Document generation method, knowledge base construction method and related products

Abstract

The invention provides a document generation method, a knowledge base construction method and a related product, and the document generation method comprises the steps: analyzing a to-be-analyzed business demand document, and obtaining document demand information; in a plurality of knowledge fragments of a target knowledge base, target fragments matched with the document demand information are determined based on the retrieval keywords and the retrieval dimensions, and technical information included in the knowledge fragments is stored according to a document hierarchical structure; and generating a target technical document corresponding to the document demand information in combination with the target fragment. According to the method, the document generation efficiency and quality can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, specifically to the field of document generation and knowledge base construction technology, and particularly to a document generation method, a knowledge base construction method, an intelligent agent, an electronic device, a readable storage medium, and a program product. Background Technology

[0002] In product development in fields such as automotive, electronics, and communications, the SOR (Statement of Requirements) document defines comprehensive requirements for product functionality, performance, quality, quantifiable metrics, and collaborative requirements. The TR (Technical Requirement) document, on the other hand, translates business requirements into a structured, implementable technical solution document. Currently, TR document generation heavily relies on manual processes. Technical personnel must manually read and analyze the SOR document, then rely on their understanding and memory of it to find reference solutions from scattered historical TR documents, integrating them to generate the TR document. This makes the conversion process from SOR to TR document cumbersome and time-consuming. When faced with complex content, key information is easily overlooked or misunderstandings arise, leading to inconsistent quality of the generated TR documents. Summary of the Invention

[0003] This disclosure provides a document generation method, a knowledge base construction method, an apparatus, an electronic device, a readable storage medium, and a program product.

[0004] In a first aspect, embodiments of this disclosure propose a document generation method, comprising: analyzing a business requirement document to be analyzed to obtain document requirement information; determining a target fragment that matches the document requirement information based on search keywords and search dimensions among multiple knowledge fragments in a target knowledge base, wherein the technical information included in the knowledge fragment is stored according to a document hierarchical structure; and generating a target technical document corresponding to the document requirement information by combining the target fragment.

[0005] Secondly, this disclosure proposes a knowledge base construction method, including: structurally splitting historical technical documents to obtain multiple knowledge fragments, wherein the technical information included in the knowledge fragments is stored according to the document hierarchical structure; and constructing a target knowledge base based on the multiple knowledge fragments.

[0006] Thirdly, this disclosure provides a document generation apparatus, comprising: an analysis module configured to analyze a business requirement document to be analyzed to obtain document requirement information; a determination module configured to determine a target fragment matching the document requirement information from multiple knowledge fragments in a target knowledge base based on search keywords and search dimensions, wherein the technical information included in the knowledge fragment is stored according to a document hierarchical structure; and a generation module configured to generate a target technical document corresponding to the document requirement information based on the target fragment.

[0007] Fourthly, this disclosure proposes a knowledge base construction apparatus, comprising: a splitting module configured to structurally split historical technical documents to obtain multiple knowledge fragments, wherein the technical information included in the knowledge fragments is stored according to the document hierarchical structure; and a construction module configured to construct a target knowledge base based on the multiple knowledge fragments.

[0008] Fifthly, embodiments of this disclosure propose an intelligent agent, comprising: an input module configured to receive input information; a processing module configured to determine a target task based on the input information received by the input module, determine a large model based on the target task, and obtain output information by calling the large model to execute a document generation method described in any implementation of the first aspect, or by calling the large model to execute a knowledge base construction method described in any implementation of the second aspect; and an output module configured to output the output information obtained by the processing module.

[0009] In a sixth aspect, embodiments of this disclosure provide an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to implement a document generation method as described in any implementation of the first aspect, or to enable the at least one processor to implement a knowledge base construction method as described in any implementation of the second aspect.

[0010] In a seventh aspect, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions, which, when executed by a computer, enable the computer to implement a document generation method as described in any implementation of the first aspect, and a knowledge base construction method as described in any implementation of the second aspect.

[0011] Eighthly, embodiments of this disclosure provide a computer program product including a computer program, which, when executed by a processor, can implement a document generation method as described in any implementation of the first aspect, and a knowledge base construction method as described in any implementation of the second aspect.

[0012] According to the document generation scheme provided in this disclosure, firstly, by analyzing the business requirement document, document requirement information is obtained, and then matching is performed in the constructed target knowledge base based on search keywords and search dimensions to obtain target fragments that match the document requirement information; finally, the target fragments are automatically combined to generate the target technical document corresponding to the document requirement information.

[0013] By using a target knowledge base, unstructured historical technical documents can be transformed into multiple knowledge fragments stored according to a hierarchical document structure. This solves the problem of disorganized and difficult-to-search historical technical documents, providing a precise database foundation for subsequent retrieval. Through search keywords and search dimensions, target fragments matching the document's requirements can be quickly filtered from the target knowledge base and combined to generate the target technical document. This transforms document generation from a time-consuming process relying on personal experience into an automated process, improving both the speed and accuracy of document generation.

[0014] According to the knowledge base construction scheme provided in the embodiments of this disclosure, historical technical documents are structurally split into knowledge fragments and technical information is stored hierarchically, making scattered historical data more organized and easy to retrieve. This allows the subsequent matching process to locate the target fragment directly without having to browse the complete document, reducing the cost of searching and filtering. At the same time, it ensures the standardization and consistency of knowledge fragments and solves the problems of low utilization and inconvenience in accessing traditional historical technical data.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0016] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is an exemplary system architecture to which this disclosure can be applied; Figure 2 A flowchart illustrating a document generation method provided in this embodiment of the disclosure; Figure 3 A flowchart for determining search keywords and search dimensions is provided for embodiments of this disclosure; Figure 4 A flowchart illustrating another document generation method provided in this disclosure embodiment; Figure 5 A flowchart illustrating a knowledge base construction method provided in this embodiment of the disclosure; Figure 6A flowchart illustrating another knowledge base construction method provided in this disclosure embodiment; Figure 7 A schematic block diagram of an intelligent agent provided in an embodiment of this disclosure; Figure 8 This is a schematic diagram of the structure of an electronic device suitable for performing a document generation method or a knowledge base construction method, provided in an embodiment of this disclosure. Detailed Implementation

[0017] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding; these should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0018] The document generation method, knowledge base construction method, and related products provided in this disclosure are mainly applied in the field of data processing technology, specifically involving the sub-fields of document generation and knowledge base construction. They are particularly suitable for complex product development scenarios such as automobiles, electronics, and communications, such as the technology requirement transformation stage of products like smart cockpits, vehicle control systems, and core components in the automotive electronics field. They can also be extended to various technical fields such as aerospace and industrial automation, where business requirement documents (e.g., SOR documents) need to be transformed into structured technical implementation plan documents (e.g., TR documents).

[0019] In the aforementioned application areas, the product development process requires a clear technical implementation path based on business requirement documents. The SOR (System Requirements Orientation) document, as the core document defining product functions, performance, quality, quantifiable metrics, and collaborative requirements, is crucial for product implementation when converted into TR (Transformation Requirements) documents. However, current technologies require technical personnel to manually interpret the complex requirements of SOR documents, then sift through scattered historical technical documents for reference and reorganization. This makes the document generation process cumbersome and time-consuming, and the analysis and reorganization process susceptible to individual experience and misunderstandings. Furthermore, converting multi-format documents requires manual, one-by-one operation, leading to version management chaos and a lack of effective progress tracking and feedback in collaborative projects, reducing the efficiency and reliability of document processing. In addition, historical technical documents are often stored in unstructured formats, lacking unified management, making it difficult to accurately locate key content during retrieval, and prone to information omissions or redundant filtering.

