A software testing analysis method and system based on multi-agent cooperation and a storage medium

Abstract

The application discloses a kind of software test analysis method, system and storage medium based on multi-agent cooperation.The method comprises: receiving test object data and carrying out semantic analysis and dependency analysis, and is disassembled into multiple test subtasks;According to the type of subtask, different agents are called to analyze and process, and subtask results are generated;The results of multiple agents are consistent, conflict detection and fusion processing are checked, and the test analysis result after fusion is generated;Based on the confidence of fusion result calculation and output structured test data.The system includes input module, task disassembly module, agent cooperation module, result fusion module, confidence calculation module and output module.The application effectively improves the consistency, stability and reliability of test analysis result through multi-agent cooperation and result fusion mechanism, and is suitable for complex software system test analysis and quality control scene.

Description

Technical Field

[0001] This invention relates to the fields of computer software engineering and artificial intelligence technology, and in particular to a software testing analysis and quality control method, system and storage medium based on multi-agent collaboration. Background Technology

[0002] In the software development process, test analysis and quality control are crucial for ensuring the stability and reliability of software systems. As the scale and complexity of software systems increase, the test objects typically include a large number of requirements documents, design specifications, code change information, and historical defect data. Test analysis tasks are characterized by large data volumes, complex structures, and numerous dependencies.

[0003] In existing technologies, test analysis primarily relies on human experience or automated analysis methods based on a single model. Because a single analytical entity is susceptible to incomplete input information or unstable model output when handling complex test objects, test analysis results often suffer from inconsistent results, unresolved conflicts, and difficulty in assessing reliability. Furthermore, while artificial intelligence (AI) technology can assist in generating test analysis results in some scenarios with its application in the testing field, existing solutions typically lack a systematic collaborative processing mechanism for test analysis tasks. This makes it difficult to effectively integrate and verify multi-source analysis results under engineering conditions, thus affecting the stability and reliability of the test analysis results.

[0004] Therefore, it is necessary to provide a new technical solution to address the problems of insufficient consistency, conflicting results, and difficulty in quantifying reliability in existing test analysis processes. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention introduces a multi-agent collaborative processing mechanism to decompose, collaboratively process, fuse results, and evaluate confidence levels for test analysis tasks, thereby improving the consistency, stability, and reliability of test analysis results.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution: a software testing and analysis method based on multi-agent collaboration, comprising the following steps:

[0009] S1. Receive test object data, wherein the received test object data includes at least one of the following: requirement document, requirement change information, design specification, code change information, or historical defect data;

[0010] S2. Perform semantic parsing and dependency analysis on the test object data, and break down the test analysis task into multiple sub-tasks;

[0011] S3. Based on the subtask type, invoke multiple agents with different processing strategies to analyze and process the subtask.

[0012] S4. Each agent performs analysis and processing on its corresponding sub-task and generates sub-task analysis result data.

[0013] S5. Perform consistency verification, conflict detection and fusion processing on the analysis results data of multiple sub-tasks to generate fused test analysis results data;

[0014] S6. Based on the consistency, evidence completeness, and historical stability of the multi-agent analysis results during the fusion process, calculate the confidence index of the test analysis results.

[0015] S7. Output the test analysis results data, confidence index, and corresponding structured report.

[0016] As a preferred embodiment, the subtasks in step S2 include at least one of the following: requirements analysis subtask, test strategy planning subtask, test case generation or adjustment subtask, test impact analysis subtask, and defect risk analysis subtask.

[0017] As a preferred embodiment, the task decomposition in step S2 is based on at least one of the following: test phase type, scope of requirement change, and scope of system impact.

[0018] As a preferred embodiment, the intelligent agent in step S3 includes at least one of the following:

[0019] The requirement parsing agent is used to extract business entities and logical rules from test objects;

[0020] A strategy planning agent is used to determine the test scope, test type, and test priority.

[0021] A test case generation agent is used to generate test steps and test data based on business logic.

[0022] A test case matching agent is used to match generated test cases with a historical test case library based on similarity.

[0023] The test execution agent is used to execute automated test scripts and capture execution logs;

[0024] A defect analysis agent is used to analyze anomaly logs and locate the root cause of defects.

[0025] A report-generating agent is used to summarize and analyze data to generate structured test reports.

[0026] As a preferred option, the consistency verification in step S4 adopts a hierarchical multi-dimensional verification algorithm, including at least one of the following levels: requirement relevance, test scope mapping, test case coverage dimension, and defect risk level.

