Support vector machine network emergency early warning method and device based on grey correlation analysis

By using the method based on gray correlation analysis and support vector machine, a network emergencies early warning model is constructed using quantitative evaluation indicators, which solves the problem of difficulty in using quantitative indicators for early warning in the existing technology, and realizes the accurate level division of network emergencies and predict.

CN118965131APending Publication Date: 2024-11-15BEIJING INFORMATION SCI & TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410980050.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Existing research is difficult to use all quantitative indicators to build an early warning index system for network emergencies, and it is difficult to effectively define and predict the level of network emergencies.

Method used

Using the support vector machine network emergencies warning method based on gray correlation analysis, a gray correlation analysis model is established by determining quantitative evaluation indicators such as subject length and publisher membership level, to analyze the correlation and impact degree between indicators, and use K Mean clustering divides the alert level, and early warning model training is carried out in combination with entropy weight method and principal component analysis method.

Benefits of technology

It has achieved scientifically divided the seriousness of network emergencies and made accurate predictions, reducing the subjectivity and arbitrary nature of early warnings and improving the efficiency of early warnings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965131A_ABST
    Figure CN118965131A_ABST
Patent Text Reader

Abstract

The invention provides a support vector machine network emergency early warning method and device based on grey correlation analysis, and the method comprises the steps: taking an emergency publisher, an emergency subject, the spread divergence and other quantitative indexes as starting points, employing a grey correlation analysis method and K-means to carry out the classification of emergencies, and carrying out the early warning of the emergencies. The method comprises the following steps: classifying the network emergencies, performing emergency early warning model training by taking a classification result as a standard, performing weight calculation on each index by utilizing an entropy weight method in a model construction process, and then determining network emergency early warning levels by combining a principal component analysis method and a support vector machine model. And the early warning level of the emergency is accurately predicted in combination with the multi-dimensional information of the emergency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a support vector machine network emergency early warning method and device based on grey correlation analysis. Background Art

[0002] Currently, few existing studies use all quantitative indicators to construct early warning indicator systems. In addition, since the crisis level of network emergencies is difficult to define, there are few methods that use all quantitative indicators to define and predict the level. Summary of the invention

[0003] The present invention aims to provide a support vector machine network emergency warning method and device based on grey correlation analysis, which overcomes the above problems or at least partially solves the above problems.

[0004] To achieve the above object, the technical solution of the present invention is specifically implemented as follows:

[0005] One aspect of the present invention provides a support vector machine network emergency early warning method based on grey correlation analysis, comprising:

[0006] Determine quantitative evaluation indicators, including: main body length, publisher membership level, publisher number of fans, forwarding volume, comment volume, like volume, reading volume, discussion volume, interaction volume, original volume, top ranking on the hot search list, and time on the list;

[0007] Acquire network emergency data, establish a grey correlation analysis model, input the network emergency data into the grey correlation analysis model, analyze the correlation and influence degree between each of the quantitative evaluation indicators, and obtain a grey correlation analysis result;

[0008] Based on the results of the grey correlation analysis, K-means clustering is used to divide emergencies into three categories based on the degree of correlation, and the alert levels corresponding to different time periods of emergencies are determined;

[0009] The alert level is used as a category label, the early warning indicator data after the event occurs at a preset time is used as a feature variable, and the entropy weight method is used to assign the indicator weight;

[0010] Using principal component analysis to reduce the dimension of the early warning indicators;

[0011] Based on the data after dimensionality reduction, the support vector machine early warning model is used to classify the emergency levels.

[0012] Optionally, the acquiring of network emergency data, establishing a grey correlation analysis model, inputting the network emergency data into the grey correlation analysis model, analyzing the correlation and influence degree between the quantitative evaluation indicators, and obtaining the grey correlation analysis results includes:

[0013] Acquire the network emergency event data by using Python crawler technology;

[0014] Determine the reference series that reflects the crisis behavior characteristics of emergencies:

[0015] X 0 =(x 0 (1),x 0 (2),...,x 0 (n));

[0016] Wherein, the reference series is the maximum value of the quantitative evaluation index in the entire emergency event;

[0017] Determine the comparative series that influences emergency crisis behavior:

[0018] X 1 =(x 1 (1),x 1 (2),...,x 1 (n))......(X 2 To X m-1 )……X m =(x m (1),x m (2),...,x m (n));

[0019] Wherein, the comparison series is an evaluation index of the emergency event;

[0020] Performing dimensionless processing on the reference number series and the comparison number series;

[0021] Calculate the correlation factors of each indicator of the reference series and comparison series of emergency crisis:

[0022]

[0023] Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution;

[0024] Calculate the correlation between the reference series and the comparison series of the emergency crisis;

[0025]

[0026] Sort by relevance.

[0027] Optionally, the dimensionless processing of the reference number series and the comparison number series includes:

[0028] Performing initialization and homogenization processing on the network emergency event data;

[0029] The initialization process is calculated using the following formula:

[0030]

[0031] The homogenization process is calculated using the following formula:

[0032]

[0033] Among them, X i ={X i (k)|k=1,2,...,n},i=1,2,..,m, is a comparison sequence; k corresponds to a time period, and i corresponds to a row in the comparison sequence.

