Label distribution learning model generation method and device based on feature selection

Abstract

The invention relates to a label distribution learning model generation method and device based on feature selection. The method comprises the following steps: dividing a data sample set into a training set and a test set, wherein the data sample set comprises K features and T tags; initializing the selected feature set to be null and the candidate feature set to be K features in the data sample; generating a fuzzy equivalence relationship among the data samples in the training set through a Gaussian kernel function; based on the fuzzy equivalence relationship, determining a target feature from the candidate feature set by using a label distribution feature selection method of fuzzy weight redundancy, and adding the target feature into a selected feature set; repeatedly determining a target feature from the candidate feature set, and adding the target feature into the selected feature set until the number of features in the selected feature set meets a threshold value; and training a machine learning model through the selected feature set to generate a label distribution learning model. According to the invention, the performance of the label distribution learner can be obviously improved.

Description

Technical Field

[0001] This application relates to the field of computer information processing. Specifically, it relates to a method and device for generating a label distribution learning model based on feature selection. Background Art

[0002] Single-label learning is a learning method that focuses on whether a label accurately describes the current instance, but it usually ignores the relative degree of description of the label for the instance, that is, the relative importance of each label. In the context of network security, this problem is particularly important. For example, in DDoS (Distributed Denial of Service) attack detection, a network traffic data may involve multiple attack labels, and these attacks have different threat levels and techniques. Traditional single-label learning methods have limitations in describing labels and cannot effectively handle multi-label scenarios. Although multi-label learning has been widely applied, it still faces challenges in dealing with label ambiguity problems in network security. For example, a network traffic event may simultaneously contain multiple DDoS attack features, and it is necessary to understand the relative importance of these features to cope with potential threats. For this reason, Label Distribution Learning (LDL) has gradually become a research hotspot and can handle learning problems with label ambiguity.

[0003] With the extensive research on label distribution learning tasks, label distribution feature selection algorithms have gradually become an important research direction. These algorithms aim to improve the performance of the learner through dimensionality reduction and provide more effective judgments for fields such as biological experiments, network security protection, and emotion prediction. Similar to traditional supervised learning, label distribution learning also faces the challenge of high-dimensional data, which contains a large number of irrelevant and redundant features. These features may not only reduce the accuracy of the learner but also waste time and space resources. Therefore, label distribution feature selection becomes crucial, which can help accurately select relevant features in high-dimensional data, improve the recognition accuracy of network security threats such as DDoS attacks, and provide more targeted protection strategies.

[0004] Existing feature selection methods mainly focus on single-label and multi-label learning, and there are few feature selection methods for continuous label distribution data. In the prior art, the label distribution feature selection method based on the positive domain and boundary domain combines the fuzzy rough set model, but there are still deficiencies. First, traditional feature selection methods cannot directly handle continuous label distribution data because the labels change from a discrete set to a continuous distribution. Second, the feature selection method combined with mutual information has a relatively broad determination of redundancy and cannot describe the dynamic change of redundancy between features under the influence of labels.

[0005] Therefore, a new method and device for generating a label distribution learning model based on feature selection are needed.

[0006] The above information disclosed in the background section is only used to enhance the understanding of the background of the present application. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] In view of this, the present application provides a method and device for generating a label distribution learning model based on feature selection, which can effectively process continuous label distribution data, dynamically describe the redundancy between features, and significantly improve the performance of the label distribution learner. It is applicable to multiple application scenarios and has broad application prospects and practical value.

[0008] Other features and advantages of the present application will become apparent through the following detailed description, or will be learned in part through the practice of the present application.

[0009] According to one aspect of the present application, a method for generating a label distribution learning model based on feature selection is proposed. The method includes: dividing a data sample set into a training set and a test set, where the data sample set contains K features and T labels; initializing the selected feature set to be empty and the candidate feature set to be the K features in the data samples; generating a fuzzy equivalence relationship between the data samples in the training set through a Gaussian kernel function; based on the fuzzy equivalence relationship, using a fuzzy weight redundancy-based label distribution feature selection method to determine target features from the candidate feature set and adding the target features to the selected feature set; repeating the process of determining target features from the candidate feature set and adding the target features to the selected feature set until the number of features in the selected feature set meets a threshold; training a machine learning model through the selected feature set to generate a label distribution learning model.

[0010] In an exemplary embodiment of the present application, it further includes: inputting the data samples in the test set into the label distribution learning model to generate label prediction results; evaluating the label distribution learning model through the label prediction results.

[0011] In an exemplary embodiment of the present application, generating a fuzzy equivalence relationship between the data samples in the training set through a Gaussian kernel function includes: calculating the fuzzy equivalence relationship between each two data samples in the training set through a Gaussian kernel function.

