Defect determination method and device, computer device, and storage medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses a defect determination method and device, computer equipment and a storage medium, which comprise the following steps: collecting software data of target software, inputting the software data into a pre-trained defect prediction model for defect risk prediction processing to obtain target risk defects with risk evaluation indexes higher than a preset index threshold; taking the target risk defects as target nodes, extracting feature data from the software data to generate associated nodes, forming a cause-effect dependency graph according to the cause-effect correlation between the target nodes and the associated nodes; based on a graph model cause-effect reasoning algorithm and the cause-effect dependency graph, screening candidate root causes of the target risk defects from the associated nodes; and determining root cause results of the target risk defects according to the associated data of the candidate root causes. The application can be specifically used in the fields of financial technology and health care, improves the identification accuracy of defects and root causes of software, and thus improves the software management efficiency.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and is applied to fintech and healthcare scenarios. In particular, it relates to a defect determination method, apparatus, computer equipment, and storage medium. Background Technology

[0002] In software development and testing within fintech and healthcare scenarios, timely defect prediction, accurate location, and effective verification are crucial for improving software quality, reducing development costs, and shortening delivery cycles. As software scales and architectures become increasingly complex (e.g., the widespread adoption of microservices and cloud-native architectures), traditional defect management methods are no longer sufficient to meet the demands for efficiency and accuracy. Specifically, relevant technical solutions typically perform defect prediction separately, such as predicting defect risks through code features and log data. Some solutions also use log analysis to pinpoint the root cause of defects during the prediction process. However, this approach struggles to guarantee the accuracy of root cause identification, easily leading to misjudgments or omissions, resulting in low accuracy and efficiency in defect remediation. Summary of the Invention

[0003] This invention provides a defect determination method, apparatus, computer equipment, and medium, aiming to solve the technical problems of difficulty in identifying defects and accurately determining the causes of defects in the current software operation and maintenance process in the financial and medical fields.

[0004] Firstly, a defect determination method is provided, including: Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. Using the target risk defect as the target node, feature data is extracted from the software data to generate associated nodes, and a causal dependency graph is formed based on the causal relationship between the target node and the associated nodes. Based on the graph model causal reasoning algorithm and the causal dependency graph, candidate root causes of the target risk defect are selected from the associated nodes; Based on the correlation data of the candidate root causes, the root cause result of the target risk defect is determined.

[0005] Secondly, a defect determination apparatus is provided, comprising: The defect prediction module is used to collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. The causal construction module is used to extract feature data from the software data to generate associated nodes, taking the target risk defect as the target node, constructing the causal relationship between the target node and the associated nodes, and forming a causal dependency graph. The candidate root cause reasoning module is used to filter out candidate root causes of the target risk defect from the associated nodes based on the graph model causal reasoning algorithm and the causal dependency graph. The root cause identification module is used to determine the root cause result of the target risk defect based on the correlation data of the candidate root causes.

[0006] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described defect determination method.

[0007] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned defect determination method.

[0008] The aforementioned defect identification method, apparatus, computer equipment, and storage medium automatically collect software data from the target software. Then, based on a defect prediction model, it identifies target risk defects with risk assessment indicators exceeding a preset threshold, improving the efficiency and accuracy of defect identification. By using the target risk defect as the target node and extracting feature data from the software data to generate associated nodes, a causal dependency graph is constructed. This establishes a visual and traceable causal relationship between abstract defects and specific software feature data. The graph model causal reasoning algorithm can filter out the candidate root causes with the strongest correlation to the target risk defect based on the strength of causal relationships between nodes in the causal dependency graph. Furthermore, the constructed causal dependency graph is only built for the identified target risk defects, eliminating the need to build it across the entire software, thus reducing the amount of processing required to generate the causal dependency graph. Based on this causal dependency graph, candidate target risk defects can be inferred more efficiently. Then, by obtaining the associated data of the candidate root causes, the confidence level of the candidate root causes is further verified, ensuring the accuracy and reliability of the root cause results. This provides a clear direction for subsequent defect remediation and improves the overall efficiency of software operation and maintenance. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the 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.

[0010] Figure 1 This is a schematic diagram of an application environment for a defect determination method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a defect determination method according to an embodiment of the present invention; Figure 3 This is another flowchart illustrating the defect determination method in one embodiment of the present invention; Figure 4 This is another flowchart illustrating the defect determination method in one embodiment of the present invention; Figure 5 This is another schematic flowchart of the defect determination method in one embodiment of the present invention; Figure 6 This is a schematic diagram of a defect determination device in one embodiment of the present invention; Figure 7 This is another structural schematic diagram of the defect determination device in one embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 9 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] The defect determination method provided in this embodiment of the invention can be applied to, for example... Figure 1In this application environment, the client can deploy target software. The server can collect and analyze software data from the client, inputting the data into a pre-trained defect prediction model for defect risk prediction. This identifies target risk defects with risk assessment indicators exceeding a preset threshold. Using the target risk defect as the target node, feature data is extracted from the software data to generate associated nodes. Based on the causal relationships between the target node and associated nodes, a causal dependency graph is formed. Then, based on the graph model causal inference algorithm and the causal dependency graph, candidate root causes of the target risk defect are selected from the associated nodes. Based on the association data of the candidate root causes, the root cause of the target risk defect is determined. This method can predict defects in advance and accurately locate root causes, effectively reducing the defect incidence rate after software deployment and improving software stability and reliability. This method is adaptable to different types and sizes of target software (such as office software, industrial control software, and internet application software). Clients can be, but are not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0013] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the defect determination method provided in this embodiment of the invention includes the following steps: S101, Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. S102, taking the target risk defect as the target node, extract feature data from the software data to generate associated nodes, and form a causal dependency graph based on the causal relationship between the target node and the associated nodes; S103, Based on the graph model causal reasoning algorithm and the causal dependency graph, candidate root causes of the target risk defect are selected from the associated nodes; S104, Based on the correlation data of the candidate root causes, determine the root cause result of the target risk defect.