[0034] Optionally, based on the result of the grey correlation analysis, K-means clustering is used to divide the emergencies into three categories based on the degree of correlation, and the alert levels corresponding to different time periods of the emergencies are determined, including:

[0035] Randomly select K points in the sample as cluster centers;

[0036] Calculate the distances of other samples in the sample from the K cluster centers respectively, and take these samples as the categories of the cluster center closest to them;

[0037] For the classified samples, the average value of each category is calculated to find the new cluster centroid;

[0038] Compared with the K cluster centroids calculated last time, if the cluster centroids have changed, return to execute the steps of calculating the distances of other samples from the K cluster centers and taking these samples as the categories of the cluster center closest to them. Otherwise, when the centroids do not change, stop and output the clustering results.

[0039] Optionally, the step of using a support vector machine early warning model to classify emergency events includes:

[0040] The parameters are optimized using a parameter optimization method based on Grid Search.

[0041] Another aspect of the present invention provides a support vector machine network emergency warning device based on grey correlation analysis, comprising:

[0042] A determination module is used to determine quantitative evaluation indicators, which include: the length of the main body, the publisher's membership level, the number of publisher fans, the number of reposts, the number of comments, the number of likes, the number of reads, the number of discussions, the number of interactions, the number of original creations, the highest ranking on the hot search list, and the time on the list;

[0043] An analysis module is used to obtain network emergency data, establish a grey correlation analysis model, input the network emergency data into the grey correlation analysis model, analyze the correlation and influence degree between each of the quantitative evaluation indicators, and obtain a grey correlation analysis result;

[0044] A classification module is used to classify the emergency events into three categories based on the results of the grey correlation analysis and the degree of correlation, and to determine the alert levels corresponding to the emergency events in different time periods;

[0045] An assignment module is used to use the alert level as a category label, the early warning indicator data after the event occurs at a preset time as a feature variable, and use the entropy weight method to assign indicator weights;

[0046] A dimension reduction module is used to perform dimension reduction processing on the warning indicators using principal component analysis;

[0047] The classification module is used to classify the emergency events based on the data after dimensionality reduction by using the support vector machine early warning model.

[0048] Optionally, the analysis module obtains network emergency data in the following manner, establishes a grey correlation analysis model, inputs the network emergency data into the grey correlation analysis model, analyzes the correlation and influence degree between each of the quantitative evaluation indicators, and obtains a grey correlation analysis result:

[0049] Acquire the network emergency event data by using Python crawler technology;

[0050] Determine the reference series that reflects the crisis behavior characteristics of emergencies:

[0051] X 0 =(x 0 (1), x 0 (2), ..., x 0 (n));

[0052] Wherein, the reference series is the maximum value of the quantitative evaluation index in the entire emergency event;

[0053] Determine the comparative series that influences emergency crisis behavior:

[0054] X 1 =(x 1 (1), x 1(2), …, x 1 (n))……(X 2 To X m-1 )……X m =(x m (1), x m (2), …, x m (n));

[0055] Wherein, the comparison series is an evaluation index of the emergency event;

[0056] Performing dimensionless processing on the reference number series and the comparison number series;

[0057] Calculate the correlation factors of each indicator of the reference series and comparison series of emergency crisis:

[0058]

[0059] Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution;

[0060] Calculate the correlation between the reference series and the comparison series of the emergency crisis;

[0061]

[0062] Sort by relevance.

[0063] Optionally, the analysis module performs dimensionless processing on the reference number series and the comparison number series in the following manner:

[0064] Performing initialization and homogenization processing on the network emergency event data;

[0065] The initialization process is calculated using the following formula:

[0066]

[0067] The homogenization process is calculated using the following formula:

[0068]

[0069] Among them, X i ={X i (k)|k=1,2,...,n},i=1,2,..,m, is a comparison sequence; k corresponds to a time period, and i corresponds to a row in the comparison sequence.

[0070] Optionally, the classification module divides the emergency events into three categories based on the result of the grey correlation analysis and the degree of correlation, and determines the alert levels corresponding to different time periods of the emergency events by using K-means clustering in the following manner:

[0071] Randomly select K points in the sample as cluster centers;

[0072] Calculate the distances of other samples in the sample from the K cluster centers respectively, and take these samples as the categories of the cluster center closest to them;

[0073] For the classified samples, the average value of each category is calculated to find the new cluster centroid;

[0074] Compared with the K cluster centroids calculated last time, if the cluster centroids have changed, return to execute the steps of calculating the distances of other samples from the K cluster centers and taking these samples as the categories of the cluster center closest to them. Otherwise, when the centroids do not change, stop and output the clustering results.

[0075] Optionally, the classification module uses a support vector machine early warning model to classify emergency events in the following manner:

[0076] The parameters are optimized using a parameter optimization method based on Grid Search.

