Interface test case generation method and system based on dual-channel relational graph convolutional network model and Bert language model
By collaboratively identifying interface dependencies using a dual-channel relational graph convolutional network model and a BERT language model, a directed dependency graph is generated, and test scripts are automatically generated. This solves the problem of insufficient accuracy and coverage in the generation of interface test cases in existing technologies, and achieves efficient interface testing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TONGJI UNIV
- Filing Date
- 2026-01-22
- Publication Date
- 2026-04-28
AI Technical Summary
Existing technologies struggle to effectively combine interface structure and semantic information when generating interface test cases, resulting in insufficient accuracy and coverage. Furthermore, existing tools have limitations in handling complex interface dependencies.
A dual-channel relational graph convolutional network model is used in conjunction with the BERT language model to identify dependencies between interfaces through structural and semantic channels. A directed dependency graph is generated by combining a multi-round filtering and fusion strategy, and test scripts are automatically generated.
It improves the accuracy and logical rationality of identifying dependencies between interfaces, supports automated processes from dependency analysis to test script generation, improves test efficiency and coverage, and controls computing resources and time overhead.
Smart Images
Figure CN121935159A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical fields of artificial intelligence and software testing, specifically to a method and system for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model. Background Technology
[0002] With the increasing popularity of web applications and microservice architectures, REST APIs have become a common way for systems to interact. In software testing, automated generation of interface test cases helps improve testing efficiency and coverage. However, existing methods still face some challenges: on the one hand, understanding the interface call logic relies to some extent on testers' manual judgment of the business scenario; on the other hand, current automated tools still need to improve the accuracy of generating test sequences and have shortcomings in semantic consistency. Traditional test case generation methods, such as those based on search or graph models, have limitations in handling complex interface dependencies; while methods based on large language models, which have emerged in recent years, have shown some potential in semantic understanding, they are still insufficient in modeling the interface structure logic. Therefore, it is necessary to explore a lightweight and efficient test case generation method that can take into account both interface structure and semantic information to achieve a better balance between accuracy, coverage, and scalability. Summary of the Invention
[0003] This invention is made to solve the above problems, and aims to provide a method and system for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model.
[0004] This invention provides a method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model. The method comprises the following steps: S1: Data input processing step, parsing the input interface definition file, extracting key metadata for each interface, and processing the key metadata to form input data for dependency identification; S2: Dual-channel collaborative identification step, based on the input data, identifying and outputting the predicted probabilities of inter-interface dependencies through structural and semantic channels respectively. Inter-interface dependencies include: unrelated, positive, and negative dependencies; S3: Relationship fusion step, fusing the predicted probabilities of inter-interface dependencies output from the structural and semantic channels and verifying directional consistency to generate a directed dependency graph; S4: Test case generation step, automatically generating test scripts adapted to the target testing framework based on the directed dependency graph and key metadata; S5: Test feedback step, executing the test scripts, evaluating the execution results based on the interface response status and assertion rules, summarizing all execution results, and outputting a structured test report.
[0005] The interface test case generation method based on the dual-channel relational graph convolutional network model and the BERT language model provided by this invention may also have the following features: In S1, the key meta-information includes: the functional module to which the interface belongs, the request method, the interface name, the URL path, the request body parameter, the response body parameter, and the type and whether the parameter is required.
[0006] The interface test case generation method based on dual-channel graph convolutional network model and BERT language model provided by the present invention may also have the following feature: In S2, the structural channel adopts the modeling method based on graph convolutional network R-GCN to capture the topological dependency features between interfaces.
[0007] The interface test case generation method based on a dual-channel relational graph convolutional network model and a BERT language model provided in this invention may also have the following features: In S2, the method for identifying and outputting the predicted probability of dependencies between interfaces through structural channels is as follows: S2-1-1: Construct an interface relational graph with interfaces as nodes and potential dependencies as edges based on the input data; S2-1-2: Input the interface relational graph into a pre-trained relational graph convolutional network R-GCN model. The relational graph convolutional network model aggregates information about neighboring nodes and edge types through a multi-layer message passing mechanism, generating a low-dimensional embedding vector for each interface node; S2-1-3: For any interface to be judged, the relational graph convolutional network model concatenates the node embedding vectors corresponding to the two interfaces and inputs them into the classifier; S2-1-4: Output the predicted probability of the dependency relationship.