[0012] In an exemplary embodiment of the present application, calculating the fuzzy equivalence relationship between each two data samples in the training set through a Gaussian kernel function includes:

[0013]

[0014] where xi and xj respectively represent the i-th data sample and the j-th data sample, rij is the fuzzy equivalence relationship between the data samples xi and xj, and σ represents the variance between xi and xj.

[0015] In an exemplary embodiment of the present application, based on the fuzzy equivalence relation, the target feature is determined from the candidate feature set by using the label distribution feature selection method with fuzzy weight redundancy, and the target feature is added to the selected feature set, including: based on the fuzzy equivalence relation, calculating the scores of all features in the candidate feature set by using the label distribution feature selection method with fuzzy weight redundancy; taking the feature corresponding to the highest score as the target feature; adding the target feature to the selected feature set.

[0016] In an exemplary embodiment of the present application, calculating the scores of all features in the candidate feature set by using the label distribution feature selection method with fuzzy weight redundancy based on the fuzzy equivalence relation, including: generating a label distribution feature selection method based on the fuzzy equivalence relation; calculating the scores of all features in the candidate feature set by using the label distribution feature selection method with fuzzy weight redundancy.

[0017] In an exemplary embodiment of the present application, generating a label distribution feature selection method based on the fuzzy equivalence relation, including: determining the correlation between the candidate feature and the label based on the fuzzy equivalence relation; determining the redundancy between the candidate feature and the selected feature based on the fuzzy equivalence relation; generating an interactive redundancy weight factor; generating a label distribution feature selection method through the correlation between the candidate feature and the label, the redundancy between the candidate feature and the selected feature, and the interactive redundancy weight factor.

[0018] In an exemplary embodiment of the present application, generating a label distribution feature selection method through the correlation between the candidate feature and the label, the redundancy between the candidate feature and the selected feature, and the interactive redundancy weight factor, including:

[0019]

[0020] where f k represents the candidate feature, f s represents the selected feature in the selected feature set S, y i represents a label in the label distribution data, J(f k ) represents the candidate feature score, (f k ; y i ) represents the correlation between the candidate feature f k and the label y i , I(f k ; f s ) represents the redundancy between f k and f s , ω(f k ) represents the interactive redundancy weight factor, and ω(f k ) is initialized to 1.

[0021] In an exemplary embodiment of the present application, repeating to determine target features from a candidate feature set and adding the target features to a selected feature set until the number of features in the selected feature set meets a threshold includes: removing the target features of the previous round from the candidate feature set to generate an updated candidate feature set; calculating scores of all features in the updated candidate feature set by using a label distribution feature selection method with fuzzy weight redundancy based on the fuzzy equivalence relation matrix; taking the feature corresponding to the highest score as the target feature of this round; and adding the target feature of this round to the selected feature set until the number of features in the selected feature set meets the threshold.

[0022] According to an aspect of the present application, there is provided a device for generating a label distribution learning model based on feature selection. The device includes: a data module for dividing a data sample set into a training set and a test set, where the data sample set includes K features and T labels; a set module for initializing the selected feature set to be empty and the candidate feature set to be the K features in the data samples; a matrix module for generating a fuzzy equivalence relation between data samples in the training set through a Gaussian kernel function; a target module for determining target features from the candidate feature set based on the fuzzy equivalence relation by using a label distribution feature selection method with fuzzy weight redundancy and adding the target features to the selected feature set; a screening module for repeating to determine target features from the candidate feature set and adding the target features to the selected feature set until the number of features in the selected feature set meets the threshold; and a training module for training a machine learning model through the selected feature set to generate a label distribution learning model.

[0023] According to an aspect of the present application, there is provided an electronic device, which includes: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0024] According to an aspect of the present application, there is provided a computer-readable medium, on which a computer program is stored, and when the program is executed by a processor, it implements the method as described above.

[0025] The method and device for generating a label distribution learning model based on feature selection according to the present application divide a data sample set into a training set and a test set. The data sample set contains K features and T labels; initialize the selected feature set to be empty and the candidate feature set to be the K features in the data sample; generate a fuzzy equivalence relationship between the data samples in the training set through a Gaussian kernel function; based on the fuzzy equivalence relationship, use the label distribution feature selection method with fuzzy weight redundancy to determine target features from the candidate feature set, and add the target features to the selected feature set; repeat determining target features from the candidate feature set and adding the target features to the selected feature set until the number of features in the selected feature set meets the threshold; by training a machine learning model through the selected feature set to generate a label distribution learning model, it can effectively process continuous label distribution data, dynamically describe the redundancy between features, and significantly improve the performance of the label distribution learner. It is applicable to multiple application scenarios and has broad application prospects and practical value.

