Test case quality control method and device, medium and product

Abstract

The invention discloses a test case quality control method and device, a medium and a product, and relates to the technical field of artificial intelligence and financial science and technology, and the method comprises the steps: obtaining a to-be-tested case and historical quantification information; converting the to-be-tested case into a vectorized test case, and combining the vectorized test case with the historical quantitative information to obtain target input data; performing dynamic weight updating on the bidirectional gating loop unit network, and inputting the vectorized test case to obtain a to-be-tested case feature; and inputting the feature into a preset local attention mechanism network to obtain dimensionality reduction use case features, inputting the dimensionality reduction use case features into a branch full-connection network to obtain a target classification confidence coefficient, obtaining a target identification result according to the target classification confidence coefficient, comparing the target identification result with a marking result, screening out an abnormal result, and when an abnormal statistical value calculated according to the abnormal result meets an early warning condition, carrying out early warning. Quality early warning is triggered, real-time monitoring of the use case is achieved, and the accuracy of use case type recognition and the stability and reliability of use case quality are ensured.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and financial technology, and in particular to a method, device, medium and product for controlling the quality of test cases. Background Technology

[0002] As financial transactions become increasingly complex and testing scales up, the importance of test case quality control is becoming more and more significant. Effective quality control not only ensures that test cases conform to business logic and requirements, but also improves testing efficiency and reduces business risks.

[0003] In existing technologies, traditional manual annotation and review methods for case quality control mainly rely on manual semantic parsing annotation and rule-based verification to determine whether test cases are compliant. However, manual annotation is highly subjective, and differences in testers' understanding of classification standards can easily lead to deviations between labels and true semantics, reducing the accuracy of test case quality identification.

[0004] Rule-based automated verification methods mainly use natural language processing models based on shallow semantic analysis for text classification and statistical models for screening to achieve quality control. However, conventional models or simple neural network models are difficult to effectively parse financial business terms, resulting in one-sided and simplistic feature extraction. Static recognition rules are difficult to adapt to complex scenario changes, have low generalization ability, and reduce the accuracy of use case quality recognition. Summary of the Invention

[0005] This invention provides a method, device, medium, and product for controlling the quality of test cases, in order to solve the problems of test case quality identification being difficult to adapt to complex scenario changes, having low generalization ability, and low accuracy.

[0006] According to one aspect of the present invention, a method for controlling the quality of test cases is provided, comprising:

[0007] Obtain the test cases and their corresponding historical quantitative information;

[0008] The test cases are converted into vectorized test cases, and the vectorized test cases and historical quantitative information are combined to obtain the target input data.

[0009] After dynamically updating the weights of the bidirectional gated recurrent unit network based on the target input data, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0010] The features of the test cases are input into a pre-set local attention mechanism network to obtain dimensionality-reduced test case features. Then, the dimensionality-reduced test case features are input into a pre-trained branch fully connected network to obtain the target classification confidence. Based on the target classification confidence, the target recognition result of the test cases is obtained.

[0011] The target identification results are compared with the labeling results for the test cases to filter out abnormal results, and a quality warning is triggered when the abnormal statistical value calculated based on the abnormal results meets the warning conditions.

[0012] According to another aspect of the present invention, a test case quality control device is provided, comprising:

[0013] The module to be tested is used to obtain the test cases and the corresponding historical quantitative information of the test cases;

[0014] The target input module is used to convert the test cases into vectorized test cases and combine the vectorized test cases with historical quantization information to obtain target input data.

[0015] The test case feature module is used to dynamically update the weights of the bidirectional gated recurrent unit network based on the target input data, and then input the vectorized test cases into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0016] The target recognition module is used to input the features of the test cases into a preset local attention mechanism network to obtain dimensionality-reduced test case features. After obtaining the dimensionality-reduced test case features, the dimensionality-reduced test case features are input into a pre-trained branch fully connected network to obtain the target classification confidence score. Based on the target classification confidence score, the target recognition result of the test cases is obtained.

[0017] The quality early warning module is used to compare the target identification results with the labeling results for the test cases, filter out abnormal results, and trigger a quality early warning when the abnormal statistical value calculated based on the abnormal results meets the early warning conditions.

[0018] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0019] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the test case quality control method according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the test case quality control method described in any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method as described in any embodiment of the present invention.

[0022] The technical solution of this invention involves acquiring the test cases and their historical quantification information, converting the test cases into vector form, and combining them with historical information to form target input data. This data is then used to dynamically update the weights of the bidirectional gated recurrent unit network (BRN), inputting the vectorized test cases to extract test case features. Dimensionality reduction features are obtained through a local attention mechanism network, and these features are input into a pre-trained fully connected branch network to obtain the target classification confidence and corresponding recognition results. The recognition results are compared with the labeling results, abnormal results are filtered, and statistical values ​​are calculated. A quality warning is triggered when certain conditions are met. The dynamically weighted BRN network processes and extracts data features, improving adaptability to complex scene changes, generalization ability, and semantic understanding of test cases. Further processing through the local attention mechanism network optimizes the interpretation of semantic information, thereby improving the accuracy of the fully connected branch network in identifying test case types. In terms of analysis dimensions, historical quantification information is comprehensively considered to ensure the comprehensiveness and accuracy of the analysis results. The system automatically identifies abnormal test cases and triggers a quality warning when the abnormal statistical value calculated based on the abnormal results meets the warning conditions. This enables real-time monitoring of test case quality and ensures the accuracy of test case type identification and the stability and reliability of test case quality.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a test case quality control method provided in Embodiment 1 of the present invention;

[0026] Figure 2 This is a schematic diagram of a branch fully connected network applicable to an embodiment of the present invention;

[0027] Figure 3 This is a flowchart of another method for controlling the quality of test cases according to Embodiment 2 of the present invention;

[0028] Figure 4 This is a schematic diagram of an auxiliary network applicable to an embodiment of the present invention;

[0029] Figure 5 This is a flowchart of another test case quality control method provided in Embodiment 3 of the present invention;

[0030] Figure 6 This is a schematic diagram of a local attention mechanism network applicable to an embodiment of the present invention;

[0031] Figure 7 This is a schematic diagram of a bidirectional gated loop unit network applicable to an embodiment of the present invention;

[0032] Figure 8 This is a schematic diagram of a bidirectional gated loop unit layer applicable to an embodiment of the present invention;

[0033] Figure 9 This is a schematic diagram of a test case quality control device provided according to Embodiment 4 of the present invention;

[0034] Figure 10 This is a schematic diagram of the structure of an electronic device that implements the test case quality control method of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0037] Example 1

[0038] Figure 1 This is a flowchart of a test case quality control method provided in Embodiment 1 of the present invention. This embodiment is applicable to situations involving the control of test case quality. The method can be executed by a test case quality control device, which can be implemented in hardware and / or software and is generally configured in an electronic device. Figure 1 As shown, the method includes:

[0039] S110. Obtain the test cases and the corresponding historical quantitative information.

[0040] In this embodiment of the invention, the test case to be tested can be specifically understood as: a set of data uploaded by the user or written and generated by the tester for testing the system. The test case to be tested may specifically include: description information of the test scenario, detailed information of the input parameters, and expected output results (labeled test case type), etc.

[0041] The test cases to be tested can include the following types: accounting-related cases, performance and capacity cases, stability cases, reliability cases, regression test cases, high-frequency trading cases, important user cases, and exception or failure cases.