[0008] The interface test case generation method based on the dual-channel relation graph convolutional network model and the Bert language model provided by the present invention may also have the following feature: In S2, the semantic channel uses the Bert Chinese language model to encode the interface description information to identify the semantic relationship between interfaces.
[0009] The interface test case generation method based on a dual-channel relation graph convolutional network model and a BERT language model provided in this invention may also have the following features: In S2, the method for identifying and outputting the predicted probability of the dependency relationship between interfaces through the semantic channel is as follows: S2-2-1: Based on the input data, construct a natural language description for each interface to be judged; S2-2-2: Concatenate the natural language descriptions of the two interfaces into a text sequence according to a preset two-sentence input format; S2-2-3: Input the text sequence into a pre-trained Chinese language model, and the model generates hidden states containing contextual information through multi-layer Transformer encoding; S2-2-4: Extract the final hidden state of the CLS tag representing the semantics of the entire input text sequence, and map it to the predicted probability of the dependency relationship through a linear classifier.
[0010] The interface test case generation method based on a dual-channel relational graph convolutional network model and a BERT language model provided by this invention may also have the following features: In step S4, the automatic generation method of the test script includes at least one of a parameter mutation strategy and a scenario path splicing strategy. The parameter mutation strategy specifically involves generating an interface call sequence based on a directed dependency graph, perturbing the request parameters of one or more interfaces, and controlling the frequency and distribution of error injection based on a configurable maximum number of errors per test case and a random mechanism to adjust the scale of the test script. The scenario path splicing strategy specifically involves extracting multiple key paths based on a directed dependency graph and splicing and combining these key paths to generate an end-to-end test script with a coherent structure and higher randomness.
[0011] The interface test case generation method based on the dual-channel relational graph convolutional network model and the BERT language model provided by this invention may also have the following features: the types of disturbances include: removing required fields, modifying data types, or violating business constraints.
[0012] A system for generating interface test cases based on a dual-channel relational graph convolutional network model and the BERT language model is characterized by comprising: a data input processing module, which parses the input interface definition file, extracts key meta-information of each interface, and processes the key meta-information to form input data for dependency identification; a dual-channel collaborative identification module, which identifies and outputs the dependencies between interfaces through structural and semantic channels based on the input data, including: unrelated, positive, and negative dependencies; a relationship fusion module, which fuses the dependencies between interfaces output by the structural and semantic channels and verifies the consistency of direction, generating a directed dependency graph; a test case generation module, which automatically generates test scripts adapted to the target testing framework based on the directed dependency graph and key meta-information; and a test feedback module, which executes the test scripts, evaluates the execution results according to the interface response status and assertion rules, summarizes the execution results of all test cases, and outputs a structured test report.
[0013] The role and effect of invention
[0014] The interface test case generation method and system based on a dual-channel relational graph convolutional network model and a BERT language model, as described in this invention, have the following beneficial effects:
[0015] 1. This invention adopts a mechanism that combines structural channels and semantic channels, taking into account the topological information of the interface and natural language description, which helps to more comprehensively identify the dependencies between interfaces.
[0016] 2. This invention integrates the prediction results of two channels through a directional consistency check and multi-round screening fusion strategy, which reduces misjudgment to a certain extent and improves the logical rationality of the dependency direction.
[0017] 3. This invention supports the generation of test sequences based on dependency graphs and combines various parameter mutation strategies (such as missing required fields, type errors, business constraint violations, etc.) to achieve an automated process from dependency analysis to test script generation.
[0018] 4. This invention emphasizes lightweight design and efficiency in the model design and test case generation process. The dual-channel model can quickly output prediction results after preprocessing. In the test script generation step, the test case size and error injection strategy can be flexibly adjusted through configuration. Compared with some existing methods, this invention controls the computational resources and time overhead while ensuring the recognition effect. Attached Figure Description
[0019] Figure 1 This is a flowchart of the interface test case generation method based on a dual-channel relational graph convolutional network model and a BERT language model in an embodiment of the present invention.
[0020] Figure 2 This is an architecture diagram of an interface test case generation method based on a dual-channel relational graph convolutional network model and a BERT language model, as described in an embodiment of the present invention.
[0021] Figure 3 This is a directed dependency graph generated in an embodiment of the present invention. Detailed Implementation
[0022] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection, an electrical connection, or a connection that allows communication between them; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0023] To make the technical means, creative features, objectives and effects of this invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the interface test case generation method and system based on a dual-channel relational graph convolutional network model and a BERT language model.
