CN105871887A - Client-side based personalized E-mail filtering system and method - Google Patents

Abstract

The invention discloses a client-based personalized email filtering system and filtering method. The system includes a receiving module, a filtering and updating module, and a display module; the receiving module receives emails and preprocesses emails; the filtering and updating module includes a database, Condition matcher and intelligent detection classifier; the database is a training data set; the condition matcher is used for users to set filter conditions, filter emails according to the filter conditions, and then use the classifier to detect and classify emails, and at the same time use the received emails to classify The training data set of the filter is updated in real time, thereby realizing personalized email filtering and classification; the display module displays the results of email filtering and classification. The technical solution provided by the invention not only has diversified filtering methods and good performance, but also meets the requirements of real-time and individualization.

Description

Client-based Personalized Email Filtering System and Filtering Method

technical field

The invention relates to mail filtering technology, in particular to a client-based personalized email filtering system and filtering method.

Background technique

Most of the current spam filtering methods are based on two feature extraction methods. Among them, one relies on traditional statistics, analyzes and processes the statistical information of the feature words to be selected, sorts them according to their distinguishability, and extracts feature words with good distinguishability; although this method can extract a large number of effective features , however, due to the lack of further processing of these features, the feature vector dimension is too high, which increases the computational complexity.

The other is based on the artificial immune system, combined with immune thinking, simulates the production process of biological antibodies, and extracts instructive features. However, this type of method focuses on the establishment of heuristic rules, and less use of statistical theory to analyze the validity of the extracted features.

The current spam filtering methods mostly use existing data sets for training, and it is difficult to update the data sets in real time based on received emails. Most of the spam filtering methods adopted by the existing mail clients are to filter on the server side, and then classify and display the mails on the client side. This method of filtering on the server side needs to collect the usage of many users before updating the mail data set, resulting in poor real-time performance. At the same time, because mail filtering is performed uniformly on the server side, the mail filtering effects of all users are similar or even the same, making it difficult to meet the personalized needs of users.

Contents of the invention

In order to overcome the deficiencies in the prior art above, the present invention provides a client-based personalized email filtering system and filtering method. By calculating the immune concentration characteristics, since the emails received by different users are different, the local client is used to In the method of training and learning, the training data set will be updated every time the user receives an email, thereby realizing personalized email filtering.

The technical scheme provided by the invention is:

A client-based personalized email filtering system, including a receiving module, a filtering and updating module, and a display module;

The receiving module is used to receive emails, then preprocess the received emails, and pass the preprocessing results to the filtering module;

Described filter and update module comprises database, condition matcher and intelligent detection classifier; Database is the training data set stored locally; Condition matcher is used for user to set filter condition, according to filter condition, the mail that receives is filtered; The intelligent detection classifier detects and classifies the received emails, obtains the classification of the received emails, and uses the received emails to update the training data set of the intelligent detection classifier in real time, so as to establish its unique training data for each user Set, so that the classifier used in intelligent detection varies from user to user, thereby realizing personalized email filtering and classification;

The display module displays the results of email filtering and classification.

The present invention specifically uses JAVA language programming to realize the above-mentioned client system; realizes classifier training and classification by calling the function library of WaikatoEnvironment for Knowledge Analysis (Weka). The filter conditions set by the user include keyword filter conditions, sender address filter conditions, and the like.

The present invention also provides a client-based personalized e-mail filtering method, which is divided into a training stage and a filtering stage. The method is based on the immune concentration feature, adopts a method of training and learning locally, and for each e-mail received by the user, through The real-time update of the local data set obtains a personalized training data set for each user, and realizes the personalized mail filtering requirements of different users, thereby solving the real-time and personalized problems of mail filtering; the specific steps are as follows:

1) In the training phase, perform the following steps:

11) For the existing email data set, according to the amount of information and the degree of tendency of words, two types of detector sets are generated, which are respectively normal mail detector set and spam detector set;

12) For the existing email data set, use step 11) to construct the detector set to construct the immune concentration feature vector, and obtain the immune concentration feature vector corresponding to each mail in the email data set;

13) Utilize the immune concentration feature vector corresponding to each mail obtained in step 12), train the classifier, and obtain the trained classifier model;

2) In the filtering stage, perform the following steps:

21) Preprocessing the received emails, parsing each received email to obtain the title, text, recipient address, sender address of the email, the title, recipient address, sender Address, setting filter conditions (including title filter conditions, sending and receiving address filter conditions, etc.), used to classify mails; word segmentation is carried out to the text, and each mail is divided into a plurality of characteristic words;