[0014] The defect identification method provided by this invention can be applied to software in various application scenarios, enabling early defect screening and risk assessment during the software development, testing, and maintenance stages. The software referred to in this application includes not only general-purpose applications but also business systems, and its scope of application is not limited by software form or deployment method. For example, in the medical field, hospitals often use Hospital Information Systems (HIS) to achieve integrated management of personnel, materials, finances, and medical services. Alternatively, they may use Picture Archiving and Communication Systems (PACS) to store, transmit, and access images such as CT, MRI, DR, ultrasound, and endoscopy. In the financial field, it can be applied to order management systems, transaction payment systems, and commodity maintenance systems to ensure stable operation of the entire business process, including order flow, transaction interaction, and data statistics.

[0015] In step S101 of some embodiments, the data acquisition and high-risk defect identification of the target software are first completed. Specifically, all software data generated during the operation of the target software is collected in real time or periodically. This software data includes multi-source heterogeneous data such as software operation logs, interface interaction data, business process operation parameters, configuration parameters, test operation data, and business interaction records. After standardized preprocessing operations such as data cleaning, format unification, and missing value imputation are completed, the processed standardized software data is input into a pre-trained and converged defect prediction model to perform defect risk prediction processing on the input data. In one feasible implementation, the defect prediction model outputs the defect type and risk assessment index of the corresponding sample. Then, each risk assessment index is compared and filtered with a pre-set index threshold. Defects with overall risk assessment indices higher than the preset index threshold are selected and identified as target risk defects to be analyzed.

[0016] In step S102 of some embodiments, the identified target risk defect is set as the target node. Then, various dimensional feature data that can directly or indirectly affect the target risk defect are extracted from the software data. Based on the business entities, operational elements, configuration conditions, and process events corresponding to the feature data, the strength of the causal relationship between the target node and each associated node is determined through mutual information calculation combined with a causal discovery algorithm. This constructs various associated nodes that match the target node. Afterwards, the topology network between the nodes is completed, resulting in a causal dependency graph.

[0017] For example, taking a financial order system as an example, in response to the high-risk defects in the order payment module, a cross-modal causal dependency graph is constructed by combining the associated data in the software data: with "payment interface exception" as the target node, the associated nodes include "payment interface code logic", "database connection pool configuration", "third-party payment interface response", "reason for test case execution failure" and "online interface response time".

[0018] In step S103 of some embodiments, after constructing the causal dependency graph, a graph model causal inference algorithm is used to perform hierarchical traversal and association analysis on the causal dependency graph to determine the candidate root causes of the target node. In one feasible implementation, first-level related nodes with direct causal relationship to the target node are first located in the causal dependency graph, and then second- and third-level related nodes with causal relationship to the first-level related nodes are extended to form a complete causal transmission link between the target node and related nodes. At the same time, the algorithm quantifies the causal influence weight of each related node on the target node, and combines the causal flow between nodes (including positive influence, negative influence, and indirect transmission) to rank all related nodes by correlation. Subsequently, invalid nodes are removed based on the ranking results. For example, invalid nodes can be nodes whose causal influence weight is lower than a preset weight threshold. Multiple related nodes that are most likely to directly or indirectly cause the target risk defect are selected, and these related nodes are uniformly classified as candidate root causes of the target risk defect.

[0019] It should be noted that graph model causal reasoning algorithms can employ Bayesian network reasoning algorithms, Markov network reasoning algorithms, structural causal reasoning algorithms, etc. All of them can rely on the constructed causal dependency graph to complete the screening of related nodes and the extraction of candidate root causes, adapting to the root cause localization needs in different software scenarios. The specific choice can be made according to actual needs.

[0020] In step S104 of some embodiments, the final root cause is confirmed based on the candidate root causes obtained from the preliminary screening. The associated data corresponding to the candidate root causes are retrieved respectively. The associated data may include original running data, configuration data, interaction logs, and process execution records, etc. The candidate root causes are verified to determine the candidate root causes that can directly lead to the occurrence of the target risk defect, and finally the unique or multiple root cause results corresponding to the target risk defect are determined.

[0021] In this embodiment, by automatically collecting software data from the target software and inputting it into a pre-trained model, target risk defects with risk assessment indicators exceeding preset thresholds can be quickly identified, effectively improving defect identification efficiency and accuracy. Then, using the selected target risk defects as target nodes, feature data is extracted from the software data to construct associated nodes and build a causal dependency graph. The associated nodes are then filtered using a graph model causal inference algorithm combined with the causal dependency graph to identify highly correlated candidate root causes of defects. This causal dependency graph is built only for identified high-risk defects, eliminating the need for modeling based on the entire software data set, significantly reducing the data processing volume for graph construction. Based on this, the associated data corresponding to the candidate root causes is retrieved to complete confidence verification, further ensuring the accuracy and reliability of the final root cause location result, and improving the overall efficiency of software daily operation and maintenance and defect management.

