Phishing email identification method and device based on email detection network model
Through the phishing email recognition method based on the email detection network model, the email features are extracted and classified and processed, and the phishing email detection network model is trained, which solves the problem that the existing technology is difficult to identify new phishing emails, and improves network security and recognition efficiency.
Patent Information
- Application Number
- CN202411186263.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-28
AI Technical Summary
Existing security protection methods are difficult to effectively identify new phishing email attacks, resulting in network security missed and reduced network security.
Using the phishing email recognition method based on the email detection network model, by obtaining the email information set and feature tags, extracting key email features, building training sample sets, performing classification processing, generating type features, and training the phishing email detection network model to achieve the Intelligent identification and blocking of phishing emails.
It improves network security, can identify phishing emails in a timely manner, reduces the consumption of computing resources and storage resources, shortens model learning time, reduces model complexity, and enhances the ability to identify new attack methods.
Smart Images

Figure CN118713930B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of phishing email identification, and specifically to a phishing email identification method and device based on an email detection network model. Background Art
[0002] Phishing email attacks have become an important way to penetrate the network, used to steal information, implant viruses, and remotely control penetration. Existing security protection measures mainly rely on technical judgments such as rule libraries, virus libraries, IP reputation libraries, domain name libraries, Bayesian, SPF, and sender verification (SPF, DKIM, DMARC), but new QR code technologies, springboard attacks, spear phishing attacks, etc., have seriously leaked existing security protection capabilities and reduced network security. Summary of the invention
[0003] The content of this application is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this application is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
[0004] Some embodiments of the present application propose a phishing email identification method, a computer device, and a computer-readable storage medium based on an email detection network model to solve the technical problems mentioned in the above background technology section.
[0005] In a first aspect, some embodiments of the present application provide a phishing email identification method based on an email detection network model, the method comprising: obtaining an email information set and feature labels corresponding to each email information in the email information set; extracting email key features of each email information in the email information set according to key email features corresponding to a pre-trained initial phishing email detection network model, and obtaining an email key feature set, wherein the initial phishing email detection network model is a decision model; constructing an email training sample set according to the email key feature set and each feature label, wherein the email training sample comprises: email key features and feature labels; for each email feature type, extracting the email key features of each email information in the email training sample set according to the key email features corresponding to the email included in the email training sample set; The email training sample set is classified according to each email sub-feature of the email feature type to generate type features corresponding to the email feature type for each type of email training sample; a preset number of email training samples are selected from the email training sample set as an email training sample group; according to the email training sample group and each type feature corresponding to the email training sample group, an initial phishing email detection network model is trained to obtain a trained phishing email detection network model; in response to receiving a target email, the target email is input into the phishing email detection network model to obtain a phishing email detection result; in response to the phishing email detection result characterizing that the target email is a phishing email, the target email is intercepted.
[0006] In a second aspect, the present application also provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the method described in any implementation of the first aspect is implemented.
[0007] In a third aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, wherein when the computer program is executed by a processor, the method described in any implementation of the first aspect is implemented.
[0008] The above-mentioned embodiments of the present application have the following beneficial effects: phishing emails can be identified in time through the trained phishing email detection network model, thereby improving the security of the network. First, an email information set and a feature label corresponding to each email information in the email information set are obtained. Thus, it is convenient to train the initial phishing email detection network model according to the email information set, so as to intelligently identify phishing emails. Secondly, according to the key email features corresponding to the pre-trained initial phishing email detection network model, the email key features of each email information in the email information set are extracted to obtain the email key feature set. Among them, the above-mentioned initial phishing email detection network model is a decision model. Thus, the key features required by the model can be extracted, providing data support for accurately identifying phishing emails. Then, according to the above-mentioned email key feature set and each feature label, an email training sample set is constructed. Among them, the email training sample includes: email key features and feature labels. For each email feature type, according to each email sub-feature corresponding to the above-mentioned email feature type included in the above-mentioned email training sample set, the above-mentioned email training sample set is classified and processed to generate type features corresponding to the above-mentioned email feature type for each type of email training sample. Therefore, the email features are converted into limited type features by classification, so that missing values can be classified separately without assuming conditions, reducing the impact of filling missing values on the model effect. Also, because the type features participate in model training when training the model, the loss of information after classification can be reduced. And because important email features are screened out when training the model, the dimension of the data can be reduced when training a complex formal model. Thus, it is possible to avoid directly using a model with larger capacity and higher complexity, thereby reducing the consumption of computing resources and storage resources, reducing the difficulty of model learning, and shortening the convergence time. Then, a preset number of email training samples are selected from the above email training sample set as an email training sample group; according to the above email training sample group and each type feature corresponding to the above email training sample group, the initial phishing email detection network model is trained to obtain a trained phishing email detection network model. Thus, each received email can be identified by the trained phishing email detection network model. Finally, in response to receiving the target email, the target email is input into the phishing email detection network model to obtain a phishing email detection result; in response to the phishing email detection result indicating that the target email is a phishing email, the target email is intercepted. Thus, the trained phishing email detection network model can be used to timely identify phishing emails, thereby improving the security of the network. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar figure labels represent the same or similar elements. It should be understood that the drawings are schematic and that the components and elements are not necessarily drawn to scale.
