Intelligent software technology defect analysis method based on multi-dimensional big data

Abstract

The invention discloses a software technology defect intelligent analysis method based on multi-dimensional big data, and relates to the technical field of software intelligent analysis, and the method comprises the steps: collecting multi-dimensional heterogeneous data in software operation, carrying out the preprocessing to obtain a standardized multi-dimensional heterogeneous data sequence, building a defect prediction model based on a time sequence analysis and depth feature extraction algorithm, generating a risk assessment result and a corresponding risk feature; analyzing a risk assessment result and corresponding risk features by using a convolutional neural network, automatically identifying and classifying defects, generating a defect assessment result, and dynamically correcting the defect assessment result in combination with historical repair data and real-time detection feedback; and constructing a causal reasoning model based on the correlation characteristics between the defect evaluation result and the defects. According to the method, the defect prediction model is constructed, and the convolutional neural network is used for dynamic correction and causal reasoning to generate the repair suggestions, so that the accuracy of software defect prediction is enhanced, and the intelligent level of repair decision is also improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of software intelligent analysis, in particular to a software technical defect intelligent analysis method based on multi-dimensional big data. BACKGROUND

[0002] In recent years, with the increasing complexity of software operation, software defects have become one of the key factors affecting software quality and stability. Traditional software defect detection and repair methods mostly rely on manual analysis and empirical rules, which often cannot respond and predict software defects in real time. For example, static code analysis and rule-based defect detection can find potential problems in programs to some extent, but often cannot cover all types of defects comprehensively, and respond slowly to problems in dynamic environments.

[0003] With the rapid development of big data technology and machine learning, data-driven defect detection methods have gradually become a research hotspot. By collecting and analyzing a large amount of runtime data in real time, using deep learning models to automatically identify potential defects, not only can improve the accuracy of defect detection, but also can realize a more efficient early warning mechanism. Especially in a multi-dimensional, big data environment, deep neural networks and convolutional neural networks are widely used in feature extraction and defect identification, which can extract valuable information from complex system logs, performance data and user behavior data. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a software technical defect intelligent analysis method based on multi-dimensional big data, which solves the problem that traditional software defect prediction methods cannot handle complex data, lack real-time dynamic correction and intelligent repair suggestions.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a software technical defect intelligent analysis method based on multi-dimensional big data, which includes, Collecting multi-dimensional heterogeneous data in software operation, preprocessing to obtain standardized multi-dimensional heterogeneous data sequence, constructing a defect prediction model based on time series analysis and deep feature extraction algorithm, generating risk assessment results and corresponding risk features; Using a convolutional neural network to analyze risk assessment results and corresponding risk features, automatically identifying and classifying defects, generating defect assessment results, and dynamically correcting defect assessment results in combination with historical repair data and real-time detection feedback; Based on the correlation between the defect evaluation result and the defect, a causal reasoning model is constructed, a repair suggestion is automatically generated, the software is repaired according to the repair suggestion, and the data in the software repair process is uploaded to the database for saving.

[0007] As a preferred scheme of the software technical defect intelligent analysis method based on multi-dimensional big data, the multi-dimensional heterogeneous data in the software running is collected, and the standardized multi-dimensional heterogeneous data sequence is obtained by preprocessing, including the following steps: The multi-dimensional heterogeneous data generated in the software running process is collected, the collected data is standardized, the standardized multi-dimensional heterogeneous data is generated, and the standardized multi-dimensional heterogeneous data sequence is obtained based on the continuous standardized multi-dimensional heterogeneous data in the collection period.

[0008] As a preferred scheme of the software technical defect intelligent analysis method based on multi-dimensional big data, the defect prediction model is constructed based on time series analysis and deep feature extraction algorithm, and the risk evaluation result and corresponding risk features are generated, including the following steps: The multi-dimensional heterogeneous data sequence is divided into samples according to the time window, and all time windows are encoded by using the time series encoder, to obtain the risk feature vector corresponding to the time window; The risk feature vector is input into the lightweight scoring head, and the scalar risk score is output; The loss of the risk score output by the scoring head is calculated using the real risk label, and the gradient descent method is used to train the parameter set used by the encoder and the scoring head until the loss is minimized, to obtain the trained parameter set; The trained parameter set is applied to the encoder and the scoring head to obtain the defect prediction model that can input the multi-dimensional heterogeneous data sequence and output the risk evaluation result and the corresponding risk features; The risk score threshold is set, and for the sample with a risk score greater than or equal to the risk score threshold, the corresponding time window is determined as risky, otherwise it is determined as non-risky.

