer in this step can include at least one of a demand folder path, a demand specification path, and a demand item path.

[0104] When the folder (i.e., the demand folder path) is selected, the engineer also needs to define the demand specification name, and when the system data is generated, the demand specification will be generated under the folder, and the demand item will be written as the demand specification content.

[0105] For example, the original data "a certain model project [type: folder]" can be used as input information.

[0106] The demand feedback information obtained by the large model is shown in Table 1.

[0107] Table 1

[0108] When the folder (i.e., the demand folder path) is selected, the demand specification name such as "product demand" is defined.

[0109] The system data generation result includes: Product demand [type: demand specification] and demand specification content such as Table 2 Table 2

[0110] When the demand specification (i.e., the demand specification path) is selected, when the system data is generated, the demand item will be written as the content into the demand specification above, such as the top.

[0111] For example, the original data "a certain model project [type: folder]" and product demand [type: demand specification] can be used as input information, wherein the demand specification is shown in Table 1.

[0112] The demand feedback information obtained by the large model is shown in Table 3.

[0113] Table 3

[0114] When selecting a requirement specification (i.e., the requirement specification path).

[0115] The system data generation results include: Table 4 Table 4

[0116] When selecting a requirement item (such as a requirement item path), the engineer also needs to define the generation as a child or sibling of that requirement item. When the system data is generated, the requirement (i.e. requirement feedback information) will be written to the corresponding child or sibling of the requirement item.

[0117] For example, the original data is "Project Model [Type: Folder]" and the product requirements are "Type: Requirements Specification". The requirements specifications are shown in Table 5 and can be used as input information.

[0118] The demand feedback information obtained through the large model is shown in Table 6.

[0119] Table 5

[0120] Table 6

[0121] When selecting a requirement item (such as a requirement item path), define the requirement item as: 2.1 Power and Range; Generation Type: Same Level.

[0122] The system data generation results include: Table 7 Table 7

[0123] When selecting a requirement item (such as a requirement item path), define the requirement item as: 2. Vehicle basic attributes; Generation type: child.

[0124] The system data generation results include: Table 8 Table 8

[0125] (2) System data generation The main purpose of this step is to write the requirement data (i.e. requirement feedback information) generated by the large language model into the requirement management system, based on the path selected by the engineer.

[0126] The step can first acquire data through the backend, and perform data verification and cleaning operation on the acquired data, aiming to eliminate incorrect, incomplete or redundant data, and guarantee data quality. After verification and cleaning, the data processing is divided into two parallel paths: one is to construct a graph data model and write the processed data into a graph database Neo4j; the second is to write data into a relational database MySQL. Subsequently, for the data writing operation of the two paths, transaction consistency processing is performed to ensure the consistency and integrity of the data in the writing process of different databases. Finally, the operation result is returned, and the entire data processing process is completed.

[0127] In some embodiments, regarding the processing of user information, it should be noted that when the user sends a message, two special processes can be performed, including knowledge base configuration and message template application.

[0128] 1. Knowledge base configuration The purpose of the configuration of the knowledge base is to tell the AI the scope of the knowledge base to be searched, so as to generate corresponding demand items according to the desired knowledge information. The knowledge base configuration needs to configure the following information.

[0129] Knowledge base scope: set the knowledge base to be searched by AI.

[0130] Metadata: select metadata and define values. For example, if the metadata of the model name is selected and the value is defined as C919, it will be clear that the AI searches the documents in the corresponding knowledge base that include the metadata model name and have a value of C919.

[0131] 2. Message template application The function of the message template is to provide commonly used scene terms in the demand generation process, helping users to quickly reuse when inputting messages, and attaching message intent to each message, so as to help AI quickly understand user intent and better generate demand items. The intent of the message includes the following.

[0132] Demand arrangement: arranges the original demand to generate structured data demand, thereby helping the user input process.

[0133] Demand capture: a process of directly obtaining original demand from external sources such as stakeholders, market, and regulations.

[0134] Demand decomposition: a process of decomposing high-level demand into low-level executable demand.

[0135] Demand derivation: a process of deriving new demand by analysis.

[0136] In some embodiments, regarding the commonly used data processing in system engineering, it should be noted that according to the user message, AI has different processing logic for different intents, so as to generate demand items that meet the user's intent.

[0137] Requirement sorting: When the intention is requirement sorting, the AI analyzes and sorts according to the user's attachments, without making additional modifications, to generate requirements. The scenarios include, but are not limited to, the input of numbered requirement documents, allowing the AI to automatically generate structured data, thereby eliminating the need for manual input.