[0077] It can be seen that the support vector machine network emergency warning method and device based on grey correlation analysis provided by the present invention takes quantitative indicators such as emergency publishers, emergency subjects and propagation breadth as the starting point, adopts grey correlation analysis and K-means to classify emergencies, and uses the classification results as the standard to train the emergency warning model. In the process of model construction, the entropy weight method is used to calculate the weight of each indicator, and then the principal component analysis method and the support vector machine model are combined to determine the network emergency warning level. The present invention can scientifically classify the severity of network emergencies, and accurately predict the warning level of emergencies in combination with the multi-dimensional information of emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0079] Figure 1 A flowchart of a support vector machine network emergency early warning method based on grey correlation analysis provided by an embodiment of the present invention;

[0080] Figure 2 A specific flow chart of a support vector machine network emergency early warning method based on grey correlation analysis provided by an embodiment of the present invention;

[0081] Figure 3 A schematic diagram of a network emergency warning evaluation index system provided by an embodiment of the present invention;

[0082] Figure 4 A flow chart of the K-means clustering algorithm provided by an embodiment of the present invention;

[0083] Figure 5 A schematic diagram of the optimal classification surface provided by an embodiment of the present invention;

[0084] Figure 6 A schematic diagram of the structure of a support vector machine network emergency warning device based on grey relational analysis provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0085] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0086] Figure 1 The flowchart of the support vector machine network emergency warning method based on grey correlation analysis provided by an embodiment of the present invention is shown. Figure 2 The specific flow chart of the support vector machine network emergency warning method based on grey correlation analysis provided by the embodiment of the present invention is shown in FIG. Figure 1 and Figure 2 The support vector machine network emergency warning method based on grey correlation analysis provided by the embodiment of the present invention includes:

[0087] S1. Determine the quantitative evaluation indicators. The quantitative evaluation indicators include: the length of the main body, the publisher's membership level, the number of publisher fans, the number of reposts, the number of comments, the number of likes, the number of reads, the number of discussions, the number of interactions, the amount of originality, the highest ranking on the hot search list, and the time on the list.

[0088] Specifically, from the three key perspectives of the publisher of the emergency, the subject of the emergency and the breadth of dissemination, the present invention selects a series of quantitative evaluation indicators to build a comprehensive and integrated network emergency warning evaluation indicator system. Specifically, twelve indicators are selected, including the subject length, publisher membership level, publisher fans, forwarding volume, comment volume, like volume, reading volume, discussion volume, interaction volume, original volume, top ranking on the hot search list and time on the list. Figure 3 The specific explanation is as follows:

[0089] Main body length: The length of a blog post refers to the number of characters in the main body of the Weibo post.

[0090] Publisher membership level: Publisher membership level refers to the level divided by Sina Weibo based on the user's membership length. Members are divided into four levels: ordinary users, ordinary members (VIP), business members (VVIP), and social members (SVIP).

[0091] Number of publisher's followers: The number of followers refers to the number of other users who follow the Weibo user.

[0092] Retweets: Retweets refer to the number of times a blog post is reposted by other users.

[0093] Number of comments: The number of comments refers to the number of comments posted by other users under the blog post.

[0094] Likes: Likes refer to the number of times other users like a blog post.

[0095] Reading volume: Reading volume refers to the number of times a blog post related to a topic is read by users.

[0096] Discussion volume: Discussion volume refers to the number of Weibo posts with a topic and the number of reposted Weibo posts.

[0097] Interaction volume: Interaction volume refers to the total number of views of all Weibo posts under the topic, the number of views by bloggers, and the number of reposts, comments, likes, and private messages between each other.

[0098] Originality: Originality refers to the number of all original blog posts under a topic.

[0099] Top ranking on the trending search list: Top ranking on the trending search list refers to the highest position of a topic on the trending search list.

[0100] Time on the list: Time on the list refers to the total time a topic is on the hot search list.

[0101] The selection of these indicators aims to comprehensively consider the development of emergencies, the attributes of the participants, and the breadth and depth of information dissemination. Through the construction of this quantitative indicator system, the crisis level and development trend of network emergencies can be more objectively evaluated.

[0102] Based on the network emergency warning indicator system constructed above, relevant indicator data is obtained, and social events 30 days after the occurrence are regarded as events that have been settled. The indicator data of such events are used to divide the warning level standards, which is highly persuasive. The gray correlation analysis method and K-means clustering method are used to divide the danger level of related events, and the division results are used as the category labels for subsequent predictions. The indicator data obtained one hour after the release of social events are used to predict the risk level, and the entropy weight method is used to assign weights to each indicator system to ensure that the model takes into account the importance of different indicators. The early warning model is constructed by combining the principal component analysis method and the support vector machine method to achieve the effect of accurately predicting the danger level one hour after the emergency occurs.

[0103] S2, obtain network emergency data, establish a grey correlation analysis model, input the network emergency data into the grey correlation analysis model, analyze the correlation and impact degree between various quantitative evaluation indicators, and obtain the grey correlation analysis results.

