Automatic system modeling method and device based on large language model

Abstract

The invention provides an automatic system modeling method and device based on a large language model. The method comprises the steps that system modeling text information is acquired; based on target system modeling demand vector representation corresponding to the system modeling text information, a target knowledge text is obtained from a multi-level knowledge base, and the multi-level knowledge base is constructed by system modeling knowledge of different knowledge levels in a layered mode; inputting the system modeling text information and the target knowledge text into a large language model to obtain a target modeling text output by the large language model; and based on a Mermaid chart generation tool, analyzing and rendering the target modeling text to obtain a system modeling result corresponding to the system modeling text information. According to the invention, the accuracy of the system modeling result is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an automated system modeling method and apparatus based on a large language model. Background Technology

[0002] Mermaid is a text-based diagram generation tool that can automatically convert text into corresponding visual graphics by parsing specific declarative syntax, such as class diagrams and flowchart definitions.

[0003] Large Language Models (LLMs) are generative artificial intelligence based on the Transformer architecture. They analyze massive amounts of text data to deeply learn the inherent rules of language. Their core capability enables them to progress from understanding simple semantic patterns to mastering complex common sense and logical reasoning. With this capability, Large Language Models can accurately and deeply analyze user intent and autonomously generate new and compliant text content, demonstrating powerful text generation and comprehension abilities.

[0004] In existing model-driven system design methodologies, UML / SysML, as the core modeling language, provides diverse tool support for system design, but it has significant shortcomings. On the one hand, it requires a high level of professional knowledge from users, increasing the barrier to entry; on the other hand, traditional document-based design methods lead to information fragmentation and low collaboration efficiency. Large language models also face many limitations in this field, such as insufficient knowledge coverage, prominent illusion problems, and the inability of general knowledge bases to meet the differentiated needs of users with different roles, resulting in low accuracy of generated system modeling results. This seriously affects the quality and reliability of system design and fails to meet the needs of practical applications. Therefore, there is an urgent need for an automated system modeling method and apparatus based on large language models to solve these problems. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides an automated system modeling method and apparatus based on a large language model.

[0006] This invention provides an automated system modeling method based on a large language model, comprising: Obtain system modeling text information; Based on the target system modeling requirement vector representation corresponding to the system modeling text information, target knowledge text is obtained from a multi-level knowledge base, wherein the multi-level knowledge base is constructed by hierarchically building system modeling knowledge at different knowledge levels; The system modeling text information and the target knowledge text are input into the large language model to obtain the target modeling text output by the large language model; Based on the Mermaid chart generation tool, the target modeling text is parsed and rendered to obtain the system modeling result corresponding to the system modeling text information.

[0007] According to the present invention, an automated system modeling method based on a large language model is provided, wherein the multi-level knowledge base is constructed through the following steps: Acquire knowledge text information from system modeling samples across different domain categories; Based on the knowledge level, the knowledge text information of the system modeling sample is stratified to obtain the system modeling knowledge stratification of different knowledge levels; Based on retrieval enhancement generation and the system modeling knowledge hierarchy, the multi-level knowledge base is constructed.

[0008] According to the present invention, an automated system modeling method based on a large language model is provided, wherein the step of stratifying the knowledge text information of the system modeling sample according to the knowledge level to obtain the system modeling knowledge stratification of different knowledge levels includes: When the system modeling sample knowledge text information is Mermaid's basic chart tutorial content and basic case content, the knowledge level of the system modeling sample knowledge text information is determined to be the basic level. Based on the system modeling sample knowledge text information of the aforementioned basic level, a system modeling knowledge foundation layer is constructed. When the knowledge text information of the system modeling sample is advanced technical content and advanced practical case content of the Mermaid chart generation tool, the knowledge level of the knowledge text information of the system modeling sample is determined to be the professional level. Based on the system modeling sample knowledge text information of the aforementioned professional level, a system modeling knowledge professional layer is constructed. When the knowledge text information of the system modeling sample is the grammar reference content and grammar standardization text content of the Mermaid chart generation tool, the knowledge level of the knowledge text information of the system modeling sample is determined to be the standardization level; Based on the system modeling sample knowledge text information of the aforementioned specification level, a system modeling knowledge specification layer is constructed. When the system modeling sample knowledge text information is the industry standard content and system modeling methodology content of system modeling, the knowledge level of the system modeling sample knowledge text information is determined to be the system level; Based on the system modeling sample knowledge text information of the system level, a system modeling knowledge system layer is constructed. Based on the system modeling knowledge foundation layer, the system modeling knowledge professional layer, the system modeling knowledge specification layer, and the system modeling knowledge system layer, the system modeling knowledge stratification of each different knowledge level is obtained.

[0009] According to the present invention, an automated system modeling method based on a large language model is provided, the method further includes: A system modeling vector library was built based on HNSWLib and OpenAI Embeddings; Based on the system modeling vector library, the system modeling sample knowledge text information in the multi-level knowledge base is converted into the corresponding system modeling requirement vector representation, and a mapping relationship between knowledge text and system modeling requirement vector is established. After obtaining the system modeling text information, the method further includes: Based on the system modeling vector library, the system modeling text information is converted into a vector representation of the target system modeling requirements.

[0010] According to the present invention, an automated system modeling method based on a large language model is provided, wherein the step of obtaining target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information includes: Based on the target system modeling requirement vector representation, determine the system modeling user categories and system modeling requirements; Based on the user categories and system modeling requirements of the system, the target knowledge text is obtained from the corresponding system modeling knowledge layers in the multi-level knowledge base.