22) Classify and filter the received mail, perform the following operations:

221) For each received email, use the detector set constructed in step 11) to reconstruct each received email into a corresponding immune concentration feature vector, and obtain the immune concentration feature vector corresponding to each received email ;

222) Utilize the classifier model described in step 13) to classify the mail to obtain the classification result;

223) According to the classification result and the filter condition set by the user, filter the received mail to obtain the filter processing result;

23) Update and display in real time according to user interaction operations, including the following situations:

23a) When a received email is classified as spam, said email enters the "junk mail box";

23b) When the received mail is classified as normal mail, said mail enters the "inbox";

23c) When the user finds that there are normal emails in the spam box, or there are spam emails in the inbox, the user can manually reclassify the misclassified emails; perform word segmentation processing on the reclassified emails to obtain word segmentation, and transfer to Step 1) Use the word segmentation to update the detector set, and sequentially rebuild the immune concentration feature vector and train the classifier.

For above-mentioned filter method, further, step 11) the information content and the degree of tendency of described word are calculated by word screening method and degree of tendency calculating method respectively; Described word screening method is specifically:

For the existing email data set, the information gain I(t) of all feature words is calculated by formula 1, and all feature words are sorted according to the size of the information gain I(t), and the features in the top m% are sorted words are added to the gene pool; in the embodiment of the present invention, preferably, the value of m is 50.

In the above formula, P(C _i ) represents the frequency of documents of category C _i in the data set; P(t) represents the probability of documents containing feature word t in the data set; Indicates the probability of a document that does not contain a feature word t in the data set; P(C _i |t) indicates the probability that a document belongs to category C _i under the premise that the feature word t appears; Indicates the probability that the document belongs to category C _i under the premise that the feature word t does not appear.

The calculation of the tendency degree is specifically: for each characteristic word in the gene pool, calculate the frequency of occurrence of the characteristic word in spam and normal mail; when the frequency of occurrence of the characteristic word in normal mail When the frequency of the characteristic word in spam is greater than the frequency of occurrence of the characteristic word in the normal mail detector set; when the frequency of the characteristic word in spam is greater than the frequency of occurrence of the characteristic word in normal mail, the The feature word is recorded in the spam detector set; when the two frequencies are equal, the feature word is not included in any detector set. This results in two types of detector sets.

For the above-mentioned filtering method, further, step 12) described in the construction of the immune concentration feature vector, the specific method is: the different feature words of each mail in the email data set appear in the spam detector set and the normal mail detector set Let N represent the number of different feature words in each email, S represent the number of feature words that appear in the spam detector set in each email, and L represent the number of feature words that appear in the normal email detector set in each email The number of feature words; construct a two-dimensional vector, denoted as (S/N, L/N), as the immune concentration feature vector, thus obtain the immune concentration feature vector corresponding to each mail in the email data set.

For the above filtering method, further, the classifier adopts a support vector machine (SVM).

For the above filtering method, further, in the training process described in step 13), the parameters of the classifier are optimized using the quadratic programming method.

Compared with prior art, the beneficial effect of the present invention is:

Most of the existing spam filtering methods use existing data sets for training, and the data sets are rarely updated in real time according to the received emails. Because the method they adopt is to filter on the server side, and the data set in the server needs to collect the usage conditions of many users before it can be updated.

The spam filtering method provided by the invention can effectively extract the features of the mail by constructing the immune concentration feature vector for each mail, thereby improving the classification performance and improving the spam filtering effect. Based on the immune concentration feature method, on the basis of better filtering effect, according to the characteristics of the mails received by each user, a personalized mail detection classifier is established locally for each user, so as to realize a personalized A modernized spam filtering client. The client system also includes other rule-based filtering methods, such as whitelists, keywords, etc., which diversify the filtering methods and improve the overall performance of the system. The client provided by the present invention performs training and learning locally, and the user will update the training data set when receiving each email. The emails received by different users are different, so the real-time update of this local data set can realize the personalized training data set that is different for each user, so as to realize the personalized spam filtering effect among different users; Solve the problems of real-time and personalization.

To sum up, the technical scheme provided by the present invention, on the one hand, has a variety of filtering methods and good performance (indicators such as the accuracy rate, recall rate, and F measurement value of spam filtering can reach more than 98%); on the other hand, it can meet real-time performance and individual requirements.