[0010] Figure 1 is a flow chart of some embodiments of a phishing email identification method based on an email detection network model according to the present application;
[0011] Figure 2 It is a schematic diagram of the structure of a computer device suitable for implementing some embodiments of the present application. DETAILED DESCRIPTION
[0012] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.
[0013] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.
[0014] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
[0015] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".
[0016] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.
[0017] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.
[0018] Figure 11 is a flowchart of some embodiments of the phishing email identification method based on the email detection network model according to the present application. It shows a process 100 of some embodiments of the phishing email identification method based on the email detection network model according to the present application. The phishing email identification method based on the email detection network model includes the following steps:
[0019] Step 101: Obtain an email information set and a feature tag corresponding to each email information in the email information set.
[0020] In some embodiments, the executing subject (for example, a computer device) of the phishing email identification method based on the email detection network model can obtain an email information set and a feature tag corresponding to each email message in the email information set. Email information may refer to information of emails sent historically, and may include: email type, user name, mail server, subject, sender address, recipient address, body, and attachments. The feature tag may indicate whether the email corresponding to the email information is a phishing email. Phishing email is a network attack method that sends emails disguised as trusted sources to trick users into replying to emails, clicking on malicious links, or opening attachments containing viruses, thereby stealing users' sensitive information, such as personal accounts, passwords, or bank accounts, or executing malicious program code on the device to carry out further network attack activities.
[0021] Furthermore, the email feature training data is generated according to the feature data belonging to the input features in the email feature exchange-in data set and the email feature exchange-out data set.
[0022] In some embodiments, the execution subject may generate email feature training data based on the feature data belonging to the input model feature in the email feature swap-in data set and the email feature swap-out data set. The input model feature is the input model feature of the initial phishing email detection network model that has the swap-in and swap-out phenomenon. For example, the execution subject may determine the feature data belonging to the input model feature in the email feature swap-in data set and the email feature swap-out data set. And these feature data may be used as email feature training data. And the email feature training data may be labeled according to the data set to which the email feature training data belongs. For example, the label of the email feature swap-in data set is 1, and the label of the email feature swap-out data set is 0, thereby obtaining the email feature training data. In this embodiment, only features related to the decision model are concerned. That is, only the influence of the input model features of the decision model on the swap-in and swap-out is examined. Therefore, the input model features are usually the input model features of the decision model that has the swap-in and swap-out phenomenon. Because only these features may cause the swap-in and swap-out phenomenon. The input model features are usually the inherent attributes of the model. After the model training is completed, the input model features of the model have been determined. The execution subject can determine the attribute features at each classification node in the decision model as the input features of the decision model. It can be understood that in the tree model, each tree fork node usually performs binary classification on the data of the node according to the value of one or more characteristic parameters. The features analyzed by these tree fork nodes are the input features of the model.
[0023] Furthermore, the preset classification model is trained by using the above-mentioned email feature training data, and based on the trained preset classification model, the indicator data of each input feature of the above-mentioned initial phishing email detection network model is determined.