[0020] To address the aforementioned issues, this disclosure proposes a document generation method, a knowledge base construction method, and related products, aiming to transform the document generation process from a time-consuming manual operation relying on personal experience into an efficient, accurate, and reusable automated process, thereby improving work efficiency and document quality.

[0021] Figure 1 An exemplary system architecture 100 is shown, in which embodiments of the document generation method, knowledge base construction method and related products of this disclosure can be applied.

[0022] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0023] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various applications for enabling information communication between the terminal devices 101, 102, and 103 and server 105 can be installed. These applications include document generation applications, knowledge base building applications, cloud storage applications, and instant messaging applications.

[0024] Terminal devices 101, 102, and 103 and server 105 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices, and can be implemented as multiple software programs or software modules, or as a single software program or software module; no specific limitation is made here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules, or as a single software program or software module; no specific limitation is made here.

[0025] Server 105 can provide various services through its built-in applications. Taking a document generation application as an example, when running this application, server 105 can achieve the following: First, by analyzing the business requirement document received from the terminal, document requirement information is obtained. Based on this, a target inference model is used to determine search keywords and search dimensions. Then, based on the search keywords and search dimensions, matching is performed in the target knowledge base to filter out target fragments that match the document requirement information. Finally, combining the determined target fragments, the target technical document corresponding to the document requirement information is generated and returned to the terminal device.

[0026] Server 105 can provide various services through its built-in applications. For example, when running a knowledge base building application, Server 105 can achieve the following effects: By structurally splitting historical technical documents into knowledge fragments and storing technical information hierarchically, the scattered historical data becomes organized and easy to retrieve. This allows the subsequent matching process to locate the target fragment directly without having to browse the complete document, reducing the cost of searching and filtering. At the same time, it ensures the standardization and consistency of knowledge fragments and solves the problems of low utilization and inconvenience in accessing traditional historical technical data.

[0027] It should be noted that, in addition to being obtained from terminal devices 101, 102, and 103 via network 104, business requirement documents can also be pre-stored locally on server 105 through various means. Therefore, when server 105 detects that this data is already stored locally (for example, by directly analyzing the locally stored business requirement documents to obtain document requirement information), it can choose to retrieve this data directly from the local storage. In this case, the exemplary system architecture 100 may not include terminal devices 101, 102, and 103 and network 104.

[0028] Since large-model-based retrieval, document generation, and knowledge base construction require significant computing resources and capabilities, the document generation or knowledge base construction methods provided in subsequent embodiments of this disclosure are generally executed by server 105, which possesses strong computing power and abundant computing resources. However, it should also be noted that when terminal devices 101, 102, and 103 also possess sufficient computing power and resources, they can also perform the aforementioned calculations performed by server 105 through their installed document generation or knowledge base construction applications, thereby outputting the same results as server 105. Especially when multiple terminal devices with different computing capabilities exist simultaneously, but the document generation or knowledge base construction application determines that its terminal device possesses strong computing power and sufficient remaining computing resources, it can allow the terminal device to perform the aforementioned calculations, thereby appropriately reducing the computing pressure on server 105. In this case, the exemplary system architecture 100 may also exclude server 105 and network 104.

[0029] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0030] Please refer to Figure 2 , Figure 2 A flowchart of a document generation method provided in this disclosure embodiment, wherein process 200 includes the following steps: Step 201: Analyze the business requirement document to be analyzed to obtain document requirement information.

[0031] In this embodiment, the business requirements document to be analyzed can be a SOR document used in the product development field. For example, in the field of automotive product development, the SOR document can define in detail the functional requirements, performance standards, quality requirements, quantitative indicators, and collaborative requirements of automotive parts. This document can be temporarily stored locally or uploaded to a designated cloud server through the system interface for persistent storage, for subsequent analysis and processing.

[0032] The aforementioned functional requirements include, but are not limited to: the detection range that the vehicle radar must meet, and the deployment response time that the airbags must meet; the aforementioned performance standards include, but are not limited to: the seats must be able to withstand 100,000 opening and closing cycles without failure; the aforementioned quality requirements include, but are not limited to: the tensile strength that the vehicle body steel should achieve, and the flaw detection level that the welding should achieve; the aforementioned quantitative indicators include, but are not limited to: the dimensional tolerances of the parts, the upper limit of weight, and the power parameters; the aforementioned coordination requirements include, but are not limited to: the delivery cycle, packaging specifications, and after-sales service terms.

[0033] After receiving the SOR document uploaded by the user, it can be analyzed using natural language processing technology or predefined rule models to understand the business requirements of the SOR document. The business requirements of the SOR document are then organized and output according to the TR architecture to obtain document requirement information. Taking the automotive industry as an example, when the content of the SOR document to be analyzed is, for example, "A high-performance intelligent cockpit is needed, with a large and clear screen, a smart voice assistant, and a car that can be used in extreme weather," the document requirement information obtained from the analysis would be, for example, "The car needs to use a Qualcomm chip, the central control screen needs to be a 12.8-inch OLED with a resolution of 2560×1600 or higher, and the operating temperature needs to be between -50°C and 60°C, etc."

[0034] It should be noted that the document format of the SOR document can be PDF, Word, PPT or Excel, etc., and there is no limitation here.

[0035] It should be noted that the above description of business requirement documents, document requirement information, and analysis methods for business requirement documents is merely exemplary. The business requirement documents, document requirement information, and analysis methods for business requirement documents protected by this application are not limited to the contents listed above. Those skilled in the art can set and plan the business requirement documents, document requirement information, and analysis methods for business requirement documents according to the actual situation, as long as they can achieve the technical principles of this application.

[0036] Step 202: Among multiple knowledge fragments in the target knowledge base, the target fragment that matches the document requirement information is determined based on the search keywords and search dimensions. The technical information included in the knowledge fragment is stored according to the document hierarchical structure.

[0037] In this embodiment, within the automotive electronics field, TR documents can be used to transform the business requirements in SOR documents (e.g., smart cockpits need to support multi-screen interaction, vehicle power supplies need to be compatible with 12V voltage, etc.) into clear and actionable technical specifications (e.g., specific device models, interface specifications, etc.). The TR knowledge base, or target knowledge base, refers to a structured database specifically storing historical TR documents (i.e., historical technical documents) in the automotive electronics field. Its storage architecture can support multiple retrieval dimensions, such as a database implemented based on the FastAPI microservice architecture. Retrieval dimensions can include one or more of the following: document hierarchy, document content, document item attributes, and document version. The target knowledge base adopts a hybrid storage strategy of local temporary storage and cloud persistent storage to ensure the security and ease of access of data within the knowledge base.

[0038] The target knowledge base stores multiple knowledge fragments, each derived from the structured splitting and reorganization of historical TR documents. That is, each knowledge fragment is the smallest unit of information with independent semantics. The technical information included in each fragment consists of text content and page storage location information. The text content may include, but is not limited to, the main chapter name, sub-chapter names, recommended content information, and technical specifications. This technical information is stored according to a document hierarchical structure. For example, the first level could be the main chapter name, the second level could be the sub-chapter names under that main chapter, and other detailed content information, such as recommended content information, technical specifications, and page storage location information.

[0039] Search keywords can be core terms, technical feature words, functional keywords, security requirement identifiers, and other information that can be associated with document requirements. Search dimensions can be classification dimensions that can represent the relationship between knowledge fragments and document requirements, and their specific types can be flexibly adapted according to the requirement scenario and knowledge base storage characteristics. Search keywords and search dimensions can be preset, determined based on user input through the user interface, directly extracted from document requirements using predefined business rules, text patterns, or information extraction templates, or determined through inference models based on analysis of document requirements.