[0011] According to the present invention, an automated system modeling method based on a large language model is provided, the method further includes: Based on regular expressions, Mermaid syntax error detection is performed on the target modeling text. If a syntax error is determined to exist in the target modeling text, an error syntax prompt message is generated. The Mermaid chart generation tool is used to parse and render the target modeling text to obtain the system modeling result corresponding to the system modeling text information, including: If a rendering failure occurs during the parsing and rendering process, a corresponding fallback solution will be generated based on the rendering failure information.

[0012] The present invention also provides an automated system modeling device based on a large language model, comprising: The input module is used to obtain system modeling text information; The retrieval module is used to retrieve target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information, wherein the multi-level knowledge base is constructed by hierarchically building system modeling knowledge at different knowledge levels; The large model processing module is used to input the system modeling text information and the target knowledge text into the large language model to obtain the target modeling text output by the large language model; The parsing and rendering module is used to parse and render the target modeling text based on the Mermaid chart generation tool to obtain the system modeling result corresponding to the system modeling text information.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the automated system modeling method based on the large language model as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the automated system modeling method based on a large language model as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the automated system modeling method based on a large language model as described above.

[0016] The present invention provides an automated system modeling method and apparatus based on a large language model, which obtains system modeling text information, retrieves corresponding target knowledge text from a multi-layered knowledge base constructed in a hierarchical manner, inputs the target knowledge text and system modeling requirements into a large language model to obtain target modeling text, and then parses and renders it using the Mermaid tool, thereby improving the accuracy of the system modeling results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the automated system modeling method based on a large language model provided by this invention; Figure 2 A schematic diagram of the system modeling results provided by this invention; Figure 3 A schematic diagram of the structure of the automated system modeling device based on a large language model provided by the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] Mermaid, as a text-based diagram generation tool, has become a major technology for quickly converting structured descriptions into diagrams due to its intuitiveness and programmability. However, its nature requires users to provide standardized text code rather than directly understanding the intent of natural language.

[0021] The field of system modeling currently faces many challenges. On the one hand, Mermaid tools rely on specialized syntax, requiring users to manually convert modeling ideas into text code, which presents a high technical barrier. Moreover, large models and Mermaid tools are independent of each other, causing a break in the "understanding-generation-rendering" process and making it impossible to achieve end-to-end automatic conversion from natural language requirements to visualized system models.

[0022] On the other hand, traditional modeling languages ​​(such as UML / SysML) have a high learning curve, making it difficult for beginners to master them quickly. Design teams also lack a deep understanding of them, making it difficult to fully realize their value, resulting in poor tool adoption.

[0023] Furthermore, general-purpose large language models lack sufficient professional knowledge in the field of system modeling, making it difficult to meet the high-precision modeling needs of complex scenarios. They are also prone to generating content that does not conform to industry standards, which can mislead design decisions. At the same time, existing modeling tools lack layered adaptation for different roles, which cannot meet the needs of multi-role collaborative modeling.

[0024] To address the problems existing in the prior art, this invention provides an automated system modeling method based on a large language model, which realizes end-to-end automatic generation from natural language description to standardized system modeling diagram, while adapting to the needs of users with different professional levels, reducing the modeling threshold, and improving modeling efficiency and professionalism.

[0025] Figure 1 This is a flowchart illustrating the automated system modeling method based on a large language model provided by the present invention, as shown below. Figure 1 As shown, this invention provides an automated system modeling method based on a large language model, comprising: Step 101: Obtain system modeling text information.

[0026] In this invention, for system modeling scenarios, various natural language descriptive information related to system modeling is collected, namely system modeling text information. This information may be textual descriptions of system functions, architecture, and processes directly input by the user. For example, if a user wants to build an e-commerce system, they might input natural language text such as "The e-commerce system needs to include functional modules such as user registration and login, product display, shopping cart, order payment, and logistics tracking, and the modules interact with each other through interfaces." This is the obtained system modeling text information, expressing the user's general needs and expectations for system modeling.

[0027] Step 102: Based on the target system modeling requirement vector representation corresponding to the system modeling text information, obtain the target knowledge text from the multi-level knowledge base, wherein the multi-level knowledge base is constructed by hierarchically building system modeling knowledge at different knowledge levels.

[0028] In this invention, a four-layer structured knowledge base system of "basic-professional-standard-system" is constructed, namely, multi-level knowledge base construction.

[0029] Specifically, the basic layer contains chart tutorials and examples for beginners. For instance, there are some simple Mermaid chart drawing tutorials that demonstrate several basic steps from start to finish, along with corresponding sample code and screenshots.

[0030] The professional layer covers advanced skills and best practices for professional users. For example, in system architecture design, it introduces architectural design patterns for complex systems, such as communication methods between services in a microservice architecture, advanced techniques for data consistency handling, and best practice cases successfully applied in different industry scenarios.

[0031] The specification layer provides Mermaid syntax references and standardized documents, detailing the syntax rules for various Mermaid chart types (such as class diagrams, sequence diagrams, Gantt charts, etc.), including how each element is represented, syntax format requirements, etc., to ensure code standardization.

[0032] The system layer includes industry standards and system modeling methodologies such as IEEE 1471. Among them, IEEE 1471 is a standard for describing system architecture. The system layer will contain the specific content of this standard and the methodology on how to apply it to actual system modeling, providing authoritative guidance and reference for system modeling.