[0042] Accounting-related cases can be specifically understood as test cases for financial accounting processes, including specific test case content in scenarios such as fund transfers, invoice generation, and expense settlement. Performance and capacity cases can be specifically understood as test cases related to system performance and capacity, including specific test case content in scenarios such as high-concurrency access, large-scale data processing, and resource consumption. Stability cases can be specifically understood as test cases for testing the stability of a system under long-term operation or specific stress conditions, including specific test case content in scenarios such as uninterrupted service operation and scheduled task execution. Reliability cases can be specifically understood as test cases for verifying the reliability of a system under various abnormal conditions, including specific test case content in scenarios such as failover, data backup and recovery, and abnormal power outages.

[0043] Regression test cases can be understood as test cases that verify whether existing functions can be used normally after system modifications or upgrades, including specific test case content related to core business processes and functional points susceptible to changes. High-frequency trading test cases can be understood as test cases designed for high-frequency trading scenarios, including specific test case content for scenarios such as high-frequency trading of financial products and high-frequency data queries. Important user test cases can be understood as test cases customized for important user groups or key customers, including specific test case content related to users' core business processes and specific needs. Anomaly or failure test cases can be understood as test cases simulating various anomaly or failure scenarios, including specific test case content for anomaly or failure scenarios such as network failures, disk fullness, and illegal input.

[0044] Historical quantitative information can be specifically understood as: quantitative information obtained by statistical calculation based on test data of historical test cases pre-stored in the database. This historical quantitative information may include: historical case annotation accuracy, case classification confidence, mislabeling type distribution, review and correction rate, and requirement coverage.

[0045] Historical test case annotation accuracy can be understood as the proportion of historical test cases that were correctly annotated. It can be calculated as: Historical Test Case Annotation Accuracy = (Number of Correctly Annotated Test Cases for a Single Requirement / Total Number of Annotated Test Cases for a Single Requirement) / Total Number of Requirements. Here, "single requirement" refers to the test case type corresponding to the test case to be tested, and "total number of requirements" refers to the total number of test case types corresponding to all test cases to be tested. The number of correctly annotated test cases for a single requirement is the number of correctly annotated test cases under that single requirement. The total number of annotated test cases for a single requirement is the total number of all annotated test cases under that single requirement.

[0046] The confidence score for test case classification can be understood as the level of confidence the model has in the test case classification results. It can be calculated as: Test Case Classification Confidence Score = (Number of correctly classified test cases for a single requirement / Total number of classified test cases for a single requirement) / Total number of requirements. Specifically, the number of correctly classified test cases for a single requirement can be understood as the number of test cases correctly classified and labeled by the model for that single requirement. The total number of classified test cases for a single requirement can be understood as the total number of test cases classified and labeled by the model for that single requirement.

[0047] The distribution of mislabeling types can be understood as the distribution of the types of mislabeling errors in historical test cases. It can be calculated by the formula: Mislabeling type distribution = Number of mislabeling errors of a single type / Total number of mislabeling errors. This helps to identify which types of mislabeling are the main problems, thus providing direction for key improvements.

[0048] The review and correction rate can be understood as the proportion of marked test cases that are corrected after review. It can be calculated as: Review and Correction Rate = (Number of Reviewed and Corrected Test Cases for a Single Requirement / Total Number of Reviewed and Corrected Test Cases for a Single Requirement) / Total Number of Requirements. Specifically, the number of reviewed and corrected test cases for a single requirement refers to the number of test cases that are corrected after review for that single requirement. The total number of reviewed and corrected test cases for a single requirement refers to the total number of all reviewed test cases for that single requirement.

[0049] Requirements coverage can be understood as the scope and extent to which test cases cover requirements. This metric can be obtained from internal performance tools. Performance tools can be understood as tools used to track and manage various data during the testing process. Performance tools can automatically collect requirements coverage information from test cases, divide it by the known total number of requirements, and thus automatically calculate requirements coverage.

[0050] Optionally, based on the above embodiments, historical quantitative information can be user-based. Specifically, this involves analyzing the test cases uploaded by users to calculate various historical quantitative indicators related to that user. For example, when calculating the historical test case labeling accuracy of a user, only the proportion of correctly labeled test cases uploaded by that user is considered; when calculating the test case classification confidence, statistics are based solely on the classification of the test cases uploaded by that user. Calculating on a user-by-user basis can reflect each user's historical performance and behavioral patterns in test case labeling and management, providing a basis for subsequent personalized analysis and improvement.

[0051] S120. Convert the test cases to vectorized test cases, and combine the vectorized test cases with historical quantization information to obtain the target input data.

[0052] Specifically, since test cases can include various types of data, such as text descriptions, numerical parameters, and images, it is necessary to convert non-numerical data in the test cases into numerical vectors to obtain vectorized test cases. This allows for mathematical operations and model processing. For example, methods such as the bag-of-words model or word embedding can be used to convert text description data into numerical vectors, pixel values ​​of image data can be arranged into a vector to convert it into a numerical vector, and numerical parameter data can be directly constructed into vectors. The vectorized test cases are then combined with historical quantization information, such as through vector concatenation, to form the target input data.

[0053] S130. After dynamically updating the weights of the bidirectional gated recurrent unit network based on the target input data, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0054] In this embodiment of the invention, the bidirectional gated recurrent unit network can be specifically understood as a network model that combines forward and reverse gated recurrent units, enabling it to learn features from past and future contextual information simultaneously.

[0055] Specifically, the target input data is fed into the weight generation network or the weights of the bidirectional gated recurrent unit (BRN) network are iteratively updated to generate dynamic weight parameters for the BRN network. These dynamic weight parameters are then used to update the network. For example, the weights of the BRN network are randomly initialized, the target input data is fed into the network, the network output is calculated, the loss function is calculated based on the network output and the true labels, the weights are updated using the adaptive moment estimation algorithm, and the forward propagation, loss function calculation, and weight update steps are repeated until the model converges, resulting in dynamic weight parameters. These dynamic weight parameters are then used to update the BRN network.

[0056] Correspondingly, the vectorized test cases are input into the weight-updated bidirectional gated recurrent unit network. The network processes the input sequence through forward and reverse gating mechanisms to extract the features of the test cases.

[0057] Understandably, the varying amounts of information contained in different use cases lead to inconsistent lengths in the resulting target input data. To ensure that the data can be uniformly input into the model, the length of all target input data needs to be standardized.

[0058] Furthermore, based on the above embodiments, after obtaining the target input data, the method may further include: for target input data with a length less than a preset length, padding the end of the target input data with a special value (such as zero) to make it reach the preset length; for target input data with a length greater than the preset length, truncating it according to the preset length, and padding the end of the truncated data with a special value to make it reach the preset length.

[0059] Accordingly, a mask is added to the input layer of the bidirectional gated recurrent unit network. For valid data locations, the mask value is 1; for zero-padded data locations, the mask value is 0. This ensures that when processing input data, the model uses the mask to determine which data is valid and which is zero-padded, thus ignoring data at zero-padded locations. Specifically, the mask can be understood as a binary vector with the same shape as the input data.

[0060] Optionally, based on the above embodiments, other network models such as Long Short-Term Memory Networks or Bidirectional Long Short-Term Memory Networks can be used to replace the Bidirectional Gated Recurrent Unit Network.