[0027] As a preferred embodiment, the hierarchical multi-dimensional verification algorithm first verifies semantic consistency by calculating semantic similarity for each level, then verifies the consistency of key parameters by rule matching, and finally uses a weighted voting mechanism to comprehensively determine the overall consistency.

[0028] As a preferred embodiment, the conflict detection in step S4 includes at least one of the following conflict types: logical conflict, parameter conflict, and range conflict;

[0029] The priority of conflict resolution rules is as follows: business rules first, majority consensus first, and expert experience first.

[0030] As a preferred embodiment, the confidence index is calculated based on the results of the consistency verification, the consistency coefficient, and the completeness of evidence supporting the test analysis results. The calculation formula is as follows:

[0031] Conf = α×A + β×B + γ×C

[0032] Where A is the consistency confidence, B is the agent confidence weight, and C is the coverage confidence; α, β, and γ are weight coefficients, and satisfy α+β+γ=1.

[0033] This invention also provides a software testing and analysis system based on multi-agent collaboration, comprising:

[0034] The input module is used to receive test object data;

[0035] The task decomposition module is used to parse the test object data and decompose the task into subtasks;

[0036] The agent collaboration module includes multiple agent units with different processing strategies for processing the subtasks in parallel or serially.

[0037] The result fusion module is used to perform conflict detection and fusion on the results output by each intelligent agent unit.

[0038] The confidence level calculation module is used to quantitatively assess the reliability of the analysis results;

[0039] The output module is used to output the final test analysis results and confidence levels.

[0040] The present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the function of a software testing and analysis method based on multi-agent collaboration.

[0041] (III) Beneficial Effects

[0042] Compared with existing technologies, this invention provides a software testing and analysis method, system, and storage medium based on multi-agent collaboration, which has the following beneficial effects:

[0043] I. This invention breaks down the test task into multiple sub-tasks through a multi-agent collaborative processing mechanism, which are processed by different professional agents. Then, through a hierarchical multi-dimensional consistency verification and conflict resolution mechanism, the results of multi-source analysis are integrated, which significantly improves the consistency and stability of the test analysis results. It is especially suitable for testing scenarios of complex software systems.

[0044] Second, by introducing various specialized intelligent agents (such as requirement analysis, strategy planning, test case generation, defect analysis, etc.), this invention realizes the automated decomposition and collaborative processing of test analysis tasks, which greatly reduces manual intervention and improves analysis efficiency. At the same time, through collaboration and knowledge sharing among intelligent agents, the overall intelligence level of test analysis is improved.

[0045] Third, this invention achieves task decomposition through semantic parsing and dependency analysis, and combines this with a multi-agent collaborative processing mechanism, comprehensively covering all aspects of test analysis, adapting to the test analysis needs of complex software systems, and effectively supporting software quality control. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the testing and analysis method of the present invention;

[0047] Figure 2 This is a schematic diagram of the test and analysis system architecture of the present invention;

[0048] Figure 3 This is a schematic diagram illustrating the logic of test result fusion and confidence calculation in this invention;

[0049] Figure 4 This is a timing diagram illustrating the multi-agent collaborative interaction of the present invention. Detailed Implementation

[0050] To better understand the purpose, structure, and function of this invention, the following description, in conjunction with the accompanying drawings and specific embodiments, will further illustrate a software testing and analysis method, system, and storage medium based on multi-agent collaboration.

[0051] Example 1

[0052] In traditional software test analysis, testers typically need to manually read and understand various test object data (such as requirements documents, design specifications, code change information, etc.) to extract key test points. This method is inefficient, easily affected by differences in personnel experience, and makes it difficult to guarantee the completeness and accuracy of the extraction. Especially when the test object is unstructured text, manual extraction is prone to missing key information, resulting in a weak foundation for subsequent test analysis.

[0053] refer to Figure 1-4 To address the aforementioned problems, embodiments of this specification propose an automated method for inputting and preprocessing test objects, comprising:

[0054] I. Test Object Input and Preprocessing

[0055] In one embodiment, requirement change test analysis uses a requirement change document as the test object input. The task decomposition module breaks down the test analysis task into multiple sub-tasks, which are then analyzed and processed by different intelligent agents. The result fusion module performs consistency verification and conflict resolution on the multiple analysis results, generates a fused test analysis result, calculates the confidence level, and outputs it. The system receives the requirement change document as test object data and performs the following preprocessing steps:

[0056] 1. Analyze the requirement change document and extract the core chapter content (such as business background, functional requirements, non-functional requirements, etc.).

