An effective defect reporting detection method, apparatus, electronic device, and storage medium

Abstract

This invention relates to intelligent defect detection technology in software engineering, and discloses a method for detecting valid defect reports, comprising: acquiring a defect report; extracting structured and unstructured features from the defect report; fusing the structured and unstructured features to obtain a fused feature; inputting the fused feature into a duplicate defect report detection model; using the duplicate defect report detection model to detect whether the defect report is a duplicate defect report; if the defect report is a duplicate defect report, the defect report is an invalid report; if the defect report is a non-duplicate defect report, the fused feature is input into an automatic defect report classification model; the automatic defect report classification model is used to classify the defect report to obtain a classification result of whether the defect report is a valid report or an invalid report. This invention also proposes a valid defect report detection device, electronic device, and storage medium. This invention can improve the efficiency and accuracy of valid defect report detection.

Description

Technical Field

[0001] This invention belongs to the field of intelligent defect detection technology in software engineering, and particularly relates to an effective defect report detection method, device, electronic device and storage medium. Background Technology

[0002] A defect report is a document that records detailed information about software product defects. It has a structured format, allowing testers to easily conduct software activities based on the report (such as predicting defect severity, assigning appropriate defect resolvers, and locating and fixing defects). Users or testers write defect reports and upload them to the defect tracking system when they encounter defects. Defect reports can cover all the basic information about a defect, aiding in detailed defect analysis, helping developers find the correct direction and methods for debugging, and helping to avoid the same defects in future versions. The content of a software defect report can be divided into descriptive text and execution information. The descriptive text includes a title, description, and comments. Other characteristic fields belong to execution information, such as priority, product, status, version, reporter, and reporting time. This information plays a significant role in classifying defect reports.

[0003] When system maintenance personnel receive a software defect report, their first step is to verify the report's validity, i.e., to eliminate invalid reports. Valid reports are then assigned to the relevant developers for fixing. Existing research on valid defect report identification relies on extracting manual features or using textual information from titles, descriptions, and some annotations to characterize defect reports. This approach fails to fully utilize the unstructured information within the reports, leading to low accuracy and efficiency in classifying valid defect reports.

[0004] The definition and labeling rules for valid defect reports contain a misclassification issue regarding duplicate defect reports. The definition of valid and invalid reports is based on the resolution field of each report. A defect report's resolution can be "DUPLICATE" (i.e., other defects reported in previous reports), "Invalid" (i.e., non-software defects), "WORKSFORME" (i.e., defects that cannot be reproduced), "INCOMPLETE" (i.e., defects lacking necessary information), "FIXED" (i.e., successfully fixed), or "WONTFIX" (i.e., a real defect that has not yet been fixed). Based on this information, defect reports resolved as fixed or unfixed are considered valid, while those resolved as duplicate, invalid, functional, or incomplete are considered invalid. However, labeling defect reports with a resolution field of "DUPLICATE" as invalid fails to consider that some duplicate defect reports describe the same valid defects as valid reports. This makes the classification of valid defect reports inconsistently effective. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and storage medium for detecting valid defect reports, which can improve the efficiency and accuracy of valid defect report detection.

[0006] The redefinition and marking of valid and invalid reports are as follows: Reports with a resolution field of "FIXED" or "WONTFIX" are considered valid defect reports, while those with a resolution field of "INVALID", "WORKSFORME", or "INCOMPLETE" are considered invalid defect reports. For duplicate defect reports (i.e., those with a resolution field of "DUPLICATE"), duplicate reports that overlap with valid reports are marked as valid, and duplicate reports that overlap with invalid reports are marked as invalid.

[0007] To achieve the above objectives, the present invention provides an effective defect report detection method, comprising:

[0008] Obtain the defect report and extract the structured and unstructured features from the defect report;

[0009] The structured features and the unstructured features are fused to obtain fused features. The fused features are then input into the duplicate defect report detection model, which is used to detect whether the defect report is a duplicate defect report.

[0010] If the defect report is a duplicate defect report, then the defect report is an invalid report;

[0011] If the defect report is a non-duplicate defect report, the fusion feature is input into the automatic defect report classification model, and the defect report is classified using the automatic defect report classification model to obtain the classification result of whether the defect report is a valid report or an invalid report.