[0026] It should be understood that the above general description and the following detailed description are only exemplary and do not limit the present application. Brief Description of the Drawings

[0027] By referring to the accompanying drawings and describing its exemplary embodiments in detail, the above and other objectives, features, and advantages of the present application will become more apparent. The following described drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0028] Figure 1 It is a flowchart of a method for generating a label distribution learning model based on feature selection shown according to an exemplary embodiment.

[0029] Figure 2 It is a flowchart of a method for generating a label distribution learning model based on feature selection shown according to an exemplary embodiment.

[0030] Figure 3 It is a flowchart of a method for generating a label distribution learning model based on feature selection shown according to another exemplary embodiment.

[0031] Figure 4 It is a block diagram of a device for generating a label distribution learning model based on feature selection shown according to an exemplary embodiment.

[0032] Figure 5 It is a block diagram of an electronic device shown according to an exemplary embodiment.

[0033] Figure 6A block diagram of a computer-readable medium is shown according to an exemplary embodiment. Detailed implementation manners

[0034] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. Like reference numerals in the figures denote like or similar parts, and thus their repetitive description will be omitted.

[0035] In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of this application. However, those skilled in the art will realize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be used. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this application.

[0036] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0037] The flowcharts shown in the drawings are only illustrative and not necessarily include all the contents and operations / steps, nor are they necessarily executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to the actual situation.

[0038] It should be understood that although terms such as first, second, and third may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the concept of this application. As used herein, the term "and / or" includes any one and all combinations of one or more of the associated listed items.

[0039] Those skilled in the art can understand that the drawings are only schematic diagrams of the example embodiments, and the modules or processes in the drawings are not necessarily essential for implementing this application, so they cannot be used to limit the protection scope of this application.

[0040] The technical abbreviations involved in this application are explained as follows:

[0041] LDL: Label Distribution Learning, which is used to process data sets with multiple descriptive labels, where each label has a different degree of description and describes an instance in the form of a distribution.

[0042] FS: Feature Selection, which screens feature subsets through feature evaluation metrics to obtain some of the most effective features to reduce the dimensionality of the data set and improve learning performance.

[0043] SA - BFGS: A label distribution learning model that can make predictions through the input training set, and obtain the performance evaluation result by comparing the prediction result with the test set.

[0044] LBP: Local Binary Patterns, which uses the central pixel value as a threshold, compares it with the gray values of the surrounding pixels, and describes the local texture features of the image by statistically analyzing the frequency histogram of the results.

[0045] Figure 1 It is a flowchart of a method for generating a label distribution learning model based on feature selection shown according to an exemplary embodiment. The method 10 for generating a label distribution learning model based on feature selection includes at least steps S102 to S112.

[0046] As Figure 1 shown, in S102, the data sample set is divided into a training set and a test set, and the data sample set contains K features and T labels.

[0047] In a specific embodiment, a data sample set U = {x 1 , x 2 , …, x n} can be set, and the set contains k features and t labels. In addition, the dimension K of the label distribution feature subset is also defined.

[0048] For example, the data sample set U can also be divided into a training set Train and a test set Test according to a ratio of 9:1 to ensure that the two parts of the data are both representative and non - overlapping.

[0049] In S104, the selected feature set is initialized to be empty, and the candidate feature set is the K features in the data sample. For example, the selected feature set S can be set to be empty, while the candidate feature set covers all k features in the training set Train.

[0050] In S106, a fuzzy equivalence relation between data samples in the training set is generated through a Gaussian kernel function. The fuzzy equivalence relation R can be used to process continuous label distribution data; the fuzzy equivalence relation R represents the similarity between different instances under the same feature.

[0051] More specifically, the fuzzy equivalence relation between every two data samples in the training set can be calculated through a Gaussian kernel function:

[0052]

[0053] where xi and xj represent the i-th data sample and the j-th data sample respectively, rij is the fuzzy equivalence relation between data samples xi and xj, and σ represents the variance between xi and xj.

[0054] In S108, based on the fuzzy equivalence relation, the target feature is determined from the candidate feature set by using the label distribution feature selection method with fuzzy weight redundancy, and the target feature is added to the selected feature set.

[0055] In one embodiment, based on the fuzzy equivalence relation, the scores of all features in the candidate feature set can be calculated by using the label distribution feature selection method with fuzzy weight redundancy; the feature corresponding to the highest score is used as the target feature; the target feature is added to the selected feature set.

[0056] More specifically, a label distribution feature selection method can be generated based on the fuzzy equivalence relation; the scores of all features in the candidate feature set are calculated by using the label distribution feature selection method with fuzzy weight redundancy. Construct a label distribution feature selection algorithm J(f k ) - based on the label distribution feature selection method with fuzzy weight redundancy Fuzzy Weight Redundant Label Distribution Feature Selection (FWRLDFS).

