Software evaluation method and system with high credibility characteristics

Abstract

The invention discloses a software evaluation method and system with high credibility characteristics, and relates to the technical field of computer security, and the method comprises the following steps: obtaining code data; preprocessing the code data to obtain a code text library; performing feature extraction on the code text library, and identifying and extracting high-credibility feature information; constructing a DBN network evaluation model based on the code text library and the high-credibility feature information; and processing and analyzing the code original data according to the DBN network evaluation model, extracting high-credibility characteristics in the code original data, and generating a software evaluation report. According to the method, feature dimensions are greatly reduced through feature selection of code data, a large number of irrelevant and redundant features in the code data are removed through an online logistic regression model, high-credibility feature information is obtained, the high-credibility feature information is used for testing, the testing accuracy is fundamentally improved, and the time consumed by an algorithm is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of software testing, and more particularly to a software testing method and system with high-confidence features. Background Art

[0002] The railway information system is an information system used to manage the railway transportation network, including functions such as train running time, ticket sales, passenger information, freight information and vehicle scheduling. The railway information system can help railway companies improve transportation efficiency and service quality, and can also help passengers easily check timetables and buy tickets. With the rapid development of railway information systems, operation and maintenance management has become the focus of railway informatization construction. However, the current railway information system is still confined to some traditional operation and maintenance management models and cannot effectively meet the current high-speed business needs. One of the main reasons is that the operation and management of the information system lacks a fault analysis mechanism, so that the railway information system cannot be fully managed and fully mastered. Only through continuous follow-up investigation and summary analysis can the information system be prevented in advance, and more complete and accurate processing results can be achieved during the operation and maintenance process.

[0003] Therefore, how to provide a software evaluation method and system with high-confidence features to generate customizable evaluation reports and indicators to evaluate and improve the quality and performance of software is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0004] In view of this, the present invention provides a software evaluation method and system with high-confidence features, which greatly reduces the feature dimension by selecting features of code data, and uses an online logistic regression model to remove a large number of irrelevant and redundant features in the code data to obtain high-confidence feature information, and uses high-confidence feature information for testing, which fundamentally improves the test accuracy and reduces the time consumed by the algorithm. The DBN network evaluation model can learn the abstract feature representation in the code data, which helps to identify and extract high-confidence feature information. Improve the accuracy and depth of software evaluation. By using the trained high-confidence feature information, the generated software evaluation report is more convincing and practical.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: a software evaluation method with high reliability features, comprising:

[0006] Get code data;

[0007] Preprocessing the code data to obtain a code text library;

[0008] Performing feature extraction on the code text library to identify and extract high-confidence feature information;

[0009] Constructing a DBN network evaluation model based on the code text library and the high-confidence feature information;

[0010] The original code data is processed according to the DBN network evaluation model, high-confidence features therein are analyzed and extracted, and a software evaluation report is generated.

[0011] Preferably, the code data is obtained by using web crawler technology. Automatic acquisition of code data is achieved, which improves the efficiency and accuracy of data acquisition. The acquired code data is annotated and irrelevant information is removed to improve the quality and availability of the code data. The preprocessed code data is annotated to provide a labeled data set for subsequent feature extraction and model training. By performing text feature extraction and preference feature extraction on the annotated data, high-confidence features in the code data are extracted, providing useful information for subsequent data analysis and model training.

[0012] Preferably, feature extraction is performed on the code text library to identify and extract high-confidence feature information, including:

[0013] Using byte N-grams method to extract features from the code text library;

[0014] Sorting the features according to the correlation between the extracted features and the preset high-credibility feature categories to generate a first credible feature set;

[0015] Deleting redundant features in the first credible feature set according to an approximate Markov blanket algorithm to obtain a second credible feature set;

[0016] The second credible feature set is predicted by an online logistic regression classifier, and the second credible feature set is evaluated according to the prediction result to select highly credible feature information.