[0022] like Figure 3 As shown, Figure 3 This is a flowchart of a defect determination method provided in another embodiment of this application; the defect determination method may include, but is not limited to, steps S201 to S202: S201, Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the defect types of the target software and the risk assessment index of the defect types. S202, the defect type that is higher than the preset indicator threshold is identified as the target risk defect.

[0023] In step S201 of some embodiments, various types of software data generated during the operation of the target software are collected in real time or periodically, including code data, runtime log data, test case data, online runtime indicator data, user feedback data, customer complaints, and other multi-source heterogeneous data. The data is then cleaned, aligned, and encoded to obtain a standardized dataset. This preprocessed standardized dataset is input into a defect prediction model that has completed convergence training. The defect prediction model performs defect risk prediction processing on the input data, outputting the various defect types actually existing in the target software, and the corresponding risk assessment indicators for each defect type. These risk assessment indicators may include quantitative evaluation parameters such as defect risk scores and defect occurrence probabilities.

[0024] In step S202 of some embodiments, indicator thresholds are set in advance based on software operation level, business importance and operation and maintenance management standards, etc. The risk assessment indicators corresponding to various defect types predicted are compared and screened one by one with the preset indicator thresholds. Defect types with risk assessment indicator values ​​higher than the preset indicator thresholds are selected as high-risk defects, and such high-risk defects are identified as target risk defects.

[0025] Furthermore, in one embodiment, the training method for the defect prediction model may include the following steps S301 to S306: S301, Obtain historical software data and corresponding defect tags of the target software; the defect tags include historical defect types and historical risk indicators; S302, Based on the multimodal feature fusion layer of the defect prediction model, different types of data features are extracted from the historical software data respectively, and multi-feature fusion processing is performed through an attention mechanism to obtain a sample fusion feature vector; S303, construct a comparison sample pair based on the sample fusion feature vector, and calculate the comparison loss; S304, The sample fusion feature vector is input into the defect prediction layer of the defect prediction model to obtain the sample defect type and sample risk assessment index; S305, compare the sample defect type and the sample risk assessment index with the historical defect type and the historical risk index respectively to determine the predicted loss; S306, Generate a joint loss based on the prediction loss and the contrast loss, and backpropagate the joint loss to the multimodal feature fusion layer and the defect prediction layer of the defect prediction model. Iterate and train the defect prediction model. When the joint loss converges to a preset threshold, a pre-trained defect prediction model is obtained.

[0026] In one embodiment, the defect prediction model is configured with a multimodal feature fusion layer and a defect prediction layer from top to bottom. The two layers are connected sequentially. The multimodal feature fusion layer serves as the front-end feature processing module, responsible for extracting and optimizing features from multi-source heterogeneous data. The defect prediction layer serves as the back-end output module, relying on the fused feature vectors to predict defect-related content. The parameters of the two modules are updated in tandem to jointly complete the overall training and optimization of the model.

[0027] In step S301 of some embodiments, historical software data retained from past operations of the target software or software of the same type as the target software is collected, and corresponding defect label datasets are matched and labeled. The defect labels include actual historical defect types and historical risk indicators. The historical software data is then linked and bound to the corresponding defect labels to form a model training dataset. The historical risk indicators can be one or more of the following: historical defect risk score, historical defect occurrence probability, and historical defect risk level, selected according to the actual situation.

[0028] In step S302 of some embodiments, historical software data from the training dataset is input into the multimodal feature fusion layer of the defect prediction model. Historical software data can be data from different sources and in different formats (such as code snippets, log text, and indicator values). The multimodal feature fusion layer incorporates a multi-branch independent feature extraction unit and a multi-head attention fusion unit. The independent feature extraction unit can extract features from different modalities within the training data, such as structured data, semi-structured data, and time-series runtime data, extracting unique core feature information for each type of data. Then, the multi-head attention mechanism is invoked to adaptively allocate weights based on the influence of different features on the defect prediction results, completing the integration and fusion of multiple single features, eliminating redundant information between modalities, and ultimately generating a sample fusion feature vector with unified dimensions.

[0029] In step S303 of some embodiments, the multimodal feature fusion layer is trained through contrastive learning to achieve cross-modal redundancy removal. Specifically, according to the defect category, the sample fusion feature vectors are constructed into positive and negative contrastive sample pairs. The fusion feature vectors corresponding to defects of the same category are designated as positive sample pairs, and the fusion feature vectors corresponding to defects of different categories are designated as negative sample pairs. The contrastive loss is calculated by calculating the feature space distance between the sample pairs. Based on the contrastive loss constraint model, the feature gap of defects of the same category is reduced, and the feature gap of defects of different categories is widened, further enhancing the model's feature discrimination ability and optimizing the multimodal feature fusion effect.

[0030] In step S304 of some embodiments, the sample fusion feature vector output by the multimodal feature fusion layer is input into the defect prediction layer. The defect prediction layer is provided with a classification prediction branch and a regression prediction branch. The classification prediction branch is used to output the sample defect type corresponding to the sample, and the regression prediction branch is used to output the sample risk assessment index corresponding to the sample.

[0031] In step S305 of some embodiments, the sample defect types and various risk assessment indicators predicted by the model are compared with the historical defect types and historical risk indicators pre-labeled in the training samples. The defect type prediction loss is determined according to the classification task loss function, and the risk indicator prediction loss is determined according to the regression task loss function. The two types of losses are integrated and summarized into a unified prediction loss to reflect the degree of deviation between the current prediction result of the model and the actual labeled result.