[0061] S140. Input the features of the test cases to be tested into the preset local attention mechanism network to obtain the dimensionality reduction test case features. Then, input the dimensionality reduction test case features into the pre-trained branch fully connected network to obtain the target classification confidence. Based on the target classification confidence, obtain the target recognition result of the test cases.

[0062] In this embodiment of the invention, the local attention mechanism network can be specifically understood as a neural network structure that calculates attention within a specific local region. By limiting the scope of attention calculation and focusing on a local region of the input data, the computational load is reduced and local feature modeling is enhanced.

[0063] A branch-fully connected network can be understood as a fully connected network structure comprising two or more branches. Each branch is an independent fully connected network, sharing the same input: dimensionality-reduced use case features from a local attention mechanism network. Each branch processes and transforms the input features from different angles or in different ways, extracting different feature representations. Finally, the outputs of each branch are merged (e.g., by concatenation, averaging, or summing) to form a comprehensive feature representation, which serves as the network's final output. Target classification confidence can be understood as a quantitative estimate by the branch-fully connected network model of the probability that the input data belongs to each use case type. This can be represented by a confidence score, ranging from 0 to 1. For example, if the model outputs a confidence score of 0.9 for a certain use case type, it means the model believes there is a 90% probability that the input data belongs to that use case type. Therefore, the final use case type of the input data can be determined based on the confidence score; typically, the use case type with the highest confidence score is chosen as the prediction result.

[0064] Specifically, the features of the test cases are input into a pre-defined local attention mechanism network. This network calculates attention scores and weights each part of the features, allowing the model to focus on the most relevant features. This achieves dimensionality reduction while preserving key information, resulting in dimensionality-reduced test case features. These dimensionality-reduced features are then input into a pre-trained fully connected branch network to obtain the target classification confidence score. Based on this confidence score, the target identification result for the test cases is obtained. For example, the test case type with the highest confidence score can be selected as the target identification result; alternatively, one or more confidence thresholds can be set based on the total number of test case types to form a test case type confidence interval, and the test case type corresponding to the confidence interval containing the target classification confidence score can be used as the target identification result; or, when a test case belongs to multiple test case types, different filtering thresholds can be set for each test case type, and test case types exceeding these thresholds can be added to the test case type set, which can then be used as the target identification result.

[0065] Optionally, based on the above embodiments, the branch fully connected network may include: a training network branch for updating parameters during the training phase, an incremental network branch for updating parameters during the online learning phase, a fully connected subnetwork connected to the training network branch and the incremental network branch respectively, and a classification confidence generation subnetwork connected to the fully connected subnetwork.

[0066] Accordingly, the dimensionality-reduced use case features are input into a pre-trained fully connected branch network to obtain the target classification confidence, which may include:

[0067] The dimensionality reduction use case features are input into the training network branch and the incremental network branch, respectively;

[0068] The network branches are trained to generate the first feature results corresponding to the features of the dimensionality reduction use cases, and the first feature results are output to the fully connected subnet.

[0069] The second feature result corresponding to the dimensionality reduction use case features is generated by incremental network branching, and the second feature result is output to the fully connected subnet;

[0070] The first feature result and the second feature result are fused through a fully connected subnet to obtain the fused result, and the fused result is output to the classification confidence generation subnet.

[0071] Subnets are generated based on classification confidence scores, and target classification confidence scores are generated based on the fusion processing results.

[0072] In this embodiment of the invention, the training network branch can be specifically understood as: during the model training phase, it is used to adjust and update the network parameters of the training network branch based on the training data to minimize prediction errors and improve model performance. The incremental network branch can be specifically understood as: used during the online learning phase to update the network parameters of the incremental network branch in real time based on newly generated data or feedback, enabling the model to continuously adapt to new information and optimize performance. The fully connected subnet can be specifically understood as: connected to both the training network branch and the incremental network branch, used to further integrate and process the features from these two branches. The classification confidence generation subnet can be specifically understood as: connected to the fully connected subnet, used to convert the processed features into classification confidence, i.e., the confidence level of the model's prediction for each use case type.

[0073] Specifically, test cases with labeled test case types are retrieved from the test case library, along with corresponding historical quantization information. These test cases are then converted into vectorized test cases. The vectorized test cases and historical quantization information are combined to obtain the target input data. Based on the target input data, the weights of the bidirectional gated recurrent unit network (BRN) are dynamically updated. The vectorized test cases are then input into the updated BRN to obtain test case features. These features are then input into a pre-defined local attention mechanism network to obtain dimensionality-reduced test case features. Finally, these dimensionality-reduced features are input into a fully connected branch network for pre-training of the training network branches. This pre-trained training network branch is obtained without using the incremental network branch. Furthermore, the initial network parameters in the incremental network branch can be customized, or set to the network parameters of the pre-trained training network branch, resulting in a pre-trained incremental network branch. This leads to the pre-trained fully connected branch network.

[0074] In this embodiment of the invention, the first feature result can be specifically understood as: the feature result generated after the training network branch processes the dimensionality-reduced use case features, reflecting the feature patterns learned from the training data. The second feature result can be specifically understood as: the feature result generated after the incremental network branch processes the dimensionality-reduced use case features, reflecting the feature patterns learned from the new data during the online learning phase.

[0075] Accordingly, the dimensionality-reduced use case features are input into the training network branch and the incremental network branch, respectively. The training network branch performs further feature extraction and transformation on the dimensionality-reduced use case features to generate the first feature result. The incremental network branch performs further feature extraction and transformation on the dimensionality-reduced use case features to generate the second feature result. The fully connected subnetwork concatenates the first and second feature results and transforms them through fully connected layers to integrate the feature information from training and incremental learning, obtaining a fused processing result, which is then output to the classification confidence generation subnetwork. The classification confidence generation subnetwork uses an activation function to transform the fused processing result and map it to the confidence scores of each use case type, generating the target classification confidence score.

[0076] Figure 2 This is a schematic diagram of a branch-fully connected network applicable to an embodiment of the present invention, wherein the output of the attention mechanism network, i.e., the dimensionality-reduced use case features, is the input of the branch-fully connected network. The number of neurons in the output layer of the classification confidence generation subnetwork is the same as the total number of use case types, that is, the number of classification confidence results is the same as the total number of use case types, and each use case type will have a corresponding classification confidence. Typically, when the total number of use case types is 38, the dimension of the classification result is 38×1.

[0077] By inputting the dimensionality-reduced use case features into the training and incremental network branches respectively, the first feature result generated by the training network branch and the second feature result generated by the incremental network branch reflect the feature patterns extracted from historical data and new online data, respectively. These two feature results are then input into a fully connected subnet for fusion processing, integrating the two feature patterns to more comprehensively capture the features of the use cases. The fusion result is then input into a classification confidence generation subnet to finally generate the target classification confidence score. By fully utilizing information from both training and new data, the comprehensiveness of the model's use case feature extraction is improved, enhancing the model's adaptability to new data, thereby improving the accuracy of test case type identification and the stability and reliability of test case quality.

[0078] S150. Compare the target identification results with the labeling results for the test cases, filter out abnormal results, and trigger a quality warning when the abnormal statistical value calculated based on the abnormal results meets the warning conditions.