[0024] In some embodiments, the execution subject may train the preset classification model through the email feature training data, and determine the index data of each input model feature of the initial phishing email detection network model based on the trained preset classification model. The index data is used to characterize the importance of the feature. The preset classification model is a distributed extreme gradient boosting tree model. The preset classification model may be any model that can perform classification prediction of email feature input data and email feature output data, such as a classification model that outputs Boolean values or a classification decision model that outputs categories. For example, the preset classification model may be a distributed gradient boosting library XGB tree model. XGB is XGBoost (eXtremeGradientBoosting). In this case, the XGB tree model may be trained using training samples. And the loss function value of the XGB tree model may be determined based on the prediction results of the training samples and the corresponding sample labels, and then the model parameters of the XGB tree model may be adjusted until the model training is completed. According to the trained preset classification model, the execution subject may also use a machine learning model interpretation algorithm (SHAP, SHapley Additive exPlanations) to determine the contribution value of each input model feature to the prediction results of the preset classification model.
[0025] In practice, the execution subject can obtain the relevant index coefficients in the distributed extreme gradient boosting tree model to obtain the importance of each input feature of the initial phishing email detection network model. The obtained relevant index coefficients are used as indicator data. The relevant index coefficients here can be such as the total gain coefficient. The larger the number, the greater the classification contribution.
[0026] Furthermore, according to the determined indicator data, a set number of input features are selected from the input features of the above-mentioned initial phishing email detection network model as target features.
[0027] In some embodiments, the execution subject may select a set number of input features from each input feature of the initial phishing email detection network model as target features according to the determined indicator data. As an example, the execution subject may select a set number of input features as target features in descending order of indicator data value (i.e., importance).
[0028] Furthermore, based on the target features, the email feature swap-in data set and the email feature swap-out data set, key features of emails that cause the initial phishing email detection network model to generate swap-in and swap-out data are analyzed.
[0029] In some embodiments, the execution subject may analyze the key features of emails that cause the initial phishing email detection network model to generate input and output data based on the target features, the email feature input data set, and the email feature output data set. As an example, the execution subject may further select less than the set number of input features from the target features as key features of emails. For another example, for each input feature, the execution subject may also count the amount of data belonging to the input feature in the input and output data sets. Then, the key features of emails may be selected from these input features in descending order of data volume, such as the top five input features.
[0030] In an actual application scenario, the execution subject can analyze the key features of the email that cause the initial phishing email detection network model to generate the swapped in and out data through the following steps:
[0031] The first step is to determine the probability distribution density of the feature data belonging to the input feature in the above target feature, in the above email feature input data set and the above email feature output data set. Let the probability of a continuous random variable falling within a certain interval be P, △x>0 is the length of the interval, then the ratio of P / △x is called the "average probability distribution density" of the random variable in the interval. If the limit of the ratio exists when the interval length △x→0, then this limit is called the probability distribution density of the random variable at point x, referred to as distribution density.
[0032] The second step is to determine the population stability index value (PSI, Population Stability Index) of the above-mentioned input data set and the above-mentioned output data set under the target feature according to the probability distribution density of the feature data of each input feature. Among them, PSI usually measures the stability of the data by comparing the fluctuations of the historical sample distribution and the current sample distribution. It is generally used to quantitatively evaluate the distribution difference between the observed sample (actual, equivalent to the input data set) and the development sample (expected, equivalent to the output data set). This method usually divides the two samples into n groups according to certain rules (equal width, equal frequency), and then compares the similarity of the distribution. That is: PSI=SUM[(observation sample proportion-development sample proportion)*ln(observation sample proportion / development sample proportion)].
[0033] The third step is to determine the group stability index value under each input feature according to the exchange-in data set of the above-mentioned email features and the exchange-out data set of the above-mentioned email features, and the input features corresponding to the group stability index value within the preset range as the key features of emails with large distribution differences. According to the PSI empirical value: PSI values between 0.25-1.0 are generally features with large distribution differences; PSI values between 0.1-0.25 are generally features with average distribution differences; PSI values between 0-0.1 are generally features with small distribution differences. Here, the execution entity can set the preset range to 0.25-1.0. At this time, the execution entity can determine the input features with PSI values between 0.25-1.0 as the key features of emails.
[0034] Step 4: for each input feature in the target feature, determine the first difference between the mail feature input data set and the mail feature output data set under the input feature, wherein the first difference is used to characterize the degree of differentiation of the input and output data.