Description of drawings

Fig. 1 is a flow chart of the filtering method based on the immune concentration feature provided by the present invention.

Fig. 2 is a structural block diagram of a spam client system based on immune concentration implemented by an embodiment of the present invention.

Fig. 3 is a screenshot of the main interface of the client system after login in the embodiment of the present invention.

Fig. 4 is a screenshot of the mail reading interface of the client system in the embodiment of the present invention.

Fig. 5 is a screenshot of the filtering function setting interface of the client system in the embodiment of the present invention.

detailed description

Below in conjunction with accompanying drawing, further describe the present invention through embodiment, but do not limit the scope of the present invention in any way.

The invention provides a spam filtering method based on the immune concentration feature, proposes a new immune concentration feature extraction method, and applies the method to an email client system. The system supports multiple accounts to log in at the same time, reads user emails from the mail server, extracts the concentration features of each email, and uses the classifier to generate corresponding email classification results. The spam filtering method based on the immune concentration feature can be divided into the training stage and the filtering stage. In the training stage, the training data set is input into the classifier, and the parameters of the classifier are learned and optimized, and finally the classifier with the optimal effect is obtained; the filtering stage Apply the trained classifier to the emails received by this client; the specific steps include:

S1) Use the existing email collection as a data set, extract the immune concentration feature vector therefrom, input the classifier and perform training and learning, and generate a classifier model; SVM is used as the classifier in the embodiment of the present invention;

S2) After each user receives the mail, the mail of each user is analyzed separately to obtain the title, text, address of the recipient and the sender of the mail;

S3) Segment the text of the mail, generate an immune concentration feature vector according to the text of the mail after word segmentation and the detector set, and use the classifier model generated in S1 to classify the mail.

The specific implementation process of the filtering method based on the immune concentration feature provided by the present invention is shown in Figure 1. Each received email is analyzed separately to obtain the email title, sender's address and email text. After parsing, the email title and sender address and other parts are filtered by matching the filter conditions set by the user, including keyword filtering, sender address filtering, etc.; after parsing the email body part, the immune concentration feature is constructed after word segmentation , to calculate the classification result. Finally, the filter conditions set by the user and the classification results of the classifier are combined to filter the mails in the client system uniformly. According to the filtering method provided by the present invention, the following examples establish a spam client system based on immune concentration, which is implemented using JAVA language programming, and calls Weka's function library to realize classifier training and classification. Fig. 2 is a structural block diagram of a spam client system based on immune concentration implemented by an embodiment of the present invention. The system mainly includes three modules: a receiving module, a filtering module and a display module. The receiving module preprocesses the received mail, and passes the preprocessing result to the filtering module. The filter module filters the emails received by users through filter conditions and intelligent detection and classification methods, and at the same time updates the classifier in real time to realize personalized classification. The display module displays the filtering results, and the spam enters the spam box. The specific implementation steps of the system are as follows:

The first step: build a detector set;

The detector set (detection set) is a set of detectors, which is divided into two types in the present invention, one is a spam detector set, and the other is a normal email detector set. Among them, by calculating the tendency of the feature words to the two types of emails, the feature words that are more likely to appear in spam emails are classified into the spam detector set, and the feature words that are more likely to appear in normal emails are classified into normal emails. collection of detectors.

In the generation stage of the detector set, the main work is to combine the word screening algorithm with the propensity function, according to the amount of information of the feature words (the information gain is used as the measure of the amount of information in this embodiment, see below for the specific calculation method) and propensity to generate an ensemble of detectors for both classes. specifically:

11) Word screening method: For the existing email data set, each feature word is obtained after word segmentation of the email body. For the word segmentation of the mail text, the specific implementation method in this embodiment is to use each Chinese character as a feature word, and each word as a feature word, for example, "city" is divided into two features of "city" and "city" word. After the word segmentation is completed, each email is divided into N feature words. Calculate the information gain I(t) of all feature words, the calculation formula is shown in formula 1, and sort all feature words according to the size of information gain I(t). Add the feature words ranked in the top m% in the information gain sorting to the gene pool, and the experiment proves that the effect is the best when m=50;

In the above formula, C _i represents the category of mail i (normal mail or spam); P(C _i ) represents the frequency of documents of C _i category (normal mail or spam) in the data set; P(t) represents the frequency of documents in the data set The probability of a document containing a feature word t; Indicates the probability of a document that does not contain a feature word t in the data set; P(C _i |t) indicates the probability that a document belongs to category C _i under the premise that the feature word t appears; Indicates the probability that the document belongs to category C _i under the premise that the feature word t does not appear.