[0057] 2. Leveraging AI big data models, and combining information such as project requirements, product chain, and testing points, test points are automatically compiled.

[0058] 3. Improve and supplement the test points, and finally output the standardized test analysis input data in JSON text format.

[0059] Through the above preprocessing, the system can quickly and accurately convert unstructured or semi-structured test object data into structured and standardized input, laying the foundation for subsequent task decomposition and agent processing.

[0060] II. Task Breakdown

[0061] In traditional test analysis, the breakdown of test tasks typically relies on the tester's experience. This breakdown process is highly subjective and struggles to systematically consider multiple dimensions such as test phase, scope of changes, and system impact. This can lead to incomplete or unreasonable task breakdowns, affecting the coverage and accuracy of subsequent test analysis.

[0062] To address the aforementioned issues, this specification proposes a task decomposition method based on semantic parsing and dependency analysis.

[0063] In one embodiment, the task decomposition module identifies key entities and logical relationships within the test objects through semantic parsing based on standardized test points, and determines the connections between various test tasks through dependency analysis. The decomposed subtasks include, but are not limited to: requirement analysis subtasks, test strategy planning subtasks, test case generation subtasks, test case matching subtasks, test execution subtasks, and defect analysis subtasks. The decomposition is based on a comprehensive consideration of the test phase type, the scope of requirement changes, and the system's impact, ensuring the comprehensiveness and rationality of the decomposition.

[0064] III. Multi-agent cooperative processing

[0065] In traditional test analysis, each testing phase is typically handled independently by different personnel or tools, lacking collaboration and information sharing, which easily leads to information silos. For example, the requirements analyst may not understand the specifics of the test execution, and the test execution personnel may not be clear about the full impact of requirement changes, resulting in biased or conflicting test analysis results.

[0066] To address the aforementioned issues, this specification proposes a multi-agent collaborative processing mechanism. In one embodiment, the system schedules corresponding specialized agents for processing based on the type of the decomposed sub-tasks:

[0067] The strategy planning agent generates corresponding test strategies based on the requirements analysis results, determining the test scope, test type, and test priority. The implementation method is as follows:

[0068] (1) Extract test risk points and test plans from historical test data and condense them into a knowledge base;

[0069] (2) Based on the standardized output test points, expand them in the form of mind maps to spread them out;

[0070] (3) Using the AI ​​big model, the test risk points are associated and supplemented in the form of knowledge base matching, and finally a complete test analysis mind map is formed.

[0071] Test Case Generation Agent: Based on the test strategy, the agent generates test analysis, organizes the test analysis content, and forms test titles with links from the root node to the leaf node. With the help of the AI ​​large model, test cases are output in JSON text format. The test case content includes: title, preconditions, steps, expectations, test case attributes, test case priority, test case tags, etc., covering various types of functional test cases such as normal, abnormal, and boundary value.

[0072] The test case matching agent's core idea is to create a knowledge base for automated test cases at the system level. Based on the semantics of each node in the test analysis, it performs knowledge base matching, while simultaneously leveraging metadata during retrieval to accurately determine the matching scope. The final output includes a set of matched automated test cases and a set of nodes not associated with any automated test cases.

[0073] Test execution agent: Obtain the automated test case IDs output by the test case matching agent, create execution sets at the system level, automatically execute test operations, and collect test execution results.

[0074] Defect Analysis Agent: Analyzes the execution results of the test execution agent, identifies potential defect information, and determines the scope of the defect's impact.

[0075] Each intelligent agent processes data based on a pre-defined test analysis strategy and domain knowledge, and achieves information sharing and collaboration through data flow, ensuring the continuity and consistency of the test analysis process.

[0076] IV. Result Fusion Processing

[0077] In traditional test analysis, when multiple analysis results differ or conflict, integration typically relies on manual judgment, which is inefficient and prone to introducing subjective bias. The lack of a systematic conflict detection and resolution mechanism makes it difficult to guarantee the reliability and consistency of test analysis results.

[0078] To address the aforementioned issues, this specification proposes a hierarchical, multi-dimensional result fusion method. The result fusion module processes analysis results from multiple test agents, including: performing consistency checks on the analysis results, identifying differences or conflicts between different analysis results, and resolving conflicting results according to preset fusion rules.