[0024] The interface test case generation method based on the dual-channel relational graph convolutional network model and the BERT language model in this embodiment includes the following steps:
[0025] Figure 1 This is a flowchart of the interface test case generation method based on a dual-channel relational graph convolutional network model and a BERT language model in an embodiment of the present invention. Figure 2 This is an architecture diagram of an interface test case generation method based on a dual-channel relational graph convolutional network model and a BERT language model, as described in an embodiment of the present invention.
[0026] like Figure 1-2 As shown, step S1 is the data input processing step, which parses the input interface definition file, extracts the key meta-information of each interface, and processes the key meta-information to form input data for dependency identification, specifically:
[0027] Taking an interface definition file (JSON format) conforming to the OpenAPI specification (OpenAPI 3.0) as input, the program parses the input interface definition file and extracts key metadata for each API interface. Key metadata includes, but is not limited to: the functional module to which the interface belongs, the HTTP request method, the interface name, the URL path, the request body parameters, the response body parameters, and the parameter types and whether they are required.
[0028] To adapt to the input requirements of the subsequent dual-channel model, the key meta-information of each interface is concatenated into a text description in a specific format. This text description serves as the input data for dependency recognition, which is then used for feature extraction of the subsequent structural and semantic channels.
[0029] At the same time, the extracted key metadata is standardized, such as unifying the path parameter format and standardizing the interface name, to form a structured set of interface metadata for use in the subsequent S4 test case generation steps.
[0030] Step S2 is a dual-channel collaborative identification step. Based on the input data, it identifies and outputs the predicted probabilities of inter-interface dependencies through structural and semantic channels, respectively. Inter-interface dependencies include: unrelated, positive, and negative dependencies.
[0031] Specifically, this step is performed by dual-channel collaborative identification, which includes a structured neural network channel (structure channel) and a semantic language model channel (semantic channel) for parallel identification of potential dependencies between interfaces.
[0032] The structural channel employs a modeling method based on the relational graph convolutional network R-GCN to capture the topological dependency features between interfaces. The method for identifying and outputting the predicted probabilities of dependencies between interfaces through the structural channel is as follows:
[0033] S2-1-1: Based on input data (such as functional modules, URL path hierarchy, request / response parameter structure, etc.), construct an interface relationship graph with interfaces as nodes and potential dependencies as edges. The edge type can be defined according to rules such as parameter passing and path inclusion.
[0034] S2-1-2: Input the interface relationship graph into the pre-trained relational graph convolutional network (R-GCN) model. The R-GCN model aggregates information about neighboring nodes and edge types through a multi-layer message passing mechanism, generating a low-dimensional embedding vector for each node in the interface. This embedding vector encodes the node's position and role in the global structure, specifically:
[0035] The overall architecture of the R-GCN model consists of a two-layer R-GCN encoder and a relation classifier. The R-GCN encoder considers the type information of the edges and aggregates different relation types using independent weight matrices during message passing.
[0036] (1)
[0037] Where R = {0, 1, 2} represents three types of relations. Let i be the set of neighbors of node i under relation r. Let be the learnable parameter matrix corresponding to relation r in layer l. Through two layers of R-GCN, the original node features are mapped to a 16-dimensional embedding space.
[0038] S2-1-3: For any two interfaces A and B to be judged, the relational graph convolutional network model concatenates the node embedding vectors corresponding to the two interfaces and inputs them into the classifier (such as MLP), specifically:
[0039] For the relation edge (u, v) to be predicted, the relation graph convolutional network model concatenates the embedding vectors of the source node and the target node and inputs them into a multilayer perceptron (MLP) classifier:
[0040] (2)
[0041] S2-1-4: Output the predicted probabilities of dependencies (unrelated, positive dependency A→B, negative dependency B→A), specifically:
[0042] Output the logits values of the three types of dependencies and obtain the predicted probabilities using the softmax function.
[0043] The semantic channel uses the BERT Chinese language model to encode interface description information to identify the semantic relationships between interfaces.
[0044] The Chinese language model is pre-trained based on the Chinese-MacBERT-large language model to perform a three-class dependency recognition task. Its output layer uses a linear classifier to map the final hidden state of the [CLS] token to the class space, thereby completing the classification prediction.