[0035] For example, the execution subject may determine the first difference between the mail feature input data set and the mail feature output data set under the input model feature by the following steps:
[0036] First, based on the index coefficient of the input feature, the above-mentioned distributed extreme gradient boosting tree model predicts the score of each email feature replacement data in the above-mentioned email feature replacement data set, and determines the total replacement prediction score of the above-mentioned email feature replacement data set under the input feature. For example, after each email feature replacement data is input into the XGB tree model, the prediction score of the email feature replacement data will be obtained. After multiplying the prediction score of each email feature replacement data with the index coefficient of each input feature, the specific prediction score of each email feature replacement data under the input feature can be obtained. In this way, the sum of the feature prediction scores of all email feature replacement data in the email feature replacement data set can be the total replacement prediction score of the replacement data set under each input feature.
[0037] Second, according to the index coefficient of the input feature, the distributed extreme gradient boosting tree model predicts the score of each email feature swap-out data in the email feature swap-out data set to determine the total predicted score of the email feature swap-out data set under the input feature.
[0038] Third, the difference between the total predicted score of the input feature and the total predicted score of the output feature is taken as the first difference.
[0039] Fourth, in response to determining that the data volume of the above-mentioned email feature swap-in data set is the same as the above-mentioned email feature swap-out data set, determine the average contribution of the input feature on the above-mentioned email feature swap-in data set as the swap-in contribution average, and determine the average contribution of the input feature on the above-mentioned email feature swap-out data set as the swap-out contribution average.
[0040] Fifth, the difference between the average value of the input contribution and the average value of the output contribution of the input feature is taken as the first difference.
[0041] The fifth step is to select a target number of input features from the target features according to the first difference as key features of emails for distinguishing the data to be swapped in and out.
[0042] Thus, the feature data of the input and output data sets on the input model features can be analyzed, so as to explain which features play an important role in causing the input and output. Specifically, training data can be generated based on the feature data belonging to the input model features in the email feature input data set and the email feature output data set, so as to train the preset classification model. Since only the input and output data are considered, only these data are used to train the model, so that the model can fully learn to regress the input and output data, reducing or avoiding the influence of other data. Then, based on the trained preset classification model, the index data of each input model feature of the decision model can be determined. Thus, with reference to the index data, the target features with higher importance among these input model features can be determined. These target features are also the prediction results of the preset classification model, that is, the features predicted as input or output data and with greater contribution. In this way, based on the target features, the email feature input data set and the email feature output data set, the key features of the email that cause the decision model to generate input and output data can be analyzed, that is, which features play an important role in the input and output can be determined.
[0043] Step 102 , extracting email key features of each email message in the email message set according to the key email features corresponding to the pre-trained initial phishing email detection network model, to obtain an email key feature set.
[0044] In some embodiments, the execution subject may extract the email key features of each email message in the email message set according to the key email features corresponding to the pre-trained initial phishing email detection network model to obtain the email key feature set. The initial phishing email detection network model is a decision model. The key email features may refer to the various features in the email used to indicate whether the email is a phishing email. For example, the key email features may include features such as email type, email server, sender address, body and attachments. That is, the various features of the corresponding key email features included in each email message may be extracted as email key features.
[0045] Step 103: construct an email training sample set based on the above email key feature set and each feature label.
[0046] In some embodiments, the execution subject may construct an email training sample set based on the email key feature set and each feature label. The email training sample includes: email key features and feature labels. For example, for each email key feature, the email key feature and the corresponding feature label may be combined into an email training sample.
[0047] Step 104, for each email feature type, the email training sample set is classified according to each email sub-feature corresponding to the email feature type included in the email training sample set to generate a type feature corresponding to the email feature type for each type of email training sample.
[0048] In some embodiments, the execution subject may classify the email training sample set for each email feature type according to each email sub-feature corresponding to the email feature type included in the email training sample set, so as to generate a type feature corresponding to the email feature type for each type of email training sample. The email key feature corresponds to a feature domain. The email feature type may be a feature type name corresponding to the email key feature. For example, the execution subject may classify the email training sample set using a preset binning algorithm to generate a type feature corresponding to the email feature type for each type of email training sample. The type feature may refer to a binning feature. The preset binning algorithm may be: WOE (Weight of Evidence) binning, equal frequency binning. The type feature may characterize the quality of a type of email training sample. For example, the type feature may be a WOE value, and the larger the WOE value, the better the quality of the sample. For example, the email feature type may include: personal email, corporate email, news email, group email, business marketing email, etc., and may also be divided into: private letters, official emails, business emails, spam emails, virus emails, etc.
[0049] Step 105 : Select a preset number of email training samples from the email training sample set as an email training sample group.