[0040] During retrieval, based on the search keywords, all knowledge fragments are filtered according to the search dimensions. The knowledge fragments that match the document's requirements are identified as target fragments. Taking document hierarchy as the search dimension as an example: among all knowledge fragments in the target knowledge base, firstly, a primary matching is performed by the main chapter name, filtering out knowledge fragments whose main chapter names match the search keywords. Then, for these filtered knowledge fragments, a secondary matching is performed by sub-chapter names and other detailed content information, filtering out knowledge fragments whose sub-chapter names and other detailed content information match the search keywords, thus obtaining the target fragments.

[0041] It should be noted that the above description of search keywords and search dimensions is merely exemplary. The search keywords and search dimensions protected by this application are not limited to those listed above. Those skilled in the art can set the determination method and quantity of search keywords and search dimensions according to the actual situation, as long as the technical principles of this application can be achieved.

[0042] In some optional implementations of the embodiments of this disclosure, the technical information includes: the main chapter name and page storage information; the page storage information includes: the project name of the historical technical document, the storage location information of the data group associated with the knowledge fragment, and the page number of the single-page document in the data group in the historical technical document; the data group includes: a single-page document in the historical technical document, and the original image and thumbnail image corresponding to the single-page document.

[0043] In this embodiment, technical information refers to a set of structured data in the target knowledge base used to characterize the core content and source of a knowledge fragment. Technical information includes at least the main chapter name and page storage information corresponding to the knowledge fragment.

[0044] The main chapter title refers to the top-level category heading after the historical TR documents are hierarchically divided. It can be used to summarize the technical topic or business area to which a knowledge segment belongs. For example, system hardware solutions, structural design solutions, functional safety and information security solutions, manufacturing solutions, etc.

[0045] Page storage information refers to the data information that identifies and locates the storage location and content of historical TR documents associated with the knowledge fragment. Specifically, it includes: the project name of the historical TR document, which can be used to identify the project or task to which the source document of the knowledge fragment belongs, making it easy to filter and trace by project, for example: ××× brand ××× chip cockpit platform technology proposal_20240705; the storage location information of the data group associated with the knowledge fragment, which can be used to point to the specific path, URI (Uniform Resource Identifier), or unique object key of the data group in the storage system (e.g., local file system or cloud server), for example: sor2tr_full / single_slides / ××× brand ××× chip cockpit platform technology proposal_20240705 / slide_002.pptx; and the page number of the single-page document in the data group in the historical technical document, which can be used to record the sequential position of the single-page document in the original historical TR document, for example: number 2 indicates that the single-page document comes from page 2 of the original historical TR document.

[0046] A data set refers to the smallest indivisible physical storage unit formed by splitting and processing historical TR documents. A data set contains: a single-page document, which is a single page file (e.g., slide_002.pptx) that is split from the original historical TR document and stored independently; the original image corresponding to the single-page document, which is a high-resolution, lossless or low-loss image file generated from the single-page document to ensure clear viewing of details; and the thumbnail image corresponding to the single-page document, which is a low-resolution, small-sized image file generated from the single-page document or the original image, used for quick previewing and improving interactive efficiency.

[0047] It should be noted that the original image and thumbnail image can be in PNG (Portable Network Graphics), JPEG (Joint Photographic Experts Group), SVG (Scalable Vector Graphics), etc. The formats of the original image and thumbnail image protected in this application are not limited to those listed above. Those skilled in the art can set the image format according to the actual situation, as long as it can realize the technical principle of this application.

[0048] By storing historical technical documents as independent data groups corresponding to individual pages, the problem of coarse granularity and difficulty in accurately locating specific content pages in traditional document management is solved. At the same time, by storing page information and main chapter names, links from knowledge fragments to their original sources are established, ensuring the accuracy and reliability of the content and providing a foundation for subsequent document retrieval and generation.

[0049] In some optional implementations of the embodiments of this disclosure, the technical information further includes at least one of: sub-chapter name, recommended content information, and technical specification information; and in the knowledge fragment, the sub-chapter name, recommended content information, technical specification information, and page storage information are stored as information of the next document hierarchy structure of the main chapter name.

[0050] In this embodiment, the technical information, in addition to the main chapter name and page storage information, also includes at least one of the following: Sub-chapter names, used for more refined content division and identification, such as a list of key components, component selection introduction, system product architecture diagram, structural dimension boundary data, air-cooled / water-cooled heat dissipation solution analysis, information security solution, site layout diagram, etc. Recommended content information, used to provide suggestive and guiding text content based on business or technical needs, such as specific suggestions based on business needs information, such as "Based on intelligent cockpit system requirements..." Technical specification information, used to provide the specific technical parameters and specifications associated with each page in the historical TR document, as well as the industry standards and key performance indicators (e.g., power consumption, temperature range, and interface type) that are conformed to.

[0051] In each knowledge segment, the sub-chapter name, recommended content information, technical specifications, and page storage information are stored as information in the next document hierarchy following the main chapter name.

[0052] By using a hierarchical document storage method, the problem of overly broad search results for single chapters is solved, ensuring that the retrieved target fragments are relevant to the document requirements and providing a reliable content foundation for generating accurate and complete target technical documents.

[0053] Step 203: Combine the target fragments to generate the target technical document corresponding to the document requirement information.

[0054] In this embodiment, based on one or more target fragments that have been retrieved and match the document requirement information, the single-page documents in the data group corresponding to the target fragments can be structurally reorganized and merged in a logical order to generate a target technical document that meets the document requirement information.

[0055] For example, based on document requirement information, two relevant target fragments have been matched and identified from the target knowledge base. The first target fragment A includes the following under its main chapter name: sub-chapter names, recommended content information, technical specifications, and page storage information; the second target fragment B includes the following under its main chapter name: sub-chapter names and page storage information.

[0056] First, the technical information stored in the target segments is read, including: main chapter name, sub-chapter name, recommended content information, technical specifications, and page storage information. Then, based on a predefined document template or business logic, the logical order of each target segment within the target technical document is determined. Following the storage location information indicated by the page storage information in each target segment, data group A corresponding to target segment A and data group B corresponding to target segment B are retrieved from the cloud server. Single-page documents are then extracted from each data group; for example, single-page document file 'a' from data group A and single-page document file 'b' from data group B. Finally, files 'a' and 'b' are combined according to the determined logical order to generate the target technical document.

[0057] The document generation method provided in this disclosure can obtain accurate business requirement information by analyzing business requirement documents. Relying on a target knowledge base that stores technical information according to document hierarchy, the method matches target fragments and generates target technical documents based on search keywords and search dimensions, thereby improving the efficiency and accuracy of document generation.

[0058] In some optional implementations of the embodiments of this disclosure, the method further includes determining search keywords and search dimensions based on document demand information using a target reasoning model.

[0059] In this embodiment, the target reasoning model is an artificial intelligence reasoning model with the capabilities of demand semantic understanding, core information mining, and deduction. Its specific model architecture and technical approach are not limited, only requiring it to fulfill the core function of outputting corresponding search keywords and search dimensions based on the input document demand information. In specific implementation, the target reasoning model can be obtained by configuring and logically constraining the initial reasoning model, enabling it to identify key information from the document demand information, transform it into search keywords that conform to the knowledge base retrieval rules, and simultaneously derive search dimensions that can characterize the matching logic between document demand information and knowledge fragments.