[0079] Specifically, the target identification results are compared with the expected output results (i.e., labeling results) in the test cases. Inconsistent identification results are statistically analyzed, and anomaly statistics are calculated, such as the proportion of anomaly-type test cases among all test cases. If the anomaly statistics exceed a preset warning threshold, the model's prediction quality is considered to have a problem, meeting the warning conditions and triggering a quality warning to remind relevant personnel to check and handle the issue.

[0080] Furthermore, based on the above embodiments, after filtering out abnormal results, the process may further include:

[0081] Review and screen abnormal results, and add the test cases with incorrect identification to the error test case set;

[0082] When the number of test cases to be tested in the error test case set is greater than or equal to the preset number of training batches, the target number of trained test cases are extracted from the test case library according to the preset rules.

[0083] The test cases from both the trained and erroneous test cases are used as incremental training data to incrementally train the incremental network branch.

[0084] In this embodiment of the invention, the number of training batches can be specifically understood as the size of the sample set used to update model parameters during model training. The trained use cases can be specifically understood as use cases that have already been used to train the model.

[0085] Specifically, after manual or automatic review of abnormal results, the incorrectly identified test cases are added to the error test case set. When the number of test cases in the error test case set is greater than or equal to the preset training batch size, a target number (preset number or a number based on the number of test cases in the error test case set multiplied by a preset ratio) of trained test cases are extracted from the test case library according to preset rules (such as a preset number or a preset ratio). These trained test cases are then merged with the error test case set to form incremental training data, which is used to incrementally train the incremental network branch to optimize model performance, improve the accuracy of the fully connected network branch, and enhance its adaptability and generalization ability to complex scene changes, thereby improving the stability and reliability of test case quality.

[0086] It is understandable that clearing the set of erroneous test cases after incremental training ensures that the next round of incremental training can focus on new erroneous test cases, avoid reusing old erroneous test cases for training, and enable the model to continuously learn new features and patterns, maintaining its adaptability and sensitivity to new data.

[0087] The technical solution of this invention involves acquiring the test cases and their historical quantification information, converting the test cases into vector form, and combining them with historical information to form target input data. This data is then used to dynamically update the weights of the bidirectional gated recurrent unit network (BRN), inputting the vectorized test cases to extract test case features. Dimensionality reduction features are obtained through a local attention mechanism network, and these features are input into a pre-trained fully connected branch network to obtain the target classification confidence and corresponding recognition results. The recognition results are compared with the labeling results, abnormal results are filtered, and statistical values ​​are calculated. A quality warning is triggered when certain conditions are met. The dynamically weighted BRN network processes and extracts data features, improving adaptability to complex scene changes, generalization ability, and semantic understanding of test cases. Further processing through the local attention mechanism network optimizes the interpretation of semantic information, thereby improving the accuracy of the fully connected branch network in identifying test case types. In terms of analysis dimensions, historical quantification information is comprehensively considered to ensure the comprehensiveness and accuracy of the analysis results. The system automatically identifies abnormal test cases and triggers a quality warning when the abnormal statistical value calculated based on the abnormal results meets the warning conditions. This enables real-time monitoring of test case quality and ensures the accuracy of test case type identification and the stability and reliability of test case quality.

[0088] Example 2

[0089] Figure 3 This is a flowchart of another test case quality control method provided in Embodiment 2 of the present invention. This embodiment is a refinement of the "dynamic weight update of the bidirectional gated recurrent unit network according to the target input data" in the above embodiment. Specifically, it may include: inputting the target input data into a preset auxiliary network to obtain the output target dynamic weight, and updating the target dynamic weight to the bidirectional gated recurrent unit network.

[0090] Correspondingly, such as Figure 3 As shown, the method includes:

[0091] S310. Obtain the test cases to be tested and obtain the historical quantitative information corresponding to the test cases.

[0092] S320. Convert the test cases to vectorized test cases, and combine the vectorized test cases with historical quantization information to obtain the target input data.

[0093] S330. Input the target input data into the preset auxiliary network to obtain the output target dynamic weights. After updating the target dynamic weights into the bidirectional gated recurrent unit network, input the vectorized test cases into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0094] In this embodiment of the invention, the auxiliary network can be specifically understood as a sub-network used to generate dynamic weights. Specifically, it can consist of multiple fully connected layers and activation functions. The input to the auxiliary network is the target input data, and the output is the dynamic weights. The weights are dynamically generated based on the characteristics of the input data and updated in the main network (bidirectional gated recurrent unit network), improving the model performance of the main network and its adaptability to input data.

[0095] Optionally, based on the above embodiments, the auxiliary network is composed of a fully connected subnet, and the network layers connected in sequence in the auxiliary network may include: an input layer, at least one hidden layer, and an output layer;

[0096] The number of neurons in the input layer is equal to the sum of the number of all use case types in the case library and the number of all indicator types in the historical quantitative information. Each neuron receives one type of input feature.

[0097] The hidden layer is used for feature dimensionality reduction;

[0098] The output layer consists of only one neuron, used to output the target dynamic weights;

[0099] Correspondingly, the target input data is input into a preset auxiliary network to obtain the output target dynamic weights, which may include:

[0100] The target input data is input into a pre-defined auxiliary network; wherein, the vectorized test cases in the target input data are input into the input layer neurons corresponding to the test case type, and the historical quantification information in the target input data is input into the input layer neurons corresponding to the indicator type;

[0101] The target input data is processed by the input layer in the auxiliary network to generate a first auxiliary result;

[0102] The first auxiliary result is used to extract features from the hidden layers in the auxiliary network to generate the second auxiliary result;

[0103] The second auxiliary result is processed by the output layer of the auxiliary network to generate the target dynamic weights.

[0104] Specifically, the number of neurons in the input layer equals the sum of the total number of all use case types in the case library and the total number of all indicator types in the historical quantification information. Each neuron receives one type of input feature, ensuring that all features are included in the computation. The hidden layer generates lower-dimensional feature representations by extracting important features and combining different features to reduce computational complexity and avoid overfitting. The output layer has only one neuron, used to output the target dynamic weights.

[0105] Correspondingly, the target input data is input into the preset auxiliary network. The vectorized test cases in the target input data are input into the input layer neurons corresponding to the test case type (if the test case type corresponding to the neuron is labeled, the neuron input is 1, otherwise it is 0). The historical quantitative information in the target input data is input into the input layer neurons corresponding to the indicator type.

[0106] The input layer of the auxiliary network performs preliminary processing on the target input data. The integrated first auxiliary result is then passed to the hidden layer of the auxiliary network. The hidden layer performs feature extraction and dimensionality reduction on the preliminary processed first auxiliary result to generate a more compact second auxiliary result. The output layer of the auxiliary network processes the second auxiliary result to generate the target dynamic weights.

[0107] In a specific example Figure 4 This is a schematic diagram of an auxiliary network applicable to an embodiment of the present invention, such as... Figure 4 As shown, each neuron receives one type of input feature. For example, the historical case annotation accuracy is 0.85, the case classification confidence is 0.91, the requirement coverage is 0.97, and the performance capacity case and regression test case are 1 each, representing the test case types labeled as performance capacity cases and regression test cases. The input layer contains multiple neurons, each corresponding to a specific input feature. Assuming there are 38 test case types, each corresponding to 38 neurons, the historical case annotation accuracy, case classification confidence, review correction rate, and requirement coverage each correspond to 1 neuron, and the review correction rate corresponds to the same number of neurons as the number of test case types, 38, for a total of 80 neurons. The number of neurons in the input layer in the figure is 80 (1×80), indicating 80 input features. Vectorized test cases and historical quantization information are mapped to different neurons in the input layer, and after processing by the hidden layer, dimensionality-reduced features of 1×34 and 1×8 dimensions are obtained respectively. Finally, the output layer generates the target dynamic weights.