[0012] Optionally, the extraction of structured and unstructured features from the defect report includes:

[0013] Identify the structured information field names in the defect report, and extract the information under the field names to obtain the structured information;

[0014] The information description text and comment text in the defect report are queried, and the text description text and comment text are normalized to obtain the unstructured information;

[0015] Structured features are extracted from the structured information using regular expressions, and unstructured features are extracted from the unstructured information after text processing.

[0016] Optionally, the step of extracting structured features from the structured information using regular expressions includes:

[0017] Based on the field names of the structured information, regular expressions are used to write feature extraction rules, and these rules are then used to extract structured features.

[0018] Optionally, the step of fusing the structured features and the unstructured features to obtain the fused features includes:

[0019] The structured features are embedded using one-hot encoding to obtain a structured feature vector;

[0020] The unstructured features are transformed into feature vectors using a pre-defined text vector model, resulting in unstructured feature vectors.

[0021] The structured feature vector and the unstructured feature vector are concatenated to obtain the fused feature.

[0022] Optionally, detecting whether a defect report is a duplicate defect report using a duplicate defect report detection model includes:

[0023] The fused features are embedded into nodes using the heterogeneous information network in the recurring defect report detection model to obtain embedded information.

[0024] The bidirectional LSTM network in the duplicate defect report detection model is used to learn the textual semantic features in the embedded information;

[0025] The similarity between the defect report and the defect reports in the preset defect report set is calculated based on the semantic features of the text, and the similarity is used to detect whether the defect report is a duplicate defect report.

[0026] Optionally, the automatic defect report classification model includes a multi-head attention network layer, a feedforward neural network layer, a pooling layer, and a fully connected layer.

[0027] Optionally, classifying the defect reports using an automatic defect report classification model includes:

[0028] A multi-head attention network layer is used to process the received fusion features to obtain a fused linear projection;

[0029] The fused linear projection is activated and normalized using a feedforward neural network layer to obtain the fused output;

[0030] The fused output is pooled using a pooling layer to obtain a fused pooled output;

[0031] The classification probability of the fusion pooling output is calculated using a fully connected layer, and the defect report classification result is determined based on the classification probability.

[0032] To address the above problems, the present invention also provides an effective defect reporting detection device, the device comprising:

[0033] The feature extraction module is used to obtain defect reports and extract structured and unstructured features from the defect reports;

[0034] The duplicate defect report detection module is used to fuse the structured features and the unstructured features to obtain fused features, input the fused features into the duplicate defect report detection model, and use the duplicate defect report detection model to detect whether the defect report is a duplicate defect report.

[0035] The defect report classification module is used to classify a defect report as invalid if it is a duplicate defect report, and to input the fused features into the automatic defect report classification model if it is a non-duplicate defect report, thereby classifying the defect report as either a valid or invalid report.

[0036] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0037] At least one processor; and,

[0038] A memory communicatively connected to the at least one processor; wherein,

[0039] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the effective defect report detection method described above.

[0040] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the effective defect report detection method described above.

[0041] This invention ensures the integrity of the defect report information by extracting both structured and unstructured features from the defect report, thereby guaranteeing that every piece of information in the defect report is fully utilized and ensuring the accuracy of subsequent defect report classification. Furthermore, by fusing the structured and unstructured features to obtain fused features, and then using a duplicate defect report detection model to detect whether the defect report is a duplicate, prioritizing the detection of duplicate defect reports can reduce the impact of duplicate defect reports on the validity classification of defect reports. In addition, using an automatic defect report classification model to classify the defect reports can also ensure the classification efficiency of valid defect reports. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating an effective defect report detection method provided in an embodiment of the present invention.

[0043] Figure 2 This is a flowchart illustrating a specific implementation of an embodiment of the effective defect report detection method provided by an embodiment of the present invention;

[0044] Figure 3 This is a functional block diagram of an effective defect reporting and detection device provided in an embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the effective defect reporting detection method according to an embodiment of the present invention.

[0046] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0047] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0048] This application provides a method for detecting valid defect reports. The executing entity of this valid defect report detection method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the valid defect report detection method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0049] Referring to Table 1, the structured and unstructured features of an effective defect reporting detection method provided in an embodiment of the present invention are described.