[0060] Specifically, the parsed document requirement information can be input into the target inference model. The model, through a deep understanding of the requirement semantics, combined with pre-set inference logic and domain-wide rules, extracts the core elements related to the subsequent generation of technical documents from the document requirement information and transforms these core elements into search keywords with search-oriented characteristics. At the same time, based on the matching logic between document requirement information and knowledge fragments, the model derives and determines the search dimensions that can achieve accurate matching.

[0061] It should be noted that the aforementioned target reasoning model can be a model with corresponding text reasoning ability obtained by training on any general model architecture. This general model architecture can include one or more combinations of models such as machine learning models, deep learning models, neural network models, and large models (including large language models that mainly process text data and multimodal large models that process multimodal data). The specific model structures of the various general model architectures mentioned above will not be elaborated here.

[0062] By automatically deriving search keywords and search dimensions through a target reasoning model, the consistency between search conditions and document requirements is ensured, and a search basis is provided for accurately locating target fragments in the target knowledge base based on search keywords and search dimensions.

[0063] Please refer to Figure 3 , Figure 3 A flowchart for determining search keywords and search dimensions is provided for embodiments of this disclosure, wherein process 300 includes the following steps: Step 301: Configure target parameters and constraint instructions for the initial inference model to obtain the target inference model. The target parameters include at least one of the following: model temperature coefficient, sampling range, output length, frequency penalty parameter, and existence penalty parameter.

[0064] In this embodiment, the initial inference model is a large language model (LLM) with natural language understanding and generation capabilities. This model has dialogue and generation capabilities. In order to make it more accurate in document generation scenarios, it needs to be configured in a targeted manner. Specifically, the key parameters called by the initial inference model can be set through configuration files, environment variables or code call interfaces to strictly control its inference behavior. The target parameters include: model temperature coefficient, which can be set to a low value (e.g., 0.0) to significantly reduce the randomness in model generation and ensure that the inference output remains highly consistent and deterministic for the same input; sampling range, which can be set to a small value (e.g., 0.1) to restrict the model to sample only from the candidate words with the highest probability when generating each word, thereby ensuring the focus and accuracy of the output content and avoiding irrelevant or divergent content; output length, which can be set to an upper limit value (e.g., 32768) to constrain the maximum length of the model's output text, prevent the generation of redundant or excessively long irrelevant information, and ensure that the output search keywords and search dimensions are concise and effective; frequency penalty parameter and existence penalty parameter, which can usually be set to 0.0 or a low value to avoid unnecessary repetition of words when the model generates search keywords and search dimensions, or an excessive tendency to mention certain common but irrelevant terms.

[0065] Furthermore, when invoking the inference model, explicit constraint instructions can be added to the input prompts. For example, the instruction could be something like, "Please analyze strictly based on the provided document requirements information, output only keywords and dimensions for knowledge base retrieval, do not make any improvisations, complete or add any explanatory text, ensure that all output comes directly from historical SOR documents, and clearly state if the information is insufficient," thus laying the foundation for accurate matching of the knowledge base in the future.

[0066] Step 302: Analyze the document requirement information using the target reasoning model to obtain search keywords and search dimensions.

[0067] In this embodiment, after obtaining document requirement information derived from business requirement document analysis (e.g., a TR document containing hardware solutions and functional safety policies for a certain chip platform), this information is used as input to the target inference model for analysis to obtain search keywords and search dimensions. Specifically, the target inference model identifies and extracts key terms representing specific technical entities, components, standards, or core requirements from the document requirement information. For example, for the aforementioned requirements, the target inference model may parse search keywords including: chip platform, chip type, technical field, solution type, and supplier. Simultaneously, based on the structure and semantics of the document requirement information, the target inference model determines which method should be used for retrieval in the target knowledge base, i.e., it determines the search dimensions, such as searching by document hierarchy and / or document content.

[0068] By configuring specific parameters and adding strict constraint instructions to the initial inference model, the target inference model can only perform deterministic, text-based inference when analyzing document requirement information. This solves the problem that large language models are prone to arbitrary interpretation and generating irrelevant or incorrect keywords when processing technical requirements, improves the accuracy and reliability of search keywords and search dimensions, and provides high-quality input for accurate matching of subsequent knowledge bases.

[0069] In some optional implementations of the embodiments of this disclosure, determining the target segment that matches the document requirement information based on search keywords and search dimensions among multiple knowledge segments in the target knowledge base includes: searching for search keywords according to search dimensions among multiple knowledge segments in the target knowledge base and determining the target segment based on the search results.

[0070] In this embodiment, the retrieval dimensions may include: document hierarchy dimension and / or document content dimension, etc. The document hierarchy dimension refers to the arrangement and organization of the technical information included in the knowledge fragment according to main chapter names, sub-chapter names, recommended content information, technical specifications information, page storage information, etc., where sub-chapter names, recommended content information, technical specifications information, and page storage information all belong to the document hierarchy structure one level below the main chapter name. During retrieval, filtering is first performed based on the search keywords under the main chapter name, and then filtering is performed on the sub-chapter names, recommended content information, technical specifications information, page storage information, etc. The document content dimension refers to the ability to filter one or more of the technical information included in the knowledge fragment based on search keywords during retrieval.

[0071] Taking document hierarchy as an example for retrieval: In the target knowledge base, first traverse the main chapter names of all knowledge fragments, filter out the fragments that match the search keywords, and then match the sub-chapter names, recommended content information, technical specifications information, page storage information, etc. of the filtered fragments that match the search keywords, and determine the successfully matched fragments as target fragments.

[0072] Taking document hierarchy as an example for retrieval: In the target knowledge base, based on the search keywords, only the main chapter names of knowledge fragments are filtered, and fragments matching the search keywords are identified as target fragments. Alternatively, based on the search keywords, only the main chapter names and recommended content information of knowledge fragments are filtered, and fragments matching the search keywords are identified as target fragments. Or, based on the search keywords, only the recommended content information, technical specifications, and page storage information of knowledge fragments are filtered, and fragments matching the search keywords are identified as target fragments. In other words, based on the search keywords, one or more of the technical information included in the knowledge fragment can be filtered, and fragments matching the search keywords are identified as target fragments.

[0073] By using search keywords and search dimensions, targeted retrieval of multiple knowledge fragments within the target knowledge base can be achieved. This allows for the rapid filtering of target fragments that meet the document's information requirements, improving the accuracy and efficiency of the retrieval and reducing interference from irrelevant information. This lays the foundation for generating high-quality target technical documents in the future.

[0074] In some optional implementations of the embodiments of this disclosure, searching for search keywords according to search dimensions and determining target segments based on search results among multiple knowledge segments in the target knowledge base includes: searching for a first search keyword among the search keywords in the main chapter names of multiple knowledge segments in the target knowledge base based on search dimensions, where search dimensions include document level dimension and document content dimension, and determining the knowledge segments for which the first search keyword is found as candidate segments, wherein the document content dimension includes at least one of the following: recommended content dimension, technical specification dimension, and page storage information dimension; searching for a second search keyword among the search keywords in the sub-chapter names of the candidate segments, and determining the candidate segments for which the second search keyword is found as target segments.