[0009] As a preferred scheme of the software technical defect intelligent analysis method based on multi-dimensional big data, the convolutional neural network is used to analyze the risk evaluation result and the corresponding risk features, to automatically identify and classify defects, and to generate defect evaluation results, including the following steps: The risk evaluation result and the corresponding risk features of the time window determined as risky are spliced to obtain an enhanced feature vector; The enhanced feature vector is rearranged into a one-dimensional feature sequence to obtain an input feature matrix of the convolutional network; The feature matrix is input into the convolutional neural network, and the single-layer convolution outputs the convolutional features; The global average pooling is performed on the convolutional features to obtain a defect feature vector with a fixed length. The defect feature vector is taken as an input of a full connection layer, and a defect category probability distribution is obtained through softmax normalization. The defect feature vector and the defect category probability distribution are taken as a defect evaluation result. The loss of defect classification output by the convolutional neural network is calculated using the real classification label, and the parameter set of the convolutional layer and the full connection layer is trained using the gradient descent method until the loss is minimized, so that the trained parameter set is obtained.

[0010] As a preferred scheme of the software technology defect intelligent analysis method based on multi-dimensional big data, wherein: the historical repair data and real-time detection feedback are combined to dynamically correct the defect evaluation result, including the following steps: The historical repair database is clustered and statistically analyzed to extract the performance rules of the same type of defects in multiple operating environments, and a correction rule library corresponding to the current defect category and features is established; The historical repair data is searched to obtain the historical defect records of the same category as the current defect, and the real-time detection feedback data in the current operating environment is collected as real-time feedback information; The defect evaluation result is dynamically corrected according to the real-time feedback information and the pre-defined correction rule library.

[0011] As a preferred scheme of the software technology defect intelligent analysis method based on multi-dimensional big data, wherein: a causal reasoning model is constructed based on the association features between the defect evaluation result and the defects to automatically generate a repair suggestion, including the following steps: According to the historical defect records, the co-occurrence frequency and time sequence relationship between the defects are calculated to obtain the association values between the defects, an association threshold is set, the defects with an association value greater than or equal to the association threshold are considered to have a causal relationship, and an association feature matrix between the defects is obtained; Based on the defect evaluation result and the association feature matrix, a causal graph structure containing a node set of associated defects and a directed edge set is constructed, and the causal graph structures of all defects are integrated to obtain a causal reasoning model; The current defect evaluation result is input into the causal reasoning model to obtain the root cause defect of the current defect, the repair scheme for repairing the root cause defect is found according to the historical repair data, and the comprehensive score of each repair scheme is calculated according to the success rate and the repair cost, and the repair scheme with the highest comprehensive score is taken as the repair scheme.

[0012] As a preferred scheme of the software technical defect intelligent analysis method based on multi-dimensional big data, wherein: the software is repaired according to the repair suggestion, and verification is performed, including the following steps: The repair scheme is applied to the currently running software, multi-dimensional heterogeneous data of the software running process is collected in real time, and the multi-dimensional heterogeneous data is input into the defect prediction model for judgment, if it is still determined to be risky, it means that the repair is unqualified, and needs to be repaired again, if it is determined to be risk-free, it means that the repair is qualified.

[0013] As a preferred scheme of the software technical defect intelligent analysis method based on multi-dimensional big data, wherein: the data in the software repair process is uploaded to the database for storage, including the following steps: The multi-dimensional heterogeneous data of the software running, the output data of the defect prediction model, the output data of the convolutional neural network, the repair scheme data and the repair result data are used as experience data, and are uploaded to the database for storage after being uniformly formatted.

[0014] In a second aspect, the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, wherein: the computer program is executed by the processor to realize any step of the software technical defect intelligent analysis method based on multi-dimensional big data according to the first aspect of the present application.

[0015] In a third aspect, the present application provides a computer readable storage medium, which stores a computer program, wherein: the computer program is executed by the processor to realize any step of the software technical defect intelligent analysis method based on multi-dimensional big data according to the first aspect of the present application.