[0050] In some embodiments, the execution subject may select a preset number of email training samples from the email training sample set as the email training sample group. A preset number of email training samples may be randomly selected from the email training sample set as the email training sample group.
[0051] Step 106: train the initial phishing email detection network model according to the email training sample group and the various type features corresponding to the email training sample group to obtain a trained phishing email detection network model.
[0052] In some embodiments, the execution subject may train the initial phishing email detection network model according to the email training sample group and the various types of features corresponding to the email training sample group to obtain a trained phishing email detection network model. The initial phishing email detection network model may include: a feature embedding layer, a convolution layer, and an output layer. The feature embedding layer may be used to embed the various types of features included in the sample. For example, the embedding process may be one-hot encoding. The convolution layer may be used to learn feature weights according to the loss of the network. For example, the convolution layer may be a squeeze layer. The output layer may be used to output the phishing email detection result.
[0053] In an actual application scenario, the above execution entity can train the initial phishing email detection network model through the following steps:
[0054] The first step is to select a target email training sample from the email training sample group. For example, an email training sample can be randomly selected from the email training sample group as the target email training sample.
[0055] In the second step, each type of feature corresponding to each email feature type of the target email training sample is input into the feature embedding layer included in the initial phishing email detection network model to obtain the corresponding email feature vector.
[0056] In the third step, the email feature vector is input into the convolution layer included in the initial phishing email detection network model to obtain the corresponding email feature convolution vector. First, the email feature subvectors corresponding to the same feature domain in the email feature vector are summed to obtain the email feature domain vectors corresponding to the feature domains. Among them, the email feature subvector can be a vector corresponding to an email feature type in the email feature vector. The email feature subvectors of a single feature domain can be summed to characterize the feature domain. Second, for each email feature domain vector, the email feature domain vector is input into the fully connected layer to obtain a cross email feature domain vector. Among them, the dimension of the cross email feature domain vector is the same as the dimension of the email feature domain vector. The fully connected layer can include two sub-fully connected layers. The first fully connected layer can be used to perform feature crossover to achieve dimensionality reduction. The second fully connected layer can be used to restore the cross-linked vector to the dimension when it is input into the fully connected layer. Third, each cross email feature domain vector obtained is normalized to obtain a vector weight matrix. First, each cross-email feature domain vector in each cross-email feature domain vector can be normalized using an activation function to obtain each normalized vector. Then, each normalized vector can be combined into a vector weight matrix. Fourth, the product of the vector weight matrix and the email feature vector is determined as the corresponding email feature convolution vector.
[0057] The fourth step is to input the above email feature convolution vector into the output layer of the initial phishing email detection network model to obtain initial phishing email detection information.
[0058] Step 5: Determine the loss value between the initial phishing email detection information and the corresponding feature label. The loss value between the initial phishing email detection information and the corresponding feature label can be determined by a preset loss function. For example, the loss function can be a cross entropy loss function or a hinge loss function.
[0059] In a sixth step, in response to determining that the above loss value is less than or equal to the preset loss value, the initial phishing email detection network model is determined as the trained phishing email detection network model.
[0060] Therefore, the email features are converted into limited type features by classification, so that missing values can be classified separately without assumptions, reducing the impact of filling missing values on the model effect. Also, because the type features are involved in model training, the loss of information after classification can be reduced. Because important email features are screened out during model training, the dimension of the data can be reduced when training complex formal models. This can avoid directly using models with larger capacity and higher complexity, thereby reducing the consumption of computing resources and storage resources, reducing the difficulty of model learning, and shortening the convergence time.
[0061] Step 107, in response to receiving the target email, input the target email into the phishing email detection network model to obtain a phishing email detection result.
[0062] In some embodiments, the execution subject may, in response to receiving a target email, input the target email into the phishing email detection network model to obtain a phishing email detection result. The phishing email detection network model may be a pre-trained neural network model that takes email as input and outputs the phishing email detection result. For example, the phishing email detection network model may be a decision model (decision tree model). The phishing email detection network model may also be a trained convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory network (LSTM), gated recurrent unit (GRU) and other models. The phishing email detection result may be an indication of whether the target email is a phishing email. The target email is an email that has just been received and has not been opened.
[0063] Step 108: In response to the phishing email detection result indicating that the target email is a phishing email, intercept the target email.