[0075] In this embodiment, document requirement information (e.g., developing a hardware solution for a smart cockpit of a new energy vehicle, determining key components, interface configurations, and functional safety requirements) is input into the target inference model. After analysis by the model, one or more search keywords are output. Taking the document level dimension as an example, when there are multiple search keywords, the target inference model will, based on the document level dimension search logic, divide the multiple search keywords into first search keywords and second search keywords according to the correspondence between main chapter association and sub-chapter association. The first search keyword refers to the keyword that corresponds to the main chapter name in the knowledge fragment, and its core function is to determine the business domain or technical category to which the document requirement information belongs. The second search keyword refers to the keyword that corresponds to the sub-chapter name in the knowledge fragment, and its core function is to accurately locate the subdivided technical direction or specific functional module corresponding to the document requirement information within the business domain or technical category determined by the first search keyword.

[0076] During the search, all knowledge fragments in the target knowledge base are traversed. Knowledge fragments containing the first search keyword are searched by main chapter name, and these fragments are identified as candidate fragments. Within each candidate fragment, fragments containing the second search keyword are searched by sub-chapter name, and these fragments are identified as target fragments.

[0077] For example, search keywords include, but are not limited to, intelligent cockpit system hardware solutions, intelligent cockpit functional safety solutions, intelligent cockpit key component lists, intelligent cockpit interface configuration specifications, and intelligent cockpit functional safety level requirements. The first search keyword for primary matching could be, for example, intelligent cockpit system hardware solutions and intelligent cockpit functional safety solutions. The second search keyword for exact matching could be, for example, intelligent cockpit key component lists, intelligent cockpit interface configuration specifications, and intelligent cockpit functional safety level requirements. First, all knowledge fragments in the target knowledge base are traversed, and fragments containing "intelligent cockpit system hardware solutions" and "intelligent cockpit functional safety solutions" are searched by main chapter name. Knowledge fragments containing the aforementioned first search keywords are identified as candidate fragments. Then, within each candidate fragment, fragments containing "key component models," "intelligent cockpit key component lists," and "intelligent cockpit interface configuration specifications" are searched by sub-chapter name. Knowledge fragments containing the aforementioned second search keywords are identified as target fragments.

[0078] Taking the document content dimension as an example, the document content dimension includes at least one of the following: recommended content dimension, technical specifications dimension, and page storage information dimension.

[0079] The aforementioned recommended content dimension refers to the filtering and matching of recommended content information from multiple knowledge fragments based on search keywords during retrieval. This recommended content information stores suggestive and guiding texts based on historical project experience and addressing specific technical issues. For example, if the search keywords include "based on intelligent cockpit system requirements," then under this dimension, the target knowledge base is traversed to search for recommended content information within knowledge fragments. Knowledge fragments containing the same or similar expressions (e.g., a knowledge fragment containing text such as "Based on intelligent cockpit system requirements, it is recommended that the main chip adopt a high-performance platform...") are identified as target fragments.

[0080] The aforementioned technical specifications dimension refers to the process of filtering and matching technical specifications information from multiple knowledge fragments based on search keywords during retrieval. Technical specifications information stores explicit technical parameters, performance indicators, industry standards, or device models, among other hard requirements. For example, when document requirements involve specific chip selection or safety standards, keywords such as "SA×××P" or "AEC-Q×××" might be used. Under this dimension, the target knowledge base is traversed to search for technical specifications information within knowledge fragments. Knowledge fragments containing the same or similar expressions (e.g., a knowledge fragment containing text such as "Specific Model: ××× Brand SA×××P" or "Device Certification Standard: AEC-Q×××") are identified as target fragments.

[0081] The aforementioned page storage information dimension refers to the filtering and matching of page storage information (including: project names of historical technical documents, storage location information of data groups associated with the knowledge fragment, and page numbers of single-page documents in the historical technical documents) across multiple knowledge fragments based on search keywords during retrieval. That is, it does not directly match the main text content of historical technical documents, but rather matches the storage attributes and source context information of associated historical technical documents. For example, when prioritizing historical data from a specific car model or project, the project name can be used for filtering and matching. Using "XXX brand XXX chip cockpit platform technology proposal_20240705" as the search keyword, multiple knowledge fragments can be directly matched based on the project name in the page storage information to filter out all knowledge fragments originating from that specific project and identify them as the target fragment.

[0082] It should be noted that the first keyword and the second keyword in the search keywords can be the same or different, and there is no limitation here. Those skilled in the art can set the search keywords according to the actual situation, as long as they can achieve the technical principles of this application.

[0083] It should be noted that searches can be performed based solely on the document hierarchy dimension, solely on the document content dimension, or both the document content dimension and the document hierarchy dimension, or one or more of the document content dimensions. No limitation is made here. Those skilled in the art can set the search dimensions according to the actual situation, as long as the technical principles of this application can be achieved.

[0084] By utilizing document hierarchy and document content dimensions, targeted filtering of knowledge fragments is achieved, avoiding interference from irrelevant fragments. This allows retrieval to delve into specific technical details, solving the problems of coarse results and inability to accurately locate specific parameters and historical materials when searching with a single structure. This improves the accuracy of target fragment matching and retrieval efficiency.

[0085] In some optional implementations of the embodiments of this disclosure, generating a target technical document corresponding to the document requirement information in conjunction with the target fragment includes: obtaining a data group associated with the target fragment based on the storage location information in the page storage information included in the target fragment; and generating a target technical document corresponding to the document requirement information based on the data group associated with the target fragment.

[0086] In this embodiment, after the target fragment is determined, a data group associated with the target fragment can be downloaded from a local or cloud server based on the storage location information in the page storage information. This data group typically includes a single-page document under that path, its high-resolution original image, and a low-resolution thumbnail image. The single-page documents (or their original images as needed) in the data group are merged in a certain logical order to obtain the target technical document.

[0087] By storing information on the target fragment's pages, the corresponding data group can be accurately obtained. By integrating the single-page documents in the data group, the target technical document can be generated, eliminating the tedious steps of manually searching and filtering historical technical documents, reducing information omissions and integration errors, and lowering the time cost and operational difficulty of document generation.

[0088] In some optional implementations of the embodiments of this disclosure, generating a target technical document corresponding to the document requirement information based on the data group associated with the target fragment includes: when there are multiple target fragments, displaying fragment information of multiple target fragments, wherein the fragment information includes: thumbnail images in the data group associated with the target fragments; determining the arrangement order of each target fragment based on the adjustment operation of the display position of the multiple fragment information; and integrating the single-page documents in the data group associated with the target fragments based on the arrangement order to generate the target technical document.

[0089] In this embodiment, after matching multiple (e.g., three) target fragments from the target knowledge base, fragment information of these fragments can be displayed to the user on a graphical user interface. The fragment information of each target fragment includes a thumbnail image from the associated data group. This thumbnail is a pre-generated low-resolution (e.g., 200×200 pixels) preview image that loads quickly, allowing the user to intuitively and quickly identify the page content represented by each target fragment, thus confirming the accuracy of the matched target fragments. If the fragment information confirms the accuracy of the target fragments, their order in the target technical document can be rearranged by adjusting the display position of the multiple fragments (e.g., by dragging the fragment information on the interface with the mouse); or their order in the target technical document can be determined based on the logical order and contextual relationships of the target fragments in the original historical technical document. Once the arrangement order is determined, the single-page document corresponding to each target segment can be downloaded from the local or cloud server based on the storage location information in the page storage information. Then, multiple single-page files can be integrated according to the above arrangement order through relevant functional components (such as Win32COM component, openpyxl component, xlwings component, etc.) to generate a TR document, i.e., the target technical document, with coherent chapters and an order that conforms to the design intent.

[0090] By displaying thumbnails of target segments and supporting adjustments to their order, users can intuitively filter and sort as needed, avoiding interference from irrelevant segments. At the same time, by integrating single-page documents in sequence, repetitive manual sorting and arrangement operations are reduced, thereby lowering the time cost and error probability of document generation.

