Software development verification system based on large model
By using a large-model-based software development verification system, the problems of low efficiency, insufficient semantic understanding, and insufficient automation support in existing technologies are solved. It achieves efficient and accurate code verification and rapid problem localization, ensuring code consistency with requirements, reducing development costs and extending the cycle time.
Patent Information
- Application Number
- CN202511009971.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-22
- Publication Date
- 2025-10-28
AI Technical Summary
Existing software development verification methods are inefficient, prone to errors due to human factors, unable to accurately understand the semantic relationship between code and requirements, lack dynamic boundary analysis and automation support, are complex to deploy, increase costs and extend development cycles.
A software development verification system based on a large model is adopted, including a neural symbol fusion unit, a cognitive alignment unit, and an evolutionary contract unit. It calculates semantic distance through a dual encoder structure, generates a mismatch index through a cross-modal attention mechanism, generates API boundary constraints, and provides real-time visual feedback by combining a trustworthy quantification subsystem and a development environment integration device.
It improves the semantic alignment accuracy between code and requirements, ensures that the code behaves as expected under different input conditions, reduces the cost of manual verification, improves development efficiency and code quality, and enables convenient deployment and rapid problem localization.
Smart Images
Figure CN120849252A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of software development technology, specifically to a software development verification system based on a large model. Background Technology
[0002] Software development refers to the process of transforming user needs into a workable software system by applying the principles of computer science and software engineering, following software development standards and processes, and utilizing programming languages, development tools, and other technical means, from requirements analysis, design, coding, testing to maintenance. Software development verification refers to the evaluation and review of software products during the software development process to determine whether they meet specified requirements and standards. The reasons for conducting software development verification are: it ensures software quality by identifying defects, errors, and vulnerabilities, allowing for timely repairs, improving software reliability and stability, and avoiding serious consequences due to quality issues after software delivery; it also ensures that requirements are met by ensuring that the software's functionality and performance meet the needs of users and stakeholders, guaranteeing that the software can solve users' actual problems; it reduces development costs by identifying and resolving software problems early, avoiding the high costs associated with fixing problems later in the development stage or after software delivery; and it ensures compliance with standards and regulations, as many industries have specific standards and regulations for software, and verification ensures that the software meets these requirements, avoiding legal risks. In addition, software development verification can enhance user confidence. Software that has undergone rigorous verification is more likely to be trusted by users, thereby increasing user acceptance and satisfaction with the software.
[0003] Existing technologies have several drawbacks in software development verification. First, traditional methods rely primarily on manual review and static code analysis, which are inefficient and prone to errors due to human factors. Second, existing technologies typically employ rule-based or pattern-matching verification methods, failing to accurately understand the semantic relationship between code and requirements, resulting in imprecise verification results. Third, existing technologies lack dynamic analysis of code input and output boundaries, failing to ensure that code behaves as expected under different input conditions, thus reducing code robustness and reliability. Furthermore, the verification process of existing technologies often lacks automation support, requiring significant manual intervention, which not only increases verification costs but also extends the development cycle. Existing technologies also lack effective visual feedback mechanisms, making it difficult for developers to quickly locate problematic modules, reducing development efficiency. Finally, the deployment and integration of existing technologies are usually complex, requiring developers to perform tedious configuration and debugging, increasing the barrier to entry. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a software development verification system based on a large model. This system solves the problems that existing technologies have significant deficiencies in verification efficiency, semantic understanding, boundary analysis, automation support, visual feedback, and convenient deployment, making it difficult to meet the requirements of efficiency and accuracy in modern software development.
[0005] To achieve the above objectives, this invention provides the following technical solution: a software development verification system based on a large model, comprising a neural symbol fusion unit, a cognitive alignment unit, and an evolutionary contract unit; wherein the neural symbol fusion unit calculates the semantic distance function between code and requirements using a dual encoder structure.
[0006] in: The feature matrix of the abstract syntax tree nodes. This represents the total number of AST nodes. For a graph neural network-based AST encoder, the output dimension is... , For the demand encoder, a pre-trained BERT model is used; The cognitive alignment unit generates a requirement-code mismatch index through a cross-modal attention mechanism: in: For trainable alignment matrices, This is a confidence correction factor dynamically calculated based on historical validation accuracy. The evolutionary contract unit generates a set of API boundary constraints:
[0007] , This is the mean squared error loss function between the predicted and expected values.
[0008] Preferably, the neural symbol fusion unit includes an invariant confidence calculation model:
[0009] in: For effective AST path coverage ( ), To predetermine the distribution entropy of the error type, To control the depth of the flow graph ( ).
[0010] Preferably, the bidirectional attention mechanism of the cognitive alignment unit is implemented as follows:
[0011] in: , , , Trainable parameter matrix : Position encoding generated by the sine function.