[0016] The present application has the following advantages: by combining time sequence analysis and deep feature extraction algorithm to construct a defect prediction model, accurate prediction of software defects can be realized, and risk assessment results and risk features are generated, which effectively improves the recognition ability of different types of software defects, especially in complex and variable running environment, higher precision risk assessment and prediction are provided. By using convolutional neural network to further analyze the risk assessment results and risk features, defects are automatically identified and classified, and the defect evaluation results are dynamically modified by combining historical repair data and real-time feedback, effectively solving the problem that the repair strategy cannot be adjusted and fed back in time in the traditional method, realizing real-time and accurate defect correction. Finally, by constructing a causal reasoning model to analyze the correlation characteristics between defects, repair suggestions are automatically generated to ensure the pertinence and effectiveness of the repair strategy, and the efficiency and accuracy of software repair are improved. At the same time, the data in the repair process is uploaded to the database, which provides continuous support for subsequent optimization and model improvement. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0018] Figure 1 A flowchart of the software technical defect intelligent analysis method based on multi-dimensional big data.

[0019] Figure 2 A schematic diagram of software defect evaluation and dynamic correction. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0021] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced in other ways different from those described herein without departing from the scope of the present application, and those skilled in the art can make similar generalizations without departing from the scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0022] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or alternative to other embodiments.

[0023] Reference Figure 1 and Figure 2 For one embodiment of the present application, the embodiment provides a software technical defect intelligent analysis method based on multi-dimensional big data, comprising the following steps: Collecting multi-dimensional heterogeneous data in software running, pre-processing to obtain standardized multi-dimensional heterogeneous data sequence, constructing defect prediction model based on time series analysis and deep feature extraction algorithm, generating risk assessment results and corresponding risk features.

[0024] Specifically, multi-dimensional heterogeneous data generated during software running is collected, the collected data is standardized to generate standardized multi-dimensional heterogeneous data, and based on continuous standardized multi-dimensional heterogeneous data within the collection period, a standardized multi-dimensional heterogeneous data sequence is obtained.

[0025] By collecting the multi-dimensional heterogeneous data generated during the running of the software and standardizing the multi-dimensional heterogeneous data sequence, the uniformity and comparability of the data are ensured, the differences between the data sources are eliminated, and the accuracy of subsequent analysis and prediction is ensured.

[0026] Further, the length of the time window is set, the multi-dimensional heterogeneous data sequence is divided into samples according to the time window, and all time windows are encoded end-to-end using a time sequence encoder to obtain a risk feature vector corresponding to the time window, and the formula is: ; Wherein, is the index of the time window, is the risk feature vector of the th time window, is the time sequence deep encoder function, is the sample of the th time window, is the parameter set used by the encoder.

[0027] The risk feature vector is input into the lightweight scoring head, and a scalar risk score is output, and the formula is: ; Wherein, is the risk score of the th time window, is a Sigmoid function used to map linear output to , is the transpose of the risk score weight matrix which can be adaptively changed, is the adaptive risk score bias term.

[0028] The loss of the risk score output by the scoring head is calculated using the real risk label, and the parameter set used by the encoder and the scoring head is trained using the gradient descent method until the loss is minimized, and the trained parameter set is obtained; The trained parameter set is applied to the encoder and the scoring head to obtain a defect prediction model that can input multi-dimensional heterogeneous data sequences and output risk assessment results and corresponding risk features; Set the risk score threshold, for samples with risk score greater than or equal to the risk score threshold, judge the corresponding time window as risky, otherwise judge as non-risky.

[0029] By setting a time window and segmenting the multi-dimensional heterogeneous data sequence, deep time-related features are extracted by combining a time encoder, enabling the defect prediction model to effectively identify and analyze patterns and changes in time series, improving the accuracy of defect prediction. The lightweight scoring head simplifies the calculation process and optimizes the parameters of the defect prediction model through gradient descent, ensuring efficient and accurate risk assessment. The trained defect prediction model can adapt to different types of data and quickly generate accurate defect prediction results.

[0030] The risk assessment results and corresponding risk features are analyzed using a convolutional neural network to automatically identify and classify defects, generating defect assessment results. Historical repair data and real-time detection feedback are combined to dynamically correct the defect assessment results.