[0017] Preferably, the feature extraction of the code text library using the byte N-grams method includes:

[0018] The code data is segmented into bytes of a preset length according to the byte stream to obtain a hash dictionary of the code data;

[0019] Comparing the features of the preset sample with the hash dictionary to obtain a feature set corresponding to the hash dictionary;

[0020] If the feature in the hash dictionary appears in the preset sample, the feature value of the corresponding position in the hash dictionary is set to 1; if not, the feature value of the corresponding position in the hash dictionary is set to 0, thereby obtaining a sparse binary feature data set.

[0021] Preferably, the step of generating the first credible feature set by sorting the features according to the correlation between the extracted features and the preset high-credible feature categories includes:

[0022] Calculate the relative density of the extracted features and the preset high-confidence feature categories;

[0023] Determining the relevance of the extracted feature to a preset high-confidence feature category according to the relative density;

[0024] The features are sorted according to the relevance to generate a first credible feature set.

[0025] Preferably, the second credible feature set is obtained by deleting redundant features in the first credible feature set according to an approximate Markov blanket algorithm, including:

[0026] For the features in the first credible feature set, K features that are most relevant to the features are selected from the first credible feature set according to the correlation coefficient;

[0027] These K features are combined into a related set M i , and the related set M i As feature F i The approximate Markov blanket calculates the feature F i score;

[0028] The redundant features in the first credible feature set are deleted according to the score to obtain a second credible feature set.

[0029] The step of deleting redundant features in the first credible feature set according to the score to obtain the second credible feature set specifically includes:

[0030] Sort the features in the initial feature subset according to the scores, and delete the feature corresponding to the minimum score;

[0031] Each step is repeated to obtain a second credible feature set according to a preset number of features.

[0032] Preferably, predicting the second credible feature set by an online logistic regression classifier and evaluating the second credible feature set according to the prediction result to select highly credible feature information includes:

[0033] The online logistic regression classifier predicts the second credible feature set using a prediction function, where the prediction function is:

[0034]

[0035] Where w is the weight, b is the bias, x is the input, P(Y|x) is the prediction result and the range is [0, 1];

[0036] For the feature input in the second credible feature set, obtaining the prediction result of the prediction function; when the prediction result P>0.5, it is low-credible feature information, and when the prediction result P≤0.5, it is high-credible feature information;

[0037] The features in the second credible feature set are evaluated according to the prediction results to extract a predetermined number of highly credible feature information with the best prediction effect on the online logistic regression classifier.

[0038] The feature selection of code data greatly reduces the feature dimension, and adopts the online logistic regression model to remove a large number of irrelevant and redundant features in the code data to obtain high-confidence feature information, and uses the said high-confidence feature information for testing, which fundamentally improves the test accuracy and reduces the time consumed by the algorithm.

[0039] Preferably, the DBN network evaluation model is composed of multiple layers of RBM, and the last layer adopts BP neural network;

[0040] Training the DBN network evaluation model through the code text library and the high-confidence feature information;

[0041] The training of the DBN network evaluation model is divided into two steps: the first step is to train each layer of the RBM network layer by layer without supervision. After the current RBM layer is fully trained, the next layer of the RBM network is trained to ensure sufficient deep learning of the training data;

[0042] The second step is to set up a BP neural network in the last layer of the network. The BP neural network receives the output feature vector of the RBM as its input feature vector, and reversely adjusts the weight of the entire network through the feedback mechanism of the BP neural network to obtain a deep belief network.

[0043] Through the deep belief network, the software can be evaluated by inputting its feature vector into the trained DBN network evaluation model.

[0044] Preferably, a software evaluation system with high reliability features includes:

[0045] Acquisition module, used to acquire code data;

[0046] A preprocessing module, used for preprocessing the code data to obtain a code text library;

[0047] A feature extraction module is used to extract features from the code text library, identify and extract high-confidence feature information;

[0048] A model building module, used to build a DBN network evaluation model based on the code text library and the high-confidence feature information;

[0049] The result generation module is used to process the original code data according to the DBN network evaluation model, analyze and extract high-confidence features therein, and generate a software evaluation report.