[0032] In step S306 of some embodiments, a joint loss is generated based on the predicted loss and the contrastive loss. Specifically, the predicted loss and the contrastive loss can be weighted and fused according to preset weights to generate the joint loss required for model training. Then, the joint loss is backpropagated sequentially to the multimodal feature fusion layer at the front end and the defect prediction layer at the back end of the defect prediction model. The core training parameters such as network weights, attention allocation weights, and learning rates within the two layers are adjusted synchronously. The iterative training process of sample input, feature fusion, loss calculation, and parameter update is executed repeatedly. The changes in the joint loss value are monitored in real time. When the joint loss value gradually decreases and stabilizes within a preset threshold, the pre-trained defect prediction model is obtained.

[0033] In addition, the performance of the defect prediction model can be verified on the validation set, such as prediction accuracy and recall. When the performance indicators also reach the preset standards, the defect prediction model training is complete.

[0034] In this embodiment, the trained defect prediction model enables automated identification of software defects, thereby quickly and accurately determining the types of defects in the software and quantifying the corresponding risk assessment indicators, efficiently screening out high-priority target risk defects. Furthermore, a training method for the defect prediction model is provided. The defect prediction model includes a multimodal feature fusion layer and a defect prediction layer. The multimodal feature fusion layer achieves differentiated feature extraction and adaptive attention fusion from multi-source heterogeneous data, fully mining effective information from different dimensions of software data. Simultaneously, contrastive learning is introduced to construct sample pairs and calculate contrastive loss, effectively eliminating redundant interference between different modalities and enhancing the feature discrimination of different defect samples. Combined with the prediction loss, a joint loss is constructed to complete global iterative optimization, significantly improving the model's feature representation capability and multi-task prediction accuracy, ensuring reliable defect type and risk score output.

[0035] like Figure 4 As shown, Figure 4 This is a flowchart of a defect determination method provided in another embodiment of this application; the transportation device control method may include, but is not limited to, steps S401 to S406: S401, Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. S402, taking the target risk defect as the target node, extract feature data from the software data to generate associated nodes, and form a causal dependency graph based on the causal relationship between the target node and the associated nodes; For details, please refer to the descriptions of steps S101 to S102 in the embodiments of the above specification, which will not be repeated here.

[0036] S403, Traverse the causal dependency graph and calculate the influence weight of each associated node on the target node based on the graph model causal inference algorithm; In step S403 of some embodiments, after generating the causal dependency graph, the influence weight of each associated node on the target node is calculated based on a graph model causal inference algorithm. Specifically, the graph model causal inference algorithm is started. The graph model causal inference algorithm can be implemented in various forms such as Bayesian network inference algorithm and structural causal inference algorithm. In this embodiment, the Bayesian network inference algorithm is preferred. Then, the causal dependency graph is completely imported into the algorithm module. The algorithm input parameters include the node topology of the causal dependency graph, the attribute data of each node (including the target node and associated nodes), and the initial value of the causal association strength between nodes (calculated by the previous causal discovery algorithm). Subsequently, a depth-first search (DFS) method is used to traverse all associated nodes in the causal dependency graph one by one to ensure that no associated node with a causal relationship with the target node is missed. The causal inference algorithm is then used to quantify the inducing effect (i.e., influence weight) of each associated node on the target risk defect.

[0037] S404, Select the associated nodes with an influence weight higher than a preset threshold as candidate root causes of the target risk defect; In step S404 of some embodiments, a reasonable impact weight threshold is preset based on the business importance of the target software, operation and maintenance management standards, and defect severity level. The preset threshold can be flexibly adjusted according to different software scenarios and different defect types. Subsequently, the impact weight of each associated node is compared with the preset threshold one by one, and associated nodes with impact weight values ​​higher than the preset threshold are screened out. These associated nodes with high impact weights are uniformly classified as candidate root causes of the target risk defect, thereby narrowing the scope of root cause investigation and improving the targeting and efficiency of root cause location.

[0038] S405, determine the root cause confidence level of the candidate root cause based on the association data of the candidate root cause; In step S405 of some embodiments, for each selected candidate root cause, its corresponding correlation data is retrieved, and the root cause confidence of the candidate root cause is analyzed based on the correlation data. This correlation data comes from the full amount of software data collected in the early stage, specifically including the original running logs, configuration parameters, interface interaction records, business execution data, and historical correlation records of similar defects corresponding to the candidate root cause.

[0039] In one feasible implementation, the correlation between relevant data and candidate root causes can be analyzed using a large model to determine the rationality of the candidate root causes, i.e., the confidence level of the candidate root causes. The following example, using the target risk defect "payment interface anomaly" in a financial order system, illustrates the implementation of S405: Assume that in the financial order system, the selected candidate root causes include "payment interface code logic," "database connection pool configuration," "third-party payment interface response," "reasons for test case execution failure," and "online interface response time," and that "insufficient maximum number of database connection pool connections" has the highest impact weight, calculated using a Bayesian network inference algorithm. The relevant data corresponding to this core candidate root cause, "insufficient maximum number of database connection pool connections," is retrieved. For example, "database connection timeout" error messages are clearly recorded in the software operation log data. Analysis shows that this error message is directly related to "insufficient maximum number of database connection pool connections," which preliminarily supports the rationality of this candidate root cause. Meanwhile, the key indicator "database connection pool occupancy rate 100%" was extracted from the online operation data. This indicator shows that the database connection pool has reached full capacity and cannot handle more concurrent requests, further confirming that "insufficient maximum number of database connection pool connections" is the cause of the target risk defect "payment interface abnormality". Thus, the root cause was determined to be "insufficient maximum number of database connection pool connections".