[0012] Preferably, the condition for the evolutionary contract unit to perform counterfactual correction is:
[0013] in: These are 128-dimensional semantic vectors representing demand samples, error samples, and correction samples, respectively. : Tolerance coefficient, the value of which is positively correlated with the complexity of the software module.
[0014] A reliable quantification subsystem, embedded in the system of claim 1, includes a dynamic scoring module and a regression test generator: the dynamic scoring module is:
[0015] The weighting coefficients are calculated using the following formula: The regression test generator is:
[0016] This is for adaptive perturbation step size.
[0017] Preferably, the complexity control strategy for generating test cases is as follows:
[0018] in: For the input parameter dimension, For symbolic execution depth ( ), This represents the number of historical verification failures for the current module.
[0019] A development environment integration device, integrating the system of claim 1, includes a real-time heatmap generator and a color encoding engine: The real-time heatmap generator is:
[0020] in: The weights are based on the module's real-time usage duration and are dynamic. Static weights configured for administrators; The color encoding engine is: .
[0021] A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the operation of the system according to any one of claims 1 to 7, specifically including: Execute neural symbol fusion unit instructions: calculate and output SEMDIST and ; Schedule the TrustScore quantification subsystem: generate TrustScore and trigger regression test generation; Call the heatmap rendering interface: Visualize the RGB verification status in the integrated development environment (IDE).
[0022] This invention provides a software development verification system based on a large model. It has the following beneficial effects: This invention provides a software development verification system based on a large model. It achieves semantic alignment between code and requirements through a neural symbolic fusion unit, accurately identifying whether the code precisely fulfills the intended requirements, avoiding errors caused by semantic comprehension biases in traditional verification methods. Secondly, the cognitive alignment unit, through a cross-modal attention mechanism, effectively captures semantic deviations between code and requirements and generates targeted correction suggestions, significantly improving code correction efficiency. The evolutionary contract unit analyzes the input and output boundary conditions of the code to generate a set of API boundary constraints, ensuring that the code behaves as expected under different input conditions, improving code robustness and reliability. The credibility quantification subsystem, through a dynamic scoring module and a regression test generator, can evaluate the credibility of the code in real time and automatically generate regression test cases after code modifications, ensuring the code remains correct throughout the iteration process. The development environment integration device displays the verification results in the form of a heatmap in the integrated development environment in real time, helping developers quickly locate problematic modules and greatly improving development efficiency. Furthermore, this technology enables convenient system deployment through computer-readable storage media; developers only need to deploy the program to the development environment to use it, without complex configuration and debugging. Attached Figure Description
[0023] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation
[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0025] like Figure 1As shown, this embodiment of the invention provides a software development verification system based on a large model, including a neural symbol fusion unit, a cognitive alignment unit, and an evolutionary contract unit: the neural symbol fusion unit calculates the semantic distance function between code and requirements through a dual encoder structure.
[0026] in: The feature matrix of the abstract syntax tree nodes. This represents the total number of AST nodes. For a graph neural network-based AST encoder, the output dimension is... , For the demand encoder, a pre-trained BERT model is used; Cognitive alignment units generate a requirement-code mismatch index through a cross-modal attention mechanism: in: For trainable alignment matrices, This is a confidence correction factor dynamically calculated based on historical validation accuracy. Evolutionary contract unit generates API boundary constraint set:
[0027] , This is the mean squared error loss function between the predicted and expected values.
[0028] The neural symbol fusion unit includes an invariant confidence calculation model:
[0029] in: For effective AST path coverage ( ), To predetermine the distribution entropy of the error type, To control the depth of the flow graph ( ).
[0030] The bidirectional attention mechanism of the cognitive alignment unit is implemented as follows:
[0031] in: , , , Trainable parameter matrix : Position encoding generated by the sine function.
[0032] The conditions under which an evolutionary contract unit performs counterfactual corrections are:
[0033] in: These are 128-dimensional semantic vectors representing demand samples, error samples, and correction samples, respectively. : Tolerance coefficient, the value of which is positively correlated with the complexity of the software module.
[0034] A reliable quantification subsystem, embedded in the system of claim 1, includes a dynamic scoring module and a regression test generator: the dynamic scoring module is as follows:
[0035] The weighting coefficients are calculated using the following formula: The regression test generator is:
[0036] This is for adaptive perturbation step size.
[0037] A reliable quantification subsystem employs the following complexity control strategy for test case generation:
[0038] in: For the input parameter dimension, For symbolic execution depth ( ), This represents the number of historical verification failures for the current module.
[0039] A development environment integration device, integrating the system of claim 1, includes a real-time heatmap generator and a color encoding engine: The real-time heatmap generator is:
[0040] in: The weights are based on the module's real-time usage duration and are dynamic. Static weights configured for administrators; The color encoding engine is: .
[0041] A computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the operation of any one of claims 1 to 7, specifically comprising: Execute neural symbol fusion unit instructions: calculate and output SEMDIST and ; Schedule the TrustScore quantification subsystem: generate TrustScore and trigger regression test generation; Call the heatmap rendering interface: Visualize the RGB verification status in the integrated development environment (IDE).