[0050] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a software evaluation method and system with high-credibility features, including: obtaining code data; preprocessing the code data to obtain a code text library; extracting features from the code text library, identifying and extracting high-credibility feature information; constructing a DBN network evaluation model based on the code text library and the high-credibility feature information; processing the original code data according to the DBN network evaluation model, analyzing and extracting the high-credibility features therein, and generating a software evaluation report. The present invention greatly reduces the feature dimension by selecting features of code data, and uses an online logistic regression model to remove a large number of irrelevant and redundant features in the code data to obtain high-credibility feature information, and uses high-credibility feature information for testing, fundamentally improving the test accuracy and reducing the time consumed by the algorithm. The DBN network evaluation model can learn the abstract feature representation in the code data, which is helpful to identify and extract high-credibility feature information. The model can more accurately analyze the code data and find potential problems or advantages. The DBN network evaluation model can learn hierarchical feature representation, which is helpful to understand the complex structure and pattern of code data, improve the characterization ability of code data, and then extract more informative features. Improve the accuracy and depth of software evaluation. By using the trained high-confidence feature information, the generated software evaluation report is more convincing and practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0052] Figure 1 A flowchart of a software evaluation method with high-confidence features provided by the present invention.

[0053] Figure 2 A schematic diagram of the structure of a software evaluation system with high credibility provided by the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The embodiment of the present invention discloses a software evaluation method and system with high-confidence features. By selecting features of code data, the feature dimension is greatly reduced, and an online logistic regression model is used to remove a large number of irrelevant and redundant features in the code data to obtain high-confidence feature information. The high-confidence feature information is used for testing, which fundamentally improves the test accuracy and reduces the time consumed by the algorithm. The DBN network evaluation model can learn the abstract feature representation in the code data, which is helpful to identify and extract high-confidence feature information. The model can more accurately analyze the code data and find potential problems or advantages. The DBN network evaluation model can learn the hierarchical feature representation, which is helpful to understand the complex structure and pattern of the code data, improve the characterization ability of the code data, and then extract more informative features. Improve the accuracy and depth of software evaluation. By using the trained high-confidence feature information, the generated software evaluation report is more convincing and practical.

[0056] The embodiment of the present invention specifically discloses a software evaluation method with high reliability features, such as Figure 1 As shown, including:

[0057] Get code data;

[0058] Preprocessing the code data to obtain a code text library;

[0059] Performing feature extraction on the code text library to identify and extract high-confidence feature information;

[0060] Constructing a DBN network evaluation model based on the code text library and the high-confidence feature information;

[0061] The original code data is processed according to the DBN network evaluation model, high-confidence features therein are analyzed and extracted, and a software evaluation report is generated.

[0062] Specifically, the code data is obtained by using web crawler technology. Automatic acquisition of code data is achieved, which improves the efficiency and accuracy of data acquisition. The obtained code data is annotated and irrelevant information is removed, thereby improving the quality and availability of the code data. The preprocessed code data is annotated to provide a labeled data set for subsequent feature extraction and model training. By performing text feature extraction and preference feature extraction on the annotated data, high-confidence features in the code data are extracted, providing useful information for subsequent data analysis and model training.

[0063] Specifically, feature extraction is performed on the code text library to identify and extract high-confidence feature information, including:

[0064] Using byte N-grams method to extract features from the code text library;

[0065] Sorting the features according to the correlation between the extracted features and the preset high-credibility feature categories to generate a first credible feature set;

[0066] Deleting redundant features in the first credible feature set according to an approximate Markov blanket algorithm to obtain a second credible feature set;

[0067] The second credible feature set is predicted by an online logistic regression classifier, and the second credible feature set is evaluated according to the prediction result to select highly credible feature information.