[0031] Specifically, the risk assessment results of the time window determined to be at risk are concatenated with the corresponding risk features to obtain an enhanced feature vector. The enhanced feature vector is rearranged into a one-dimensional feature sequence to obtain an input feature matrix for the convolutional network. The feature matrix is input into the convolutional neural network, and the single-layer convolution outputs convolutional features, with the formula being: ; wherein, is the convolutional feature output by the single-layer convolution, is a one-dimensional convolution operation, is a feature matrix obtained based on the enhanced feature vector, is an adaptively changing convolution kernel parameter, is an adaptively changing convolution layer bias term.

[0032] Global average pooling is performed on the convolutional features to obtain a fixed-length defect representation vector. The defect representation vector is input into the fully connected layer, and the defect class probability distribution is obtained by softmax normalization, with the formula being: ; wherein, is the defect class probability distribution, is a normalization function, is the fixed-length defect representation vector, is an adaptively changing defect classification weight matrix, is an adaptively changing defect classification bias term.

[0033] The defect representation vector and the defect class probability distribution are used as the defect assessment results. The loss of defect classification output by the convolutional neural network is calculated using real classification labels, and the parameter set of the convolutional layer and the fully connected layer is trained using gradient descent until the loss is minimized, obtaining a trained parameter set. The trained parameter set is applied to the convolutional layer and the fully connected layer to obtain a convolutional neural network that can input an enhanced feature vector and output a defect evaluation result.

[0034] By rearranging the enhanced feature vector into a one-dimensional feature sequence as the input feature matrix of the convolutional neural network, the convolutional neural network can more effectively process and extract key features, improving the computational efficiency and response speed of the convolutional neural network. Local features can be efficiently extracted through single-layer convolution operations, providing efficient and discriminative feature representations for subsequent defect classification. Global average pooling operations integrate convolutional features to obtain fixed-length defect representation vectors, reducing information loss while preserving key information required by the model. Defect representation vectors are input into the fully connected layer and normalized by softmax to generate probability distributions of defect categories, achieving accurate classification of defect types and providing accurate decision-making basis for subsequent repair strategy generation.

[0035] Further, the historical repair database is clustered and statistically analyzed to extract the performance rules of the same type of defects in multiple operating environments, and a correction rule library corresponding to the current defect category and features is established. The historical repair data is searched to obtain historical defect records of the same category as the current defect, and real-time detection feedback data in the current operating environment is collected as real-time feedback information. The defect evaluation result is dynamically corrected according to the real-time feedback information and the predefined correction rule library.

[0036] By clustering and statistically analyzing the historical repair database, the performance rules of the same type of defects in different operating environments are extracted, and a correction rule library is established, which can provide accurate repair strategies for the current defect based on historical data. Searching historical repair data, combining the current defect category and real-time environmental feedback, further analyzing and adjusting the defect evaluation result in real time ensures the dynamic adaptability of the repair strategy, making each defect repair suggestion more accurate and efficient to meet the needs of the current environment.

[0037] Based on the correlation between the defect evaluation result and the defect, a causal reasoning model is constructed to automatically generate repair suggestions, the software is repaired according to the repair suggestions, and the data in the software repair process is uploaded to the database for saving.

[0038] Specifically, according to the historical defect records, the co-occurrence frequency and time sequence relationship between defects are calculated to obtain the correlation values between defects, a correlation threshold is set, defects with correlation values greater than or equal to the correlation threshold are considered to have a causal relationship, and a correlation feature matrix between defects is obtained. Based on the defect evaluation results and the correlation feature matrix, a causal graph structure containing a node set and a directed edge set of correlated defects is constructed, and the causal graph structures of all defects are integrated to obtain a causal reasoning model. The current defect evaluation results are input into the causal reasoning model to obtain the root cause defects of the current defect, the repair schemes for repairing all root cause defects are found according to the historical repair data, and the comprehensive scores of each repair scheme are calculated according to the success rate and repair cost, and the repair scheme with the highest comprehensive score is selected as the repair scheme.

[0039] By calculating the defect co-occurrence frequency and time sequence relationship in the historical defect records, the correlation values between defects are obtained and a correlation threshold is set to obtain the potential causal relationship between defects, which enhances the accuracy and depth of defect analysis. The causal graph model constructed based on the defect evaluation results and the correlation feature matrix converts the correlation between defects into a visual structure, which can comprehensively integrate the causal links of all defects, improve the pertinence and accuracy of defect prediction and repair strategy. By screening the optimal repair scheme from the historical repair data, the efficiency and low cost of the repair process are ensured, blind repair is avoided, and the intelligent level and overall repair efficiency of the repair decision are improved.