[0079] In one embodiment, the result fusion module performs the following steps:

[0080] a. Consistency Verification: A hierarchical, multi-dimensional verification algorithm is adopted, decomposing the analysis results into four core levels: requirement relevance, test scope mapping, test case coverage dimension, and defect risk level. For each level, semantic similarity is first verified by calculating semantic similarity (based on feature vectors extracted from the BERT pre-trained model, and cosine similarity algorithm is used to quantify text similarity).

[0081] Then, through rule matching verification, the consistency of key parameters at each level (such as module ID of the test scope, priority label of the test case, and severity level of the defect) is compared, and a similarity threshold (such as semantic similarity ≥ 0.8, parameter matching rate ≥ 90%) is set to determine the consistency of a single level.

[0082] Finally, a weighted voting mechanism was adopted (the weights of each level were preset according to the importance of the business, such as a defect risk level weight of 0.4 and a test scope mapping weight of 0.3) to comprehensively determine the overall consistency of the results of the multi-agent system.

[0083] b. Difference and Conflict Identification: Based on the hierarchical results of consistency verification, a difference tree comparison algorithm is used to construct the result difference tree, traversing each level node from top to bottom to locate the difference location (such as module-level difference, use case-level difference, parameter-level difference).

[0084] At the same time, a conflict rule base is introduced to define three core conflict types (logical conflict: such as agent A determines that a certain module needs to be added to the test, while agent B determines that no test is needed).

[0085] Parameter conflict: such as the defect severity being judged as "high" or "medium";

[0086] Scope conflicts: such as contradictions between the inclusion and exclusion of test case coverage. Explicit conflicts are identified by rule matching and threshold judgment, while implicit conflicts (such as indirect scope conflicts caused by misunderstandings of requirements) are identified by contextual analysis.

[0087] c. Conflict resolution: Conflict resolution is performed according to the preset fusion rules. The priority of the rules is as follows from high to low: business rules first, majority consensus first, and expert experience first.

[0088] For logical conflicts, the system invokes a pre-defined business rule base (such as "when a requirement change involves a core transaction module, the testing scope must be expanded") to make a decision.

[0089] To address parameter conflicts, a weighted average method (based on the agent's historical accuracy) is used to calculate the optimal parameter values.

[0090] To address scope conflicts, a source comparison is performed using the original requirements document to select the result that best matches the requirements description.

[0091] In this embodiment, consistency verification employs a hierarchical, multi-dimensional verification algorithm. The analysis results are broken down into four core levels: requirement relevance, test scope mapping, test case coverage dimension, and defect risk level. Semantic similarity calculation and rule matching are used to verify consistency at each level, and a weighted voting mechanism is ultimately used to determine overall consistency. Difference and conflict identification uses a difference tree comparison algorithm to locate differences and combines this with a conflict rule base to identify types such as logical conflicts, parameter conflicts, and scope conflicts. Conflict resolution is handled according to preset fusion rules (business rules first, majority consensus first, expert experience first) to resolve conflict results, ensuring the rationality and consistency of the final fusion result.

[0092] Through the above processing, the integrated test analysis results are generated, which reflect the comprehensive impact of requirement changes on the test scope, test cases, and potential defects.

[0093] V. Confidence Calculation and Output

[0094] Traditional test analysis results often lack reliability assessment, making it difficult for testers to judge the credibility of the analysis results. This often requires a complete review or blind trust, affecting test efficiency and quality.

[0095] To address the aforementioned issues, this specification proposes a multi-level weighted fusion confidence model. In this embodiment, after generating the fused test analysis results, the system can also assess the reliability of the test analysis results based on the consistency of the multi-agent analysis results, generating corresponding confidence information.

[0096] Its confidence level calculation model adopts a multi-level weighted fusion confidence level model, and the calculation formula is as follows:

[0097] Conf = α×A + β×B + γ×C

[0098] Wherein, A is the consistency confidence score, calculated based on the hierarchical consistency verification score of the multi-agent results, and is the weighted sum of the consistency scores at each level, with the weights consistent with the weights of the consistency verification levels; B is the agent credibility weight, calculated based on the historical task accuracy of each agent participating in the analysis, with agents having higher historical accuracy having a larger weight in their results; C is the coverage confidence score, which quantifies the completeness of the analysis results in covering the change points of the requirements, and is the ratio of the number of covered change points to the total number of change points; α, β, and γ are model weight coefficients, dynamically adjusted according to the business scenario, with default values ​​of α=0.5, β=0.3, and γ=0.2, and satisfying α+β+γ=1.