[0040] In one feasible implementation, a root cause confidence evaluation system can be constructed, setting three evaluation indicators and corresponding weights: the causal association stability between the candidate root cause and the target risk defect, the anomaly frequency of the candidate root cause, and the induction probability of the candidate root cause in similar historical defects. Based on this evaluation system, multi-dimensional verification of candidate root causes can be achieved. Based on the above evaluation system, a weighted summation method is used to calculate the root cause confidence of each candidate root cause, with the confidence value ranging from [0,1]. The specific calculation formula is: Root cause confidence = (Association stability score × First weight) + (Anomaly frequency score × Second weight) + (Historical induction probability score × Third weight). Each score is calculated by the algorithm based on the association data of the candidate root cause (score range [0,1]). For example, if a candidate root cause is a core induction in the last three similar defects, the historical induction probability score is 0.9. Finally, the root cause confidence of each candidate root cause is output. The higher the confidence, the greater the probability that the candidate root cause is the true root cause of the target risk defect.

[0041] Taking a financial order system as an example, for instance, querying historical data on similar defects reveals that 4 out of the last 5 "payment interface anomaly" defects were triggered by "payment interface call timeout," with a historical trigger probability score of 0.8. Combined with a correlation stability score of 0.9 (interface timeout and defect occur simultaneously) and an anomaly frequency score of 0.85 (timeouts ≥ 5 times per hour), and setting the first weight to 0.4, the second weight to 0.3, and the third weight to 0.3, the root cause confidence is calculated using the weighted summation formula: Root cause confidence = (0.9 × 0.4) + (0.85 × 0.3) + (0.8 × 0.3) = 0.36 + 0.255 + 0.24 = 0.855, which is 85.5%.

[0042] S406, if the confidence level of the root cause is greater than the confidence level threshold, then obtain the causal path from the target risk defect to the candidate root cause in the causal dependency graph, and determine the causal path and the candidate root cause as the root cause result of the target risk defect.

[0043] In step S406 of some embodiments, the calculated root cause confidence of each candidate root cause is compared with the threshold; for candidate root causes with a root cause confidence greater than the confidence threshold, they are determined to be potential root causes with high reliability, and the causal path is further determined; for candidate root causes with a root cause confidence less than or equal to the confidence threshold, they are determined to be non-core causes with low reliability and are eliminated.

[0044] Based on the topology of the causal dependency graph, a reverse tracing method is used. Starting from the target node (target risk defect), the causal relationship between nodes is traced back to the corresponding candidate root cause, completely extracting the entire causal transmission path and forming a visualized causal path. For example: target risk defect "payment interface anomaly" → intermediate node "database connection timeout" → candidate root cause "inappropriate database connection pool configuration". By extracting the causal path, technical personnel can clearly understand the inducing process of the defect.

[0045] Finally, the selected candidate root causes are associated with and bound to their corresponding causal paths to be jointly identified as the root cause of the target risk defect.

[0046] Understandably, if multiple candidate root causes have a root cause confidence level greater than the confidence threshold, the causal path corresponding to each candidate root cause will be traced separately, and all candidate root causes and their corresponding causal paths that meet the criteria will be included in the root cause results. Furthermore, the root causes can be sorted from high to low confidence level, and the priority of each root cause can be marked, making it easier for technical personnel to prioritize the rectification of core root causes and improve the efficiency of defect rectification.

[0047] In this embodiment, based on the previously constructed causal dependency graph, a graph model causal inference algorithm is used to quantify the influence weight of associated nodes, screen candidate root causes, and combine the associated data of candidate root causes to verify the confidence of root causes and trace causal paths, ultimately determining the root cause result. This solves the problems of inefficiency and incompleteness in traditional root cause localization, reduces operation and maintenance costs, and ensures that the root cause results are accurate and reliable with clear causal logic.

[0048] like Figure 5 As shown, Figure 5 This is a flowchart of a defect determination method provided in another embodiment of this application; the transportation device control method may include, but is not limited to, steps S501 to S504: S501, Build an isolated test environment consistent with the production environment of the target software, and copy the runtime traffic of the target software to the isolated test environment through shadow traffic technology to simulate the defect occurrence scenario; S502, in the isolated test environment, the suspicious code fragments and / or associated software modules corresponding to the root cause results are tested to determine whether the target risk defect is reproduced; S503, if the target risk defect is reproduced, and the defect disappears after the root cause result is repaired, then the root cause verification result is verified as passed. S504, if the target risk defect is not reproduced or the target risk defect is reproduced but the defect still exists after the root cause result is repaired, then the root cause verification result is verification failure.

[0049] In step S501 of some embodiments, the hardware and software configuration, service ports, data interaction rules, third-party interface call permissions, system operating parameters, etc. are uniformly replicated with reference to the production environment where the target software is officially launched and running. An isolated test environment is built that is completely consistent with the production environment architecture, operating logic, and resource configuration. This isolated test environment is independent and isolated from the official production environment and will not cause any interference to the normal operation of online business.