[0138] Requirement capture: When the intention is requirement capture, the AI analyzes and sorts according to the user's attachments and message content, without making additional extensions, to capture the content requirements for generation. The applicable scenarios include, but are not limited to, market survey questionnaires, meeting records, regulations and standards, and other input requirement capture.

[0139] Requirement decomposition: When the intention is requirement decomposition, the AI will decompose the top-level requirements based on the user's input, thereby decomposing the top-level requirements, retrieving structured domain knowledge base, or utilizing general large model capabilities to decompose the top-level requirements.

[0140] Requirement derivation: When the intention is requirement derivation, the AI analyzes the content of the user's input, retrieves structured domain knowledge base, or utilizes general large model capabilities to identify implicit requirements for generation, thereby filling in the gaps of the original requirements.

[0141] In some embodiments, regarding the output of structured data, it needs to be noted that when the AI generates requirement items, the output data format is limited to JSON format, and the output content meets the requirements of requirement item management. The content of the requirement items generated by the AI includes the following.

[0142] Parent node: Describes the parent of the current requirement item, which is the first level when the parent node of the requirement item is empty. The position of the requirement item is located through the parent node of the requirement and the generation order. Requirement title. Requirement text. Requirement attributes: The requirement item may contain multiple attributes, such as priority, verification method, and other requirement attributes.

[0143] For example, the content includes: 1. Provide a physical cooling expansion accessory to improve the overheating problem of the machine body during the game process; realize the function of the physical cooling expansion accessory to improve the overheating problem of the machine body during the game process. Priority: high.

[0144] In this embodiment, the large language model is deeply integrated with the requirement management system, and the automatically generated requirement content, such as requirement items, is continuously improved.

[0145] Corresponding to the foregoing application function implementation method embodiment, the present application also provides a requirement information feedback device, an electronic device, and corresponding embodiments.

[0146] Figure 3 is a structural schematic diagram of the requirement information feedback device according to an embodiment of the present application.

[0147] Referring to Figure 3 The requirement information feedback apparatus 300 of this embodiment comprises a determination module 310, an obtaining module 320, a processing module 330 and a first generation module 340.

[0148] The determination module 310 is configured to determine the user's intention according to the input information in response to receiving the input information of the user.

[0149] The obtaining module 320 is configured to obtain the domain-related knowledge information from the structured domain knowledge base according to the user's intention and the input information in a case where it is determined that the input information represents a knowledge question and answer related to the domain.

[0150] The processing module 330 is configured to process the input information into prompt word data based on the domain-related knowledge information.

[0151] The first generation module 340 is configured to generate the requirement feedback information for the user by means of a large model based on the prompt word data.

[0152] In some embodiments, the apparatus further comprises a second generation module configured to determine a target path according to the deployment instruction in response to receiving the deployment instruction of the user, and write the requirement feedback information into the system to generate a requirement item according to the target path.

[0153] As to the apparatus in the above-mentioned embodiments, the specific manners in which the respective modules perform operations have been described in detail in the embodiments related to the method, and thus will not be described in detail here.

[0154] According to an embodiment of the present application, any of the modules of the determining module 310, the obtaining module 320, the processing module 330 and the first generating module 340 can be combined in one module, or any of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of the modules can be combined with at least part of the functions of the other modules, and implemented in one module. According to an embodiment of the present application, at least one of the determining module 310, the obtaining module 320, the processing module 330 and the first generating module 340 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system on package, an application specific integrated circuit (ASIC), or any other reasonable manner of integrating or packaging a circuit, etc. or implemented by hardware or firmware, or implemented in any one of software, hardware and firmware or in a proper combination of any of them. Alternatively, at least one of the determining module 310, the obtaining module 320, the processing module 330 and the first generating module 340 can be at least partially implemented as a computer program module which can perform the corresponding functions when the computer program module is run.

[0155] Figure 4 FIG. 1 is a structural schematic diagram of an electronic device according to an embodiment of the present application.

[0156] Referring to FIG. 4, Figure 4 The electronic device 400 includes a memory 410 and a processor 420.

[0157] The processor 420 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0158] The memory 410 can include various types of storage units, such as a system memory, a read-only memory (ROM), and a permanent storage device. Among them, the ROM can store static data or instructions required by the processor 420 or other modules of the computer. The permanent storage device can be a read-write storage device. The permanent storage device can be a non-volatile storage device that does not lose stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a mass storage device (such as a magnetic or optical disk, a flash memory) as a permanent storage device. In some other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, an optical drive). The system memory can be a read-write storage device or a volatile read-write storage device, such as a dynamic random access memory. The system memory can store some or all of the instructions and data required by the processor during runtime. In addition, the memory 410 can include a combination of any computer readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), magnetic disks and / or optical disks. In some embodiments, the memory 410 can include a read and / or write removable storage device, such as a compact disc (CD), a read-only digital versatile disc (such as DVD-ROM, double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a min SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer readable storage medium does not include a carrier wave and an instantaneous electronic signal transmitted by wireless or wired transmission.