[0057] In one embodiment, the correlation between the candidate feature and the label can be determined based on the fuzzy equivalence relation; the redundancy between the candidate feature and the selected feature can be determined based on the fuzzy equivalence relation; an interactive redundancy weight factor is generated; a label distribution feature selection method is generated through the correlation between the candidate feature and the label, the redundancy between the candidate feature and the selected feature, and the interactive redundancy weight factor.

[0058]

[0059] where f k represents the candidate feature, f s represents the selected feature in the selected feature set S, and y iRepresents a label in the label distribution data, J(f k ) represents the candidate feature score, (f k ; y i ) represents the correlation between the candidate feature f k and the label y i ; I(f k ; f s ) represents the redundancy between f k and f s ; ω(f k ) represents the interactive redundancy weight factor, ω(f k ) is initialized to 1.

[0060] The steps of the label distribution feature selection method J(f k ) are calculated as follows:

[0061] Definition 1 Fuzzy equivalence class: The fuzzy equivalence class consists of the fuzzy equivalence relations in the label distribution dataset and is defined as follows:

[0062]

[0063] where r i1 represents the fuzzy equivalence relation between the i-th sample and the first sample, and n represents the total number of samples in the label distribution dataset.

[0064] Definition 2 Fuzzy information entropy: Fuzzy information entropy is used to measure the degree of chaos of the fuzzy equivalence relation R corresponding to continuous data and is defined as follows:

[0065]

[0066] where log is the logarithmic function, generally with base 2.

[0067] Definition 3 Fuzzy mutual information: Fuzzy mutual information is used to describe the degree of correlation between features and labels in the label distribution data and is defined as follows:

[0068]

[0069] where the fuzzy conditional entropy H(R A |R B ) is calculated by

[0070] Definition 4 Fuzzy interaction information: Fuzzy interaction information can be used to describe the degree of common information between the fuzzy equivalence relations R A , R B and R C and is defined as follows:

[0071] FI(R A ; RB ; R C ) = FI(R A , R C ) - FI(R A ; R C |R B )

[0072] In summary, each feature in the label distribution dataset can correspond to a fuzzy equivalence relation, and then the relationships among candidate features, selected features, and the label set can be constructed. The specific formula of the label distribution feature selection algorithm J(f k ) in Step 4 is given as follows:

[0073]

[0074] Among them, the interactive redundancy weight factor is initialized to 1.

[0075] Initialize the interactive redundancy weight, and then evaluate all candidate features using the label distribution feature selection algorithm, select the feature with the highest correlation with the label, add its index to the selected feature set S, and remove it from the training set.

[0076] In S110, repeatedly determine the target feature from the candidate feature set and add the target feature to the selected feature set until the number of features in the selected feature set meets the threshold. More specifically, the threshold can be K.

[0077] The target feature of the previous round can be removed from the candidate feature set to generate an updated candidate feature set; based on the fuzzy equivalence relation matrix, calculate the scores of all features in the updated candidate feature set using the label distribution feature selection method with fuzzy weight redundancy; take the feature corresponding to the highest score as the target feature of this round; add the target feature of this round to the selected feature set until the number of features in the selected feature set meets the threshold.

[0078] Use the label distribution feature selection method with fuzzy weight redundancy to evaluate the label distribution training set after removing the feature with the largest correlation, add the feature with the highest evaluation score among the remaining features to the selected feature set S. If the number of elements in the selected feature set S is equal to the dimension K of the initially specified feature subset, stop; otherwise, keep repeating the extraction.

[0079] In S112, train the machine learning model through the selected feature set to generate a label distribution learning model. Input the constructed feature subset into the label distribution learner SA - BFGS model, train the model from the feature subset, put the test set into the trained learner model, and obtain the final prediction result.

[0080] The method for generating a label distribution learning model based on feature selection according to the present application divides a data sample set into a training set and a test set. The data sample set contains K features and T labels. Initialize the selected feature set as empty and the candidate feature set as the K features in the data samples. Generate a fuzzy equivalence relationship between the data samples in the training set through a Gaussian kernel function. Based on the fuzzy equivalence relationship, use the label distribution feature selection method with fuzzy weight redundancy to determine the target features from the candidate feature set, and add the target features to the selected feature set. Repeat the process of determining the target features from the candidate feature set and adding the target features to the selected feature set until the number of features in the selected feature set meets the threshold. By training a machine learning model with the selected feature set to generate a label distribution learning model, it can effectively process continuous label distribution data, dynamically describe the redundancy between features, and significantly improve the performance of the label distribution learner. It is applicable to multiple application scenarios and has broad application prospects and practical value.

[0081] It should be clearly understood that this application describes how to form and use specific examples, but the principles of this application are not limited to any details of these examples. Instead, based on the teachings of the content disclosed in this application, these principles can be applied to many other embodiments.