[0033] Furthermore, in this invention, the OpenAI Embeddings tool can be used to vectorize the text content in the knowledge base and the system modeling text information input by the user, thereby converting the text into a numerical vector form that the computer can understand and process. For example, the knowledge text "e-commerce system user registration process" and the natural language requirement text input by the user regarding e-commerce system registration are both converted into numerical vectors of a specific dimension using OpenAI Embeddings. Then, based on the vector library built on HNSWLib, the vector most similar to the target vector is quickly found in a large amount of vector data. Next, a three-dimensional adaptation logic of "role-knowledge base-retrieval" is designed to dynamically match the knowledge base range according to the user's role (beginner / senior engineer) and adjust the retrieval parameters (such as the topK value). For example, if the user is a beginner, the basic knowledge base may be matched mainly, with a larger topK value to obtain more relevant introductory knowledge; if the user is a senior engineer, the professional and system-level knowledge bases will be matched more often, with a relatively smaller topK value to obtain more accurate professional knowledge. Finally, the target knowledge text corresponding to the system modeling text information is obtained from the multi-level knowledge base.

[0034] Step 103: Input the system modeling text information and the target knowledge text into the large language model to obtain the target modeling text output by the large language model.

[0035] In this invention, the system modeling text information and target knowledge text obtained in the above embodiments are input into the large language model.

[0036] This invention employs a combination of rules and a large language model to parse user natural language requirements. The rule layer first identifies the modeling type (e.g., flowcharts, class diagrams), core entities (e.g., users and products in an e-commerce system), and constraints (e.g., requiring users to fill in their real names during registration). The large language model layer further parses implicit requirements to ensure a complete understanding of them. Then, based on retrieved professional knowledge, the large language model generates modeling text that conforms to Mermaid syntax specifications. During the generation process, it incorporates modeling domain specifications and industry standards to avoid the "illusion" problem. The "illusion" problem refers to the large model generating content that appears reasonable but is actually inconsistent with reality or industry standards. For example, the large language model will not generate code that violates Mermaid syntax rules, nor will it generate process designs that do not conform to e-commerce system industry standards, thus ensuring the accuracy and standardization of the generated text, ultimately outputting the target modeling text.

[0037] Step 104: Based on the Mermaid chart generation tool, parse and render the target modeling text to obtain the system modeling result corresponding to the system modeling text information.

[0038] Mermaid is a text-based diagram generation tool that can parse specific declarative syntax (such as Mermaid syntax code contained in the generated target modeling text) and automatically render corresponding visualizations. For example, if the target modeling text is user registration process code for an e-commerce system written in Mermaid syntax, the Mermaid tool will parse and render it into an intuitive flowchart, showing the various steps and decision points a user takes from opening the registration page to completing registration. This visualization is the system modeling result corresponding to the system modeling text information, making it easier for users to understand and evaluate the system design scheme more clearly.

[0039] The automated system modeling method based on a large language model provided by this invention obtains system modeling text information, retrieves corresponding target knowledge text from a multi-layered knowledge base constructed in a hierarchical manner, inputs the target knowledge text and system modeling requirements into a large language model to obtain target modeling text, and then parses and renders it using the Mermaid tool, ultimately improving the accuracy of the system modeling results.

[0040] Based on the above embodiments, the multi-level knowledge base is constructed through the following steps: Acquire knowledge text information from system modeling samples across different domain categories; Based on the knowledge level, the knowledge text information of the system modeling sample is stratified to obtain the system modeling knowledge stratification of different knowledge levels; Based on retrieval enhancement generation and the system modeling knowledge hierarchy, the multi-level knowledge base is constructed.

[0041] In this invention, system modeling is the process of abstractly describing and designing a system. Systems in different fields (such as finance, healthcare, e-commerce, and manufacturing) have their own unique characteristics and requirements, so the methods and content of system modeling will also differ. Therefore, the first step is to collect sample knowledge text information related to system modeling that already exists in various fields. This sample knowledge text information can include system design documents, modeling case analyses, system architecture descriptions, and flowchart illustrations, etc., which record the specific practices and experiences of system modeling in different fields.

[0042] For example, in the financial sector, documents on the modeling of core banking systems may be collected, which detail the modeling ideas and implementation methods of modules such as account management, transaction processing, and risk control; in the medical sector, modeling cases of hospital information management systems may be obtained, including modeling information on processes such as patient registration, consultation, examination, and medication dispensing; in the e-commerce sector, system modeling materials for e-commerce platforms may be available, covering modeling content in areas such as product management, user management, order processing, and payment settlement.

[0043] Knowledge levels are a classification of the complexity, professional level, and applicable objects of system modeling knowledge. Different knowledge levels correspond to different system modeling knowledge content. This invention, by stratifying the collected system modeling sample knowledge text information, can group knowledge of similar difficulty and scope of application into one category, forming a hierarchical system modeling knowledge system of different knowledge levels, thereby improving the utilization efficiency of the knowledge base.

[0044] In this invention, when constructing the knowledge base, Retrieval Augmented Generation (RAG) technology is incorporated to achieve dynamic retrieval and role-based adaptation of modeling expertise. Based on RAG technology, relevant information fragments can be retrieved from a large amount of knowledge text according to the user's query requirements. This retrieved information is then used as context input into the text generation model to generate more accurate and targeted answers or content. This technology can fully utilize existing knowledge resources to improve the quality and relevance of the generated content.

[0045] Furthermore, based on the pre-defined knowledge levels, a comprehensive knowledge repository is built using retrieval-enhanced generation technology, thereby providing knowledge support at different levels and in different fields according to the different needs and roles of users.