[0099] Confidence assessment system: The confidence score is divided into 5 levels, corresponding to different reliability levels and manual review strategies, as shown in Table 1 below:

[0100] Table 1

[0101] Confidence level Score range Reliability level Manual review strategy S-Class [0.9,1.0] Extremely high reliability No manual review required, results are output directly. Grade A [0.8,0.9) High reliability Sampling review (review rate ≤ 10%) Grade B [0.7,0.8) Medium reliability Partial review (review rate 30%-50%) Class C [0.6,0.7) Lower reliability Comprehensive review (review rate ≥ 80%) Class D [0,0.6) Low reliability Re-execute the multi-agent analysis process

[0102] This confidence level information characterizes the stability of the test analysis results and can serve as a reference for determining whether further manual review is needed. Ultimately, the system output includes test analysis conclusions, test case adjustment information, defect analysis results, and the corresponding test report.

[0103] Example 2

[0104] Traditional software testing and analysis systems typically employ a monolithic architecture or a simple modular design, resulting in high coupling between functional modules and a lack of flexibility and scalability. Specifically:

[0105] Single functional module: Traditional systems usually integrate test analysis functions into a single module, which makes it difficult to cope with complex and ever-changing test scenarios.

[0106] Lack of collaboration mechanism: The lack of effective collaboration and information sharing mechanisms between functional modules leads to a fragmented testing and analysis process and low information transmission efficiency.

[0107] Lack of reliability assessment: Traditional systems lack a mechanism for assessing the reliability of analysis results, making it difficult for users to judge the reliability of the analysis results.

[0108] To address the aforementioned issues, this specification proposes a software testing and analysis system based on multi-agent collaboration. This system employs a modular and loosely coupled design philosophy, and through agent collaboration mechanisms and result fusion algorithms, achieves automation, collaboration, and intelligence in the testing and analysis process.

[0109] refer to Figure 2 The following will provide a detailed explanation of each module in conjunction with the accompanying drawings:

[0110] I. Input Module

[0111] Traditional test analysis systems typically require users to input test data in a fixed format, and have limited ability to process unstructured or semi-structured data. Users need to spend a lot of time on data preparation and format conversion, and the completeness and accuracy of the input data are difficult to guarantee.

[0112] In this embodiment, the input module is designed to support a unified access and preprocessing platform for multi-source heterogeneous data. In one embodiment, the input module includes the following functional components:

[0113] Data interface adapter: Supports access to various data formats and protocols, including but not limited to document files (Word, PDF), database connections, API interfaces, version control systems, etc.

[0114] Data parsing engine: Based on natural language processing and machine learning technologies, it automatically parses the structure and content of test object data and extracts key information.

[0115] Standardized converter: Converts the parsed data into a standardized format (such as JSON) that is consistent with the system's internal standards, ensuring consistency in subsequent processing.

[0116] Quality validator: Automatically validates the integrity and quality of input data, identifies issues such as missing fields and format errors, and provides repair suggestions.

[0117] Through the above design, the input module can automatically process various test object data, reduce manual intervention, and improve data input efficiency and accuracy.

[0118] II. Task Decomposition Module

[0119] Traditional test analysis systems typically treat the entire test analysis task as a whole, lacking the ability to fine-grainedly decompose the task. This makes it difficult for the system to cope with complex test scenarios and unable to fully utilize parallel processing capabilities.

[0120] In this embodiment, the task decomposition module employs a task decomposition algorithm based on semantic parsing and dependency analysis. In one embodiment, this module includes the following core components:

[0121] Semantic parser: Based on pre-trained deep learning models (such as BERT, GPT, etc.), it analyzes the semantic information of test object data and identifies key entities, relationships and intentions.

[0122] Dependency Analyzer: Constructs a dependency graph between elements in the test object data to analyze the impact of changes and test coverage requirements.

[0123] Task decomposition engine: Based on semantic parsing and dependency analysis results, the overall test analysis task is broken down into multiple relatively independent sub-tasks. The decomposition strategy comprehensively considers multiple dimensions such as test phase type, scope of requirement changes, and system impact.

[0124] Task scheduler: Assigns priority and resource requirements to each subtask, providing a basis for subsequent agent scheduling.