[0082] Figure 2 It is a flowchart of a method for generating a label distribution learning model based on feature selection shown according to an exemplary embodiment. Figure 2 The process 20 shown is for Figure 1 a detailed description of S110 in the process shown, which is "Repeat the process of determining the target features from the candidate feature set and adding the target features to the selected feature set until the number of features in the selected feature set meets the threshold".

[0083] As Figure 2 shown, in S202, remove the target features of the previous round from the candidate feature set to generate an updated candidate feature set.

[0084] More specifically, remove the target features selected in the previous round from the candidate feature set to generate an updated candidate feature set: the updated candidate feature set contains all the remaining candidate features.

[0085] In S204, based on the fuzzy equivalence relationship matrix, use the label distribution feature selection method with fuzzy weight redundancy to calculate the scores of all features in the updated candidate feature set.

[0086] In S206, take the feature corresponding to the highest score as the target feature of this round. Sort all the features in the updated candidate feature set according to the scores. Select the feature with the highest score as the target feature of this round.

[0087] In S208, add the target feature of this round to the selected feature set.

[0088] In S210, determine whether the number of features in the selected feature set meets the threshold.

[0089] In S212, if it is satisfied, stop the calculation; if not, calculate again.

[0090] The technical effects of this application are as follows:

[0091] This application uses fuzzy mutual information and fuzzy conditional mutual information to replace the traditional mutual information, effectively overcoming the limitations of the traditional mutual information in dealing with continuous label distribution data;

[0092] This application uses fuzzy interaction information as the redundancy weight factor to deeply explore the dynamic changes of redundancy between features under the influence of labels, providing a more applicable special method for feature selection in label distribution learning;

[0093] The larger the interaction redundancy weight factor of this application indicates that the redundancy between features is enhanced under the influence of labels, and the smaller the interaction redundancy weight factor indicates that the learning ability of the learner can be improved when two features appear simultaneously; the correlation and redundancy between features and labels are defined through fuzzy mutual information, and a label distribution feature selection algorithm named FWRLDFS is defined to effectively select feature subsets and improve the performance of the label distribution learner.

[0094] Figure 3 It is a diagram showing a method for generating a label distribution learning model based on feature selection according to another exemplary embodiment. Figure 3 The shown process 30 is a supplementary description of Figure 1 the shown process.

[0095] As Figure 3 shown, in S302, preprocess the original data set.

[0096] In S304, divide it into a training set and a test set.

[0097] In S306, initialize the weights.

[0098] In S308, screen features according to the FWRLDFS method.

[0099] In S310, screen according to the obtained feature subset.

[0100] In S312, train and verify on the feature distribution learner.

[0101] In S314, obtain the trained model.

[0102] In a specific application, observing the label distribution datasets in the real world, the Yeast_spoem, Yeast_spo, Yeast_cold, and Yeast_heat datasets are from biological experiments on the budding yeast Saccharomyces cerevisiae. Each dataset contains 2465 yeast genes, and each gene is represented by a phylogenetic profile vector of length 24. Among them, the labels correspond to discrete time points during the experiment, reflecting the dynamic changes in gene expression levels over time. After normalization, the gene expression levels can be used as a natural metric to describe the degree of gene description under specific conditions. The SJAFFE dataset contains 213 grayscale facial expression images, all of which are presented by 10 Japanese female models. To accurately extract the key features of each image, the Local Binary Pattern (LBP) method is adopted to convert each image into a 243-dimensional feature vector. Among them, the average score of each emotion (happiness, sadness, surprise, fear, anger, and disgust) is used to represent the intensity of that emotion in the image. Table 1 summarizes the label distribution datasets used in the experiments to verify the present invention.

[0103] Table 1 Details description of experimental datasets

[0104] Serial number Data set Number of instances Number of features Number of labels 1 Yeast_spoem 2465 24 2 2 Yeast_spo 2465 24 3 3 Yeast_cold 2465 24 4 4 Yeast_heat 2465 24 6 5 SJAFFE 213 243 6

[0105] Steps of the label distribution feature selection process diagram according to the technology of the present application. The dimension K of the input feature subset is 75% of the total number of dataset features. Then, a feature subset S is created based on the selected feature set, and the SA-BFGS learner model is trained by the feature subset to obtain the model SA-BFGS_FWR.

[0106] First, set the dimension K of the feature subset, with a value of 75% of the total number of dataset features, to ensure that the selected features can fully represent the original data. Subsequently, the FWRLDFS algorithm is used to screen the features. This algorithm can accurately evaluate the importance of each feature and select the most representative features to form a subset based on the evaluation results. Next, the SA-BFGS learner model is trained using the selected feature subset to obtain the optimized SA-BFGS_FWR model. Finally, the performance of the model is evaluated through the test set to ensure its better accuracy and stability in practical applications.