[0046] Based on the above embodiments, the step of stratifying the system modeling sample knowledge text information according to the knowledge level to obtain system modeling knowledge stratification at different knowledge levels includes: When the system modeling sample knowledge text information is Mermaid's basic chart tutorial content and basic case content, the knowledge level of the system modeling sample knowledge text information is determined to be the basic level. Based on the system modeling sample knowledge text information of the aforementioned basic level, a system modeling knowledge foundation layer is constructed. When the knowledge text information of the system modeling sample is advanced technical content and advanced practical case content of the Mermaid chart generation tool, the knowledge level of the knowledge text information of the system modeling sample is determined to be the professional level. Based on the system modeling sample knowledge text information of the aforementioned professional level, a system modeling knowledge professional layer is constructed. When the knowledge text information of the system modeling sample is the grammar reference content and grammar standardization text content of the Mermaid chart generation tool, the knowledge level of the knowledge text information of the system modeling sample is determined to be the standardization level; Based on the system modeling sample knowledge text information of the aforementioned specification level, a system modeling knowledge specification layer is constructed. When the system modeling sample knowledge text information is the industry standard content and system modeling methodology content of system modeling, the knowledge level of the system modeling sample knowledge text information is determined to be the system level; Based on the system modeling sample knowledge text information of the system level, a system modeling knowledge system layer is constructed. Based on the system modeling knowledge foundation layer, the system modeling knowledge professional layer, the system modeling knowledge specification layer, and the system modeling knowledge system layer, the system modeling knowledge stratification of each different knowledge level is obtained.

[0047] In this invention, when the system modeling sample knowledge text information consists of Mermaid's basic diagram tutorial content and basic case content, the knowledge level of this sample knowledge text information is determined to be the basic level. Mermaid, as a tool for defining diagrams and flowcharts using text, allows beginners to quickly get started with its basic diagram tutorial content through simple syntax rules. In this invention, the basic case content can be simple examples combined with real-world application scenarios, such as using Mermaid to draw a simple project management process case, helping beginners understand how to apply their learned knowledge to practice. This content is the most basic and easiest to understand part of the entire knowledge system, and therefore is classified as the basic level. Furthermore, based on these system modeling sample knowledge text information at the basic level, a system modeling knowledge foundation layer is constructed.

[0048] When the system modeling sample knowledge text information comprises advanced technical content and advanced practical case studies of the Mermaid chart generation tool, the knowledge level of this sample knowledge text information is determined to be professional level. In this invention, the advanced technical content involves more advanced and complex functions and techniques of the Mermaid tool, such as how to use Mermaid to achieve more complex chart interaction effects and customize chart styles. Advanced practical cases are designed for professional users, combining complex business scenarios in actual work, such as business process optimization in large enterprises and complex system architecture design, using Mermaid for detailed modeling and visualization. Furthermore, based on the professional-level system modeling sample knowledge text information, a professional layer of system modeling knowledge is constructed to address more complex system modeling tasks.

[0049] When the system modeling sample knowledge text information consists of the grammar reference content and grammar standardization text content of the Mermaid diagram generation tool, the knowledge level of this sample knowledge text information is determined to be the normative level. In this invention, the grammar reference content is a detailed description of all grammar rules of the Mermaid tool, allowing users to easily refer to the specific usage and meaning of a particular grammar element. The grammar standardization text content ensures that the grammar used when drawing diagrams using Mermaid conforms to a unified standard, avoiding diagrams that cannot be correctly generated or understood due to non-standard grammar. For example, for the grammar used to define flowcharts in Mermaid, the normative level content clearly specifies the function and usage format of each grammar symbol, ensuring that the modeling process is carried out according to a unified standard. This content guarantees the accuracy and consistency of system modeling.

[0050] Furthermore, based on the knowledge text information of system modeling samples at the specification level, a system modeling knowledge specification layer is constructed. This layer provides standards and specifications for the entire system modeling process, ensuring that a unified standard is followed when using Mermaid tools, thereby improving the quality and readability of the modeling.

[0051] When the knowledge text information of the system modeling samples constitutes industry standard content and system modeling methodology content for system modeling, the knowledge level of this sample knowledge text information is determined as the system level. In this invention, industry standard content refers to the unified specifications and requirements followed for system modeling within a specific industry. For example, the IEEE 1471 standard provides a framework and guidelines for describing and modeling system architecture, ensuring that different enterprises and teams can follow unified standards when modeling systems, facilitating communication and collaboration. System modeling methodology content refers to a series of theories, methods, and steps on how to perform system modeling, such as object-oriented modeling methods and structured modeling methods. These methodologies provide systematic ideas and methods for system modeling and have a guiding role in the entire field of system modeling.

[0052] Furthermore, based on the system-level system modeling sample knowledge text information, a system modeling knowledge system layer is constructed. In the subsequent system modeling process, this layer can provide macro-level guidance and reference for the large language model, enabling an understanding of the latest industry standards and development trends, thereby constructing a high-quality system model that meets industry requirements.

[0053] Finally, based on the aforementioned system modeling knowledge foundation layer, system modeling knowledge professional layer, system modeling knowledge specification layer, and system modeling knowledge system layer, a hierarchical system modeling knowledge hierarchy for each different knowledge level is obtained. These four layers are interconnected and progressive, forming a complete multi-layered knowledge base system that provides comprehensive and systematic support for system modeling knowledge.

[0054] Based on the above embodiments, the method further includes: A system modeling vector library was built based on HNSWLib and OpenAI Embeddings; Based on the system modeling vector library, the system modeling sample knowledge text information in the multi-level knowledge base is converted into the corresponding system modeling requirement vector representation, and a mapping relationship between knowledge text and system modeling requirement vector is established. After obtaining the system modeling text information, the method further includes: Based on the system modeling vector library, the system modeling text information is converted into a vector representation of the target system modeling requirements.