[0050] Specifically, by combining containerization technology (Docker) and container orchestration system (Kubernetes, K8s) technologies, an isolated test environment completely identical to the target software's production environment is built, ensuring that the environment configuration, dependent services, and runtime parameters are fully matched to the production environment. Docker is a containerization tool that can quickly build standardized, portable runtime environments without repeatedly deploying complex hardware, reducing environment setup costs. K8s (Kubernetes) is a container orchestration system that enables dynamic scheduling and resource allocation of the isolated test environment, ensuring its stability and consistency. Specifically, a basic test environment is first built using Docker containers, configuring software dependencies, interface permissions, and runtime parameters completely identical to the production environment, ensuring that the test environment's runtime logic and resource configuration are completely consistent with the production environment. Then, K8s is used to implement resource scheduling and load balancing in the test environment, avoiding deviations in test results caused by environmental differences.

[0051] Based on this, traffic mirroring technology is used to copy the real traffic (including normal and abnormal traffic) in the target software's production environment and import it into the established isolated test environment at a preset ratio (such as 1:1) to simulate the real scenario of defect occurrence.

[0052] In step S502 of some embodiments, based on the root cause results determined above, the suspicious code segment corresponding to the root cause and the associated software module with anomalies are identified, and simulation testing is performed in an isolated test environment. Specifically, the environment is continuously driven by the imported real shadow traffic, and the running status, data flow, interface response status, and business execution results of the isolated test environment are observed to determine whether the previously identified target risk defect can be reproduced in this simulation scenario.

[0053] In step S503 of some embodiments, when the target risk defect can be reproduced in the isolated test environment, the target risk defect is repaired according to the rectification plan corresponding to the determined root cause result. This includes logical correction or code optimization of suspicious code segments; or, parameter adjustment, functional reconstruction, configuration optimization, and interface adaptation rectification of the problematic related software modules. After rectification, the isolated test environment is run again using the same shadow traffic driver. If the same type of target risk defect disappears after rectification, the root cause verification result is determined to be successful, indicating that the root cause result obtained from the current location is true and valid.

[0054] In step S504 of some embodiments, if the target risk defect cannot be reproduced in the isolated test environment based on real operating traffic, it indicates that the previously identified root cause does not match the actual fault cause; or, although the target risk defect can be successfully reproduced, the target risk defect continues to appear and cannot be eliminated after completing the corresponding code and software module repair operations according to the predetermined root cause results, it indicates that the currently determined root cause has a deviation or judgment error, and the root cause verification result is determined to be verification failure. It is necessary to perform defect feature extraction, causal relationship construction, and graph model causal reasoning and tracing operations again to re-investigate the root cause of the defect.

[0055] This embodiment utilizes a homogeneous isolated testing environment combined with shadow traffic redirection to simulate and reproduce defect scenarios outside of online production operations. This approach ensures that the daily operation of the target software is not affected, while maximizing the replication of real-world fault scenarios and guaranteeing the authenticity and reliability of the verification environment. Furthermore, by conducting reproduction and repair tests on the identified root causes of defects, the accuracy of the root cause determination can be effectively verified, improving the rigor and accuracy of defect root cause identification.

[0056] In one embodiment, after testing the suspicious code fragments and / or associated software modules corresponding to the root cause results in the isolated testing environment to determine whether the target risk defect is reproduced, the method further includes: S601, Obtain the root cause verification results, the repair data of the target risk defect, and the running data of the target software after the target risk defect is repaired; S602, Based on the root cause verification results, the repair data and the running data, a new training dataset for the defect prediction model is generated, so as to retrain the defect prediction model based on the new training dataset; S603, adjust the node association relationships in the root cause dependency graph based on the root cause verification results, the repair data, and the running data.

[0057] In step S601 of some embodiments, the root cause verification results, the repair data corresponding to the target risk defect, and the operational data generated by the actual operation of the target software after the defect repair is completed are obtained. The root cause verification results include two categories of judgment conclusions: verification passed and verification failed. Simultaneously, abnormal judgment reasons such as root cause misjudgment, causal path deviation, and low confidence level are recorded. The repair data includes the defect repair scheme adopted, the overall repair time, and the corresponding defect risk scores before and after the defect repair. The software operation data after repair includes the defect occurrence rate after repair, the fluctuation of various online business operation indicators, etc., providing effective data for subsequent iterative optimization.

[0058] In step S602 of some embodiments, the collected root cause verification results, defect repair data, and target risk defect repair post-repair operation data are standardized and integrated to generate a new training dataset that can be used for model training. This new training dataset is then merged into the original sample dataset, or a new sample dataset is directly used for secondary retraining and optimization of the defect prediction model. During retraining, the operating parameters related to the multimodal feature extraction weights and attention mechanism within the model can be adjusted based on actual test feedback data to correct the model's original prediction bias, continuously optimize the model's ability to identify various software defects and assess risks, and steadily improve the overall defect prediction accuracy of the model.

[0059] In step S603 of some embodiments, the root cause localization method can be adjusted and optimized based on the root cause verification results, defect repair effects, and online operation status data. To address issues such as root cause verification failure, root cause matching errors, insufficient decision confidence, and deviations in causal transmission paths, the connection relationships between nodes within the causal dependency graph, the connection strength of causal relationships between nodes, and the causal correlation strength between nodes can be adjusted to correct structural deviations in the original graph.

[0060] Optionally, based on the root cause verification results, defect repair effects, and online operation status data, the causal reasoning algorithm's operating parameters can be optimized, the weight screening threshold, root cause confidence judgment threshold, and invalid node pruning threshold can be reasonably adjusted, the root cause screening and discrimination logic can be improved, false root causes can be filtered out, redundant root causes of the same kind can be integrated and merged, the ranking weight of true root causes can be increased, and the overall root cause screening logic can be continuously optimized.