[0125] Through the task decomposition module, complex test analysis tasks are broken down into multiple subtasks that can be processed in parallel, which greatly improves the system's processing efficiency and flexibility.

[0126] III. Intelligent Agent Collaboration Module

[0127] Traditional test analysis systems typically employ a single algorithm or model to handle all test analysis tasks, making it difficult to adapt to the specific needs of different testing scenarios. Even when multiple algorithms are used, there is a lack of effective collaboration mechanisms, hindering information sharing and collaborative decision-making among the algorithms.

[0128] In this embodiment, the agent collaboration module adopts a multi-agent system architecture, enabling specialized agents to work collaboratively. In one embodiment, this module includes the following core components:

[0129] Agent Manager: Responsible for agent registration, discovery, status monitoring, and lifecycle management. Supports dynamic loading and unloading of agents, providing good extensibility.

[0130] Agent scheduler: Dynamically schedules suitable agents to execute tasks based on subtask type and resource requirements. The scheduling strategy comprehensively considers agent capability matching, current load, and historical performance.

[0131] Intelligent Agent Communication Bus: Provides standardized communication interfaces and protocols, supporting asynchronous message passing and data sharing between intelligent agents. The communication bus adopts a publish-subscribe model, reducing the coupling between intelligent agents.

[0132] IV. Results Fusion Module

[0133] Traditional test analysis systems typically output the results of each analysis module directly, lacking the ability to fuse results from multiple sources. When conflicts arise in the outputs of different modules, the system cannot automatically resolve these conflicts, requiring manual intervention and reducing the system's automation level and efficiency.

[0134] In this embodiment, the result fusion module employs a hierarchical multi-dimensional fusion algorithm to achieve automatic fusion and conflict resolution of multi-source analysis results. In one embodiment, this module includes the following core components:

[0135] Consistency Verifier: Employs a hierarchical, multi-dimensional verification algorithm to break down the analysis results into multiple core levels (such as requirement relevance, test scope mapping, test case coverage dimension, and defect risk level), and performs consistency verification for each level.

[0136] Conflict detector: Based on the difference tree comparison algorithm and conflict rule base, it identifies differences and conflicts between analysis results, including types such as logical conflicts, parameter conflicts and range conflicts.

[0137] Conflict resolver: Resolves conflict outcomes based on preset fusion rules. These rules are prioritized as follows: business rules first, majority consensus first, and expert experience first.

[0138] The result fusion processor merges the conflict-resolved analysis results to generate a unified test analysis result. The fusion algorithm comprehensively considers the analysis results of each agent, confidence weights, and historical performance.

[0139] Through the results fusion module, the system can automatically handle the differences and conflicts in multi-source analysis results and generate consistent and reliable test analysis results.

[0140] V. Confidence Calculation Module

[0141] Traditional test analysis systems typically do not provide credibility assessments of their analysis results, making it difficult for users to determine the reliability of these results. Even if some systems offer simple credibility metrics, they are often calculated based on a single dimension and cannot comprehensively reflect the reliability of the analysis results.

[0142] In this embodiment, the confidence calculation module employs a multi-level weighted fusion confidence model to provide a scientific reliability assessment of the test analysis results. In one embodiment, this module includes the following core components:

[0143] Confidence Calculation Engine: Calculates the confidence level of test analysis results based on the following formula:

[0144] Conf = α×A + β×B + γ×C

[0145] Where A is the consistency confidence score (calculated based on the hierarchical consistency verification score of the multi-agent results), B is the agent credibility weight (calculated based on the weighted average of the historical task accuracy of each agent participating in the analysis), and C is the coverage confidence score (quantifying the completeness of the analysis results in covering the change points of the requirements).

[0146] Confidence evaluator: The confidence score is divided into multiple levels (such as S, A, B, C, and D), each level corresponding to a different level of reliability and a manual review strategy.

[0147] Historical performance tracker: continuously tracks the historical task accuracy of each agent, dynamically adjusts the agent's confidence weight, and ensures the accuracy of confidence calculation.

[0148] Confidence Visualizer: Presents the confidence calculation results to users in an intuitive way, including confidence score, level description, and review suggestions.

[0149] Through the confidence calculation module, the system provides users with a scientific basis for reliability assessment, helping them to better understand and trust the test analysis results.

[0150] VI. Output Module

[0151] In this embodiment, the output module is designed as a flexible output platform that supports multi-format and multi-dimensional output.