[0055] In this invention, OpenAI Embeddings is a technique for converting text data into vector representations. In the context of system modeling knowledge processing, it can transform various texts related to system modeling, such as basic tutorials, professional skills, and industry standards, into vectors in a high-dimensional space. These vectors contain semantic information of the text; texts with similar semantics are closer together in the vector space, while texts with significant semantic differences are farther apart. For example, a text about a basic Mermaid chart drawing tutorial and another similar basic tutorial text will have vectors that are relatively close in position in the vector space after being converted by OpenAI Embeddings; however, a basic tutorial text and a text about advanced system modeling algorithms will have vectors that are farther apart after conversion. In this way, text information in the system modeling knowledge base can be transformed from natural language form into a vector form that computers can understand and process.

[0056] Furthermore, when constructing a system modeling vector library using HNSWLib, the vast amount of system modeling knowledge text vectors generated by OpenAI Embeddings can be organized and stored to form an optimized index structure, thereby establishing a mapping relationship between knowledge text and system modeling requirement vectors. This structure enables the rapid and accurate retrieval of vectors most similar to the target vector during subsequent vector searches. For example, when it is necessary to find knowledge vectors related to a specific problem from system modeling knowledge vectors, HNSWLib can utilize its efficient index structure to avoid comparing all vectors one by one, greatly improving retrieval efficiency.

[0057] In this invention, OpenAI Embeddings is used to vectorize all text in the system modeling knowledge base, resulting in a large number of system modeling knowledge text vectors. These vectors are then imported into HNSWLib, which constructs a multi-level index structure based on the vectors' characteristics and distribution, thereby completing the construction of the system modeling vector library.

[0058] The multi-layered knowledge base in this invention is a collection of system modeling knowledge constructed according to different knowledge levels and content types. It includes different levels of knowledge such as the basic layer (beginner chart tutorials and examples), the professional layer (advanced techniques and best practices), the normative layer (Mermaid syntax references and standardized documents), and the system layer (industry standards and system modeling methodologies). Each layer has its own specific system modeling sample knowledge text information. For each system modeling sample knowledge text information in the multi-layered knowledge base, a pre-built system modeling vector library (which includes OpenAI Embeddings conversion rules and HNSWLib's index structure) is used to input it into the corresponding conversion process, thereby converting each sample knowledge text into a corresponding knowledge text vector representation. For example, for a Mermaid basic chart drawing example text in the basic layer, after processing with OpenAI Embeddings, a vector of a specific dimension is obtained. This vector represents the semantic information of the example text and can be stored and retrieved in the vector library built by HNSWLib. In this way, all system modeling sample knowledge text information in the multi-layered knowledge base is converted into vector form, facilitating subsequent retrieval and matching.

[0059] When a user makes a natural language request related to system modeling, such as "How to draw a complex system architecture diagram using Mermaid", the input text is the system modeling text information. This text information reflects the user's specific problem and needs, and requires finding relevant knowledge from the system modeling knowledge base.

[0060] In this invention, in order to accurately and efficiently retrieve knowledge that matches user needs from the system modeling vector library, it is necessary to convert the system modeling text information input by the user into vector form, namely, the target system modeling requirement vector representation, so as to measure the degree of correlation between user needs and knowledge in the knowledge base by calculating the distance or similarity between vectors.

[0061] In this invention, similar to the process of converting sample knowledge text in a multi-level knowledge base into vectors, OpenAI Embeddings technology is used to vectorize the system modeling text information input by the user. The processed system modeling requirement vector representation is stored in a suitable data structure, establishing a mapping relationship between knowledge text and system modeling requirement vectors, so as to facilitate subsequent retrieval and matching operations with vectors in the system modeling vector library. For example, by calculating the cosine similarity between the target knowledge text vector and each knowledge text vector in the vector library, the vectors with the highest similarity are found, and the knowledge text corresponding to these vectors is the knowledge content most likely to meet the user's needs.

[0062] Through the above steps, the text information and user requirement text in the system modeling knowledge base are converted into vector form, and an efficient vector library is constructed. Based on this, the present invention also constructs a three-dimensional adaptation logic of "role-knowledge base-retrieval," dynamically matching the knowledge base scope and adjusting retrieval parameters (such as topK values) according to the user's role (e.g., beginner / senior engineer). This enables more accurate retrieval of professional knowledge matching user needs from the vector library, providing structured knowledge support for large-scale model generation, effectively solving the problem of insufficient professional knowledge coverage in large-scale models, and improving the accuracy and efficiency of system modeling knowledge retrieval and application.

[0063] Based on the above embodiments, the step of obtaining target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information includes: Based on the target system modeling requirement vector representation, determine the system modeling user categories and system modeling requirements; Based on the user categories and system modeling requirements of the system, the target knowledge text is obtained from the corresponding system modeling knowledge layers in the multi-level knowledge base.

[0064] In this invention, the system modeling text information primarily originates from user input. This could be natural language descriptions entered by the user through the interactive interface of a relevant system modeling platform or tool, or via chat windows. For example, a user might input, "I'm a beginner in system modeling and want to understand how to draw simple flowcharts using Mermaid," or "I'm a senior system architect and need to find information about system boundary definitions in the IEEE 1471 standard." This text information is presented in natural language and includes the user's identity information and specific problem requirements.

[0065] Furthermore, the system modeling text information is converted into vectors for semantic analysis (such as similarity calculation and semantic clustering) to identify keywords and phrases that reflect the user's identity and knowledge level. For example, if words such as "beginner," "novice," or "just starting to learn" appear in the text, it can be preliminarily determined that the user belongs to the beginner category; if expressions such as "senior engineer," "senior architect," or "has many years of experience" appear, the user is more likely to be classified as an advanced user.

[0066] In addition to keywords, this invention also considers the overall context of the text. Some users may not directly state their identity, but this can be indirectly inferred from the complexity of the questions they ask and the depth of their professional knowledge. For example, a user asking "How to use Mermaid to implement a multi-layered visualization of a complex system architecture, considering performance optimization" involves high levels of professional knowledge and skills, indicating an advanced user; while a user asking "How to draw a simple sequence diagram in Mermaid" is more likely a beginner.