[0091] In some optional implementations of the embodiments of this disclosure, the fragment information further includes: technical information included in the target fragment and the original image in the data group associated with the target fragment.

[0092] In this embodiment, when multiple retrieved target fragments are displayed to the user on the graphical user interface, the fragment information of each target fragment includes not only a thumbnail image for quick identification, but also technical information. The technical information can display a summary of the target fragment's content, such as the main chapter name, sub-chapter name, recommended content information, technical specifications information, and page storage information, so that the user can quickly understand the core content of the target fragment without opening any historical technical files.

[0093] In addition to thumbnail images used for quick identification, fragment information also includes high-resolution original images (e.g., 400×400 pixels). The original images are high-fidelity visual renderings generated from single-page documents. On the graphical user interface, users can view them by hovering to zoom in for a preview or clicking to view a larger image, allowing for pixel-level confirmation of details in the target fragment.

[0094] By providing users with both the technical information of the target segment and a high-resolution original image, users can not only quickly understand the key points of the content, but also accurately confirm the details of the target segment by viewing the original image, thus ensuring the accuracy of the final generated document.

[0095] For a deeper understanding, please refer to Figure 4 , Figure 4 A flowchart of another document generation method provided in this disclosure embodiment, wherein process 400 includes the following steps: Step 401: Start executing the document generation method.

[0096] Step 402: After receiving the SOR document uploaded by the user, it can be analyzed using natural language processing technology or predefined rule models to understand the business requirements of the SOR document and obtain the TR architecture analysis results, i.e., document requirement information.

[0097] Step 403: After obtaining the TR architecture analysis results, they can be input into the target reasoning model. The model performs semantic understanding on the TR architecture analysis results, combines the preset reasoning logic and domain-general rules, and extracts the core elements related to the subsequent technical document generation. These core elements are then transformed into search keywords with search-oriented characteristics. At the same time, based on the matching logic between the TR architecture analysis results and knowledge fragments, the model derives and determines the search dimensions that can achieve accurate matching.

[0098] Step 404: In the TR knowledge base, based on the search keywords, search all knowledge fragments in the knowledge base according to the search dimensions, and filter the knowledge base content that matches the TR architecture analysis results.

[0099] Step 405: Determine whether knowledge base content matching the TR architecture analysis results was filtered out in step 404. If no matching content is found, proceed to step 406; if matching content is found, proceed to step 408.

[0100] Step 406: Terminate the process, or return to step 403 to optimize search conditions such as search keywords and search dimensions.

[0101] Step 407: Display the content summaries, single-page documents, original images, and thumbnail images corresponding to the selected knowledge base content (i.e., target fragments) that match the TR architecture analysis results to the user through a graphical user interface to confirm the accuracy of the content. Simultaneously, the logical order of the content in the final generated TR document can be rearranged by adjusting its display position (e.g., by dragging fragment information on the interface with the mouse).

[0102] Step 408: Submit the content and logical order that were finally displayed in Step 407 through the graphical user interface on the graphical user interface.

[0103] Step 409: If, in step 408, in response to the detection that the knowledge base content matched by the user-submitted representation is accurate, the corresponding single-page document can be downloaded from the local or cloud server based on the storage location information, and then merged according to the logical order confirmed by the user or its logical order in historical TR documents and contextual relationships through relevant functional components to generate the final TR document.

[0104] Step 410: After generating the final TR document, end the document generation method.

[0105] Please refer to Figure 5 , Figure 5 A flowchart of a knowledge base construction method provided in this disclosure embodiment, wherein process 500 includes the following steps: Step 501: Structure the historical technical documents to obtain multiple knowledge fragments, wherein the technical information included in the knowledge fragments is stored according to the document hierarchical structure.

[0106] In this embodiment, the uploaded historical TR documents, i.e. historical technical documents, are paginated to obtain multiple single-page documents, and each single-page document contains a knowledge fragment with independent semantics and structure. The technical information contained in each knowledge fragment (e.g., descriptions, parameters, associated historical technical documents, etc. under a certain technical topic) is stored according to the document hierarchy structure. For the same parts, please refer to the corresponding parts of the above embodiment, which will not be repeated here.

[0107] Step 502: Construct the target knowledge base based on multiple knowledge fragments.

[0108] In this embodiment, after structurally splitting a batch of historical technical documents (e.g., TR proposal PPTs for multiple different vehicle models), multiple structured knowledge fragments are obtained, each containing its own technical information. When constructing the target knowledge base, these knowledge fragments can be temporarily stored locally in a unified format (e.g., JSON documents or database records). To save local storage space, these knowledge fragments can also be stored on a cloud server (e.g., a database or distributed file system) in a unified format, and each knowledge fragment can be assigned an identifier to achieve persistent and standardized data management. For example, indexes can be created for main chapter names, sub-chapter names, recommended content information, technical specifications, and page storage information to facilitate filtering and matching by document hierarchy or document content.

[0109] By structurally breaking down historical technical documents into knowledge fragments and storing technical information hierarchically, scattered historical data becomes organized and easy to retrieve. This allows subsequent matching processes to locate target fragments directly without having to browse through the entire document, reducing the cost of searching and filtering. At the same time, it ensures the standardization and consistency of knowledge fragments and solves the problems of low utilization and inconvenience in accessing traditional historical technical data.

[0110] In some optional implementations of the embodiments of this disclosure, the technical information includes text content information and page storage information; the historical technical documents are structurally split to obtain multiple knowledge fragments, including: paginating the historical technical documents to obtain multiple single-page documents; extracting content from the original images of the multiple single-page documents using a multimodal large model to obtain the text content information of each single-page document; for each single-page document among the multiple single-page documents, storing the single-page document and its corresponding original image and thumbnail image as a data group and generating the page storage information of the single-page document, wherein the page storage information includes the storage location information of the data group; and combining the text content information and page storage information of each single-page document according to the document hierarchy structure rules to obtain multiple knowledge fragments.

[0111] In this embodiment, taking a historical technical document named "XXX Brand XXX Chip Cockpit Platform Technical Proposal_20240705.pptx" as an example: This PPT document is paginated to obtain multiple single-page documents (e.g., `slide_001.pptx`, `slide_002.pptx`, etc.), each single-page document being an independent file unit corresponding to a knowledge fragment. For each single-page document, the multi-threaded mechanism of Multimodal Large Model (MLM) is used to concurrently perform visual and text analysis on its page image (i.e., the original image), identifying the title, body text, lists, chart labels, etc., and understanding the semantic association between page layout, charts, and text, extracting the text content information of the single-page document. For each split single-page document, the corresponding original image and thumbnail image are generated, and the single-page document, original image, and thumbnail image are packaged as a logical data group and uploaded to a cloud server for persistent storage. Simultaneously, the page storage information for this data group is generated, which includes: the source project name (e.g., ××× brand ××× chip cockpit platform technical proposal_20240705), the storage location information of the data group in the cloud server, and the page number of this single-page document in the original historical technical document.

[0112] Each single-page document is not an isolated fragment; its corresponding text content and page storage information combine to form technical information. Following the original internal logic of historical technical documents, this information is systematically combined according to the document hierarchy to obtain the knowledge fragment corresponding to that single-page document. For example, several consecutive pages of single-page documents introducing hardware might be grouped under a unified main chapter title (e.g., System Hardware Solution). Under this main chapter, further subdivisions could be made into sub-chapter titles (e.g., Key Component List), and detailed descriptions extracted from the corresponding pages could be used as recommended content information or technical specifications. In the knowledge fragment data structure, the main chapter title serves as the first-level structure of the document hierarchy, under which are associated sub-chapter titles, recommended content information, technical specifications, and page storage information pointing to specific pages.