[0040] Further, the repair scheme is applied to the currently running software, multi-dimensional heterogeneous data of the software running process are collected in real time, and the multi-dimensional heterogeneous data are input into the defect prediction model for judgment. If it is still determined to be risky, it means that the repair is unqualified and needs to be repaired again. If it is determined to be risk-free, it means that the repair is qualified.

[0041] By inputting the real-time collected multi-dimensional heterogeneous data into the defect prediction model for judgment, the repair effect can be evaluated in time to ensure the stability and reliability of the software, which combines the repair process and real-time data feedback to improve the dynamic adaptability of the repair scheme and ensure the continuous optimization and risk control of software maintenance.

[0042] Further, the multi-dimensional heterogeneous data of the software running, the output data of the defect prediction model, the output data of the convolutional neural network, the repair scheme data, and the repair result data are used as experience data, which are saved in the database after being formatted.

[0043] Through comprehensive management of the data, all key data in the repair process can be systematically stored and traced, the usability and analyzability of the data are enhanced, and the effect of different repair schemes can be compared and analyzed, which provides more accurate reference for future defect prediction and repair decision.

[0044] The embodiment also provides a computer device suitable for the case of the intelligent analysis method for software technical defects based on multi-dimensional big data, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the intelligent analysis method for software technical defects based on multi-dimensional big data proposed in the above embodiment.

[0045] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0046] The embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to realize the intelligent analysis method for software technical defects based on multi-dimensional big data proposed in the above embodiment. The storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0047] To sum up, the application can realize accurate prediction of software defects, generate risk assessment results and risk features, effectively improve the identification ability of different types of software defects, especially in complex and variable operating environments, and provide higher precision risk assessment and prediction by combining time series analysis with deep feature extraction algorithm to construct a defect prediction model. By using a convolutional neural network to further analyze the risk assessment results and risk features, automatically identifying and classifying defects, and combining historical repair data and real-time feedback to dynamically correct the defect assessment results, the problem of the traditional method that the repair strategy cannot be adjusted and fed back in time is effectively solved, and real-time and accurate defect correction is realized. Finally, by constructing a causal reasoning model to analyze the correlation characteristics between defects, automatically generating repair suggestions to ensure the pertinence and effectiveness of the repair strategy, the efficiency and accuracy of software repair are improved. At the same time, the data in the repair process is uploaded to the database, providing continuous support for subsequent optimization and model improvement.

[0048] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered by the scope of the claims of the present application.