[0107] To verify the effect of the technology of this application, four distance measures, namely Canberra distance, Euclidean distance, Clark distance, and Wave-Hedges distance, are used as the criteria for evaluating the label distribution learning model. The comparative experiment is to directly train the label distribution learner SA-BFGS model using the training set Train, and directly obtain the model SA-BFGS_Train. Then, the test set Test is put into the trained model, and finally, the four distance evaluation indicators of the SA-BFGS_Train model are obtained. The above data are aggregated into a table as follows:

[0108] Table 2 Results of Feature Subset S and Original Feature Dataset Yeast_spoem under Four Distance Measures

[0109] Label distribution learner Clark Canberra Euclidean Wave-Hedges SA-BFGS_FWR 0.1297 0.1790 0.1232 0.8693 SA-BFGS_Train 0.1299 0.1794 0.1234 0.8695

[0110] Table 3 Results of Feature Subset S and Original Feature Dataset Yeast_spo under Four Distance Measures

[0111] Label distribution learner Clark Canberra Euclidean Wave-Hedges SA-BFGS_FWR 0.2503 0.5139 0.0821 0.8990 SA-BFGS_Train 0.2505 0.5142 0.0822 0.8994

[0112] Table 4 Results of Feature Subset S and Original Feature Dataset Yeast_cold under Four Distance Measures

[0113] Label distribution learner Clark Canberra Euclidean Wave-Hedges SA-BFGS_FWR 0.1393 0.2399 0.0681 0.4354 SA-BFGS_Train 0.1395 0.2401 0.0682 0.4356

[0114] Table 5 Results of Feature Subset S and Original Feature Dataset Yeast_heat under Four Distance Measures

[0115] Label distribution learner Clark Canberra Euclidean Wave-Hedges SA-BFGS_FWR 0.1825 0.3638 0.0591 0.6582 SA-BFGS_Train 0.1827 0.3643 0.0593 0.6591

[0116] Table 6 Results of Feature Subset S and Original Feature Dataset SJAFFE under Four Distance Measures

[0117]

[0118]

[0119] Tables 2 to 6 show the results of comparative experiments on five different datasets using four distance measure indicators. These distance measure indicators are used to measure the closeness between the predicted value and the actual value, and the smaller the value, the more accurate the prediction result. The experimental data fully prove the effectiveness of the FWRLDFS label distribution feature selection algorithm in this application. This algorithm can significantly improve the performance of the label distribution learning model and provide strong support for research and applications such as network security protection, biological experiments, and emotion prediction.

[0120] Those skilled in the art can understand that all or part of the steps to implement the above embodiments are realized as a computer program executed by a CPU. When the computer program is executed by the CPU, the above functions defined by the above method provided by this application are executed. The program can be stored in a computer-readable storage medium, which can be a read-only memory, a magnetic disk or an optical disc, etc.

[0121] In addition, it should be noted that the above drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of this application, rather than for limiting purposes. It is easy to understand that the processes shown in the above drawings do not indicate or limit the chronological order of these processes. Additionally, it is also easy to understand that these processes can be executed synchronously or asynchronously in, for example, multiple modules.

[0122] The following is an embodiment of the device of this application, which can be used to execute the method embodiment of this application. For details not disclosed in the device embodiment of this application, please refer to the method embodiment of this application.

[0123] Figure 4 is a block diagram of a device for generating a label distribution learning model based on feature selection shown according to an exemplary embodiment. As Figure 4 shown, the device 40 for generating a label distribution learning model based on feature selection includes: a data module 402, a set module 404, a matrix module 406, a target module 408, a screening module 410, and a training module 412.

[0124] The data module 402 is used to divide the data sample set into a training set and a test set, and the data sample set contains K features and T labels.

[0125] The set module 404 is used to initialize the selected feature set as empty and the candidate feature set as the K features in the data samples.

[0126] The matrix module 406 is used to generate a fuzzy equivalence relationship between data samples in the training set through a Gaussian kernel function; the matrix module 406 is also used to calculate the fuzzy equivalence relationship between every two data samples in the training set through a Gaussian kernel function.

[0127] The target module 408 is used to determine target features from the candidate feature set based on the fuzzy equivalence relationship by using a label distribution feature selection method with fuzzy weight redundancy, and add the target features to the selected feature set; the target module 408 is also used to calculate the scores of all features in the candidate feature set based on the fuzzy equivalence relationship by using a label distribution feature selection method with fuzzy weight redundancy; take the feature corresponding to the highest score as the target feature; and add the target features to the selected feature set.