[0067] After identifying the user categories for system modeling, the core issues of concern to users are determined from the system modeling text information. For example, in the question "I want to use Mermaid to draw an order processing flowchart for an e-commerce system, but I don't know how to represent the transitions between different states," the key issues are using Mermaid to draw specific flowcharts and represent state transitions. Further analysis of the specific details of user needs is needed, including the scope, depth, and application scenarios of the required knowledge. For instance, if the user requirement is "to understand commonly used graphic symbols in system modeling and their meanings so that they can be accurately used when drawing project flowcharts," the scope of the requirement is clearly defined as commonly used graphic symbols, the depth as understanding their meanings, and the application scenario as drawing project flowcharts.

[0068] In this invention, a multi-layered knowledge base is constructed according to different levels and content types of system modeling knowledge, including a basic layer, a professional layer, a specification layer, and a system layer. The basic layer is mainly aimed at beginners, containing beginner diagram tutorials and examples, such as the basic syntax of Mermaid and simple diagram drawing methods; the professional layer is aimed at professional users, covering advanced techniques and best practices, such as modeling methods and optimization strategies for complex system architectures; the specification layer provides Mermaid syntax references and standardized documents, as well as industry specifications and standards for system modeling; the system layer includes industry standards such as IEEE 1471 and system modeling methodologies, providing macro-level guidance and theoretical support for advanced users.

[0069] Within the defined knowledge hierarchy, based on the key questions and details of user needs, the corresponding text can be accurately retrieved in the vector space. For example, searching for "complex system architecture modeling and optimization" in the professional layer will yield relevant tutorials, case studies, and best practice documents.

[0070] Based on the above embodiments, the method further includes: Based on regular expressions, Mermaid syntax error detection is performed on the target modeling text. If a syntax error is determined to exist in the target modeling text, an error syntax prompt message is generated. The Mermaid chart generation tool is used to parse and render the target modeling text to obtain the system modeling result corresponding to the system modeling text information, including: If a rendering failure occurs during the parsing and rendering process, a corresponding fallback solution will be generated based on the rendering failure information.

[0071] In this invention, regular expressions can be used for syntax error detection when the system processes target modeling text. A regular expression is an expression that can match strings according to specific rules. By pre-setting a series of rule patterns that conform to Mermaid syntax specifications, the target modeling text is scanned and matched character by character or sentence by sentence.

[0072] For example, Mermaid syntax has specific keyword and structural requirements for different types of diagrams (such as flowcharts, class diagrams, etc.). For flowcharts, node definitions typically use a specific format, and regular expressions can precisely match these keywords and structures. If a part of the target modeling text does not conform to these preset rule patterns, it is judged to have a syntax error.

[0073] When a syntax error is detected in the target modeling text, error syntax hints are generated. These hints specify the location of the error (e.g., which line and column of the text it is in) and the type of error (e.g., missing keywords, formatting errors, etc.) to enable quick location and correction of the error.

[0074] In this invention, the Mermaid diagram generation tool is used to parse and render the target modeling text that has passed the syntax check (whether it passes or not, it is processed directly if it passes, and processed after correction if there are errors), and convert it into a visual diagram form, such as a flowchart, class diagram, sequence diagram, state diagram, ER diagram and requirement diagram, so as to obtain the system modeling result corresponding to the system modeling text information.

[0075] In this invention, rendering may fail for various reasons during the parsing and rendering process. These reasons may include complex and difficult-to-handle grammatical structures in the target modeling text, insufficient system resources causing rendering interruptions, and unresolved compatibility issues in the Mermaid chart generation tool itself.

[0076] When rendering failures occur, corresponding fallback solutions can be generated based on the rendering failure information. The rendering failure information records detailed information such as the cause of the failure and its location. Analysis and judgment are made based on this information. For example, if the rendering failure is due to a complex syntax structure, the fallback solution might be to simplify the syntax structure, converting it into a simpler form that can be rendered correctly. If the problem is insufficient system resources, the fallback solution might be to adjust rendering parameters to reduce resource requirements, or prompt the user to close other resource-intensive programs before attempting rendering again. If the problem is a compatibility issue with the Mermaid chart generation tool, the fallback solution might be to try rendering with other similar chart generation tools, or to convert the target modeling text into other common chart description languages ​​before rendering. These fallback solutions maximize the chances of users obtaining visualized system modeling results, improving system stability and user experience.

[0077] Figure 2 This is a schematic diagram of the system modeling results provided by the present invention, which can be used as a reference. Figure 2 As shown, taking the construction of an automatic feeder system as an example, the diagrams, supported by a multi-layered knowledge base, encompass both basic hardware knowledge (such as sensor types) and specialized control logic (such as motor drives), addressing the issue of insufficient professional knowledge coverage in large models. Furthermore, the Mermaid tool incorporates standards such as IEEE 1471 to ensure that module naming and relationship descriptions conform to industry norms, avoiding misunderstandings caused by unstructured output. The graph TD represents the target modeling text corresponding to the automatic feeder system modeling results output by the Mermaid tool.

[0078] In this invention, when interacting with the model generated by the Mermaid tool, users can directly drag and drop various elements within the model using the mouse on the visual interface. For example, they can drag different process nodes in a flowchart or drag the class frames in a class diagram. Through this operation, users can flexibly adjust the positions of elements in the model according to their needs, re-plan the layout, and make the model better suited to their design ideas or display requirements. For example, in Figure 2 In the flowchart, if a user feels that the position of a key step is not suitable, they can easily move it to a more appropriate position by dragging and dropping, making the whole process clearer and easier to understand.