[0113] It should be noted that the aforementioned multimodal large model can be a model with visual and text analysis capabilities obtained by training on any general model architecture. This general model architecture can include one or more combinations of models such as machine learning models, deep learning models, neural network models, and large language models that mainly process text data. The specific model structures of the various general model architectures mentioned above will not be elaborated here.

[0114] By breaking down historical technical documents into structured knowledge fragments, the inefficiency and difficulty in ensuring the accuracy of content extraction caused by manual document splitting are solved. A multimodal large model accurately extracts text content from the original image, replacing error-prone manual transcription. By packaging single-page documents, original images, and thumbnail images into data groups and generating storage information, it is ensured that each knowledge fragment can be traced back to a complete and usable original historical technical document, thereby guaranteeing the accuracy of the target knowledge base.

[0115] In some optional implementations of this disclosure, the page storage information further includes: the project name of the historical technical document corresponding to the data group and the page number of the single-page document in the historical technical document. The text content information includes the main chapter name corresponding to the single-page document and at least one of the following: sub-chapter name, recommended content information, and technical specification information; and in the knowledge fragment, the sub-chapter name, recommended content information, technical specification information, and page storage information are stored as information of the next document hierarchy structure after the main chapter name.

[0116] In this embodiment, please refer to the corresponding parts of the above embodiment for the same content, and it will not be repeated here.

[0117] For a deeper understanding, please refer to Figure 6 , Figure 6 A flowchart of another knowledge base construction method provided in this disclosure embodiment, wherein process 600 includes the following steps: Step 601: Begin executing the knowledge base construction method.

[0118] Step 602: Create a TR knowledge base.

[0119] Step 603: Upload historical TR documents, i.e., upload historical technical documents. The TR knowledge base is generated based on the uploaded historical TR documents.

[0120] Step 604: The uploaded historical TR documents, i.e. historical technical documents, can be automatically paginated using relevant functional components to generate multiple single-page documents, as well as the original image and thumbnail image corresponding to each single-page document. Each single-page document is an independent file unit, corresponding to a knowledge fragment.

[0121] Step 605: Package the single-page document, the original image, and the thumbnail image into a logical data group and upload it to the cloud server for persistent storage.

[0122] Step 606: For each single-page document, use the multi-threading mechanism of Multimodal Large Model (MLM) to perform visual and text analysis on its page image (i.e., the original image) concurrently to identify content information such as titles, body text, lists, and chart labels on the page.

[0123] Step 607: Organize the recognition results from Step 606 and input them into the large model. Through the large model, understand the semantic relationship between the page layout, charts and text, and generate complete, coherent and logical text content information corresponding to the single-page document.

[0124] Step 608: Combine the text content information corresponding to each single-page document with the page storage information to form technical information. According to the original internal logic of the historical TR documents, organize the combination in a hierarchical structure to obtain the knowledge fragment corresponding to the single-page document, and write it into the TR knowledge base created above.

[0125] Step 609: The TR knowledge base construction is complete, ending the knowledge base construction method.

[0126] This disclosure also provides an intelligent agent. For example... Figure 7 As shown, the intelligent agent 700 includes: an input module 701, a processing module 702, and an output module 703.

[0127] The input module 701 is configured to receive input information. The processing module 702 is configured to determine a target task based on the input information received by the input module 701, determine a large model based on the target task, and execute the methods mentioned in the above embodiments by calling the large model to obtain output information. The output module 703 is configured to output the output information obtained by the processing module 702.

[0128] In the embodiments of this disclosure, the above-mentioned large model can be used to call the target reasoning model, multimodal large model, etc. mentioned in the above embodiments, and use the called corresponding model to execute the schemes related to document generation or knowledge base construction in the methods mentioned in the above embodiments to obtain the corresponding output results, which will not be repeated here.

[0129] According to embodiments of this disclosure, this disclosure also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to implement the document generation method described in any of the above embodiments, or enable the at least one processor to implement the knowledge base construction method described in any of the above embodiments.

[0130] According to embodiments of this disclosure, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions that enable a computer to implement the document generation method described in any of the above embodiments when executed, or that enable a computer to implement the knowledge base construction method described in any of the above embodiments when executed.

[0131] For example, the computer instructions corresponding to a document generation method or knowledge base construction method in this disclosure embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the computer instructions corresponding to a document generation method or knowledge base construction method in the storage medium are read or executed by a computer, the document generation method or knowledge base construction method as described in any of the above embodiments can be implemented.

[0132] According to embodiments of this disclosure, this disclosure also provides a computer program product including a computer program, which, when executed by a processor, can implement the document generation method described in any of the above embodiments, or the knowledge base construction method described in any of the above embodiments, when executed by a processor.

[0133] Figure 8A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device 800 proposed in embodiments of the present disclosure includes a processor 801 and a memory 802 storing an executable computer program. The processor 801, when executing the executable computer program stored in the memory 802, implements the document generation method or knowledge base construction method provided in embodiments of the present disclosure.

[0134] In some optional implementations of the embodiments of this disclosure, the electronic device 800 may further include a communication interface 803 and a bus 804 for connecting the processor 801, the memory 802 and the communication interface 803.

[0135] In some optional implementations of the embodiments of this disclosure, bus 804 is used to connect communication interface 803, processor 801 and memory 802 to realize mutual communication between these devices.

[0136] In some optional implementations of the embodiments of this disclosure, the processor 801 may be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that, for different devices, the electronic device used to implement the functions of the processor 801 may also be other types, and the embodiments of this disclosure do not specifically limit its use.

[0137] The aforementioned memory 802 is used to store executable computer programs and data. The executable computer program includes computer operation instructions. Memory 802 may include high-speed RAM and may also include non-volatile memory, such as at least two disk drives. In practical applications, the aforementioned memory 802 can be volatile memory, such as Random-Access Memory (RAM); or non-volatile memory, such as Read-Only Memory (ROM), flash memory, Hard Disk Drive (HDD), or Solid-State Drive (SSD); or a combination of the above types of memory, and provides executable computer programs and data to the processor.

[0138] Furthermore, the functional modules in this embodiment can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.

[0139] If the integrated units described above are implemented as software functional modules and not sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this disclosure embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] According to the technical solution of this disclosure, by analyzing business requirement documents, accurate business requirement information can be obtained. Relying on a target knowledge base that stores technical information according to a hierarchical document structure, target fragments are matched and target technical documents are generated based on search keywords and search dimensions, improving document generation efficiency and accuracy. Furthermore, by structurally splitting historical technical documents into knowledge fragments and storing technical information hierarchically, scattered historical data becomes organized and easily searchable. This allows subsequent matching processes to directly locate target fragments without needing to peruse the entire document, reducing search and filtering costs. Simultaneously, it ensures the standardization and consistency of knowledge fragments and solves the problems of low utilization and inconvenient access to traditional historical technical data.

[0141] By structurally breaking down historical technical documents into knowledge fragments and storing technical information hierarchically, scattered historical data becomes organized and easy to retrieve. This allows subsequent matching processes to locate target fragments directly without having to browse through the entire document, reducing the cost of searching and filtering. At the same time, it ensures the standardization and consistency of knowledge fragments and solves the problems of low utilization and inconvenience in accessing traditional historical technical data.