[0108] By inputting the target input data into a pre-defined auxiliary network, and through processing in the input layer, hidden layer, and output layer, dynamic target weights are generated. This allows the auxiliary network to dynamically adjust the weights of the main network based on the target input data, enabling the model to focus more on important features and thus improving its adaptability to input data and the accuracy of use case type identification.

[0109] S340. Input the features of the test cases to be tested into the preset local attention mechanism network to obtain the dimensionality-reduced test case features. Then, input the dimensionality-reduced test case features into the pre-trained branch fully connected network to obtain the target classification confidence. Based on the target classification confidence, obtain the target recognition result of the test cases.

[0110] S350. Compare the target identification results with the labeling results for the test cases, filter out abnormal results, and trigger a quality warning when the abnormal statistical value calculated based on the abnormal results meets the warning conditions.

[0111] The technical solution of this invention, after acquiring the test cases and their historical quantification information, converts the test cases into vector form and combines them with historical information to form target input data. The target input data is then input into a preset auxiliary network to generate target dynamic weights that match the input data features. These weights are then updated in a bidirectional gated recurrent unit network, improving the model's semantic understanding ability, enhancing its handling of domain terminology and logical dependencies, and avoiding the one-sidedness of feature extraction. Furthermore, the dynamic weight update mechanism can capture implicit vulnerability features of different test cases, reducing key point omissions, lowering label bias, improving classification accuracy, enhancing the model's adaptability to complex scenarios, improving generalization ability, and enhancing the interpretability and dynamism of warning rules. Vectorized test cases are input into the bidirectional gated recurrent unit network to extract test case features; dimensionality-reduced test case features are obtained through a local attention mechanism network; these dimensionality-reduced test case features are then input into a pre-trained fully connected branch network to obtain the target classification confidence and corresponding recognition results. By comparing the identification and labeling results, abnormal results are filtered and statistical values ​​are calculated. A quality alert is triggered when certain conditions are met. Further processing via a local attention mechanism network optimizes the interpretation of semantic information, thereby improving the accuracy of the fully connected branch network in identifying test case types. In terms of analysis dimensions, historical quantitative information is comprehensively considered to ensure the comprehensiveness and accuracy of the analysis results. The system automatically identifies abnormal test cases and triggers a quality alert when the calculated abnormal statistical values ​​meet the alert conditions, achieving real-time monitoring of test case quality and ensuring the accuracy of test case type identification and the stability and reliability of test case quality.

[0112] Example 3

[0113] Figure 5 This is a flowchart of another test case quality control method provided in Embodiment 3 of the present invention. This embodiment is a refinement of the above embodiment's step of "inputting the features of the test cases to be tested into a preset local attention mechanism network to obtain dimensionality-reduced test case features". Specifically, it may include: inputting the features of the test cases to be tested into a preset local attention mechanism network; using the local attention mechanism network to perform dimensionality changes on the features of the test cases according to a preset local window size to obtain transformed data blocks; using the local attention mechanism network to perform local attention mechanism operations on the transformed data blocks according to a preset segmented local window to obtain intermediate pooled window data; and using the local attention mechanism network to continue performing local attention mechanism operations on each window of the intermediate pooled window data until dimensionality-reduced test case features of a preset dimension are obtained.

[0114] Correspondingly, such as Figure 5 As shown, the method includes:

[0115] S510. Obtain the test cases to be tested and obtain the historical quantitative information corresponding to the test cases.

[0116] S520. Convert the test cases to vectorized test cases, and combine the vectorized test cases with historical quantization information to obtain the target input data.

[0117] S530. After dynamically updating the weights of the bidirectional gated recurrent unit network based on the target input data, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0118] Optionally, based on the above embodiments, the bidirectional gated cyclic unit network may include at least one bidirectional gated cyclic unit layer, at least one fully connected network layer, and an output layer connected in sequence.

[0119] In this layer, the input part of the bidirectional gated recurrent unit layer is used to receive input features or the input part is connected to the output part of the fully connected network layer.

[0120] The input part of the fully connected network layer is connected to the output part of the bidirectional gated loop unit, and the output part is connected to the input part of the bidirectional gated loop unit layer, or to the output layer.

[0121] The bidirectional gated loop unit layer includes a forward propagation gated loop unit layer and a backward propagation gated loop unit layer;

[0122] Among them, the forward propagation gated recurrent unit layer is used to process the data of each time step in chronological order starting from the beginning of the sequence, and calculate and output the forward hidden state features of the current time step.

[0123] The backpropagation gated recurrent unit layer is used to process the data of each time step in reverse chronological order, starting from the end of the sequence, to calculate and output the back hidden state feature of the current time step; the forward hidden state feature and the back hidden state feature are fused to obtain the inter-layer feature output by the bidirectional gated recurrent unit layer.

[0124] Accordingly, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases, which may include:

[0125] The vectorized test cases are input into the updated bidirectional gated recurrent cell network;

[0126] Vectorized test cases are processed through a bidirectional gated recurrent unit layer to generate inter-layer features.

[0127] The first inter-layer features are processed by a fully connected network layer to generate the second inter-layer features;

[0128] The second inter-layer features are processed through the remaining sequentially connected bidirectional gated recurrent unit layers and fully connected network layers in the network to generate the third inter-layer features;

[0129] The output layer processes the features between the third layer to generate features for the test cases.

[0130] Specifically, the bidirectional gated recurrent unit network includes at least one bidirectional gated recurrent unit layer, at least one fully connected network layer, and an output layer. These layers are connected sequentially to form a network architecture for extracting features from vectorized test cases.

[0131] The input portion of a bidirectional gated recurrent unit (NRU) layer can receive input features from external sources or be connected to the output portion of a fully connected network layer to receive processed features. Each NRU layer includes a forward propagation gated recurrent unit (RPU) layer and a backward propagation gated recurrent unit (RPU) layer. The forward propagation RPU layer calculates and outputs the forward hidden state features for each time step, starting from the beginning of the sequence and proceeding in chronological order. The backward propagation RPU layer calculates and outputs the backward hidden state features for each time step, starting from the end of the sequence and proceeding in reverse chronological order. The forward and backward hidden state features are then fused (e.g., concatenated or summed) to form the inter-layer features output by the NRU layer.

[0132] The input of the fully connected layer is connected to the output of the bidirectional gated recurrent unit layer, receiving inter-layer features and performing linear transformations and nonlinear mappings on the input features to extract higher-level feature representations. The output of the fully connected layer can be connected to the input of another bidirectional gated recurrent unit layer for further feature extraction, or directly connected to the output layer to output features. The output layer performs final processing on the input features to generate the features for the test cases.

[0133] Accordingly, vectorized test cases are input into the updated bidirectional gated recurrent unit (BRN) network. The BRN layers compute the forward and backward hidden state features at each time step and fuse them to form the first inter-layer features. The fully connected layer receives the first inter-layer features as input and performs feature transformations through linear transformations and nonlinear activation functions to extract higher-level feature representations, generating the second inter-layer features. The second inter-layer features are then further processed by the remaining BRN layers in the network to extract features from a time-series perspective, and by the fully connected layers for feature transformation and dimensionality reduction, generating the third inter-layer features. The output layer receives the third inter-layer features and performs final processing (e.g., nonlinear transformations or normalization through activation functions) to generate the final test case features.