[0079] Simultaneously, users can zoom in and out of the generated model. When users want to view detailed parts of the model, they can zoom in to see the detailed information of each element more clearly, such as attributes and method names in the figure, and node descriptions in the flowchart. Conversely, when users want to grasp the overall picture of the model, they can zoom out to display the entire model on the screen, allowing them to have a macroscopic understanding of the model's overall structure and relationships. For example, in a large system architecture diagram, users can first zoom out to view the module composition and interrelationships of the entire system, and then zoom in to view the detailed structure of a specific module.

[0080] Optionally, this invention provides users with the ability to view the model's source code, allowing them to see the underlying code representation of the generated model, such as Mermaid syntax code. Simultaneously, it supports user editing of the source code, enabling direct modification of the model's descriptive code within the source code editing interface. After modification, the visual model is updated in real-time, re-rendering the modified code into a new model for display. For example, if users find adjusting the model through the visual interface inconvenient, they can directly modify relevant parameters or structures in the source code and immediately see the effect of the modified model.

[0081] SVG (Scalable Vector Graphics) is a markup language used to describe two-dimensional vector graphics. The models generated by this invention are typically rendered and displayed in SVG format. When users perform the aforementioned interactive operations (such as dragging, zooming, etc.), the layout of the SVG image is dynamically adjusted. For example, when a user drags an element, the relative positional relationship between that element and other elements, as well as the overall layout structure of the model, are recalculated to ensure that the connection relationships and hierarchical relationships between elements remain correct, avoiding problems such as element overlap and disordered connecting lines. This ensures that the model maintains a clear and reasonable layout under different operating states, improving the visualization effect.

[0082] Meanwhile, this invention dynamically adjusts the size of the container holding the model based on the model's size and the user's operational needs. When a user zooms in on the model, if the container size is fixed, the model may not be fully displayed or scrollbars may appear, affecting the user experience. Therefore, by monitoring changes in the model's size in real time, the container size is automatically adjusted to ensure the model is fully displayed within the container. Similarly, when a user zooms out on the model, the container size is reduced accordingly to avoid excessive blank areas, making the interface more compact and aesthetically pleasing. By dynamically adjusting the container size, problems such as missing graphics rendering (e.g., some model elements cannot be displayed due to the container being too small) or logical inconsistencies (e.g., the model layout is disordered due to an inappropriate container size) can be avoided, further improving the visualization effect and user interaction experience.

[0083] In this invention, a role selection interface can also be set up. Before performing system modeling, users need to select their role, such as a beginner or a senior engineer. This role selection module is the foundation for the system to implement differentiated services, providing different functions and content to users based on their selected role.

[0084] For example, for beginners, the content provides multi-step guided modeling tutorials and basic teaching information. These multi-step guided modeling tutorials break down the complex modeling process into simple steps, guiding beginners step-by-step to create their models.

[0085] For senior engineers, we provide comprehensive professional content and complex modeling solutions. This comprehensive content covers all aspects of the modeling field, including the latest technological trends, advanced modeling techniques, and complex model architecture designs, meeting the depth and breadth needs of senior engineers.

[0086] This invention addresses the issue that existing modeling tools do not adequately consider the diverse needs of users in different roles, and offer relatively limited functions and content, failing to meet the different requirements of beginners and senior engineers. By setting up a role selection module, it enables differentiated interaction and knowledge provision for users in different roles, providing the most suitable functions and services based on the user's role. This solves the problem of insufficient role adaptation in existing tools, and improves the applicability and user satisfaction of the system.

[0087] The following describes the automated system modeling apparatus based on a large language model provided by the present invention. The automated system modeling apparatus based on a large language model described below can be referred to in correspondence with the automated system modeling method based on a large language model described above.

[0088] Figure 3 This is a schematic diagram of the structure of the automated system modeling device based on a large language model provided by the present invention, as shown below. Figure 3 As shown, this invention provides an automated system modeling device based on a large language model, including an input module 301, a retrieval module 302, a large model processing module 303, and a parsing and rendering module 304. The input module 301 is used to acquire system modeling text information; the retrieval module 302 is used to acquire target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information, wherein the multi-level knowledge base is constructed hierarchically by system modeling knowledge at different knowledge levels; the large model processing module 303 is used to input the system modeling text information and the target knowledge text into the large language model to obtain the target modeling text output by the large language model; the parsing and rendering module 304 is used to parse and render the target modeling text based on a Mermaid chart generation tool to obtain the system modeling result corresponding to the system modeling text information.

[0089] The automated system modeling device based on a large language model provided by this invention obtains system modeling text information, retrieves corresponding target knowledge text from a multi-layered knowledge base constructed in a hierarchical manner, inputs the target knowledge text and system modeling requirements into a large language model to obtain target modeling text, and then parses and renders it using the Mermaid tool, ultimately improving the accuracy of the system modeling results.

[0090] The system provided in this embodiment of the invention is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0091] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4 As shown, the electronic device may include: a processor 401, a communications interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communications interface 402, and the memory 403 communicate with each other via the communication bus 404. The processor 401 can call logical instructions in the memory 403 to execute an automated system modeling method based on a large language model. This method includes: acquiring system modeling text information; acquiring target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information, wherein the multi-level knowledge base is constructed hierarchically by system modeling knowledge at different knowledge levels; inputting the system modeling text information and the target knowledge text into a large language model to obtain the target modeling text output by the large language model; and parsing and rendering the target modeling text using a Mermaid chart generation tool to obtain the system modeling result corresponding to the system modeling text information.

[0092] Furthermore, the logical instructions in the aforementioned memory 403 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a 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.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. 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.