Claims

1. A method for intelligent analysis of software technical defects based on multi-dimensional big data, characterized in that: include, Collect multidimensional heterogeneous data during software operation, preprocess it to obtain a standardized multidimensional heterogeneous data sequence, construct a defect prediction model based on time series analysis and deep feature extraction algorithm, and generate risk assessment results and corresponding risk characteristics. The convolutional neural network is used to analyze the risk assessment results and corresponding risk characteristics, automatically identify and classify defects, generate defect assessment results, and dynamically correct the defect assessment results by combining historical repair data and real-time detection feedback. Based on the correlation characteristics between defect assessment results and defects, a causal reasoning model is constructed to automatically generate repair suggestions. The software is then repaired according to the repair suggestions, verified, and the data from the software repair process is uploaded to the database for storage. 2.The multi-dimensional big data based intelligent software technical defect analysis method according to claim 1, characterized in that: The multidimensional heterogeneous data collected by the acquisition software is preprocessed to obtain a standardized multidimensional heterogeneous data sequence, including the following steps: The software collects multidimensional heterogeneous data generated during operation, standardizes the collected data to generate standardized multidimensional heterogeneous data, and obtains a standardized multidimensional heterogeneous data sequence based on continuous standardized multidimensional heterogeneous data within the collection period. 3.The multi-dimensional big data based software technical defect intelligent analysis method of claim 2, wherein: The defect prediction model, constructed based on time series analysis and deep feature extraction algorithms, generates risk assessment results and corresponding risk characteristics, including the following steps: Set the time window length, divide the multidimensional heterogeneous data sequence into samples according to the time window, and use a time encoder to encode each time window end-to-end to obtain the risk feature vector of the corresponding time window. Input the risk feature vector into the lightweight scoring head and output a scalar risk score; The loss of the risk score output by the scoring head is calculated using real risk labels, and the parameter set used by the encoder and the scoring head is trained using gradient descent until the loss is minimized, thus obtaining the trained parameter set. The trained parameter set is applied to the encoder and scoring head to obtain a defect prediction model that can take in multidimensional heterogeneous data sequences and output risk assessment results and corresponding risk characteristics. Set a risk score threshold. For samples with a risk score greater than or equal to the risk score threshold, their corresponding time window is judged as risky; otherwise, it is judged as risk-free. 4.The multi-dimensional big data based software technical defect intelligent analysis method of claim 3, wherein: The process of using convolutional neural networks to analyze risk assessment results and corresponding risk characteristics, automatically identify and classify defects, and generate defect assessment results includes the following steps: The risk assessment results of the time window identified as risky are concatenated with the corresponding risk characteristics to obtain the enhanced feature vector; The enhanced feature vectors are rearranged into a one-dimensional feature sequence to obtain the input feature matrix of the convolutional network; The feature matrix is ​​input into the convolutional neural network, and a single convolutional layer outputs convolutional features. Global average pooling is applied to the convolutional features to obtain a fixed-length defect representation vector; The defect representation vector is used as the input to the fully connected layer, and the defect category probability distribution is obtained by softmax normalization. The defect characterization vector and the probability distribution of defect classification are used as the defect evaluation results; The loss of defect classification output by the convolutional neural network is calculated using real classification labels, and the parameter set of the convolutional layer and the fully connected layer is trained using gradient descent until the loss is minimized, obtaining a trained parameter set. The trained parameter set is applied to the convolutional layer and the fully connected layer to obtain a convolutional neural network that can input an enhanced feature vector and output a defect evaluation result. 5.The multi-dimensional big data based software technical defect intelligent analysis method according to claim 4, characterized in that: The defect evaluation result is dynamically corrected by combining historical repair data with real-time detection feedback, including the following steps: Cluster and statistically analyze the historical repair database to extract the performance rules of the same type of defects in various operating environments, and establish a correction rule library corresponding to the current defect category and characteristics; Retrieve historical repair data to obtain historical defect records of the same category as the current defect, and collect real-time detection feedback data in the current operating environment as real-time feedback information; According to the real-time feedback information and the pre-defined correction rule library, the defect evaluation result is dynamically corrected. 6.The multi-dimensional big data based software technical defect intelligent analysis method according to claim 5, characterized in that: Based on the association characteristics between the defect evaluation result and the defect, a causal reasoning model is constructed to automatically generate repair suggestions, including the following steps: According to the historical defect records, the co-occurrence frequency and time sequence relationship between the defects are calculated to obtain the association value between the defects. Set the association threshold value. Defects with an association value greater than or equal to the association threshold value are considered to have a causal relationship, and an association feature matrix between the defects is obtained. Based on the defect evaluation result and the association feature matrix, a causal graph structure containing a node set of associated defects and a directed edge set is constructed, and the causal graph structures of all defects are integrated to obtain a causal reasoning model. Input the current defect evaluation result into the causal reasoning model to obtain the root cause defect of the current defect. According to the historical repair data, find all repair solutions for the root cause defect, and calculate the comprehensive score of each repair solution according to the success rate and repair cost. The one with the highest comprehensive score is selected as the repair solution. 7.The multi-dimensional big data based software technical defect intelligent analysis method according to claim 6, wherein: The software is repaired according to the repair suggestion and verified, including the following steps: Apply the repair solution to the currently running software, collect multi-dimensional heterogeneous data during software operation in real time, and input the multi-dimensional heterogeneous data into the defect prediction model for judgment. If it is still determined to be risky, it means that the repair is unqualified and needs to be repaired again. If it is determined to be risk-free, it means that the repair is qualified. 8.The multi-dimensional big data based software technical defect intelligent analysis method of claim 7, wherein: The data during the software repair process is uploaded to the database for saving, including the following steps: The multi-dimensional heterogeneous data of software operation, the output data of the defect prediction model, the output data of the convolutional neural network, the repair solution data, and the repair result data are saved in the database as experience data after being formatted. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the software technology defect intelligent analysis method based on multi-dimensional big data according to any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the software technology defect intelligent analysis method based on multi-dimensional big data according to any one of claims 1-8.

Images