[0152] Through the above implementation methods, this invention achieves automation, collaboration, and intelligence in the test analysis process, effectively improving the consistency, stability, and reliability of test analysis results. It is suitable for test analysis and quality control scenarios in complex software systems. Compared to test analysis methods based on a single analysis model or human experience, this invention, through multi-agent collaboration and result fusion mechanisms, effectively reduces the probability of inconsistent test analysis results and improves the stability and reliability of test analysis results under complex requirement change scenarios.

[0153] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A software testing and analysis method based on multi-agent collaboration, characterized in that, Includes the following steps: S1. Receive test object data, wherein the received test object data includes at least one of the following: requirement document, requirement change information, design specification, code change information, or historical defect data; S2. Perform semantic parsing and dependency analysis on the test object data, and break down the test analysis task into multiple sub-tasks; S3. Based on the subtask type, invoke multiple agents with different processing strategies to analyze and process the subtask. S4. Each agent performs analysis and processing on its corresponding sub-task and generates sub-task analysis result data. S5. Perform consistency verification, conflict detection and fusion processing on the analysis results data of multiple sub-tasks to generate fused test analysis results data; S6. Based on the consistency, evidence completeness, and historical stability of the multi-agent analysis results during the fusion process, calculate the confidence index of the test analysis results. S7. Output the test analysis results data, confidence index, and corresponding structured report.

2. The software testing and analysis method based on multi-agent collaboration according to claim 1, characterized in that, The subtasks mentioned in step S2 include at least one of the following: requirements analysis subtask, test strategy planning subtask, test case generation or adjustment subtask, test impact analysis subtask, and defect risk analysis subtask.

3. The software testing and analysis method based on multi-agent collaboration according to claim 2, characterized in that, The task breakdown in step S2 is based on at least one of the following: test phase type, scope of requirement changes, and scope of system impact.

4. The software testing and analysis method based on multi-agent collaboration according to claim 1, characterized in that, The intelligent agent mentioned in step S3 includes at least one of the following: The requirement parsing agent is used to extract business entities and logical rules from test objects; A strategy planning agent is used to determine the test scope, test type, and test priority. A test case generation agent is used to generate test steps and test data based on business logic. A test case matching agent is used to match generated test cases with a historical test case library based on similarity. The test execution agent is used to execute automated test scripts and capture execution logs; A defect analysis agent is used to analyze anomaly logs and locate the root cause of defects. A report-generating agent is used to summarize and analyze data to generate structured test reports.

5. The software testing and analysis method based on multi-agent collaboration according to claim 1, characterized in that, The consistency verification in step S4 adopts a hierarchical multi-dimensional verification algorithm, including at least one of the following levels: requirement relevance, test scope mapping, test case coverage dimension, and defect risk level.

6. The software testing and analysis method based on multi-agent collaboration according to claim 5, characterized in that, In the hierarchical multi-dimensional verification algorithm, semantic consistency is first verified by semantic similarity calculation for each level, then the consistency of key parameters is verified by rule matching, and finally the overall consistency is determined by a weighted voting mechanism.

7. A software testing and analysis method based on multi-agent collaboration according to claim 5, characterized in that, The conflict detection in step S4 includes at least one of the following conflict types: logical conflict, parameter conflict, and range conflict. The priority of conflict resolution rules is as follows: business rules first, majority consensus first, and expert experience first.

8. The software testing and analysis method based on multi-agent collaboration according to claim 1, characterized in that, The confidence index is calculated based on the results of the consistency verification, the consistency coefficient, and the completeness of evidence supporting the test analysis results. Its calculation formula is as follows: Conf = α×A + β×B + γ×C Where A is the consistency confidence, B is the agent confidence weight, and C is the coverage confidence; α, β, and γ are weight coefficients, and satisfy α+β+γ=1.

9. A software testing and analysis system based on multi-agent collaboration, characterized in that, include: The input module is used to receive test object data; The task decomposition module is used to parse the test object data and decompose the task into subtasks; The agent collaboration module includes multiple agent units with different processing strategies for processing the subtasks in parallel or serially. The result fusion module is used to perform conflict detection and fusion on the results output by each intelligent agent unit. The confidence level calculation module is used to quantitatively assess the reliability of the analysis results; The output module is used to output the final test analysis results and confidence levels.

10. 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 method described in any one of claims 1 to 8.

Images