[0045] Specifically, the method for identifying and outputting the predicted probability of dependencies between interfaces through semantic channels is as follows:
[0046] S2-2-1: Based on the input data, construct a natural language description for each interface A and B to be judged. The description format for each interface includes four fields: module, method, interface name, and URL. For example: "Module = User Management, Method = POST, Interface Name = Add User, URL = / api / user".
[0047] S2-2-2: Concatenate the natural language descriptions of the two interfaces into a text sequence according to a preset two-sentence input format, specifically:
[0048] The description of interface A is taken as Sentence A, and the description of interface B is taken as Sentence B. They are concatenated into a text sequence according to the two-sentence input format (sentence A + sentence B) of pre-trained language models such as BERT.
[0049] S2-2-3: Input the text sequence into a pre-trained Chinese language model. The model generates hidden states containing contextual information through multi-layer Transformer encoding, represented as follows:
[0050] (3)
[0051] Among them, X a With X b These are natural language descriptions of interface A and interface B, respectively. This is the hidden state of the CLS tag, which represents the semantics of the entire sequence after encoding.
[0052] S2-2-4: Extract the final hidden state of the CLS markers representing the semantics of the entire input text sequence, and map it to the predicted probabilities of dependencies using a linear classifier:
[0053] (4)
[0054] in, and For classifier parameters, This represents the predicted probabilities of the three types of dependencies in the final output.
[0055] Step S3 is the relationship fusion step, which fuses the predicted probabilities of inter-interface dependencies output by the structural and semantic channels and verifies their directional consistency to generate a logically clear set of interface dependencies and a directed dependency graph. Specifically:
[0056] The receiver receives the predicted probabilities of three types of dependencies for the same interface pair output from the structural and semantic channels. To improve the accuracy and logical consistency of dependency directions, a direction consistency fusion strategy is used to process the results from both channels.
[0057] For the interface pair (A, B), obtain the structural channel prediction results R1(A→B) and R1(B→A), and the semantic channel prediction results R2(A→B) and R2(B→A).
[0058] * First round of screening: If both R1(A→B) and R2(A→B) tend to be "positive relationship" (e.g., the probability is higher than the set threshold), and both R1(B→A) and R2(B→A) tend to be "negative relationship", then a clear dependency relationship A → B is confirmed.
[0059] * Second round of screening: If at least one of R1(A→B) and R2(A→B) tends to be a "positive relationship", while neither R1(B→A) nor R2(B→A) tends to be a "negative relationship", then a dependency relationship A → B is also considered to exist.
[0060] * Third round of screening: If neither R1(A→B) nor R2(A→B) tends to be a "positive relationship", but at least one of R1(B→A) and R2(B→A) tends to be a "negative relationship", then the dependency relationship B → A is confirmed to exist.
[0061] Figure 3 This is a directed dependency graph generated in an embodiment of the present invention.
[0062] like Figure 3 As shown, by combining the results of the three rounds of screening, a final set of directed dependencies with clearly defined directions is obtained. This set constitutes a directed dependency graph (ODG), where nodes are interfaces and directed edges represent the direction of dependency.
[0063] Step S4 is the test case generation step. Based on the directed dependency graph and key metadata, test scripts adapted to the target testing framework are automatically generated, specifically as follows:
[0064] Based on the ODG diagram and original interface metadata set generated in the previous step, Python test scripts adapted to mainstream testing frameworks are automatically generated, that is, test cases are built using the HttpRunner framework combined with the Python language.
[0065] In terms of construction strategy, two generation routes are adopted in parallel:
[0066] Data mutation path
[0067] Sequence Construction: A first-to-last traversal strategy is used on the ODG graph to generate a valid sequence of interface calls. Sequence construction ensures that the output of the predecessor interface can serve as the input of subsequent interfaces, guaranteeing the correctness and executability of the call logic.
[0068] Parameter Mutation and Linked Injection: To improve test coverage and fault detection rate, a parameter mutation strategy is introduced when generating scripts. This includes perturbing request parameters, such as removing required fields, modifying data types, and setting values that violate business rules. Simultaneously, a multi-interface linked error injection strategy is supported. This means that within a test case sequence, multiple interfaces can be configured to inject different types of errors simultaneously or sequentially to simulate more complex exception scenarios. The type, number, and location of error injection can be flexibly controlled through configuration.
[0069] This approach focuses on validating the complete business processes and logical dependencies across multiple interfaces to ensure correct overall behavior in both normal and edge scenarios.