[0068] Specifically, the byte N-grams method is used to extract features from the code text library, including:

[0069] The code data is segmented into bytes of a preset length according to the byte stream to obtain a hash dictionary of the code data;

[0070] Comparing the features of the preset sample with the hash dictionary to obtain a feature set corresponding to the hash dictionary;

[0071] If the feature in the hash dictionary appears in the preset sample, the feature value of the corresponding position in the hash dictionary is set to 1; if not, the feature value of the corresponding position in the hash dictionary is set to 0, thereby obtaining a sparse binary feature data set.

[0072] Specifically, the step of generating the first credible feature set by sorting the features according to the correlation between the extracted features and the preset high-credible feature categories includes:

[0073] Calculate the relative density of the extracted features and the preset high-confidence feature categories; the details are as follows:

[0074]

[0075] Among them, F is the feature set, F i is the i-th feature in the feature set, C is the preset high-confidence feature category set, C l is the lth category in the category set, Category C l The number of samples contained, L is the total number of categories, M is the total number of features, Indicates that when the i-th feature value is 1, relative to category C l The relative density of

[0076] Determining the relevance of the extracted feature to a preset high-confidence feature category according to the relative density;

[0077] The features are sorted according to the relevance to generate a first credible feature set.

[0078] The step of determining the relevance of the extracted features to the preset mail category according to the relative density specifically includes:

[0079] The correlation is calculated based on the relative density, and the formula is as follows:

[0080]

[0081] Among them, W(F i ) diff The range is [0, 1], It represents the relative density of category C1 when the i-th feature value is 1. represents the relative density of category C0 when the ith feature value is 1, and when W(F i ) diff =0, indicating feature F i Least relevant to the category, when W(F i ) diff =1, indicating feature F i Most relevant to the category;

[0082] W(F i ) diff As an evaluation criterion, the W(F i ) diff Compare with the preset threshold ω to determine the feature F i The correlation with the preset high-confidence feature category.

[0083] Specifically, the second credible feature set is obtained by deleting redundant features in the first credible feature set according to the approximate Markov blanket algorithm, including:

[0084] For the feature F in the first trusted feature set i According to the correlation coefficient, the feature F is selected from the first credible feature set. iThe most relevant K features; the calculation formula of the correlation coefficient is as follows:

[0085]

[0086] Among them, f i and c i They are feature F i and Category C i , n is the number of samples, and It is feature F i and Category C i The mean is calculated as follows:

[0087]

[0088] These K features are combined into a related set M i , and the related set M i As feature F i The approximate Markov blanket calculates the feature F i score; score δ G (F i |M i ) is calculated as follows:

[0089]

[0090] Among them, D KL Relative entropy is an indicator of the similarity between variables. The calculation formula is as follows:

[0091]

[0092] The redundant features in the first credible feature set are deleted according to the score to obtain a second credible feature set.

[0093] The step of deleting redundant features in the first credible feature set according to the score to obtain the second credible feature set specifically includes:

[0094] Sort the features in the initial feature subset according to the scores, and delete the feature corresponding to the minimum score;

[0095] Each step is repeated to obtain a second credible feature set according to a preset number of features.

[0096] Specifically, predicting the second credible feature set by an online logistic regression classifier and evaluating the second credible feature set according to the prediction result to select highly credible feature information includes:

[0097] The online logistic regression classifier predicts the second credible feature set using a prediction function, where the prediction function is:

[0098]

[0099] Where w is the weight, b is the bias, x is the input, P(Y|x) is the prediction result and the range is [0, 1];

[0100] For the feature input in the second credible feature set, obtaining the prediction result of the prediction function; when the prediction result P>0.5, it is low-credible feature information, and when the prediction result P≤0.5, it is high-credible feature information;

[0101] The features in the second credible feature set are evaluated according to the prediction results to extract a predetermined number of highly credible feature information with the best prediction effect on the online logistic regression classifier.