[0042] The implementation of this invention first requires the construction of a software development verification system based on a large model. This system consists of three core modules: a neural symbolic fusion unit, a cognitive alignment unit, and an evolutionary contract unit. The neural symbolic fusion unit is responsible for semantically aligning the abstract syntax tree with the requirement text. It uses a dual-encoder structure to encode both the code's syntax tree and the requirement text into vector representations and calculates the semantic distance between them to determine whether the code accurately implements the requirement intent. The cognitive alignment unit uses a cross-modal attention mechanism to bidirectionally align code features with requirement features, calculates a mismatch index, identifies semantic deviations between the code and the requirements, and generates correction suggestions. The evolutionary contract unit analyzes the input and output boundary conditions of the code to generate a set of API boundary constraints, ensuring that the code behaves as expected under different input conditions. To improve the system's credibility, this invention also includes a credibility quantification subsystem. This system calculates the code's credibility score through a dynamic scoring module and generates regression test cases based on historical verification results, ensuring that the code remains correct after modification. To facilitate use by developers, this technology also provides a development environment integration device. This device displays the verification results in real-time in the form of a heatmap within the integrated development environment, helping developers quickly locate problematic modules. Finally, all the functions of this technology can be implemented through a program stored on a computer-readable storage medium, which developers can use simply by deploying it to the development environment. Through the above steps, this technology can effectively improve the verification efficiency of software development, reduce the cost of manual verification, and ensure a high degree of consistency between the code and the requirements.
[0043] The advantages of this invention lie in its comprehensive improvement of the verification efficiency and accuracy of software development. Firstly, the neural symbolic fusion unit achieves semantic alignment between code and requirements, accurately identifying whether the code accurately implements the intended requirements, avoiding errors caused by semantic comprehension biases in traditional verification methods. Secondly, the cognitive alignment unit, through a cross-modal attention mechanism, effectively captures semantic deviations between code and requirements and generates targeted correction suggestions, significantly improving code correction efficiency. The evolutionary contract unit analyzes the input and output boundary conditions of the code to generate a set of API boundary constraints, ensuring that the code behaves as expected under different input conditions, improving code robustness and reliability. Finally, the trustworthiness quantification subsystem, through a dynamic scoring module and a regression test generator, can evaluate the trustworthiness of the code in real time and automatically generate regression tests after code modifications. For example, this technology ensures that the code remains correct throughout the iteration process. The integrated development environment (IDE) displays the verification results in real-time as a heatmap, helping developers quickly locate problematic modules and greatly improving development efficiency. Furthermore, this technology enables convenient system deployment through computer-readable storage media; developers only need to deploy the program to the development environment to use it, without complex configuration and debugging. Finally, through automated verification and intelligent correction suggestions, this technology significantly reduces the cost of manual verification, improving the overall efficiency and quality of software development. In summary, this technology has significant advantages in semantic alignment, deviation identification, boundary constraints, credibility assessment, visual feedback, and convenient deployment, providing comprehensive verification support for software development, ensuring a high degree of consistency between code and requirements, and significantly improving development efficiency and code quality.
[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A software development verification system based on a large model, characterized in that, It includes a neural symbol fusion unit, a cognitive alignment unit, and an evolutionary contract unit: the neural symbol fusion unit calculates the semantic distance function of code-requirement through a dual encoder structure. ; The cognitive alignment unit generates a requirement-code mismatch index through a cross-modal attention mechanism: ; The evolutionary contract unit generates a set of API boundary constraints: .
2. The software development verification system based on a large model according to claim 1, characterized in that: The neural symbol fusion unit includes an invariant confidence calculation model: .
3. The software development verification system based on a large model according to claim 1, characterized in that: The bidirectional attention mechanism of the cognitive alignment unit is implemented as follows: .
4. The software development verification system based on a large model according to claim 1, characterized in that: The condition under which the evolutionary contract unit performs counterfactual correction is: .
5. A reliable quantification subsystem, embedded in the system of claim 1, characterized in that, It includes a dynamic scoring module and a regression test generator: the dynamic scoring module is as follows: ; The weighting coefficients are calculated using the following formula: ; The regression test generator is: .
6. A trustworthy quantification subsystem according to claim 5, characterized in that: The complexity control strategy for generating the test cases is as follows: .
7. A development environment integration device, integrating the system of claim 1, characterized in that, Includes a real-time heatmap generator and a color encoding engine: The real-time heatmap generator is: ; The color encoding engine is: .
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the operation of the system according to any one of claims 1 to 7, specifically including: Execute neural symbol fusion unit instructions: calculate and output SEMDIST and ; Schedule the TrustScore quantification subsystem: generate TrustScore and trigger regression test generation; Call the heatmap rendering interface: Visualize the RGB verification status in the integrated development environment.