[0104] Specifically, the 30-day network emergency data can be dimensionless processed to ensure that each indicator has comparability and uniform dimensions. This step establishes a grey correlation analysis model: a grey correlation analysis model is established using quantified data to explore the correlation and impact between each indicator.

[0105] As an optional implementation of the embodiment of the present invention, network emergency data is obtained, a grey correlation analysis model is established, the network emergency data is input into the grey correlation analysis model, the correlation and influence degree between various quantitative evaluation indicators are analyzed, and the grey correlation analysis results obtained include:

[0106] Obtain network emergency data through Python crawler technology;

[0107] Determine the reference series that reflects the crisis behavior characteristics of emergencies:

[0108] X 0 =(x 0 (1), x 0 (2), …, x 0 (n));

[0109] Among them, the reference series is the maximum value of the quantitative evaluation index in the entire emergency;

[0110] Determine the comparative series that influences emergency crisis behavior:

[0111] X 1 =(x 1 (1), x 1 (2), …, x 1 (n))……(X2 To X m-1 )……X m =(x m (1), x m (2), …, x m (n));

[0112] Among them, the comparison series is the evaluation index of the emergency;

[0113] Make the reference series and the comparison series dimensionless;

[0114] Calculate the correlation factors of each indicator of the reference series and comparison series of emergency crisis:

[0115]

[0116] Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution;

[0117] Calculate the correlation between the reference series and the comparison series of the emergency crisis;

[0118]

[0119] Sort by relevance.

[0120] In the specific implementation, the maximum value of the selected 12 evaluation indicators in the entire emergency is used as the reference series, and the evaluation indicators of different emergencies are used as the comparison series. Then, the grey correlation coefficient and correlation degree between each emergency and the reference series are calculated, and the risk level of each emergency is divided accordingly. The specific calculation steps are as follows:

[0121] (1) Determine the reference series (evaluation indicators) that reflect the characteristics of emergency crisis behavior:

[0122] X 0 =(x 0 (1), x 0 (2), …, x 0 (n))

[0123] Determine the comparative series that influences emergency crisis behavior:

[0124] X 1 =(x 1 (1), x 1 (2), …, x 1 (n))……(X 2 To X m-1 )

[0125] X m =(x m (1), x m (2), …, xm (n))

[0126] (2) Dimensionless processing is performed on the reference series and the comparison series.

[0127] (3) Calculate the correlation factors of the reference series and the comparison series of the emergency crisis. i (i=1, 2, ..., m) and the reference sequence X 0 The correlation factor is defined as:

[0128]

[0129] Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution is.

[0130] (4) Calculation of the correlation between the reference sequence and the comparison sequence of the emergency crisis. i (k) is the correlation factor value of each component between the reference series and the comparison series. The results are numerous and not easy to compare. Therefore, it is necessary to concentrate the correlation factors of each indicator of each sub-emergency factor vector into one value, which is called the correlation degree.

[0131] The reference number series of network emergency crisis is X 0 , the comparison series of network emergency crisis is X i (i=1,2,…,m), then X 0 With X i The association degree of (i=1,2,...,m) is defined as:

[0132]

[0133] (5) Relevance ranking. The degree of relevance between emergencies is mainly described by the order of magnitude of the relevance, rather than simply the magnitude of the relevance. Arranging the relevance of m sub-emergencies to the same parent emergency in order of magnitude forms a relevance sequence, denoted as {X}, which reflects the "superiority" relationship of each sub-emergency to the parent emergency. If Y i >Y j , then {X i}For the same parent incident {X i} is better than {X j}, denoted as {X i}>{X j}.

[0134] As an optional implementation of the present invention, dimensionless processing of the reference sequence and the comparison sequence includes:

[0135] Perform initialization and homogenization processing on network emergency event data;

[0136] The initialization process is calculated using the following formula:

[0137]

[0138] The homogenization process is calculated using the following formula:

[0139]

[0140] Among them, X i ={X i (k)|k=1,2,…,n},i=1,2,…,m,is a comparison sequence; k corresponds to a time period, and i corresponds to a row in the comparison sequence.

[0141] Specifically, since the data in each factor column may have different dimensions, it is not easy to compare or it is difficult to get the correct conclusion when comparing. Therefore, when performing grey relational analysis, it is generally necessary to perform dimensionless processing of the data. There are mainly two methods:

[0142] ① Initialization processing:

[0143]

[0144] ②Homogenization treatment:

[0145]

[0146] Where k corresponds to the time period, and i corresponds to a row in the comparison series (i.e., a feature).

[0147] S3, based on the results of grey correlation analysis and the degree of correlation, K-means clustering was used to divide emergencies into three categories, and the alert levels corresponding to different time periods of emergencies were determined.

[0148] Specifically, this step performs K-means clustering: based on the results of grey correlation analysis and the degree of correlation, K-means clustering is used to divide emergencies into three categories, thereby determining the alert levels corresponding to different time periods of emergencies.