[0093] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the automated system modeling method based on a large language model provided by the above methods, the method comprising: acquiring system modeling text information; acquiring target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information, wherein the multi-level knowledge base is constructed by hierarchically constructing system modeling knowledge at different knowledge levels; inputting the system modeling text information and the target knowledge text into a large language model to obtain target modeling text output by the large language model; and parsing and rendering the target modeling text based on a Mermaid chart generation tool to obtain the system modeling result corresponding to the system modeling text information.

[0094] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the automated system modeling method based on a large language model provided in the above embodiments. The method includes: acquiring system modeling text information; acquiring target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information, wherein the multi-level knowledge base is constructed hierarchically by system modeling knowledge at different knowledge levels; inputting the system modeling text information and the target knowledge text into a large language model to obtain target modeling text output by the large language model; and parsing and rendering the target modeling text using a Mermaid chart generation tool to obtain the system modeling result corresponding to the system modeling text information.

[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for modeling automated systems based on a large language model, characterized in that, include: Obtain system modeling text information; Based on the target system modeling requirement vector representation corresponding to the system modeling text information, target knowledge text is obtained from a multi-level knowledge base, wherein the multi-level knowledge base is constructed by hierarchically building system modeling knowledge at different knowledge levels; The system modeling text information and the target knowledge text are input into the large language model to obtain the target modeling text output by the large language model; Based on the Mermaid chart generation tool, the target modeling text is parsed and rendered to obtain the system modeling result corresponding to the system modeling text information.

2. The automated system modeling method based on a large language model according to claim 1, characterized in that, The multi-level knowledge base is constructed through the following steps: Acquire knowledge text information from system modeling samples across different domain categories; Based on the knowledge level, the knowledge text information of the system modeling sample is stratified to obtain the system modeling knowledge stratification of different knowledge levels; Based on retrieval enhancement generation and the system modeling knowledge hierarchy, the multi-level knowledge base is constructed.

3. The automated system modeling method based on a large language model according to claim 2, characterized in that, The step of stratifying the system modeling sample knowledge text information according to the knowledge level to obtain system modeling knowledge stratification at different knowledge levels includes: When the system modeling sample knowledge text information is Mermaid's basic chart tutorial content and basic case content, the knowledge level of the system modeling sample knowledge text information is determined to be the basic level. Based on the system modeling sample knowledge text information of the aforementioned basic level, a system modeling knowledge foundation layer is constructed. When the knowledge text information of the system modeling sample is advanced technical content and advanced practical case content of the Mermaid chart generation tool, the knowledge level of the knowledge text information of the system modeling sample is determined to be the professional level. Based on the system modeling sample knowledge text information of the aforementioned professional level, a system modeling knowledge professional layer is constructed. When the knowledge text information of the system modeling sample is the grammar reference content and grammar standardization text content of the Mermaid chart generation tool, the knowledge level of the knowledge text information of the system modeling sample is determined to be the standardization level; Based on the system modeling sample knowledge text information of the aforementioned specification level, a system modeling knowledge specification layer is constructed. When the system modeling sample knowledge text information is the industry standard content and system modeling methodology content of system modeling, the knowledge level of the system modeling sample knowledge text information is determined to be the system level; Based on the system modeling sample knowledge text information of the system level, a system modeling knowledge system layer is constructed. Based on the system modeling knowledge foundation layer, the system modeling knowledge professional layer, the system modeling knowledge specification layer, and the system modeling knowledge system layer, the system modeling knowledge stratification of each different knowledge level is obtained.

4. The automated system modeling method based on a large language model according to claim 1, characterized in that, The method further includes: A system modeling vector library was built based on HNSWLib and OpenAI Embeddings; Based on the system modeling vector library, the system modeling sample knowledge text information in the multi-level knowledge base is converted into the corresponding system modeling requirement vector representation, and a mapping relationship between knowledge text and system modeling requirement vector is established. After obtaining the system modeling text information, the method further includes: Based on the system modeling vector library, the system modeling text information is converted into a vector representation of the target system modeling requirements.

5. The automated system modeling method based on a large language model according to claim 1, characterized in that, The target system modeling requirement vector representation based on the system modeling text information is obtained from a multi-level knowledge base, including: Based on the target system modeling requirement vector representation, determine the system modeling user categories and system modeling requirements; Based on the user categories and system modeling requirements of the system, the target knowledge text is obtained from the corresponding system modeling knowledge layers in the multi-level knowledge base.

6. The automated system modeling method based on a large language model according to any one of claims 1 to 5, characterized in that, The method further includes: Based on regular expressions, Mermaid syntax error detection is performed on the target modeling text. If a syntax error is determined to exist in the target modeling text, an error syntax prompt message is generated. The Mermaid chart generation tool is used to parse and render the target modeling text to obtain the system modeling result corresponding to the system modeling text information, including: If a rendering failure occurs during the parsing and rendering process, a corresponding fallback solution will be generated based on the rendering failure information.

7. An automated system modeling device based on a large language model, characterized in that, include: The input module is used to obtain system modeling text information; The retrieval module is used to retrieve target knowledge text from a multi-level knowledge base based on the target system modeling requirement vector representation corresponding to the system modeling text information, wherein the multi-level knowledge base is constructed by hierarchically building system modeling knowledge at different knowledge levels; The large model processing module is used to input the system modeling text information and the target knowledge text into the large language model to obtain the target modeling text output by the large language model; The parsing and rendering module is used to parse and render the target modeling text based on the Mermaid chart generation tool to obtain the system modeling result corresponding to the system modeling text information.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the automated system modeling method based on a large language model as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the automated system modeling method based on a large language model as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the automated system modeling method based on a large language model as described in any one of claims 1 to 6.

Images