[0134] In a specific example Figure 6 This is a schematic diagram of a bidirectional gated cyclic unit network applicable to an embodiment of the present invention, as shown below. Figure 6 As shown, the database stores multiple test cases. Quantization information (historical quantization information) and test cases to be tested are obtained from the database, and the test cases to be tested are transformed into vectorized test cases (vectorized test cases). The vectorized test cases and historical quantization information are input into the auxiliary network to obtain dynamic weights, which are then updated into the bidirectional gated recurrent unit network. The vectorized test cases are input into the bidirectional gated recurrent unit network, and after processing by the bidirectional gated recurrent unit layer and the fully connected network, the output layer finally outputs the feature vector, which is the feature of the test case.

[0135] Figure 7 This is a schematic diagram of a bidirectional gated loop unit layer applicable to an embodiment of the present invention, as shown below. Figure 7 As shown, the input sequence is represented in time steps as T-1, T, T+1, and T+2, etc., with each time step corresponding to one input data point. The time order from left to right is forward propagation, and the time order from right to left is backward propagation. The gated recurrent unit (ROU) for forward propagation starts from the beginning of the sequence, T-1, and processes the data of each time step sequentially according to the time order, calculating and outputting the forward hidden state feature of the current time step. The gated recurrent unit for backward propagation starts from the end of the sequence, T+2, and processes the data of each time step in reverse chronological order, calculating and outputting the backward hidden state feature of the current time step. The forward and backward hidden state features are fused and used as the output of the bidirectional gated recurrent unit layer.

[0136] By inputting vectorized test cases into an updated bidirectional gated recurrent unit network, which then processes the data sequentially through bidirectional gated recurrent unit layers, fully connected network layers, and subsequent sequentially connected network layers, forward and reverse data features are fused. Features are extracted and reduced layer by layer, and finally, the output layer generates the features of the test cases. By refining the features of the test cases layer by layer, the semantic understanding of the test cases is improved, ensuring the comprehensiveness and accuracy of feature extraction. This, in turn, ensures the accuracy of test case type identification and the stability and reliability of test case quality.

[0137] S540. Input the features of the test cases to be tested into the preset local attention mechanism network.

[0138] S550 uses a local attention mechanism network to perform dimensional changes on the features of the test cases according to a preset local window size, resulting in transformed data blocks.

[0139] Specifically, the features of the test cases are input into a pre-defined local attention mechanism network, which then modifies the dimensions of the input features according to a pre-defined local window size. For example, the feature sequence is divided into multiple local windows, each containing a pre-defined number of time steps or feature dimensions (corresponding to the pre-defined local window size), thus obtaining the transformed data block.

[0140] S560. The local attention mechanism network performs local attention mechanism operations on the transformed data blocks according to the preset segmented local windows to obtain intermediate pooled window data.

[0141] In this embodiment of the invention, the preset segmented local window can be specifically understood as: the range of data areas to be processed each time in a data block, defined according to the specific task and data characteristics.

[0142] Specifically, based on the preset segmented local windows, attention mechanism operations are performed independently within each segmented local window. The features of each local window are compressed and pooled to form a dimensionality-reduced feature representation, resulting in intermediate pooled window data.

[0143] S570. The local attention mechanism network is used to continue to perform local attention mechanism operations on each window of the intermediate pooling window data until the dimensionality reduction test case features of the preset dimension are obtained. The dimensionality reduction test case features are then input into the pre-trained branch fully connected network to obtain the target classification confidence and the target recognition result of the test case is obtained based on the target classification confidence.

[0144] Specifically, through repeated computation and pooling by the local attention mechanism network, the feature dimensionality is reduced while retaining key information in the features, until a dimensionality-reduced use case feature with a preset dimension is obtained. Then, the dimensionality-reduced use case feature is input into a pre-trained branch fully connected network to obtain the target classification confidence, and the target recognition result of the test case is obtained based on the target classification confidence.

[0145] In a specific example Figure 8 This is a schematic diagram of a local attention mechanism network applicable to an embodiment of the present invention, as shown below. Figure 8As shown, the vectorized test cases (the features of the test cases output by the bidirectional gated recurrent unit network) are input into the local attention mechanism network in the form of a one-dimensional vector, including multiple feature dimensions (such as 1, 2, 3, ..., n), assumed to be 400×1. Since the network configuration of the bidirectional gated recurrent unit network in the previous step is a temporal sequence, in order to maintain the temporal features, the features of the test cases are dimensionally transformed according to the preset local window size (4×100), and the data is arranged vertically to obtain transformed data blocks (4×100 two-dimensional matrix blocks). According to the preset segmented local window (4×4×25), attention mechanism operations are performed on each window of the transformed data block. The first pooling transforms the data dimension to 2×2×25 to obtain intermediate pooled window data. The attention mechanism operation is then performed on each window. The second pooling transforms the data to 1×1×25 to obtain the dimensionality-reduced test case features of the preset dimension (1×25).

[0146] S580. Compare the target identification results with the labeling results for the test cases, filter out abnormal results, and trigger a quality warning when the abnormal statistical value calculated based on the abnormal results meets the warning conditions.

[0147] The technical solution of this invention involves acquiring the test cases and their historical quantization information, converting the test cases into vector form, and combining them with historical information to form target input data. This data is then used to dynamically update the weights of the bidirectional gated recurrent unit network, inputting the vectorized test cases to extract features from the test cases. After inputting the test case features into a preset local attention mechanism network, the network performs dimensionality transformation on the features according to a preset local window size, obtaining adapted transformed data blocks. Based on a preset segmented local window, the network performs local attention mechanism operations on the transformed data blocks to obtain intermediate pooling window data, achieving initial dimensionality reduction and key feature extraction. Local attention mechanism operations are then performed on each window of the intermediate pooling window data, gradually deepening feature extraction until finally obtaining dimensionality-reduced test case features that conform to the preset dimensions. The dimensionality-reduced test case features are then input into a pre-trained fully connected branch network to obtain the target classification confidence and the corresponding recognition result. By comparing the identification and labeling results, anomalies are filtered and statistical values ​​are calculated. Quality alerts are triggered when certain conditions are met. A bidirectional gated recurrent unit network with dynamically updated weights processes and extracts data features, improving adaptability to complex scene changes, generalization ability, and semantic understanding of test cases. Further processing via a local attention mechanism network optimizes the interpretation of semantic information, reduces data dimensionality, and enhances the model's ability to capture key features, thereby improving the accuracy of the fully connected branch network in identifying test case types. In terms of analysis, historical quantitative information is comprehensively considered to ensure the comprehensiveness and accuracy of the analysis results. The system automatically identifies abnormal test cases and triggers quality alerts when the calculated anomaly statistics meet the alert conditions, enabling real-time monitoring of test case quality and ensuring the accuracy of test case type identification and the stability and reliability of test case quality.

[0148] Example 4

[0149] Figure 9 This is a schematic diagram of a test case quality control device provided in Embodiment 4 of the present invention. Figure 9 As shown, the device includes: a module to be tested 910, a target input module 920, a test case feature module 930, a target recognition module 940, and a quality early warning module 950, wherein:

[0150] The module to be tested, 910, is used to obtain the test cases and the historical quantitative information corresponding to the test cases.