[0149] As an optional implementation of the embodiment of the present invention, based on the results of grey correlation analysis and based on the degree of correlation, K-means clustering is used to divide emergencies into three categories, and the alert levels corresponding to different time periods of emergencies are determined, including:

[0150] Randomly select K points in the sample as cluster centers;

[0151] Calculate the distances of other samples in the sample from the K cluster centers respectively, and take these samples as the categories of the cluster center closest to them;

[0152] For the classified samples, the average value of each category is calculated to find the new cluster centroid;

[0153] Compared with the K cluster centroids calculated last time, if the cluster centroids have changed, return to execute the steps of calculating the distances of other samples from the K cluster centers and taking these samples as the categories of the cluster center closest to them. Otherwise, when the centroids do not change, stop and output the clustering results.

[0154] In the specific implementation, the K-means cluster analysis method is selected to classify network emergencies. The K-Means algorithm divides the feature matrix X of a group of N samples into K non-intersecting clusters. Intuitively, clusters are groups of data gathered together, and the data in a cluster are considered to be of the same type. The mean of all data in a cluster is usually called the "center of mass" of the cluster. In a two-dimensional plane, the horizontal coordinate of the center of mass of a cluster of data points is the mean of the horizontal coordinates of the cluster of data points, and the vertical coordinate of the center of mass is the mean of the vertical coordinates of the cluster of data points.

[0155] In the K-Means algorithm, the number of clusters K is a hyperparameter that needs to be determined manually. The core task of K-Means is to find the K optimal centroids based on the set K, and assign the data closest to these centroids to the clusters represented by these centroids. The specific process can be summarized as follows:

[0156] a. First, randomly select K points in the sample as cluster centers;

[0157] b. Calculate the distances between other samples in the sample and the K cluster centers, and use these samples as the categories of the cluster center closest to them;

[0158] c. Calculate the average value of each category for the samples classified above and solve the new cluster centroid;

[0159] d. Compare with the K cluster centroids calculated last time. If the cluster centroids change, go to process b, otherwise go to process e.

[0160] e. When the centroid does not change (when we find a centroid, the samples assigned to this centroid in each iteration are consistent, that is, each newly generated cluster is consistent, all sample points will no longer be transferred from one cluster to another, and the centroid will not change), stop and output the clustering results. The calculation process of the K-Means algorithm is as follows Figure 4 shown.

[0161] Based on the above steps, the present invention uses the correlation between each sub-emergency and the parent emergency as a clustering index, and classifies and grades the emergencies at each moment according to the cluster center, thereby obtaining the risk level results of each emergency.

[0162] S4, taking the alert level as the category label, taking the warning indicator data after the preset time of the event as the feature variable, and using the entropy weight method to assign the indicator weight.

[0163] Specifically, this step assigns indicator weights: the alert level is used as the category label, the early warning indicator data one hour after the event occurs is used as the characteristic variable, and the entropy weight method is used to assign indicator weights to ensure that the influence of each indicator in the early warning model is consistent.

[0164] S5, use principal component analysis to reduce the dimension of early warning indicators.

[0165] Specifically, this step performs principal component analysis: principal component analysis is used to reduce the dimensionality of early warning indicators, reduce data dimensions, and improve the computational efficiency and prediction accuracy of the model.

[0166] S6, based on the data after dimensionality reduction, the support vector machine early warning model is used to classify the emergency events.

[0167] Specifically, this step performs support vector machine classification: based on the data after dimensionality reduction, the support vector machine method is used to classify the emergency events so as to achieve accurate early warning of the emergency events.

[0168] As an optional implementation of the embodiment of the present invention, using the support vector machine early warning model to classify the emergency level includes:

[0169] The parameters are optimized using a parameter optimization method based on Grid Search.

[0170] In specific implementation, the early warning model of the present invention mainly selects the support vector machine (SVM) method for risk level classification. The support vector machine (SVM) is generated from the study of classification problems. The SVM adopts the criterion of structural risk minimization to find a plane that maximizes the distance between data points of different classes, that is, the optimal hyperplane. The optimal hyperplane is usually linear, but not all optimal hyperplanes for classification problems are linear, so according to the difference in the optimal hyperplane, the support vector machine can be divided into two types: linear support vector classifier and nonlinear support vector classifier.

[0171] Figure 5 There is a surface H in it, which separates the two types of data, and the distance from the nearest point in the two types of samples to itself is maximized. H is the optimal hyperplane we are looking for, and the points closest to H are the support vectors.

[0172] The optimal classification decision function is:

[0173]

[0174] Among them, sign is the sign function, l is the number of support vectors, and b* is the classification threshold. When constructing the optimal hyperplane, the kernel function of the linearly separable case is the linear kernel function. If the optimal hyperplane is nonlinearly separable, we must find a way to transform it into a linear one, and then find the optimal hyperplane. Specifically, all sample data are mapped to a high-dimensional space that makes it linearly separable through a complex nonlinear function φ(x).

[0175] K(x i , x j )=(x i ·x j )

[0176] The nonlinear function φ(x) here is the kernel function, that is, K(x i ,x j ).