[0050]

[0051]

[0052] Reference Figure 1 The diagram shown is a flowchart illustrating a valid defect report detection method according to an embodiment of the present invention. In this embodiment, the valid defect report detection method includes:

[0053] S1. Obtain the defect report and extract the structured and unstructured features from the defect report.

[0054] In this embodiment of the invention, the defect report is a document that records detailed information about defects in a software product. It has a structured format and can facilitate testers to carry out software activities based on the defect report.

[0055] As an embodiment of the present invention, the extraction of structured and unstructured features from the defect report includes:

[0056] Identify the structured information field names in the defect report, and extract the information under the field names to obtain the structured information;

[0057] The information description text and comment text in the defect report are queried, and the text description text and comment text are normalized to obtain the unstructured information;

[0058] Structured features are extracted from the structured information using regular expressions, and unstructured features are extracted from the unstructured information after text processing.

[0059] In this embodiment of the invention, the structured information refers to information in a specific format in the defect report, which has specific field names. For example, the structured information includes, but is not limited to, BID (bug_id), DID (dup_id), PRO (product), COM (component), VER (version), SEV (severity), PRI (priority), REP (repoter), and ATT (attachment).

[0060] In this embodiment of the invention, the unstructured information refers to the title text, summary text, and comment text in each defect report.

[0061] In this embodiment of the invention, the regular expression refers to a tool that can perform text matching using a specially written string, and can extract the required data or information from complex text.

[0062] Furthermore, the extraction of structured features from the structured information using regular expressions includes:

[0063] Based on the field names of the structured information, regular expressions are used to write feature extraction rules, and these rules are then used to extract structured features.

[0064] S2. The structured features and the unstructured features are fused to obtain fused features. The fused features are then input into the duplicate defect report detection model. The duplicate defect report detection model is used to detect whether the defect report is a duplicate defect report.

[0065] As an embodiment of the present invention, the step of fusing the structured features and the unstructured features to obtain the fused features includes:

[0066] The structured features are embedded using one-hot encoding to obtain a structured feature vector;

[0067] The unstructured features are transformed into feature vectors using a pre-defined text vector model, resulting in unstructured feature vectors.

[0068] The structured feature vector and the unstructured feature vector are concatenated to obtain the fused feature.

[0069] Specifically, the step of using a duplicate defect report detection model to detect whether the defect report is a duplicate defect report includes:

[0070] The fused features are embedded into nodes using the heterogeneous information network in the recurring defect report detection model to obtain embedded information.

[0071] The bidirectional LSTM network in the duplicate defect report detection model is used to learn the textual semantic features in the embedded information;

[0072] The similarity between the defect report and the defect reports in the preset defect report set is calculated based on the semantic features of the text, and the similarity is used to detect whether the defect report is a duplicate defect report.

[0073] Specifically, if the similarity with any defect report in the preset defect report set is greater than the preset similarity threshold, the defect report is considered a duplicate defect report; if the similarity with all defect reports in the preset defect report set is not greater than the preset similarity threshold, the defect report is considered not a duplicate defect report.

[0074] In this embodiment of the invention, the heterogeneous information network refers to a network composed of nodes and relationships of different types. In a heterogeneous information network, nodes and relationships can have different types and attributes.

[0075] In this embodiment of the invention, the bidirectional LSTM network refers to a variant of a recurrent neural network that can effectively capture semantic features in text sequences.

[0076] In this embodiment of the invention, the preset defect report set may be historical defect reports or other defect reports to be detected.

[0077] S3. If the defect report is a duplicate defect report, then the defect report is an invalid report.

[0078] S4. If the defect report is a non-duplicate defect report, the fusion feature is input into the automatic defect report classification model, and the defect report is classified using the automatic defect report classification model to obtain the classification result of whether the defect report is a valid report or an invalid report.

[0079] As an embodiment of the present invention, the automatic defect report classification model includes a multi-head attention network layer, a feedforward neural network layer, a pooling layer, and a fully connected layer.

[0080] In this embodiment of the invention, the automatic defect report classification model refers to a neural network model that can classify defect reports into valid defect reports and invalid defect reports.