[0159] The memory 410 stores executable code, which, when processed by the processor 420, can cause the processor 420 to perform part or all of the above-mentioned methods.

[0160] In addition, the method according to the present application can also be implemented as a computer program or computer program product, which includes computer program code instructions for executing part or all of the steps of the above-mentioned methods of the present application.

[0161] Alternatively, the present application can also be implemented as a computer readable storage medium (or non-transitory machine readable storage medium or machine readable storage medium) having executable code (or computer program or computer instruction code) stored thereon, which, when executed by a processor of an electronic device (or server, etc.), causes the processor to execute part or all of the steps of the above-mentioned methods according to the present application.

[0162] Having described various embodiments of the application, it is to be understood that the above description is meant to be illustrative only, and that many modifications and variations of the embodiments described herein are possible. It is therefore to be understood that within the scope of the appended claims, and their equivalents, many alternatives to the embodiments described herein are possible. The selection of terms to be used in the description is not intended to limit the scope of the embodiments described herein, but rather to best explain the principles of the embodiments, practical application, or improvement over the technology in the art, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for feedback of demand information, characterized in that, include: In response to receiving user input, determine the user's intent based on the input; If the input information is determined to represent domain-related knowledge questions and answers, domain-related knowledge information is obtained from a structured domain knowledge base based on the user intent and the input information. Based on the relevant knowledge information in the aforementioned field, the input information is processed into prompt word data; Based on the aforementioned prompt word data, a large model is used to generate feedback information tailored to the user's needs.

2. The method according to claim 1, characterized in that, The method further includes: In response to receiving a deployment instruction from a user, the system determines a target path based on the deployment instruction and writes the requirement feedback information into the system based on the target path to generate a requirement entry.

3. The method according to claim 2, characterized in that, The target path includes at least one of the following: the requirement folder path, the requirement specification path, and the requirement item path; In response to receiving a deployment instruction from a user, the system determines a target path based on the deployment instruction, and writes the requirement feedback information into the system based on the target path to generate a requirement entry, including at least one of the following: In response to receiving a user's deployment instruction, if the target path is determined to include the requirement folder path according to the deployment instruction, the user-defined requirement specification name is obtained. When generating the requirement item, the requirement specification is generated in the folder corresponding to the requirement specification name, and the requirement item is written as the content of the requirement specification. In response to receiving a user's deployment instruction, and if it is determined from the deployment instruction that the target path includes the requirement specification path, when generating the requirement entry, the requirement entry is written as content above the requirement specification in the corresponding requirement specification. In response to receiving a deployment instruction from a user, and if the target path is determined to include the path of the requirement item based on the deployment instruction, the user-defined child or sibling level of the corresponding item is obtained. When generating the requirement item, the requirement feedback information is written into the child or sibling level of the corresponding item.

4. The method according to claim 1, characterized in that, The method further includes: Construct the structured domain knowledge base; wherein the construction method includes: Data preparation and structuring are performed to obtain the target data; Construct a vector index based on the target data; An integrated retrieval enhancement generation RAG model is used to perform multi-dimensional enhancement processing on the query results obtained through the vector index.

5. The method according to claim 1, characterized in that, The input information includes the current input information and historical dialogue information.

6. The method according to claim 5, characterized in that, The step of responding to receiving user input information and determining the user's intent based on the input information includes: In response to receiving user input, an intent tag is determined based on the current input and historical dialogue information.

7. A demand information feedback device, characterized in that, include: The determination module is used to determine the user's intent based on the received user input information in response to the input information. The acquisition module is used to obtain domain-related knowledge information from a structured domain knowledge base based on the user intent and the input information, provided that the input information represents a domain-related knowledge question and answer. The processing module is used to process the input information into prompt word data based on the domain-related knowledge information; The first generation module is used to generate user-specific feedback information based on the prompt word data and through a large model.

8. The apparatus according to claim 7, characterized in that, The device further includes: The second generation module is used to respond to the user's deployment instruction, determine the target path according to the deployment instruction, and write the requirement feedback information into the system according to the target path to generate requirement entries.

9. An electronic device, characterized in that, include: processor; as well as A memory having executable code stored thereon, which, when executed by the processor, causes the processor to perform the method as described in any one of claims 1-6.

10. A computer-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the method as described in any one of claims 1-6.

Images