[0177] Algorithm steps:

[0178] (1) Provide a training sample space;

[0179] (2) Study the relationship and characteristics between variables and select appropriate kernel functions;

[0180] (3) Constructing a quadratic programming problem based on the risk function;

[0181] (4) Calculate the optimal solution;

[0182] (5) Construct the final regression function.

[0183] The event classification results are used as the category labels of the early warning model training data, and the data of each indicator one hour after the event occurs are used as the characteristic variables. The entropy weight method is first used to determine the indicator weight for each indicator to ensure that the importance of each indicator in the early warning model is taken into account. After each indicator is weighted, principal component analysis is performed to reduce the attributes, that is, to reduce the number of input indicators of the support vector machine. On this basis, the principal component analysis and SVM methods are connected in series, and the processing results of the principal component analysis are used as the input of the support vector machine to establish a network emergency crisis early warning model based on SVM, realizing the relevant indicators X of the input event. i , output the corresponding crisis level Y i .

[0184] Let x be the change value of the indicator that affects the early warning, and y be the level value of the severity of the crisis. Applying the knowledge of mathematical statistics regression, the regression function is:

[0185] y i =f(x)=w·Φ(x ij )+b i

[0186] Among them, Φ(x ij ) is a nonlinear mapping. ij Refers to the size of the jth indicator at the i-th time node. The support vector machine model used in this article is:

[0187]

[0188] Among them, ε is the allowable error value, Represents the average value. The early warning model based on SVM is as follows:

[0189]

[0190] Through the above steps, the danger level of the event can be accurately predicted one hour after the event occurs.

[0191] It can be seen that the support vector machine network emergency early warning method based on grey correlation analysis provided by the embodiment of the present invention analyzes the influencing factors of network emergency crisis, selects 12 quantitative factors such as the main text length, publisher membership level, forwarding volume, and comment volume as indicator factors affecting the degree of network emergency crisis, uses the grey correlation analysis method to calculate the correlation between different emergencies and reference series events, and identifies and classifies different network emergencies through cluster analysis, thereby determining and dividing the risk level of network emergencies. On this basis, the indicator data one hour after the event is used as the early warning input, and the entropy weight method is used to assign different weights to each indicator, and the principal component analysis and support vector machine are connected in series to realize the automatic early warning identification of the degree of crisis of network emergencies. The grey correlation analysis method is used for correlation analysis comparison, and the subjectivity and arbitrariness of the traditional network emergency classification are eliminated by establishing the size of the correlation and classifying the categories. The joint principal component analysis method and support vector machine early warning identification model based on the entropy weight method can effectively improve the early warning efficiency.

[0192] Figure 6 The structure diagram of the support vector machine network emergency warning device based on grey correlation analysis provided by an embodiment of the present invention is shown. The support vector machine network emergency warning device based on grey correlation analysis applies the above method. The following is only a brief description of the structure of the support vector machine network emergency warning device based on grey correlation analysis. For other matters not covered, please refer to the relevant description in the support vector machine network emergency warning method based on grey correlation analysis, see Figure 6The support vector machine network emergency warning device based on grey correlation analysis provided by the embodiment of the present invention includes:

[0193] The determination module is used to determine the quantitative evaluation indicators, which include: the length of the main body, the publisher's membership level, the number of publisher fans, the number of reposts, the number of comments, the number of likes, the number of reads, the number of discussions, the number of interactions, the number of original works, the highest ranking on the hot search list, and the time on the list;

[0194] The analysis module is used to obtain network emergency data, establish a grey correlation analysis model, input the network emergency data into the grey correlation analysis model, analyze the correlation and influence degree between various quantitative evaluation indicators, and obtain the grey correlation analysis results;

[0195] The classification module is used to classify emergencies into three categories based on the results of grey correlation analysis and the degree of correlation, and to determine the alert levels corresponding to different time periods of emergencies;

[0196] The assignment module is used to use the alert level as the category label, the warning indicator data after the preset time of the event as the feature variable, and use the entropy weight method to assign the indicator weight;

[0197] Dimensionality reduction module, used to reduce the dimension of early warning indicators using principal component analysis;

[0198] The classification module is used to classify the emergency events based on the data after dimensionality reduction by using the support vector machine early warning model.

[0199] As an optional implementation of the embodiment of the present invention, the analysis module obtains network emergency data in the following manner, establishes a grey correlation analysis model, inputs the network emergency data into the grey correlation analysis model, analyzes the correlation and influence degree between various quantitative evaluation indicators, and obtains the grey correlation analysis result:

[0200] Obtain network emergency data through Python crawler technology;

[0201] Determine the reference series that reflects the crisis behavior characteristics of emergencies:

[0202] X 0 =(x 0 (1), x 0 (2), ..., x 0 (n));

[0203] Among them, the reference series is the maximum value of the quantitative evaluation index in the entire emergency;

[0204] Determine the comparative series that influences emergency crisis behavior:

[0205] X1 =(x 1 (1), x 1 (2), …, x 1 (n))……(X 2 To X m-1 )……X m =(x m (1), x m (2), …, x m (n));

[0206] Among them, the comparison series is the evaluation index of the emergency;

[0207] Make the reference series and the comparison series dimensionless;

[0208] Calculate the correlation factors of each indicator of the reference series and comparison series of emergency crisis:

[0209]

[0210] Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution;

[0211] Calculate the correlation between the reference series and the comparison series of the emergency crisis;

[0212]

[0213] Sort by relevance.