[0070] Scenario-based approach
[0071] The scenario-based approach decomposes the Dependency Graph (ODG) into multiple subgraphs, and then concatenates these subgraphs end-to-end through permutations and combinations to generate a coherent and more randomized end-to-end test script.
[0072] Topology traversal: Starting from the node with an in-degree of 0 (such as "user login"), perform a depth-first or breadth-first search along the dependencies to build a complete business flow from the start point to the end point (e.g., login → create project → query project list → logout), ensuring that the core critical path is generated.
[0073] Random walk: All core critical paths are combined into a set, and four critical paths are selected without repetition each time. These paths are then combined in a permutation and combination manner to form an end-to-end test scenario that closely resembles real user behavior.
[0074] This approach extracts core critical paths through topological traversal and combines multiple paths with random walks to generate end-to-end test cases that are structurally coherent and closely resemble real user behavior. It focuses on verifying the complete business process across interfaces and its logical dependencies to ensure the overall correctness of behavior in normal and edge scenarios.
[0075] Step S5 is the test feedback step. The test script is executed, the execution results are evaluated based on the interface response status and assertion rules, all execution results are summarized, and a structured test report is output. Specifically:
[0076] * Execution and Monitoring: When the test script runs, it automatically sends HTTP requests to the target API interface and receives responses.
[0077] * Result Evaluation: The execution result (pass / fail) of a single interface call or the entire test case sequence is determined based on preset assertion rules (such as HTTP status codes 2xx indicating success and 5xx indicating server error) and the validation of the response body content (such as field values, data structures, etc.).
[0078] * Report Generation: Automatically summarizes the execution results of all test cases, calculates key metrics such as API call coverage, effective response rate, and number of faults found, and generates structured test reports to provide data support for subsequent test analysis and system optimization.
[0079] This embodiment also provides an interface test case generation system based on a dual-channel relational graph convolutional network model and the BERT language model, including:
[0080] The data input processing module is used to implement step S1, namely: parsing the input interface definition file, extracting the key meta-information of each interface, and processing the key meta-information to form input data for dependency identification.
[0081] The dual-channel collaborative recognition module is used to implement step S2, namely: based on the input data, it identifies and outputs the inter-interface dependency relationship through the structural channel and the semantic channel respectively. The inter-interface dependency relationship includes: unrelated, positive dependency relationship and negative dependency relationship.
[0082] The relationship fusion module is used to implement step S3, namely: to fuse the dependencies between the interfaces output by the structural channel and the semantic channel and verify the consistency of the directions, and generate a directed dependency graph;
[0083] The test case generation module is used to implement step S4, namely: automatically generating test scripts adapted to the target test framework based on the directed dependency graph and key meta information.
[0084] The test feedback module is used to implement step S5, namely: execute the test script, evaluate the execution results based on the interface response status and assertion rules, summarize the execution results of all test cases, and output a structured test report.
[0085] The role and effect of the embodiments
[0086] The interface test case generation method and system based on the dual-channel relational graph convolutional network model and the BERT language model according to the present invention have the following beneficial effects:
[0087] This invention aims to improve the ability to identify dependencies between REST API interfaces and the quality of test script generation by integrating dual-channel intelligent recognition and automated test case generation, thereby significantly improving the efficiency and accuracy of black-box testing. Specifically:
[0088] 1. This invention adopts a mechanism that combines structural channels and semantic channels, taking into account the topological information of the interface and natural language description, which helps to more comprehensively identify the dependencies between interfaces.
[0089] 2. This invention integrates the prediction results of two channels through a directional consistency check and multi-round screening fusion strategy, which reduces misjudgment to a certain extent and improves the logical rationality of the dependency direction.
[0090] 3. This invention supports the generation of test sequences based on dependency graphs and combines various parameter mutation strategies (such as missing required fields, type errors, business constraint violations, etc.) to achieve an automated process from dependency analysis to test script generation.
[0091] 4. This invention emphasizes lightweight design and efficiency in the model design and test case generation process. The dual-channel model can quickly output prediction results after preprocessing. In the test script generation step, the test case size and error injection strategy can be flexibly adjusted through configuration. Compared with some existing methods, this invention controls the computational resources and time overhead while ensuring the recognition effect.
[0092] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.