[0061] This application provides an end-to-end closed-loop solution from defect prediction, root cause localization, root cause verification to model feedback iteration, improving defect management efficiency and accuracy while reducing R&D costs. Furthermore, by integrating root cause verification results, repair data, and other feedback data into the model iteration in real time, the defect prediction model and root cause localization logic can be continuously optimized, enhancing the solution's generalization ability and long-term applicability, and adapting to different types and scales of software scenarios.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0063] In one embodiment, a defect determination apparatus is provided, which corresponds one-to-one with the defect determination methods described in the above embodiments. For example... Figure 6 As shown, the defect determination device 600 includes a defect prediction module 10, a causal construction module 20, a candidate root cause reasoning module 30, and a root cause confirmation module 40. Detailed descriptions of each functional module are as follows: The defect prediction module 10 is used to collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. Causal construction module 20 is used to extract feature data from the software data to generate associated nodes with the target risk defect as the target node, construct the causal relationship between the target node and the associated nodes, and form a causal dependency graph. The candidate root cause reasoning module 30 is used to filter out candidate root causes of the target risk defect from the associated nodes based on the graph model causal reasoning algorithm and the causal dependency graph. The root cause identification module 40 is used to determine the root cause result of the target risk defect based on the correlation data of the candidate root causes.

[0064] In one embodiment, the candidate root cause reasoning module 30 is specifically used for: Traverse the causal dependency graph and calculate the influence weight of each associated node on the target node based on the graph model causal inference algorithm; The associated nodes with an influence weight higher than a preset threshold are selected as candidate root causes of the target risk defect.

[0065] In one embodiment, the defect prediction module 10 is specifically used for: Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the defect types of the target software and the risk assessment index of the defect types. Defect types whose risk assessment indicators exceed preset threshold values ​​are identified as target risk defects.

[0066] In one embodiment, the defect prediction module 10 is further configured to: Obtain historical software data and corresponding defect tags for the target software; the defect tags include historical defect types and historical risk indicators. Based on the multimodal feature fusion layer of the defect prediction model, different types of data features are extracted from the historical software data, and multi-feature fusion processing is performed through an attention mechanism to obtain the sample fusion feature vector. Construct contrast sample pairs based on the sample fusion feature vectors and calculate the contrast loss; The sample fusion feature vector is input into the defect prediction layer of the defect prediction model to obtain the sample defect type and sample risk assessment index. The sample defect type and the sample risk assessment index are compared with the historical defect type and the historical risk index, respectively, to determine the predicted loss; A joint loss is generated based on the prediction loss and the contrast loss. The joint loss is then backpropagated to the multimodal feature fusion layer and the defect prediction layer of the defect prediction model. The defect prediction model is trained iteratively. When the joint loss converges to a preset threshold, a pre-trained defect prediction model is obtained.

[0067] In one embodiment, the root cause identification module 40 is specifically used for: The root cause confidence level of the candidate root cause is determined based on the association data of the candidate root causes; If the confidence level of the root cause is greater than the confidence level threshold, then the causal path from the target risk defect to the candidate root cause in the causal dependency graph is obtained, and the causal path and the candidate root cause are determined as the root cause result of the target risk defect.

[0068] like Figure 7 As shown, another defect determination device 700 is provided. In one embodiment, the defect determination device further includes a verification module 50 and an iteration module 60. The functional modules are described in detail below: The verification module 50 is specifically used to: build an isolated test environment consistent with the production environment of the target software, and copy the runtime traffic of the target software to the isolated test environment through shadow traffic technology to simulate the defect occurrence scenario; In the isolated test environment, the suspicious code fragments and / or associated software modules corresponding to the root cause results are tested to determine whether the target risk defect is reproduced. If the target risk defect is reproduced, and the defect disappears after the root cause is fixed, then the root cause verification result is verified as passed. If the target risk defect is not reproduced, or if the target risk defect is reproduced but the defect still exists after the root cause is fixed, then the root cause verification result is verification failure.

[0069] Iteration module 60 is specifically used for: Obtain the root cause verification results, the repair data of the target risk defect, and the running data of the target software after the target risk defect is repaired; A new training dataset for the defect prediction model is generated based on the root cause verification results, the repair data, and the running data, so as to retrain the defect prediction model based on the new training dataset. The node relationships in the root cause dependency graph are adjusted based on the root cause verification results, the repair data, and the running data.

[0070] This invention provides a defect identification device that automatically collects software data from target software and inputs it into a pre-trained model. This allows for the rapid identification of target risk defects whose risk assessment indicators exceed preset thresholds, effectively improving defect identification efficiency and accuracy. Subsequently, using the selected target risk defects as target nodes, feature data is extracted from the software data to construct associated nodes and build a causal dependency graph. A graph model causal inference algorithm, combined with the causal dependency graph, is used to filter associated nodes, identifying highly correlated candidate root causes of defects. This causal dependency graph is built only for identified high-risk defects, eliminating the need for modeling based on the entire software data set, significantly reducing the data processing volume for graph construction. Based on this, the associated data corresponding to the candidate root causes is retrieved to complete confidence verification, further ensuring the accuracy and reliability of the final root cause location result and improving the overall efficiency of software daily operation and maintenance and defect management.

[0071] Specific limitations regarding the defect determination device can be found in the limitations of the intelligent question-answering method described above, and will not be repeated here. Each module in the aforementioned defect determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0072] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a defect determination method on the server side.

[0073] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements the functions or steps of a defect determination method on the client side.