[0081] As an embodiment of the present invention, classifying the defect reports using an automatic defect report classification model includes:

[0082] A multi-head attention network layer is used to process the received fusion features to obtain a fused linear projection;

[0083] The fused linear projection is activated and normalized using a feedforward neural network layer to obtain the fused output;

[0084] The fused output is pooled using a pooling layer to obtain a fused pooled output;

[0085] The classification probability of the fusion pooling output is calculated using a fully connected layer, and the defect report classification result is determined based on the classification probability.

[0086] For example, the calculation of the fused features using the multi-head attention network layer involves the following steps:

[0087] Input representation S i The linear projections learned through the three individuals and V vectors are shown below:

[0088] Q = S i W Q K = S i W K V = S i W V

[0089] in, d e This represents the dimension of each vector. The output of the attention network is calculated as follows:

[0090]

[0091] Based on this, repeating the calculation h times yields the output of the multi-head self-attention network, resulting in different linear projections, where h is the number of attention heads. The dimensions of Q, K, and V are expressed as d. q d k and d v The dimensions of Q, K, and V are all equal to d. e / h. Therefore, the output O of the multi-head self-attention network. M The calculation is as follows:

[0092] O M =MultiHead(Q,K,V)=Concat(head1,…,head h W O

[0093] Concat is a join operation that joins each self-attention head. j (1≤j≤h), projection head j The calculation formula is as follows:

[0094]

[0095] The multi-head self-attention network is residually connected, and then layer normalization is performed. The output O of the normalization layer... N It can be represented as:

[0096] O N =LayerNorm(inputs+O M )

[0097] Furthermore, the activation and normalization of the fused linear projection using a feedforward neural network layer is performed as follows:

[0098] The output value O of the feedforward neural network layer F The calculation method is as follows:

[0099] O F =ReLU(O N W1)W2+b2

[0100] Where b2 is a constant, and the parameter matrix is... Finally, the outputs are calculated using a normalization layer:

[0101] outputs = LayerNorm(O F +O F )

[0102] In this embodiment of the invention, the pooling operation in the pooling layer adopts global average pooling, and the pooling operation steps are as follows:

[0103] Global average pooling is used for calculation, for the output z i =[z1,z2,…,z n The calculation formula is as follows:

[0104]

[0105] Where G is the vector after global average pooling.

[0106] For example, the calculation of the classification probability of the fused pooling output using a fully connected layer employs the following steps:

[0107] By concatenating the vectors obtained from the global average pooling of the pooling layers, we get G = [g1, g2, ..., g k ];

[0108] Based on the vector obtained after global average pooling, a linear classifier and a sigmoid activation function in the fully connected layer are used for classification prediction. The classification prediction can be calculated using the following formula:

[0109]

[0110] Among them, y predict For the classification prediction results, w i Let b be the weight vector and b be the bias parameter.

[0111] This invention, through extracting structured and unstructured features from the defect report, ensures the completeness of the obtained defect report information, thereby guaranteeing the full utilization of each piece of information in the defect report and ensuring the accuracy of subsequent defect report classification. Furthermore, by fusing the structured and unstructured features to obtain fused features, a duplicate defect report detection model is used to detect whether the defect report is a duplicate report. Prioritizing the detection of duplicate defect reports reduces their impact on the effective classification of defect reports. Additionally, using an automatic defect report classification model further ensures the efficiency of classifying effective defect reports.

[0112] Table 2 shows a comparison of the experimental results of the effective defect reporting detection method—BugVal in this invention—with the CNN-based method in the prior art.

[0113] Table 2

[0114]

[0115] The table above compares the experimental results of the present invention's valid defect report detection method—BugVal—with CNN-based methods. On the Eclipse dataset, the present invention's method improves the baseline in AUC and F1 score by 8.8% and 8.9%, respectively. On the Mozilla dataset, the present invention's method improves AUC and F1 score by 10.9% and 7.1%, respectively. On the Netbeans dataset, the present invention's method improves AUC and F1 score by 24.0% and 13.9%, respectively. On average, across the five datasets, the present invention's method achieves 96.3% and 94.1% in AUC and F1 score, respectively. Compared to the baseline, AUC, F1, and F1 score are improved by 14.1% and 9.9%, respectively. This indicates that the present invention's method has high performance in predicting accurate method names and can correctly detect valid defect reports.