[0151] The target input module 920 is used to convert the test cases into vectorized test cases and combine the vectorized test cases with historical quantization information to obtain target input data.

[0152] The test case feature module 930 is used to dynamically update the weights of the bidirectional gated recurrent unit network based on the target input data, and then input the vectorized test cases into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0153] The target recognition module 940 is used to input the features of the test cases to a preset local attention mechanism network to obtain the dimensionality reduction test case features, and then input the dimensionality reduction test case features to a pre-trained branch fully connected network to obtain the target classification confidence, and obtain the target recognition result of the test cases based on the target classification confidence.

[0154] The quality early warning module 950 is used to compare the target identification results with the labeling results for the test cases, filter out abnormal results, and trigger a quality early warning when the abnormal statistical value calculated based on the abnormal results meets the early warning conditions.

[0155] The technical solution of this invention involves acquiring the test cases and their historical quantification information, converting the test cases into vector form, and combining them with historical information to form target input data. This data is then used to dynamically update the weights of the bidirectional gated recurrent unit network (BRN), inputting the vectorized test cases to extract test case features. Dimensionality reduction features are obtained through a local attention mechanism network, and these features are input into a pre-trained fully connected branch network to obtain the target classification confidence and corresponding recognition results. The recognition results are compared with the labeling results, abnormal results are filtered, and statistical values ​​are calculated. A quality warning is triggered when certain conditions are met. The dynamically weighted BRN network processes and extracts data features, improving adaptability to complex scene changes, generalization ability, and semantic understanding of test cases. Further processing through the local attention mechanism network optimizes the interpretation of semantic information, thereby improving the accuracy of the fully connected branch network in identifying test case types. In terms of analysis dimensions, historical quantification information is comprehensively considered to ensure the comprehensiveness and accuracy of the analysis results. The system automatically identifies abnormal test cases and triggers a quality warning when the abnormal statistical value calculated based on the abnormal results meets the warning conditions. This enables real-time monitoring of test case quality and ensures the accuracy of test case type identification and the stability and reliability of test case quality.

[0156] Based on the above embodiments, the branch fully connected network may include: a training network branch for updating parameters during the training phase, an incremental network branch for updating parameters during the online learning phase, a fully connected subnetwork connected to the training network branch and the incremental network branch respectively, and a classification confidence generation subnetwork connected to the fully connected subnetwork.

[0157] Accordingly, based on the above embodiments, the target recognition module 940 is specifically used for:

[0158] The dimensionality reduction use case features are input into the training network branch and the incremental network branch, respectively;

[0159] The network branches are trained to generate the first feature results corresponding to the features of the dimensionality reduction use cases, and the first feature results are output to the fully connected subnet.

[0160] The second feature result corresponding to the dimensionality reduction use case features is generated by incremental network branching, and the second feature result is output to the fully connected subnet;

[0161] The first feature result and the second feature result are fused through a fully connected subnet to obtain the fused result, and the fused result is output to the classification confidence generation subnet.

[0162] Subnets are generated based on classification confidence scores, and target classification confidence scores are generated based on the fusion processing results.

[0163] Furthermore, based on the above embodiments, the test case quality control device may further include: a review and screening module, a trained module, and an incremental training module, wherein:

[0164] The review and filtering module is used to review and filter abnormal results after they are filtered out, and to add the test cases with errors to the error test case set.

[0165] The trained module is used to extract a target number of trained test cases from the test case library according to preset rules when the number of test cases to be tested in the error test case set is greater than or equal to the preset number of training batches.

[0166] The incremental training module is used to use the test cases from both the trained test cases and the error test cases as incremental training data to perform incremental training on the incremental network branches.

[0167] Based on the above embodiments, the use case feature module 930 is specifically used for:

[0168] The target input data is input into a preset auxiliary network to obtain the output target dynamic weights, and the target dynamic weights are updated into the bidirectional gated recurrent unit network.

[0169] Based on the above embodiments, the auxiliary network is composed of a fully connected subnet, and the network layers connected sequentially in the auxiliary network may include: an input layer, at least one hidden layer, and an output layer.

[0170] The number of neurons in the input layer is equal to the sum of the number of all use case types in the case library and the number of all indicator types in the historical quantitative information. Each neuron receives one type of input feature.

[0171] The hidden layer is used for feature dimensionality reduction;

[0172] The output layer consists of only one neuron, used to output the target dynamic weights;

[0173] Accordingly, based on the above embodiments, the use case feature module 930 is further used for:

[0174] The target input data is input into a pre-defined auxiliary network; wherein, the vectorized test cases in the target input data are input into the input layer neurons corresponding to the test case type, and the historical quantification information in the target input data is input into the input layer neurons corresponding to the indicator type;

[0175] The target input data is processed by the input layer in the auxiliary network to generate a first auxiliary result;

[0176] The first auxiliary result is used to extract features from the hidden layers in the auxiliary network to generate the second auxiliary result;

[0177] The second auxiliary result is processed by the output layer of the auxiliary network to generate the target dynamic weights.

[0178] Based on the above embodiments, the bidirectional gated cyclic unit network may include at least one bidirectional gated cyclic unit layer, at least one fully connected network layer, and an output layer connected in sequence.

[0179] In this layer, the input part of the bidirectional gated recurrent unit layer is used to receive input features or the input part is connected to the output part of the fully connected network layer.

[0180] The input part of the fully connected network layer is connected to the output part of the bidirectional gated loop unit, and the output part is connected to the input part of the bidirectional gated loop unit layer, or to the output layer.

[0181] The bidirectional gated loop unit layer includes a forward propagation gated loop unit layer and a backward propagation gated loop unit layer;

[0182] Among them, the forward propagation gated recurrent unit layer is used to process the data of each time step in chronological order starting from the beginning of the sequence, and calculate and output the forward hidden state features of the current time step.

[0183] The backpropagation gated recurrent unit layer is used to process the data of each time step in reverse chronological order, starting from the end of the sequence, to calculate and output the back hidden state feature of the current time step; the forward hidden state feature and the back hidden state feature are fused to obtain the inter-layer feature output by the bidirectional gated recurrent unit layer.

[0184] Accordingly, based on the above embodiments, the use case feature module 930 is further used for:

[0185] The vectorized test cases are input into the updated bidirectional gated recurrent cell network;

[0186] Vectorized test cases are processed through a bidirectional gated recurrent unit layer to generate inter-layer features.

[0187] The first inter-layer features are processed by a fully connected network layer to generate the second inter-layer features;

[0188] The second inter-layer features are processed through the remaining sequentially connected bidirectional gated recurrent unit layers and fully connected network layers in the network to generate the third inter-layer features;

[0189] The output layer processes the features between the third layer to generate features for the test cases.

[0190] Based on the above embodiments, the target recognition module 940 is further configured to:

[0191] The features of the test cases are input into a pre-defined local attention mechanism network;

[0192] The local attention mechanism network transforms the features of the test cases according to the preset local window size to obtain transformed data blocks.

[0193] The local attention mechanism network performs local attention mechanism operations on the transformed data blocks according to the preset segmented local windows to obtain intermediate pooled window data;

[0194] The local attention mechanism network is used to continue the local attention mechanism operation on each window of the intermediate pooled window data until the dimensionality reduction use case features of the preset dimension are obtained.