[0214] As an optional implementation of the embodiment of the present invention, the analysis module performs dimensionless processing on the reference sequence and the comparison sequence in the following manner:

[0215] Perform initialization and homogenization processing on network emergency event data;

[0216] The initialization process is calculated using the following formula:

[0217]

[0218] The homogenization process is calculated using the following formula:

[0219]

[0220] Among them, X i ={X i (k)|k=1,2,...,n},i=1,2,..,m, is a comparison sequence; k corresponds to a time period, and i corresponds to a row in the comparison sequence.

[0221] As an optional implementation of the embodiment of the present invention, the classification module divides the emergency events into three categories based on the results of the grey correlation analysis and the degree of correlation, and determines the alert levels corresponding to the emergency events in different time periods:

[0222] Randomly select K points in the sample as cluster centers;

[0223] Calculate the distances of other samples in the sample from the K cluster centers respectively, and take these samples as the categories of the cluster center closest to them;

[0224] For the classified samples, the average value of each category is calculated to find the new cluster centroid;

[0225] Compared with the K cluster centroids calculated last time, if the cluster centroids have changed, return to execute the steps of calculating the distances of other samples from the K cluster centers and taking these samples as the categories of the cluster center closest to them. Otherwise, when the centroids do not change, stop and output the clustering results.

[0226] As an optional implementation of the embodiment of the present invention, the classification module uses the support vector machine warning model to classify the emergency level in the following manner:

[0227] The parameters are optimized using a parameter optimization method based on Grid Search.

[0228] It can be seen that, through the support vector machine network emergency early warning device based on grey correlation analysis provided by the embodiment of the present invention, by analyzing the influencing factors of network emergency crisis, 12 quantitative factors such as the main body length, publisher membership level, forwarding volume, and comment volume are selected as index factors affecting the degree of network emergency crisis, and the grey correlation analysis method is used to calculate the correlation between different emergencies and reference series events, and different network emergencies are identified and classified through cluster analysis, so as to determine and divide the risk level of network emergencies. On this basis, the indicator data one hour after the event is used as the early warning input, and the entropy weight method is used to assign different weights to each indicator, and the principal component analysis and support vector machine are connected in series to realize the automatic early warning identification of the degree of network emergency crisis. The grey correlation analysis method is used for correlation analysis comparison, and the subjectivity and arbitrariness of the traditional network emergency classification are eliminated by establishing the size of the correlation and classifying the categories. The joint principal component analysis method and support vector machine early warning identification model based on the entropy weight method can effectively improve the early warning efficiency.

[0229] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A support vector machine network emergency warning method based on grey correlation analysis, characterized in that: include: Determine quantitative evaluation indicators, including: main body length, publisher membership level, publisher number of fans, forwarding volume, comment volume, like volume, reading volume, discussion volume, interaction volume, original volume, top ranking on the hot search list, and time on the list; Acquire network emergency data, establish a grey correlation analysis model, input the network emergency data into the grey correlation analysis model, analyze the correlation and influence degree between each of the quantitative evaluation indicators, and obtain a grey correlation analysis result; Based on the results of the grey correlation analysis, K-means clustering is used to divide emergencies into three categories based on the degree of correlation, and the alert levels corresponding to different time periods of emergencies are determined; The alert level is used as a category label, the early warning indicator data after the event occurs at a preset time is used as a feature variable, and the entropy weight method is used to assign the indicator weight; Using principal component analysis to reduce the dimension of the early warning indicators; Based on the data after dimensionality reduction, the support vector machine early warning model is used to classify the emergency levels.

2. The method according to claim 1, characterized in that The acquisition of network emergency data, the establishment of a grey correlation analysis model, the input of the network emergency data into the grey correlation analysis model, the analysis of the correlation and influence degree between the quantitative evaluation indicators, and the acquisition of the grey correlation analysis results include: Acquire the network emergency event data by using Python crawler technology; Determine the reference series that reflects the crisis behavior characteristics of emergencies: X0=(x0(1),x0(2),...,x0(n)); Wherein, the reference series is the maximum value of the quantitative evaluation index in the entire emergency event; Determine the comparative series that influences emergency crisis behavior: X1=(x1(1),x1(2),...,x1(n))......(X2 to X m-1 )……X m =(x m (1),x m (2),...,x m (n)); Wherein, the comparison series is an evaluation index of the emergency event; Performing dimensionless processing on the reference number series and the comparison number series; Calculate the correlation factors of each indicator of the reference series and comparison series of emergency crisis: Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution; Calculate the correlation between the reference series and the comparison series of the emergency crisis; Sort by relevance.