[0116] Table 3 shows the impact of structured features on the Eclipses dataset on the performance of BugVal, the effective defect report detection method in this invention.

[0117] Table 3

[0118] AUC% Accuracy% Precision% Recall% F1_score% TFVBRD 95.1 91.1 93.2 95.7 94.4 w\o-PRO 93.9 90.3 92.7 94.2 92.9 w\o-COM 93.7 90.4 92.4 94.6 93.1 w\o-VER 94.9 91.0 93.1 95.3 94.2 w\o-SEV 95.0 90.5 93.0 95.7 94.3 w\o-PRI 95.0 90.9 92.9 94.2 94.0 w\o-REP 94.0 90.3 92.6 94.5 93.0 w\o-ATT 93.6 89.9 91.5 94.2 92.6 <![CDATA[BugVal stru ]]> 66.9 53.6 75.6 65.1 61.4 <![CDATA[BugVal text ]]> 92.7 88.3 92.3 93.2 92.4

[0119] The table above indicates that w\o-PRO means removing the PRO structured feature and using the other six structured features combined with text features to represent the defect report (the same applies to w\o-COM, w\o-VER, w\o-SEV, w\o-PRI, w\o-REP, and w\o-ATT). BugVal stru BugVal indicates that only structured features are used to represent defect reports. text This indicates that only unstructured features, i.e., text, are used to represent defect reports. Experimental results show that removing individual structured features leads to varying degrees of performance degradation in the overall model. The model with no-REP features (removing REP features) shows the largest performance drop, with AUC, Recall, and F1 score decreasing by 1.1%, 1.2%, and 1.4%, respectively. However, when all structured features are removed and only text features are used, BugVal... text Its performance is worse than any model that has had its single structured feature removed. Compared to the original BugVal model, BugVal... text The AUC, Accuracy, and Recall decreased by 2.4%, 2.8%, and 2.0%, respectively. This indicates that structured information is supplementary to the textual description in the report, and each of the seven structured features (PRO, COM, VER, SEV, PRI, REP, and ATT) is crucial for enriching the information representation of defect reports. Combining all structured features with unstructured text features can better represent defect reports, resulting in better model performance. In comparative experiments using structured features alone and using unstructured text features alone, the results show that using text features alone (BugVal)... text The performance of the model will decrease if only all structured features are used without text features (BugVal). stru The performance of the structured feature was the worst, with a decrease of 28.2% in AUC and 30.6% in Recall. This indicates that using structured features alone is not a good way to characterize defect reports.

[0120] Table 4 shows the impact of the duplicate defect report pre-classification module on the Eclipses dataset on the performance of the effective defect report detection method—BugVal in this invention.

[0121] Table 4

[0122] AUC% Accuracy% Precision% Recall% F1_score% BugVal 95.1 91.1 93.2 95.7 94.4 w\o-DBRP 88.7 85.4 81.1 93.2 90.5

[0123] The table above shows that w\o-DBRP removes the duplicate defect report pre-classification module and, following the previous method of labeling valid and invalid defect reports, marks duplicate defect reports with a resolution field of "DUPLICATE" as invalid. Then, the defect report representation obtained by combining structured and unstructured features from the first stage is directly used as input to TransformerEncode for deep feature learning and report validity prediction in the second stage. Experimental results show that the performance of the w\o-DBRP model is lower than that of the original BugVal, decreasing by 6.4%, 5.7%, 12.1%, and 3.9% in AUC, Accuracy, Precision, and F1 score, respectively. This indicates that labeling and classifying duplicate defect reports as valid is crucial. If all duplicate defect reports are marked as invalid, the deep learning model may be misled by incorrect feedback when learning defect report features, leading to a decline in model performance.

[0124] like Figure 2 The diagram shown is a flowchart illustrating a specific implementation of an embodiment of the effective defect report detection method provided by an embodiment of the present invention.

[0125] like Figure 3 The diagram shown is a functional block diagram of an effective defect reporting and detection device provided in an embodiment of the present invention.

[0126] The effective defect report detection device 100 of the present invention can be installed in an electronic device. Depending on the functions implemented, the effective defect report detection device 100 may include a feature extraction module 101, a duplicate defect report detection module 102, and a defect report classification module 103.