[0128] The screening module 410 is used to repeatedly determine target features from the candidate feature set and add the target features to the selected feature set until the number of features in the selected feature set meets the threshold; the screening module 410 is further used to remove the target features of the previous round from the candidate feature set to generate an updated candidate feature set; based on the fuzzy equivalence relation matrix, calculate the scores of all features in the updated candidate feature set by using the label distribution feature selection method with fuzzy weight redundancy; take the feature corresponding to the highest score as the target feature of this round; add the target feature of this round to the selected feature set until the number of features in the selected feature set meets the threshold.

[0129] The training module 412 is used to train a machine learning model through the selected feature set to generate a label distribution learning model.

[0130] According to the label distribution learning model generation device based on feature selection of the present application, by dividing the data sample set into a training set and a test set, the data sample set contains K features and T labels; initialize the selected feature set to be empty and the candidate feature set to be the K features in the data sample; generate the fuzzy equivalence relation between the data samples in the training set through the Gaussian kernel function; based on the fuzzy equivalence relation, use the label distribution feature selection method with fuzzy weight redundancy to determine the target features from the candidate feature set and add the target features to the selected feature set; repeatedly determine the target features from the candidate feature set and add the target features to the selected feature set until the number of features in the selected feature set meets the threshold; by training the machine learning model through the selected feature set to generate the label distribution learning model, it can effectively process continuous label distribution data, dynamically describe the redundancy between features, and significantly improve the performance of the label distribution learner. It is applicable to multiple application scenarios and has broad application prospects and practical value.

[0131] Figure 5 It is a block diagram of an electronic device shown according to an exemplary embodiment.

[0132] Next, refer to Figure 5 to describe the electronic device 500 according to this embodiment of the present application. Figure 5 The shown electronic device 500 is only an example and should not bring any limitations to the functions and usage scopes of the embodiments of the present application.

[0133] As Figure 5 shown, the electronic device 500 is presented in the form of a general computing device. The components of the electronic device 500 may include but are not limited to: at least one processing unit 510, at least one storage unit 520, a bus 530 connecting different system components (including the storage unit 520 and the processing unit 510), a display unit 540, etc.

[0134] Among them, the storage unit stores program code, and the program code can be executed by the processing unit 510, so that the processing unit 510 executes the steps according to various exemplary embodiments of the present application described in this specification. For example, the processing unit 510 can execute as Figure 1 , Figure 2 , Figure 3 the steps shown in.

[0135] The storage unit 520 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 5201 and / or a cache storage unit 5202, and may further include a read-only storage unit (ROM) 5203.

[0136] The storage unit 520 may further include a program / utility 5204 having a set (at least one) of program modules 5205. Such program modules 5205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0137] The bus 530 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0138] The electronic device 500 may also communicate with one or more external devices 500' (such as a keyboard, a pointing device, a Bluetooth device, etc.), so that the device that enables a user to interact with the electronic device 500 communicates, and / or the electronic device 500 can communicate with any device that can communicate with one or more other computing devices (such as a router, a modem, etc.). Such communication can be carried out through the input / output (I / O) interface 550. And, the electronic device 500 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 560. The network adapter 560 can communicate with other modules of the electronic device 500 through the bus 530. It should be understood that although not shown in the figure, other hardware and / or software modules can be used in combination with the electronic device 500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0139] Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described herein can be implemented by software, or can be implemented by a manner of software combined with necessary hardware. Therefore, as Figure 6As shown, the technical solution according to the embodiments of the present application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, server, or network device, etc.) to execute the above-mentioned method according to the embodiments of the present application.

[0140] The software product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0141] The computer-readable storage medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than the readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted by any appropriate medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination of the above.

[0142] The program code for performing the operations of the present application can be written in any combination of one or more programming languages. The programming languages include object-oriented programming languages - such as Java, C++, etc., and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0143] The above computer-readable medium carries one or more programs, which, when executed by the device, cause the computer-readable medium to implement the following functions: dividing a data sample set into a training set and a test set, where the data sample set contains K features and T labels; initializing the selected feature set to be empty and the candidate feature set to be the K features in the data samples; generating a fuzzy equivalence relation between the data samples in the training set through a Gaussian kernel function; based on the fuzzy equivalence relation, determining target features from the candidate feature set using a fuzzy weight redundancy label distribution feature selection method, and adding the target features to the selected feature set; repeating the process of determining target features from the candidate feature set and adding the target features to the selected feature set until the number of features in the selected feature set meets a threshold; training a machine learning model through the selected feature set to generate a label distribution learning model.

[0144] Those skilled in the art can understand that the above-mentioned modules can be distributed in the device according to the description of the embodiments, or can be correspondingly changed and distributed in one or more devices that are uniquely different from this embodiment. The modules of the above embodiments can be combined into one module, or can be further split into multiple sub-modules.

[0145] Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described here can be implemented by software, or can be implemented by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on the network, including several instructions to cause a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0146] The above specifically shows and describes the exemplary embodiments of the present application. It should be understood that the present application is not limited to the detailed structures, setting methods or implementation methods described here; on the contrary, the present application is intended to cover various modifications and equivalent settings included in the spirit and scope of the appended claims.