[0142] It should be understood that if this disclosure references any user data and personal information (including but not limited to device information, behavioral data, location information, etc.) and before applying the technical solutions described in the embodiments of this disclosure, the relevant products or services should comply with the laws and regulations concerning the protection of user data and personal information, strictly process users' personal information and data in accordance with the provisions of applicable laws and regulations throughout the entire data processing lifecycle, follow the principles of legality, legitimacy, necessity, good faith, openness, and transparency, and adopt reasonable privacy design schemes and technical measures to ensure the security of user data and personal information, protect users' legitimate rights and interests, and prevent the risks of leakage, theft, or tampering of user data and personal information.

[0143] Specifically, the company must publish and display its privacy policy in a prominent position on the user interface, clearly informing users of the types, purposes, uses, and methods of processing personal information, as well as other matters that should be disclosed as required by laws and regulations; obtain users' prior informed consent or explicit authorization regarding data processing through user-initiated interaction (such as confirmation pop-ups); process or store user data securely within the legally required timeframe; adopt a series of security technologies and management measures, including but not limited to data encryption and access control; share and transfer user data within the scope permitted by law and in a legally required manner; and process user rights, including the rights to query, access, correct, delete, withdraw authorization and consent, cancel registration, and obtain copies of personal information, within the legally required timeframe.

[0144] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0145] It should also be understood that expressions such as "comprising," "including," "having," "containing," and / or "comprising" are open-ended rather than closed-ended expressions in this disclosure, indicating the presence of the stated features, elements, and / or components, but not excluding the presence of one or more other features, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features, not just individual elements in the list. Additionally, when describing embodiments of this disclosure, the word "may" is used to mean "one or more embodiments of this disclosure." And the term "exemplary" is intended to refer to an example or illustration.

[0146] Unless otherwise specified, all terms used herein (including engineering and technical terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It should also be understood that, unless expressly stated in this disclosure, terms as defined in common dictionaries shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or overly formalized meaning.

[0147] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A document generation method, comprising: Analyze the business requirement documents to be analyzed to obtain document requirement information; Among multiple knowledge fragments in the target knowledge base, the target fragment that matches the document requirement information is determined based on search keywords and search dimensions, wherein the technical information included in the knowledge fragment is stored according to the document hierarchical structure; Based on the target fragment, a target technical document corresponding to the document requirement information is generated.

2. The method according to claim 1, wherein, The technical information includes: the main chapter name and page storage information; The page storage information includes: the project name of the historical technical document, the storage location information of the data group associated with the knowledge fragment, and the page number of the single-page document in the data group in the historical technical document; The data set includes: a single-page document from the historical technical documents, and the original image and thumbnail image corresponding to the single-page document.

3. The method according to claim 2, wherein, The technical information also includes at least one of the following: sub-chapter name, recommended content information, and technical specifications; and In the knowledge fragment, the sub-chapter name, the recommended content information, the technical specifications information, and the page storage information are stored as information of the next document hierarchy structure of the main chapter name.

4. The method according to claim 2 or 3, wherein, The method further includes: Based on the document requirement information, the target reasoning model determines the search keywords and the search dimensions; and The step of identifying the target fragment that matches the document requirement information based on search keywords and search dimensions from multiple knowledge fragments in the target knowledge base includes: Among the multiple knowledge fragments in the target knowledge base, the search keywords are searched according to the search dimensions, and the target fragment is determined based on the search results.

5. The method according to claim 4, wherein, The step of determining the search keywords and search dimensions based on the document demand information using the target reasoning model includes: Configure target parameters and constraint instructions on the initial inference model to obtain the target inference model, wherein the target parameters include at least one of the following: model temperature coefficient, sampling range, output length, frequency penalty parameter and existence penalty parameter; The document requirement information is analyzed and reasoned using the target reasoning model to obtain the search keywords and search dimensions.

6. The method according to claim 4, wherein, The step of searching for the search keywords according to the search dimension among multiple knowledge fragments in the target knowledge base and determining the target fragment based on the search results includes: The first search keyword is searched for in the main chapter names of multiple knowledge fragments in the target knowledge base based on search dimensions. The search dimensions include document level dimension and document content dimension. The knowledge fragments in which the first search keyword is found are determined as candidate fragments. The document content dimension includes at least one of the following: recommended content dimension, technical specification dimension, and page storage information dimension. Search for the second search keyword among the search keywords in the sub-chapter names of the candidate segments, and determine the candidate segments for which the second search keyword is found as the target segments.

7. The method according to claim 2 or 3, wherein, The step of combining the target fragment to generate the target technical document corresponding to the document requirement information includes: Based on the storage location information in the page storage information included in the target segment, obtain the data group associated with the target segment; Based on the data group associated with the target fragment, the target technical document corresponding to the document requirement information is generated.

8. The method according to claim 7, wherein, The step of generating the target technical document corresponding to the document requirement information based on the data group associated with the target fragment includes: When there are multiple target segments, segment information of multiple target segments is displayed, wherein the segment information includes: thumbnail images in the data group associated with the target segments; Based on the adjustment operation of the display position of multiple fragment information, the arrangement order of each target fragment is determined; Based on the arrangement order, the single-page documents in the data group associated with the target fragment are integrated to generate the target technical document.

9. The method according to claim 8, wherein, The fragment information also includes: the technical information included in the target fragment and the original image in the data group associated with the target fragment.

10. A method for constructing a knowledge base, comprising: Historical technical documents are structurally broken down to obtain multiple knowledge fragments, wherein the technical information included in the knowledge fragments is stored according to the document hierarchical structure; Based on multiple knowledge fragments, a target knowledge base is constructed.

11. The method according to claim 10, wherein, The technical information includes text content information and page storage information; the structured decomposition of historical technical documents yields multiple knowledge fragments, including: The historical technical documents are paginated to obtain multiple single-page documents; The original images of the multiple single-page documents are extracted using a multimodal large model to obtain the text content information of each single-page document; For each of the plurality of single-page documents, the single-page document, along with its corresponding original image and thumbnail image, is stored as a data group and page storage information for that single-page document is generated, wherein the page storage information includes the storage location information of the data group; According to the document hierarchy structure rules, the text content information and page storage information of each single-page document are combined to obtain multiple knowledge fragments.

12. The method according to claim 11, wherein, The page storage information also includes: the project name of the historical technical document corresponding to the data group and the page number of the single-page document in the data group in the historical technical document.

13. The method according to claim 11, wherein, The text content information includes the main chapter name corresponding to the single-page document and at least one of the following: sub-chapter name, recommended content information, and technical specifications information; and In the knowledge fragment, the sub-chapter name, the recommended content information, the technical specifications information, and the page storage information are stored as information of the next document hierarchy structure of the main chapter name.

14. An intelligent agent, comprising: The input module is configured to receive input information; The processing module is configured to determine a target task based on the input information received by the input module, determine a large model based on the target task, and obtain output information by calling the large model to execute the document generation method of any one of claims 1 to 9, or by calling the large model to execute the knowledge base construction method of any one of claims 10 to 13. The output module is configured to output the output information obtained by the processing module.

15. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the document generation method of any one of claims 1-9, or to enable the at least one processor to perform the knowledge base construction method of any one of claims 10-13.

16. A non-transitory computer-readable storage medium storing computer instructions, the computer instructions being configured to cause the computer to perform the document generation method of any one of claims 1-9, or the computer instructions being configured to cause the computer to perform the knowledge base construction method of any one of claims 10-13.

17. A computer program product comprising a computer program that, when executed by a processor, implements the document generation method according to any one of claims 1-9, or the knowledge base construction method according to any one of claims 10-13 when executed by a processor.

Images