[0127] The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0128] In this embodiment, the functions of each module / unit are as follows:

[0129] The feature extraction module 101 is used to obtain a defect report and extract the structured and unstructured features from the defect report.

[0130] In this embodiment of the invention, the defect report is a document that records detailed information about defects in a software product. It has a structured format and can facilitate testers to carry out software activities based on the defect report.

[0131] As an embodiment of the present invention, the extraction of structured and unstructured features from the defect report includes:

[0132] Identify the structured information field names in the defect report, and extract the information under the field names to obtain the structured information;

[0133] The information description text and comment text in the defect report are queried, and the text description text and comment text are normalized to obtain the unstructured information;

[0134] Structured features are extracted from the structured information using regular expressions, and unstructured features are extracted from the unstructured information after text processing.

[0135] In this embodiment of the invention, the structured information refers to information in a specific format in the defect report, which has specific field names. For example, the structured information includes, but is not limited to, BID (bug_id), DID (dup_id), PRO (product), COM (component), VER (version), SEV (severity), PRI (priority), REP (repoter), and ATT (attachment).

[0136] In this embodiment of the invention, the unstructured information refers to the title text, summary text, and comment text in each defect report.

[0137] In this embodiment of the invention, the regular expression refers to a tool that can perform text matching using a specially written string, and can extract the required data or information from complex text.

[0138] Furthermore, the extraction of structured features from the structured information using regular expressions includes:

[0139] Based on the field names of the structured information, regular expressions are used to write feature extraction rules, and these rules are then used to extract structured features.

[0140] The duplicate defect report detection module 102 is used to fuse the structured features and the unstructured features to obtain fused features, input the fused features into the duplicate defect report detection model, and use the duplicate defect report detection model to detect whether the defect report is a duplicate defect report.

[0141] As an embodiment of the present invention, the step of fusing the structured features and the unstructured features to obtain the fused features includes:

[0142] The structured features are embedded using one-hot encoding to obtain a structured feature vector;

[0143] The unstructured features are transformed into feature vectors using a pre-defined text vector model, resulting in unstructured feature vectors.

[0144] The structured feature vector and the unstructured feature vector are concatenated to obtain the fused feature.

[0145] Specifically, the step of using a duplicate defect report detection model to detect whether the defect report is a duplicate defect report includes:

[0146] The fused features are embedded into nodes using the heterogeneous information network in the recurring defect report detection model to obtain embedded information.

[0147] The bidirectional LSTM network in the duplicate defect report detection model is used to learn the textual semantic features in the embedded information;

[0148] The similarity between the defect report and the defect reports in the preset defect report set is calculated based on the semantic features of the text, and the similarity is used to detect whether the defect report is a duplicate defect report.

[0149] Specifically, if the similarity with any defect report in the preset defect report set is greater than the preset similarity threshold, the defect report is considered a duplicate defect report; if the similarity with all defect reports in the preset defect report set is not greater than the preset similarity threshold, the defect report is considered not a duplicate defect report.

[0150] In this embodiment of the invention, the heterogeneous information network refers to a network composed of nodes and relationships of different types. In a heterogeneous information network, nodes and relationships can have different types and attributes.

[0151] In this embodiment of the invention, the bidirectional LSTM network refers to a variant of a recurrent neural network that can effectively capture semantic features in text sequences.

[0152] In this embodiment of the invention, the preset defect report set may be historical defect reports or other defect reports to be detected.

[0153] The defect report classification module 103 is used to classify a defect report as invalid if it is a duplicate defect report, and to input the fusion features into the automatic defect report classification model if it is a non-duplicate defect report, and to classify the defect report using the automatic defect report classification model to obtain a classification result of whether the defect report is a valid report or an invalid report.

[0154] If the defect report is a duplicate defect report, then the defect report is invalid.

[0155] As an embodiment of the present invention, the automatic defect report classification model includes a multi-head attention network layer, a feedforward neural network layer, a pooling layer, and a fully connected layer.

[0156] In this embodiment of the invention, the automatic defect report classification model refers to a neural network model that can classify defect reports into valid reports and invalid reports.