[0064] In some embodiments, the execution subject may intercept the target email in response to the phishing email detection result indicating that the target email is a phishing email, that is, block the target email and prohibit opening the email.
[0065] The present application also provides a computer device 200. Figure 2 As shown, the computer device 200 includes: a bus 201, a processor 202, a memory 203 and a communication interface 204. The processor 202, the memory 203 and the communication interface 204 communicate with each other through the bus 202. The computer device 200 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computer device 200.
[0066] The bus 202 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 The bus 202 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 202 may include a path for transmitting information between various components of the computer device 200 (for example, the memory 202, the processor 203, and the communication interface 204).
[0067] The processor 203 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).
[0068] The memory 202 may include a volatile memory, such as a random access memory (RAM). The memory 202 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).
[0069] The memory 202 stores executable program codes, and the processor 202 executes the executable program codes to respectively implement the functions of the aforementioned acquisition module, sampling module, determination module, and mixing module, thereby implementing the aforementioned phishing email identification method based on the email detection network model. That is, the memory 202 stores instructions for executing the aforementioned phishing email identification method based on the email detection network model.
[0070] The communication interface 204 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computer device 200 and other devices or a communication network.
[0071] An embodiment of the present application also provides a chip, which includes a processor and a data interface. The processor reads instructions stored in a memory through the data interface to execute the above-mentioned phishing email identification method based on the email detection network model.
[0072] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute the above-mentioned phishing email identification method based on the email detection network model.
[0073] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0074] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the protection scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for identifying phishing emails based on an email detection network model, the method comprising: Acquire an email information set and a feature tag corresponding to each email information in the email information set; Extracting email key features of each email message in the email message set according to key email features corresponding to the pre-trained initial phishing email detection network model to obtain an email key feature set, wherein the initial phishing email detection network model is a decision model; Constructing an email training sample set according to the email key feature set and each feature label, wherein the email training sample includes: email key features and feature labels; For each email feature type, classifying the email training sample set according to each email sub-feature corresponding to the email feature type included in the email training sample set, so as to generate a type feature corresponding to the email feature type for each type of email training sample; Selecting a preset number of email training samples from the email training sample set as an email training sample group; Training the initial phishing email detection network model according to the email training sample group and various type features corresponding to the email training sample group to obtain a trained phishing email detection network model; In response to receiving a target email, inputting the target email into the phishing email detection network model to obtain a phishing email detection result; In response to the phishing email detection result indicating that the target email is a phishing email, the target email is intercepted.
2. The phishing email identification method based on the email detection network model according to claim 1 is characterized in that: The initial phishing email detection network model is trained according to the email training sample group and each type feature corresponding to the email training sample group to obtain a trained phishing email detection network model, including: Selecting a target email training sample from the email training sample group; Inputting each type of feature corresponding to each email feature type of the target email training sample into a feature embedding layer included in the initial phishing email detection network model to obtain a corresponding email feature vector; Inputting the email feature vector into a convolutional layer included in the initial phishing email detection network model to obtain a corresponding email feature convolutional vector; Inputting the email feature convolution vector into the output layer of the initial phishing email detection network model to obtain initial phishing email detection information; Determining a loss value between the initial phishing email detection information and a corresponding feature tag; In response to determining that the loss value is less than or equal to the preset loss value, the initial phishing email detection network model is determined as a trained phishing email detection network model.
3. The phishing email identification method based on the email detection network model according to claim 1 is characterized in that: Before extracting the email key features of each email message in the email message set according to the key email features corresponding to the pre-trained initial phishing email detection network model to obtain the email key feature set, the method further includes: Generate email feature training data based on feature data belonging to the input model features in the email feature swap-in data set and the email feature swap-out data set, wherein the input model features are the input model features of the initial phishing email detection network model with the swap-in and swap-out phenomena; Training a preset classification model using the email feature training data, and determining the index data of each input feature of the initial phishing email detection network model based on the trained preset classification model, wherein the index data is used to characterize the importance of the feature; According to the determined indicator data, a set number of input features are selected from each input feature of the initial phishing email detection network model as target features; Based on the target feature, the email feature swap-in data set and the email feature swap-out data set, key features of emails that cause the initial phishing email detection network model to generate swap-in and swap-out data are analyzed.
4. A computer device, wherein: The computer device comprises a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 3 are implemented.
5. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.