[0074] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. Using the target risk defect as the target node, feature data is extracted from the software data to generate associated nodes, and a causal dependency graph is formed based on the causal relationship between the target node and the associated nodes. Based on the graph model causal reasoning algorithm and the causal dependency graph, candidate root causes of the target risk defect are selected from the associated nodes; Based on the correlation data of the candidate root causes, the root cause result of the target risk defect is determined.

[0075] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. Using the target risk defect as the target node, feature data is extracted from the software data to generate associated nodes, and a causal dependency graph is formed based on the causal relationship between the target node and the associated nodes. Based on the graph model causal reasoning algorithm and the causal dependency graph, candidate root causes of the target risk defect are selected from the associated nodes; Based on the correlation data of the candidate root causes, the root cause result of the target risk defect is determined.

[0076] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0079] It should be noted that any software tools or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0080] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A defect determination method, characterized in that, include: Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. Using the target risk defect as the target node, feature data is extracted from the software data to generate associated nodes, and a causal dependency graph is formed based on the causal relationship between the target node and the associated nodes. Based on the graph model causal reasoning algorithm and the causal dependency graph, candidate root causes of the target risk defect are selected from the associated nodes; Based on the correlation data of the candidate root causes, the root cause result of the target risk defect is determined.

2. The defect determination method as described in claim 1, characterized in that, The graph-based causal inference algorithm and the causal dependency graph filter candidate root causes of the target risk defect from the associated nodes, including: Traverse the causal dependency graph and calculate the influence weight of each associated node on the target node based on the graph model causal inference algorithm; The associated nodes with an influence weight higher than a preset threshold are selected as candidate root causes of the target risk defect.

3. The defect determination method as described in claim 1, characterized in that, The software data of the target software is collected, and the software data is input into a pre-trained defect prediction model for defect risk prediction processing to obtain target risk defects whose risk assessment indicators are higher than a preset indicator threshold, including: Collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the defect types of the target software and the risk assessment index of the defect types. Defect types whose risk assessment indicators exceed preset threshold values ​​are identified as target risk defects.

4. The defect determination method as described in claim 3, characterized in that, Before collecting software data from the target software, inputting the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtaining a target risk defect whose risk assessment index is higher than a preset index threshold, the process further includes: Obtain historical software data and corresponding defect tags for the target software; the defect tags include historical defect types and historical risk indicators. Based on the multimodal feature fusion layer of the defect prediction model, different types of data features are extracted from the historical software data, and multi-feature fusion processing is performed through an attention mechanism to obtain the sample fusion feature vector. Construct contrast sample pairs based on the sample fusion feature vectors and calculate the contrast loss; The sample fusion feature vector is input into the defect prediction layer of the defect prediction model to obtain the sample defect type and sample risk assessment index. The sample defect type and the sample risk assessment index are compared with the historical defect type and the historical risk index, respectively, to determine the predicted loss; A joint loss is generated based on the prediction loss and the contrast loss. The joint loss is then backpropagated to the multimodal feature fusion layer and the defect prediction layer of the defect prediction model. The defect prediction model is trained iteratively. When the joint loss converges to a preset threshold, a pre-trained defect prediction model is obtained.

5. The defect determination method as described in claim 1, characterized in that, The step of determining the root cause result of the target risk defect based on the correlation data of the candidate root causes includes: The root cause confidence level of the candidate root cause is determined based on the association data of the candidate root causes; If the confidence level of the root cause is greater than the confidence level threshold, then the causal path from the target risk defect to the candidate root cause in the causal dependency graph is obtained, and the causal path and the candidate root cause are determined as the root cause result of the target risk defect.

6. The defect determination method according to any one of claims 1-5, characterized in that, The method further includes: An isolated test environment identical to the production environment of the target software is set up, and the runtime traffic of the target software production is copied to the isolated test environment through shadow traffic technology to simulate the defect occurrence scenario; In the isolated test environment, the suspicious code fragments and / or associated software modules corresponding to the root cause results are tested to determine whether the target risk defect is reproduced. If the target risk defect is reproduced, and the defect disappears after the root cause is fixed, then the root cause verification result is verified as passed. If the target risk defect is not reproduced, or if the target risk defect is reproduced but the defect still exists after the root cause is fixed, then the root cause verification result is verification failure.

7. The defect determination method as described in claim 6, characterized in that, After testing the suspicious code segments and / or associated software modules corresponding to the root cause results in the isolated testing environment to determine whether the target risk defect is reproduced, the method further includes: Obtain the root cause verification results, the repair data of the target risk defect, and the running data of the target software after the target risk defect is repaired; Based on the root cause verification results, the repair data, and the running data, a new training dataset is generated for the defect prediction model, and the defect prediction model is retrained based on the new training dataset. The node relationships in the root cause dependency graph are adjusted based on the root cause verification results, the repair data, and the running data.

8. A defect determination device, characterized in that, include: The defect prediction module is used to collect software data of the target software, input the software data into a pre-trained defect prediction model for defect risk prediction processing, and obtain the target risk defect whose risk assessment index is higher than the preset index threshold. The causal construction module is used to extract feature data from the software data to generate associated nodes, taking the target risk defect as the target node, constructing the causal relationship between the target node and the associated nodes, and forming a causal dependency graph. The candidate root cause reasoning module is used to filter out candidate root causes of the target risk defect from the associated nodes based on the graph model causal reasoning algorithm and the causal dependency graph. The root cause identification module is used to determine the root cause result of the target risk defect based on the correlation data of the candidate root causes.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the defect determination method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the defect determination method as described in any one of claims 1 to 7.

Images