Claims

1. A method for generating a label distribution learning model based on feature selection, characterized in that: include: Divide the data sample set into a training set and a test set, wherein the data sample set contains K features and T labels; Initialize the selected feature set to be empty and the candidate feature set to be the K features in the data sample; Generate a fuzzy equivalence relationship between the data samples in the training set by using a Gaussian kernel function; Based on the fuzzy equivalence relationship, a fuzzy weight redundant label distribution feature selection method is used to determine a target feature from a candidate feature set, and the target feature is added to a selected feature set; Repeating determining a target feature from the candidate feature set and adding the target feature to the selected feature set until the number of features in the selected feature set meets a threshold; The machine learning model is trained using the selected feature set to generate a label distribution learning model.

2. The method according to claim 1, characterized in that Also includes: Inputting the data samples in the test set into the label distribution learning model to generate label prediction results; The label distribution learning model is evaluated through label prediction results.

3. The method according to claim 1, characterized in that Generating a fuzzy equivalence relationship between the data samples in the training set by using a Gaussian kernel function includes: The fuzzy equivalence relationship between every two data samples in the training set is calculated by a Gaussian kernel function.

4. The method according to claim 3, characterized in that Calculating the fuzzy equivalence relationship between every two data samples in the training set by using a Gaussian kernel function includes: Among them, xi and xj represent the i-th data sample and the j-th data sample respectively, rij is the fuzzy equivalence relationship between data samples xi and xj, and σ represents the variance between xi and xj.

5. The method according to claim 1, characterized in that Based on the fuzzy equivalent relationship, a fuzzy weight redundant label distribution feature selection method is used to determine a target feature from a candidate feature set, and the target feature is added to the selected feature set, including: Based on the fuzzy equivalence relationship, the scores of all features in the candidate feature set are calculated using a fuzzy weight redundant label distribution feature selection method; The feature corresponding to the highest score is used as the target feature; Add the target feature to the selected feature set.

6. The method according to claim 5, characterized in that Based on the fuzzy equivalence relationship, the scores of all features in the candidate feature set are calculated using a fuzzy weight redundant label distribution feature selection method, including: Generate a label distribution feature selection method based on the fuzzy equivalence relationship; The scores of all features in the candidate feature set are calculated using a fuzzy weight redundant label distribution feature selection method.

7. The method according to claim 6, characterized in that The method for selecting label distribution features based on the fuzzy equivalence relationship includes: Determining the correlation between the candidate features and the labels based on the fuzzy equivalence relationship; Determining redundancy between candidate features and selected features based on the fuzzy equivalence relationship; generating interactive redundancy weight factors; The label distribution feature selection method is generated through the correlation between candidate features and labels, the redundancy between candidate features and selected features, and the interactive redundancy weight factor.

8. The method according to claim 7, characterized in that The label distribution feature selection method is generated by the correlation between candidate features and labels, the redundancy between candidate features and selected features, and the interactive redundancy weight factor, including: Among them, f k represents the candidate features, f s represents the selected features in the selected feature set S, y i represents a label in the label distribution data, J(f k ) represents the candidate feature score, (f k ;y i ) represents the candidate feature f k and label y i The correlation between k ;f s ) indicates f k and f s The redundancy between k ) represents the interactive redundancy weight factor, ω(f k ) is initialized to 1.

9. The method according to claim 1, characterized in that Repeating determining a target feature from a candidate feature set and adding the target feature to a selected feature set until the number of features in the selected feature set meets a threshold, including: Remove the target features of the previous round from the candidate feature set to generate an updated candidate feature set; Based on the fuzzy equivalence relationship matrix, the scores of all features in the update candidate feature set are calculated using a fuzzy weight redundant label distribution feature selection method; The feature corresponding to the highest score is used as the target feature of this round; The target features of this round are added to the selected feature set until the number of features in the selected feature set meets the threshold.

10. A label distribution learning model generation device based on feature selection, characterized in that: include: A data module, used to divide a data sample set into a training set and a test set, wherein the data sample set includes K features and T labels; The set module is used to initialize the selected feature set to be empty and the candidate feature set to be K features in the data sample; A matrix module, used for generating a fuzzy equivalent relationship between data samples in the training set through a Gaussian kernel function; A target module is used to determine a target feature from a candidate feature set based on the fuzzy equivalence relationship and using a fuzzy weight redundant label distribution feature selection method, and add the target feature to a selected feature set; A screening module, used for repeatedly determining a target feature from a candidate feature set and adding the target feature to a selected feature set until the number of features in the selected feature set meets a threshold; A training module is used to train the machine learning model through the selected feature set to generate a label distribution learning model.

Images