3. The method according to claim 2, characterized in that The dimensionless processing of the reference number series and the comparison number series comprises: Performing initialization and homogenization processing on the network emergency event data; The initialization process is calculated using the following formula: The homogenization process is calculated using the following formula: Among them, X i ={X i (k)|k=1,2,...,n},i=1,2,..,m, is a comparison sequence; k corresponds to a time period, and i corresponds to a row in the comparison sequence.

4. The method according to claim 3, characterized in that Based on the results of the grey correlation analysis, K-means clustering is used to divide emergencies into three categories based on the degree of correlation, and the alert levels corresponding to different time periods of emergencies are determined, including: Randomly select K points in the sample as cluster centers; Calculate the distances of other samples in the sample from the K cluster centers respectively, and take these samples as the categories of the cluster center closest to them; For the classified samples, the average value of each category is calculated to find the new cluster centroid; Compared with the K cluster centroids calculated last time, if the cluster centroids have changed, return to execute the steps of calculating the distances of other samples from the K cluster centers and taking these samples as the categories of the cluster center closest to them. Otherwise, when the centroids do not change, stop and output the clustering results.

5. The method according to claim 4, characterized in that The use of the support vector machine early warning model to classify emergency events includes: The parameters are optimized using a parameter optimization method based on Grid Search.

6. A support vector machine network emergency warning device based on grey correlation analysis, characterized in that: include: A determination module is used to determine quantitative evaluation indicators, which include: the length of the main body, the publisher's membership level, the number of publisher fans, the number of reposts, the number of comments, the number of likes, the number of reads, the number of discussions, the number of interactions, the number of original creations, the highest ranking on the hot search list, and the time on the list; An analysis module is used to obtain network emergency data, establish a grey correlation analysis model, input the network emergency data into the grey correlation analysis model, analyze the correlation and influence degree between each of the quantitative evaluation indicators, and obtain a grey correlation analysis result; A classification module is used to classify the emergency events into three categories based on the results of the grey correlation analysis and the degree of correlation, and to determine the alert levels corresponding to the emergency events in different time periods; An assignment module is used to use the alert level as a category label, the early warning indicator data after the event occurs at a preset time as a feature variable, and use the entropy weight method to assign indicator weights; A dimension reduction module is used to perform dimension reduction processing on the warning indicators using principal component analysis; The classification module is used to classify the emergency events based on the data after dimensionality reduction by using the support vector machine early warning model.

7. The device according to claim 6, characterized in that The analysis module obtains network emergency data in the following manner, establishes a grey correlation analysis model, inputs the network emergency data into the grey correlation analysis model, analyzes the correlation and influence degree between the quantitative evaluation indicators, and obtains the grey correlation analysis result: Acquire the network emergency event data by using Python crawler technology; Determine the reference series that reflects the crisis behavior characteristics of emergencies: X0=(x0(1),x0(2),...,n0(n)); Wherein, the reference series is the maximum value of the quantitative evaluation index in the entire emergency event; Determine the comparative series that influences emergency crisis behavior: X1=(x1(1),x1(2),...,x1(n))......(X2 to X m-1 )……X m =(x m (1),x m (2),...,x m (n)); Wherein, the comparison series is an evaluation index of the emergency event; Performing dimensionless processing on the reference number series and the comparison number series; Calculate the correlation factors of each indicator of the reference series and comparison series of emergency crisis: Among them, ρ∈(0,+∞) is called resolution, the smaller ρ is, the greater the resolution; Calculate the correlation between the reference series and the comparison series of the emergency crisis; Sort by relevance.

8. The device according to claim 7, characterized in that The analysis module performs dimensionless processing on the reference sequence and the comparison sequence in the following manner: Performing initialization and homogenization processing on the network emergency event data; The initialization process is calculated using the following formula: The homogenization process is calculated using the following formula: Among them, X i ={X i (k)|k=1,2,...,n},i=1,2,..,m, is a comparison sequence; k corresponds to a time period, and i corresponds to a row in the comparison sequence.

9. The device according to claim 8, characterized in that The classification module divides the emergencies into three categories based on the results of the grey correlation analysis and the degree of correlation, and determines the alert levels corresponding to different time periods of emergencies by using K-means clustering: Randomly select K points in the sample as cluster centers; Calculate the distances of other samples in the sample from the K cluster centers respectively, and take these samples as the categories of the cluster center closest to them; For the classified samples, the average value of each category is calculated to find the new cluster centroid; Compared with the K cluster centroids calculated last time, if the cluster centroids have changed, return to execute the steps of calculating the distances of other samples from the K cluster centers and taking these samples as the categories of the cluster center closest to them. Otherwise, when the centroids do not change, stop and output the clustering results.

10. The device according to claim 9, characterized in that The classification module uses the support vector machine early warning model to classify the emergency events in the following way: The parameters are optimized using a parameter optimization method based on Grid Search.