[0157] As an embodiment of the present invention, classifying the defect reports using an automatic defect report classification model includes:

[0158] A multi-head attention network layer is used to process the received fusion features to obtain a fused linear projection;

[0159] The fused linear projection is activated and normalized using a feedforward neural network layer to obtain the fused output;

[0160] The fused output is pooled using a pooling layer to obtain a fused pooled output;

[0161] The classification probability of the fusion pooling output is calculated using a fully connected layer, and the defect report classification result is determined based on the classification probability.

[0162] For example, the calculation of the fused features using the multi-head attention network layer involves the following steps:

[0163] Input representation S i The linear projections learned through the three individuals and V vectors are shown below:

[0164] Q = S i W Q K = S i W K V = S i W V

[0165] in, d e This represents the dimension of each vector. The output of the attention network is calculated as follows:

[0166]

[0167] Based on this, repeating the calculation h times yields the output of the multi-head self-attention network, resulting in different linear projections, where h is the number of attention heads. The dimensions of Q, K, and V are expressed as d. q d k and d v The dimensions of Q, K, and V are all equal to d. e / h. Therefore, the output O of the multi-head self-attention network.M The calculation is as follows:

[0168] O M =MultiHead(Q,K,V)=Concat(head1,…,head h W O

[0169] Concat is a join operation that joins each self-attention head. j (1≤j≤h), projection head j The calculation formula is as follows:

[0170]

[0171] The multi-head self-attention network is residually connected, and then layer normalization is performed. The output O of the normalization layer... N It can be represented as:

[0172] O N =LayerNorm(inputs+O M )

[0173] Furthermore, the activation and normalization of the fused linear projection using a feedforward neural network layer is performed as follows:

[0174] The output value O of the feedforward neural network layer F The calculation method is as follows:

[0175] O F =ReLU(O N W1)W2+b2

[0176] Where b2 is a constant, and the parameter matrix is... Finally, the outputs are calculated using a normalization layer:

[0177] outputs = LayerNorm(O N +O F )

[0178] In this embodiment of the invention, the pooling operation in the pooling layer adopts global average pooling, and the pooling operation steps are as follows:

[0179] Global average pooling is used for calculation, for the output z i =[z1,z2,…,z n The calculation formula is as follows:

[0180]

[0181] Where G is the vector after global average pooling.

[0182] For example, the calculation of the classification probability of the fused pooling output using a fully connected layer employs the following steps:

[0183] By concatenating the vectors obtained from the global average pooling of the pooling layers, we get G = [g1, g2, ..., g k ];

[0184] Based on the vector obtained after global average pooling, a linear classifier and a sigmoid activation function in the fully connected layer are used for classification prediction. The classification prediction can be calculated using the following formula:

[0185]

[0186] Among them, y predict For the classification prediction results, w i Let b be the weight vector and b be the bias parameter.

[0187] like Figure 4 The diagram shown is a structural schematic of an electronic device that implements an effective defect reporting detection method according to an embodiment of the present invention.

[0188] The electronic device may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13. It may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a valid defect report detection method program.

[0189] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., executing a valid defect report detection method program) and calls data stored in the memory 11 to perform various functions of the electronic device and process data.

[0190] The memory 11 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 11 can include both internal and external storage units of the electronic device. The memory 11 can be used not only to store application software and various types of data installed on the electronic device, such as the code of an effective defect report detection method program, but also to temporarily store data that has been output or will be output.

[0191] The communication bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0192] The communication interface 13 is used for communication between the aforementioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), or, optionally, a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0193] Figure 4 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 4The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0194] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0195] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0196] The effective defect report detection method program stored in the memory 11 of the electronic device is a combination of multiple instructions, which, when run in the processor 10, can achieve the following:

[0197] Obtain the defect report and extract the structured and unstructured features from the defect report;

[0198] The structured features and the unstructured features are fused to obtain fused features. The fused features are then input into the duplicate defect report detection model, which is used to detect whether the defect report is a duplicate defect report.

[0199] If the defect report is a duplicate defect report, then the defect report is an invalid report;

[0200] If the defect report is a non-duplicate defect report, the fusion feature is input into the automatic defect report classification model, and the defect report is classified using the automatic defect report classification model to obtain the classification result of whether the defect report is a valid report or an invalid report.

[0201] Specifically, the specific implementation method of the processor 10 for the above instructions can be referred to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, and will not be repeated here.