[0195] The test case quality control device provided in this embodiment of the invention can execute the test case quality control method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0196] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0197] In the technical solution disclosed herein, the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse.

[0198] In the technical solution disclosed herein, if automated decision-making is involved, a corresponding operation entry will be provided to the user, allowing the user to choose to agree to or reject the automated decision result; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0199] Example 5

[0200] Figure 10 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0201] like Figure 10 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0202] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0203] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as test case quality control methods, i.e.:

[0204] Obtain the test cases and their corresponding historical quantitative information;

[0205] The test cases are converted into vectorized test cases, and the vectorized test cases and historical quantitative information are combined to obtain the target input data.

[0206] After dynamically updating the weights of the bidirectional gated recurrent unit network based on the target input data, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases.

[0207] The features of the test cases are input into a pre-set local attention mechanism network to obtain dimensionality-reduced test case features. Then, the dimensionality-reduced test case features are input into a pre-trained branch fully connected network to obtain the target classification confidence. Based on the target classification confidence, the target recognition result of the test cases is obtained.

[0208] The target identification results are compared with the labeling results for the test cases to filter out abnormal results, and a quality warning is triggered when the abnormal statistical value calculated based on the abnormal results meets the warning conditions.

[0209] In some embodiments, the test case quality control method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the test case quality control method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the test case quality control method by any other suitable means (e.g., by means of firmware).

[0210] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0211] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0212] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0213] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0214] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0215] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0216] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0217] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for controlling the quality of test cases, characterized in that, include: Obtain the test cases and their corresponding historical quantitative information; The test cases are converted into vectorized test cases, and the vectorized test cases and historical quantitative information are combined to obtain the target input data. After dynamically updating the weights of the bidirectional gated recurrent unit network based on the target input data, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases. The features of the test cases are input into a pre-set local attention mechanism network to obtain dimensionality-reduced test case features. Then, the dimensionality-reduced test case features are input into a pre-trained branch fully connected network to obtain the target classification confidence. Based on the target classification confidence, the target recognition result of the test cases is obtained. The target identification results are compared with the labeling results for the test cases to filter out abnormal results, and a quality warning is triggered when the abnormal statistical value calculated based on the abnormal results meets the warning conditions.

2. The method according to claim 1, characterized in that, The fully connected branch network includes: a training network branch for updating parameters during the training phase, an incremental network branch for updating parameters during the online learning phase, fully connected subnetworks connected to the training network branch and the incremental network branch respectively, and a classification confidence generation subnetwork connected to the fully connected subnetworks. Accordingly, the dimensionality-reduced use case features are input into a pre-trained fully connected branch network to obtain the target classification confidence, including: The dimensionality reduction use case features are input into the training network branch and the incremental network branch, respectively; The network branches are trained to generate the first feature results corresponding to the features of the dimensionality reduction use cases, and the first feature results are output to the fully connected subnet. The second feature result corresponding to the dimensionality reduction use case features is generated by incremental network branching, and the second feature result is output to the fully connected subnet; The first feature result and the second feature result are fused through a fully connected subnet to obtain the fused result, and the fused result is output to the classification confidence generation subnet. Subnets are generated based on classification confidence scores, and target classification confidence scores are generated based on the fusion processing results.

3. The method according to claim 1, characterized in that, After filtering out abnormal results, the process also includes: Review and screen abnormal results, and add the test cases with incorrect identification to the error test case set; When the number of test cases to be tested in the error test case set is greater than or equal to the preset number of training batches, the target number of trained test cases are extracted from the test case library according to the preset rules. The test cases from both the trained and erroneous test cases are used as incremental training data to incrementally train the incremental network branch.

4. The method according to claim 1, characterized in that, Dynamic weight updates are performed on the bidirectional gated recurrent unit network based on the target input data, including: The target input data is input into a preset auxiliary network to obtain the output target dynamic weights, and the target dynamic weights are updated into the bidirectional gated recurrent unit network.

5. The method according to claim 4, characterized in that, The auxiliary network consists of fully connected subnets. The network layers connected in sequence in the auxiliary network include: an input layer, at least one hidden layer, and an output layer. The number of neurons in the input layer is equal to the sum of the number of all use case types in the case library and the number of all indicator types in the historical quantitative information. Each neuron receives one type of input feature. The hidden layer is used for feature dimensionality reduction; The output layer consists of only one neuron, used to output the target dynamic weights; Accordingly, the target input data is input into a preset auxiliary network to obtain the output target dynamic weights, including: The target input data is input into a pre-defined auxiliary network; wherein, the vectorized test cases in the target input data are input into the input layer neurons corresponding to the test case type, and the historical quantification information in the target input data is input into the input layer neurons corresponding to the indicator type; The target input data is processed by the input layer in the auxiliary network to generate a first auxiliary result; The first auxiliary result is used to extract features from the hidden layers in the auxiliary network to generate the second auxiliary result; The second auxiliary result is processed by the output layer of the auxiliary network to generate the target dynamic weights.

6. The method according to claim 1, characterized in that, A bidirectional gated cyclic cell network comprises at least one bidirectional gated cyclic cell layer, at least one fully connected network layer, and an output layer connected in sequence. In this layer, the input part of the bidirectional gated recurrent unit layer is used to receive input features or the input part is connected to the output part of the fully connected network layer. The input part of the fully connected network layer is connected to the output part of the bidirectional gated loop unit, and the output part is connected to the input part of the bidirectional gated loop unit layer, or to the output layer. The bidirectional gated loop unit layer includes a forward propagation gated loop unit layer and a backward propagation gated loop unit layer; Among them, the forward propagation gated recurrent unit layer is used to process the data of each time step in chronological order starting from the beginning of the sequence, and calculate and output the forward hidden state features of the current time step. The backpropagation gated recurrent unit layer is used to process the data of each time step in reverse chronological order, starting from the end of the sequence, to calculate and output the back hidden state feature of the current time step; the forward hidden state feature and the back hidden state feature are fused to obtain the inter-layer feature output by the bidirectional gated recurrent unit layer. Accordingly, the vectorized test cases are input into the updated bidirectional gated recurrent unit network to obtain the features of the test cases, including: The vectorized test cases are input into the updated bidirectional gated recurrent cell network; Vectorized test cases are processed through a bidirectional gated recurrent unit layer to generate inter-layer features. The first inter-layer features are processed by a fully connected network layer to generate the second inter-layer features; The second inter-layer features are processed through the remaining sequentially connected bidirectional gated recurrent unit layers and fully connected network layers in the network to generate the third inter-layer features; The output layer processes the features between the third layer to generate features for the test cases.

7. The method according to claim 1, characterized in that, The features of the test cases are input into a pre-defined local attention mechanism network to obtain dimensionality-reduced test case features, including: The features of the test cases are input into a pre-defined local attention mechanism network; The local attention mechanism network transforms the features of the test cases according to the preset local window size to obtain transformed data blocks. The local attention mechanism network performs local attention mechanism operations on the transformed data blocks according to the preset segmented local windows to obtain intermediate pooled window data; The local attention mechanism network is used to continue the local attention mechanism operation on each window of the intermediate pooled window data until the dimensionality reduction use case features of the preset dimension are obtained.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the test case quality control method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the test case quality control method according to any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the test case quality control method according to any one of claims 1-7.

Images