Claims
1. A method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model, characterized in that, Includes the following steps: S1: Data input processing step, parse the input interface definition file, extract the key meta information of each interface, and process the key meta information to form input data for dependency identification; S2: Dual-channel collaborative identification step, based on the input data, identifies and outputs the predicted probability of inter-interface dependencies through structural and semantic channels respectively, the inter-interface dependencies include: unrelated, positive dependencies and negative dependencies; S3: Relationship fusion step, fusing and verifying the direction consistency of the predicted probabilities of the inter-interface dependency relationships output by the structural channel and the semantic channel to generate a directed dependency graph; S4: Test case generation step, based on the directed dependency graph and the key meta information, automatically generate test scripts adapted to the target test framework; S5: Test feedback step, execute the test script, evaluate the execution results based on the interface response status and assertion rules, summarize all the execution results and output a structured test report.
2. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model according to claim 1, characterized in that: in, In S1, the key metadata includes: the functional module to which the interface belongs, the request method, the interface name, the URL path, the request body, the response body, and the type and whether the parameters are required.
3. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model according to claim 1, characterized in that: in, In S2, the structural channel adopts a modeling method based on relational graph convolutional network R-GCN to capture the topological dependency features between interfaces.
4. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model according to claim 3, characterized in that: in, In step S2, the method for identifying and outputting the predicted probability of inter-interface dependencies through structural channels is as follows: S2-1-1: Construct an interface relationship graph based on the input data, with interfaces as nodes and potential dependencies as edges; S2-1-2: Input the interface relationship graph into the pre-trained relational graph convolutional network R-GCN model. The relational graph convolutional network model aggregates the information of neighboring nodes and edge types through a multi-layer message passing mechanism to generate a low-dimensional embedding vector for each interface node. S2-1-3: For any interface to be judged, the relational graph convolutional network model concatenates the node embedding vectors corresponding to the two interfaces and inputs them into the classifier. S2-1-4: Output the predicted probability of the dependency.
5. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model according to claim 1, characterized in that: in, In step S2, the semantic channel uses the Bert Chinese language model to encode the interface description information, which is used to identify the semantic relationship between interfaces.
6. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model according to claim 5, characterized in that: in, In step S2, the method for identifying and outputting the predicted probability of inter-interface dependencies through semantic channels is as follows: S2-2-1: Based on the input data, construct a natural language description for each interface to be judged; S2-2-2: Concatenate the natural language descriptions of the two interfaces into a text sequence according to the preset two-sentence input format; S2-2-3: Input the text sequence into a pre-trained Chinese language model. The model generates hidden states containing contextual information through multi-layer Transformer encoding. S2-2-4: Extract the final hidden state of the CLS tag representing the semantics of the entire input text sequence, and map it to the predicted probability of the dependency through a linear classifier.
7. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model as described in claim 2. Its features are: In step S4, the automatic generation method for the test script includes at least one of a parameter mutation strategy and a scene path concatenation strategy. The parameter mutation strategy is as follows: Based on the directed dependency graph, an interface call sequence is generated, and the request parameters of one or more interfaces are perturbed. At the same time, based on the configurable maximum number of errors per test case and the random mechanism, the frequency and distribution of error injection are controlled to adjust the size of the test script. The scenario path splicing strategy is as follows: extract multiple critical paths based on the directed dependency graph, and splice and combine the multiple critical paths to generate an end-to-end test script with a coherent structure and higher randomness.
8. The method for generating interface test cases based on a dual-channel relational graph convolutional network model and a BERT language model according to claim 7, characterized in that: in, The types of disturbances include: removing required fields, modifying data types, or violating business constraints.
9. An interface test case generation system based on a dual-channel relational graph convolutional network model and a BERT language model, characterized in that, include: The data input processing module parses the input interface definition file, extracts the key meta-information of each interface, and processes the key meta-information to form input data for dependency identification. The dual-channel collaborative recognition module, based on the input data, identifies and outputs the inter-interface dependencies through the structural channel and the semantic channel respectively. The inter-interface dependencies include: unrelated, positive, and negative dependencies. The relationship fusion module fuses and verifies the directional consistency of the dependency relationships between the interfaces output by the structural channel and the semantic channel, and generates a directed dependency graph. The test case generation module automatically generates test scripts adapted to the target testing framework based on the directed dependency graph and the key meta information. The test feedback module executes the test script, evaluates the execution results based on the interface response status and assertion rules, summarizes the execution results of all test cases, and outputs a structured test report.