12) Calculation of propensity: For each feature word in the gene pool, calculate the frequency of occurrence of the feature word in each detection set (in this embodiment, spam and normal mail).

The feature words that appear more frequently in spam will be included in the spam detector set D _S ; the feature words that appear more frequently in normal emails will be included in the normal email detector set D _L ; The feature words that appear more frequently in normal emails should belong to the spam detector set; the feature words that appear more frequently in normal emails should belong to the normal email detector set).

The second step: construct the immune concentration feature vector;

For the existing email data set, count the number of different feature words of each email appearing in the spam detector set DS _S and the normal email detector set _DSL . Let N represent the number of different feature words in each email, S represent the number of feature words in each email that appear in the spam detector set, and L represent the number of feature words in each email that appear in the normal email detector set. Then the constructed immune concentration feature vector is a two-dimensional vector: (S/N, L/N).

Step 3: Train the classifier

In the previous step, each email was reconstructed into the corresponding immune concentration feature vector, and the classifier was trained using these feature vectors. The classifier in this example is a support vector machine (SVM). During the training process, the parameters of the classifier model are optimized using the quadratic programming method.

Step 4: Preprocessing of received mail by the client system

Fig. 4 is the screenshot of the mail reading interface of the client system in the embodiment of the present invention. As shown in Fig. 4, after the client system receives the mail, it parses the mail to obtain the title, text, addressee and sender of the mail Address; the title and sending and receiving addresses can be filtered based on the filter conditions set by the user; after the text of the email is word-segmented, it is used for the classifier model trained in the previous step;

Step 5: The client system classifies and filters emails

In the previous step, each email in the client system was divided into multiple feature words. Enable the filtering function of the client system, as shown in Figure 5. At this time, use the detector set built in the first step again, and reconstruct each email into the corresponding immune concentration feature vector according to the method in the second step, and use the classifier model trained in the third step to classify the email sort. Finally, according to the classification results and the filter conditions set by the user on the title, sending and receiving addresses, etc. (such as matching whether the sender's address exists in the blacklist, or whether the title contains certain keywords), the client system The mail is filtered, and the result is displayed in the client as shown in Figure 3.

Step 6: Real-time update according to the characteristics of user interaction operation

The filtering result of the previous step is displayed, and the email classified as spam goes into the "junk mail box", and the normal email goes into the "inbox". However, when the user finds that there are normal emails in the junk mail box or spam emails in the inbox, the user can manually reclassify the misclassified emails. At the same time, after processing the word segmentation of these emails, jump back to the first step, use these word segmentation to update the detector set, and rebuild the immune concentration feature vector and train the classifier in turn. The update method of the detector set takes an email manually marked as spam by the user as an example: all the feature words in the email that do not belong to the normal email detector set are added to the spam detector set. Similarly, for emails manually marked as normal emails, all the feature words that do not belong to the spam detector set are added to the normal email detector set.

It should be noted that the purpose of the disclosed embodiments is to help further understand the present invention, but those skilled in the art can understand that various replacements and modifications are possible without departing from the spirit and scope of the present invention and the appended claims of. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the protection scope of the present invention is subject to the scope defined in the claims.

Claims (10)