[0202] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0203] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0204] Obtain the defect report and extract the structured and unstructured features from the defect report;

[0205] The structured features and the unstructured features are fused to obtain fused features. The fused features are then input into the duplicate defect report detection model, which is used to detect whether the defect report is a duplicate defect report.

[0206] If the defect report is a duplicate defect report, then the defect report is an invalid report;

[0207] If the defect report is a non-duplicate defect report, the fusion feature is input into the automatic defect report classification model, and the defect report is classified using the automatic defect report classification model to obtain the classification result of whether the defect report is a valid report or an invalid report.

[0208] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0209] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0210] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0211] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0212] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0213] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0214] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0215] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0216] Finally, it should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An effective defect report detection method, characterized by, The method comprises: obtaining a defect report, extracting structured features and unstructured features in the defect report; performing feature fusion on the structured features and the unstructured features to obtain fused features, wherein the feature fusion on the structured features and the unstructured features to obtain fused features comprises: performing structured embedding on the structured features using one-hot encoding to obtain a structured feature vector; converting the unstructured features into a feature vector using a preset text vector model to obtain an unstructured feature vector; concatenating the structured feature vector and the unstructured feature vector to obtain the fused features, and inputting the fused features into a repeated defect report detection model to detect whether the defect report is a repeated defect report using the repeated defect report detection model, wherein the detection of whether the defect report is a repeated defect report using the repeated defect report detection model comprises: performing node embedding on the fused features using a heterogeneous information network in the repeated defect report detection model to obtain embedded information; learning text semantic features in the embedded information using a bidirectional LSTM network in the repeated defect report detection model; calculating a similarity of the defect report to defect reports in a preset defect report set according to the text semantic features, and detecting whether the defect report is a repeated defect report according to the similarity; if the defect report is a repeated defect report, the defect report is an invalid report; if the defect report is a non-repeated defect report, inputting the fused features into a defect report automatic classification model, wherein the defect report automatic classification model comprises a multi-head attention network layer, a feedforward neural network layer, a pooling layer, and a fully connected layer, classifying the defect report using the defect report automatic classification model to obtain a classification result that the defect report is a valid report or an invalid report, wherein the classification of the defect report using the defect report automatic classification model comprises: processing the received fused features using the multi-head attention network layer to obtain a fused linear projection; performing activation processing and normalization processing on the fused linear projection using the feedforward neural network layer to obtain a fused output; performing pooling processing on the fused output using the pooling layer to obtain a fused pooling output; calculating a classification probability of the fused pooling output using the fully connected layer, and determining a defect report classification result of the defect report according to the classification probability.

2. The effective defect report detection method of claim 1, wherein, The extraction of the structured features and the unstructured features in the defect report comprises: identifying a structured information field name in the defect report, and extracting information under the field name to obtain the structured information; querying information description text and comment text in the defect report, and performing text standardization operation on the information description text and the comment text to obtain the unstructured information; extracting structured features from the structured information using a regular expression, and extracting unstructured features after text processing of the unstructured information.

3. The effective defect report detection method of claim 2, wherein, The extraction of the structured features from the structured information using a regular expression comprises: A feature extraction rule is written according to a field name of the structured information by using a regular expression writing rule, and a structured feature is extracted by using the feature extraction rule.

4. An effective defect report detection apparatus characterized by comprising: The device can implement the effective defect report detection method according to any one of claims 1 to 3, and the device comprises: a feature extraction module, configured to acquire a defect report, and extract a structured feature and an unstructured feature in the defect report; a repeated defect report detection module, configured to perform feature fusion on the structured feature and the unstructured feature to obtain fused features, input the fused features into a repeated defect report detection model, and detect whether the defect report is a repeated defect report by using the repeated defect report detection model; a defect report classification module, configured to, if the defect report is a repeated defect report, determine that the defect report is an invalid report; and if the defect report is a non-repeated defect report, input the fused features into a defect report automatic classification model, classify the defect report by using the defect report automatic classification model, and obtain a classification result that the defect report is an effective report or an invalid report.

5. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the effective defect report detection method according to any one of claims 1 to 3.

6. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to implement the effective defect report detection method according to any one of claims 1 to 3.

Images