[0102] The embodiment of the present invention selects features of code data, greatly reduces the feature dimension, and adopts an online logistic regression model to remove a large number of irrelevant and redundant features in the code data to obtain high-confidence feature information, and uses the high-confidence feature information for testing, which fundamentally improves the test accuracy and reduces the time consumed by the algorithm.

[0103] Specifically, the DBN network evaluation model is composed of multiple layers of RBM, and the last layer adopts BP neural network;

[0104] Training the DBN network evaluation model through the code text library and the high-confidence feature information;

[0105] The training of the DBN network evaluation model is divided into two steps: the first step is to train each layer of the RBM network layer by layer without supervision. After the current RBM layer is fully trained, the next layer of the RBM network is trained to ensure sufficient deep learning of the training data;

[0106] The second step is to set up a BP neural network in the last layer of the network. The BP neural network receives the output feature vector of the RBM as its input feature vector, and reversely adjusts the weight of the entire network through the feedback mechanism of the BP neural network to obtain a deep belief network.

[0107] Through the deep belief network, the software can be evaluated by inputting its feature vector into the trained DBN network evaluation model.

[0108] Specifically, the original code data is processed according to the DBN network evaluation model, and high-reliability features therein are analyzed and extracted, and a software evaluation report is generated, including:

[0109] Prepare a software evaluation report based on the results of feature analysis. The report includes an assessment of software quality, performance, security, and other aspects, as well as suggestions and improvement measures for potential problems. Present the software evaluation report to relevant teams or decision makers for discussion and exchange, and jointly develop improvement plans and optimize software quality.

[0110] Specifically, in the embodiment of the present invention, the software may be a railway information system, specifically including:

[0111] Obtain code data, the code data includes: system operation data, user feedback and other information. Clean, convert and standardize the data for subsequent analysis and evaluation. Preprocess the code data to obtain a code text library. Perform feature extraction on the code text library, identify and extract high-trust feature information; the high-trust features include: system stability: including system availability, fault tolerance, etc.; performance efficiency: including system response time, processing throughput, etc.; security: including data protection, access control, etc.; user experience: including interface friendliness, ease of operation, etc. Construct a DBN network evaluation model based on the code text library and the high-trust feature information; train the model and perform performance evaluation to ensure that the model has high accuracy and reliability. Process the original code data according to the DBN network evaluation model, analyze and extract the high-trust features therein, and generate a software evaluation report. Run the evaluation model to evaluate the software quality and performance of the railway information system and obtain the evaluation results. Generate customizable evaluation reports and indicators and provide them to relevant stakeholders. Regularly evaluate the railway information system, optimize the evaluation model and system functions based on the evaluation results, and ensure that the system continues to meet high quality and performance requirements.

[0112] The embodiment of the present invention realizes a high-reliability feature software evaluation method for a railway information system, helping railway companies to monitor and optimize system operating status and improve service quality and efficiency.

[0113] The embodiment of the present invention also discloses a software evaluation system with high reliability features, such as Figure 2 As shown, including:

[0114] Acquisition module, used to acquire code data;

[0115] A preprocessing module, used for preprocessing the code data to obtain a code text library;

[0116] A feature extraction module is used to extract features from the code text library, identify and extract high-confidence feature information;

[0117] A model building module, used to build a DBN network evaluation model based on the code text library and the high-confidence feature information;

[0118] The result generation module is used to process the original code data according to the DBN network evaluation model, analyze and extract high-confidence features therein, and generate a software evaluation report.

[0119] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0120] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A software evaluation method with high-confidence features, characterized in that: include: Get code data; Preprocessing the code data to obtain a code text library; Performing feature extraction on the code text library to identify and extract high-confidence feature information; Constructing a DBN network evaluation model based on the code text library and the high-confidence feature information; The original code data is processed according to the DBN network evaluation model, high-confidence features therein are analyzed and extracted, and a software evaluation report is generated.

2. According to the software evaluation method with high reliability features described in claim 1, it is characterized in that: The code data is obtained by using a web crawler technology.