1. A client-based personalized email filtering system is characterized in that it includes a receiving module, filtering and updating module, and a display module; The receiving module is used to receive emails, then preprocess the received emails, and pass the preprocessing results to the filtering module; Described filter and update module comprises database, condition matcher and intelligent detection classifier; Database is the training data set stored locally; Condition matcher is used for user to set filter condition, according to filter condition, the mail that receives is filtered, reuses The intelligent detection classifier detects and classifies the received emails to obtain the classification of the received emails; at the same time, it uses the received emails to update the local training data set of the intelligent detection classifier in real time, thereby realizing personalized email filtering and classification ; The display module displays the results of email filtering and classification. 2. The personalized email filtering system as claimed in claim 1, characterized in that, the JAVA language programming is used to realize the personalized email filtering system; the intelligent detection classifier is realized by calling the function library of Weka. 3. The personalized email filtering system according to claim 1, wherein the filter conditions set by the user include keyword filter conditions and sender address filter conditions. 4. A client-based personalized email filtering method, comprising a training phase and a filtering phase; the training phase inputs the training data set into a classifier, learns and optimizes the parameters of the classifier, and obtains an optimal classifier; In the filtering stage, the optimal classifier trained is used for the mail received in the client; the mail filtering method is based on the immune concentration feature, and obtains a personalized training data set for each user through the real-time update of the local data set of the client , to realize the personalized spam filtering requirements of different users; specifically, the following steps are included: 1) In the training phase, perform the following steps: 11) for the existing email data set, according to the amount of information and the tendency of word segmentation to generate a detector set; the detector set includes a normal mail detector set and a spam detector set; 12) For the existing email data set, use step 11) to construct the detector set to construct the immune concentration feature vector, and obtain the immune concentration feature vector corresponding to each mail in the email data set; 13) Utilize the immune concentration feature vector corresponding to each mail obtained in step 12), train the classifier, and obtain the trained classifier model; 2) In the filtering stage, perform the following steps: 21) Preprocessing the received emails, parsing each received email to obtain the title, text, recipient address, sender address of the email, the title, recipient address, sender address, setting filter conditions for mail classification; word segmentation for the text, each mail is divided into multiple feature words; 22) Classify and filter the received mail, perform the following operations: 221) For each received email, use the detector set constructed in step 11) to reconstruct each received email into a corresponding immune concentration feature vector, and obtain the immune concentration feature vector corresponding to each received email ; 222) Using the classifier model described in step 13) to classify the mail, the classification includes spam or normal mail, thus obtaining the classification result; 223) According to the classification result and the filter condition set by the user, filter the received mail, and further obtain a filtering result; the result is that the received mail is classified as spam or normal mail; 23) Real-time update and display according to user interaction operation. 5. mail filtering method as claimed in claim 4, is characterized in that, step 23) carries out real-time update and display according to user interaction operation and comprises following situation: 23a) When a received email is classified as spam, said email enters the "junk mail box"; 23b) When the received mail is classified as normal mail, said mail enters the "inbox"; 23c) When the user finds that there are normal mails in the "junk mail box" or spam mails in the "inbox", the user can manually reclassify the misclassified mails; perform word segmentation processing on the reclassified mails to obtain Segmentation, turn to the training phase of step 1), use the word segmentation to update the detector set, and rebuild the immune concentration feature vector and train the classifier in turn. 6. mail filtering method as claimed in claim 4, it is characterized in that, step 11) obtains detector set according to the information amount and the tendency degree of participle, the information amount and the tendency degree of described participle are calculated by word screening method and tendency degree respectively calculated by the method; The word screening method is specifically: For the existing email data set, the information gain I(t) of all feature words is calculated by formula 1, and all feature words are sorted according to the size of the information gain I(t), and the top m% features are sorted words added to the gene pool; $\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>$ (Formula 1) In the above formula, P(C _i ) represents the frequency of documents of category C _i in the data set; P(t) represents the probability of documents containing feature word t in the data set; Indicates the probability of a document that does not contain a feature word t in the data set; P(C _i |t) indicates the probability that a document belongs to category C _i under the premise that the feature word t appears; Indicates the probability that the document belongs to category C _i under the premise that the feature word t does not appear; The calculation of the tendency degree is specifically: for each characteristic word in the gene pool, calculate the frequency of occurrence of the characteristic word in spam and normal mail; when the frequency of occurrence of the characteristic word in normal mail When the frequency of the feature word in spam is greater than the frequency of the feature word in spam, the feature word is recorded in the normal mail detector set; when the frequency of the feature word in spam is greater than the frequency of the feature word in normal mail, This feature word is entered into the spam detector set; thus two types of detector sets are generated. 7. The mail filtering method according to claim 6, wherein the value of m is 50. 8. mail filtering method as claimed in claim 4, is characterized in that, step 12) described concrete method of constructing immune concentration feature vector is: Count the number of different feature words appearing in the spam detector set and the ham detector set for each email in the email data set; Let N represent the number of different feature words in each email, S represent the number of feature words that appear in the spam detector set in each email, and L represent the number of feature words that appear in the normal email detector set in each email Construct and obtain a two-dimensional vector, denoted as (S/N, L/N), as the immune concentration feature vector, thus obtaining the immune concentration feature vector corresponding to each email in the email data set. 9. The mail filtering method according to claim 4, wherein the classifier adopts a support vector machine (SVM). 10. The mail filtering method according to claim 4, characterized in that, in the training process of step 13), the parameter optimization of the classifier is performed using a quadratic programming method.