3. The software evaluation method with high reliability features according to claim 1 is characterized in that: Perform feature extraction on the code text library to identify and extract high-confidence feature information, including: Using byte N-grams method to extract features from the code text library; Sorting the features according to the correlation between the extracted features and the preset high-credibility feature categories to generate a first credible feature set; Deleting redundant features in the first credible feature set according to an approximate Markov blanket algorithm to obtain a second credible feature set; The second credible feature set is predicted by an online logistic regression classifier, and the second credible feature set is evaluated according to the prediction result to select highly credible feature information.

4. A software evaluation method with high credibility features according to claim 3, characterized in that: The method of using byte N-grams to extract features from the code text library includes: The code data is segmented into bytes of a preset length according to the byte stream to obtain a hash dictionary of the code data; Comparing the features of the preset sample with the hash dictionary to obtain a feature set corresponding to the hash dictionary; If the feature in the hash dictionary appears in the preset sample, the feature value of the corresponding position in the hash dictionary is set to 1; if not, the feature value of the corresponding position in the hash dictionary is set to 0, thereby obtaining a sparse binary feature data set.

5. The software evaluation method with high reliability features according to claim 3 is characterized in that: The step of generating a first credible feature set by sorting the features according to the correlation between the extracted features and the preset high-credible feature categories includes: Calculate the relative density of the extracted features and the preset high-confidence feature categories; Determining the relevance of the extracted feature to a preset high-confidence feature category according to the relative density; The features are sorted according to the relevance to generate a first credible feature set.

6. A software evaluation method with high credibility features according to claim 3, characterized in that: The second credible feature set is obtained by deleting redundant features in the first credible feature set according to the approximate Markov blanket algorithm, including: For the features in the first credible feature set, K features that are most relevant to the features are selected from the first credible feature set according to the correlation coefficient; These K features are combined into a related set M i , and the related set M i As feature F i The approximate Markov blanket calculates the feature F i score; The redundant features in the first credible feature set are deleted according to the score to obtain a second credible feature set.

7. A software evaluation method with high reliability features according to claim 3, characterized in that: Predicting the second credible feature set by an online logistic regression classifier and evaluating the second credible feature set according to the prediction result to select highly credible feature information, including: The online logistic regression classifier predicts the second credible feature set using a prediction function, where the prediction function is: Where w is the weight, b is the bias, x is the input, P(Y|x) is the prediction result and the range is [0, 1]; For the feature input in the second credible feature set, obtaining a prediction result of the prediction function; The features in the second credible feature set are evaluated according to the prediction results to extract a predetermined number of highly credible feature information with the best prediction effect on the online logistic regression classifier.

8. A software evaluation method with high reliability features according to claim 1, characterized in that: The DBN network evaluation model is composed of multiple layers of RBM, and the last layer adopts BP neural network; Training the DBN network evaluation model through the code text library and the high-confidence feature information; The training of the DBN network evaluation model is divided into two steps: the first step is to train each layer of the RBM network layer by layer without supervision. After the current RBM layer is fully trained, the next layer of the RBM network is trained to ensure sufficient deep learning of the training data; The second step is to set up a BP neural network in the last layer of the network. The BP neural network receives the output feature vector of the RBM as its input feature vector, and reversely adjusts the weight of the entire network through the feedback mechanism of the BP neural network to obtain a deep belief network. Through the deep belief network, the feature vector of the software is input into the trained DBN network evaluation model to evaluate it.

9. A software evaluation system with high credibility features, using a software evaluation method with high credibility features as claimed in any one of claims 1 to 8, characterized in that: include: Acquisition module, used to obtain code data; A preprocessing module, used for preprocessing the code data to obtain a code text library; A feature extraction module is used to extract features from the code text library, identify and extract high-confidence feature information; A model building module, used to build a DBN network evaluation model based on the code text library and the high-confidence feature information; The result generation module is used to process the original code data according to the DBN network evaluation model, analyze and extract high-confidence features therein, and generate a software evaluation report.

Images