CN100561988C - A kind of method and system of anti-rubbish mail - Google Patents
A kind of method and system of anti-rubbish mail Download PDFInfo
- Publication number
- CN100561988C CN100561988C CNB2006100339800A CN200610033980A CN100561988C CN 100561988 C CN100561988 C CN 100561988C CN B2006100339800 A CNB2006100339800 A CN B2006100339800A CN 200610033980 A CN200610033980 A CN 200610033980A CN 100561988 C CN100561988 C CN 100561988C
- Authority
- CN
- China
- Prior art keywords
- sample set
- spam
- level
- mail sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Abstract
The present invention is applicable to computer communication field, a kind of method and system of anti-rubbish mail are provided, said method comprising the steps of: A. handles level by first and according to the first identification principle original e-mail sample set is discerned processing, exports the final recognition result of non-spam and the first mail sample set; B. by second handle level according to the second identification principle to the described first mail sample set and append the mail sample set and further discern processing, export the final recognition result of non-spam and the second mail sample set; C. the described second mail sample set is exported as the final recognition result of spam, perhaps continued to input to and export final recognition result of non-spam and the final recognition result of spam after next processing level is discerned processing.The present invention takes that the mail sample is carried out multistage identification and handles, and skewed popularity identification and specific aim correction is combined, and further by feedback mechanism, improved the interception rate of spam, has reduced the False Rate of spam.
Description
Technical field
The invention belongs to computer communication field, relate in particular to a kind of method and system of anti-rubbish mail.
Background technology
At present aspect Email, spam spread unchecked be one can not be ignored, problem demanding prompt solution.A large amount of spams not only expends user's processing time, and the precious resources of waste mailing system, hindered the process that the user obtains useful information.Spam transmission technology and anti-spam technologies are constantly growing up in the antagonism, anti-spam technologies mainly adopt " study=〉 classification=feedback " frame structure that combines, wherein learning process can be finished also and can be finished by algorithm by the people, the method of taking comprises that Mail Contents filters and transmission behavior filtration, and extensively adopt the means of machine learning, thereby has the ability of continuous self-study and corrigendum.The spammer is also in the transmission means of bringing in constant renewal in them simultaneously, technical characterstic at the system of anti-rubbish mail is taked jamming countermeasure, for example change the content of spam, the boundary etc. of fuzzy spam and normal email painstakingly, make traditional anti-rubbish mail means DeGrain or inefficacy.
The existing algorithm that is used for anti-rubbish mail has multiple, based on these algorithms different anti-rubbish mail means are arranged, for example rule-based keyword string is filtered means, is sent the behavior control devices based on interception means, flow restriction etc. such as classification means of identification, sender's black and white lists such as algorithm of adding up such as Bayesian (Bayes), these methods can be used separately also and can be used in combination, because the mail sample of original input does not change in each processing links, its processor of comprehensively regarding a spam as can be treated usually when being used in combination.
Above-mentioned common spam identifying has following common trait:
1) single data promptly all are that single data set is as public input
2) uniprocessing
3), after handling, obtain corresponding recognition result for new mail
Adopt such scheme, no matter the content with spam can effectively operate as input as the behavior that input still sends rubbish, such as the anti-spam technologies based on Bayesian of extensive use a few days ago.
When by above-mentioned prior art spam being handled, the ratio of spam and normal email is away from equilibrium valve, and spam often far away more than good mail, reaches 10: 1 ratio, on the large-scale mailing system even higher.Simultaneously, when same or analogous mail sample appears at the quality set simultaneously, directly influence the differentiation result of mail, cause the interception rate of spam to descend and the False Rate rising.Address the above problem and need manually keep the balance sample collection, but can not truly represent the mail sample distribution situation of nature, this moment to repeat, the weighting of similar mail or subtract difficult setting of numerical value of power.
The root that the problems referred to above occur comes from currently used spam recognition methods has strictness to the balance of sample requirement.For middle-size and small-size mailing system, the False Rate rising problem of bringing thus still can be accepted, but for large-scale mailing system, after number of users reaches certain rank, spam is more serious to the situation that identical, similar mail belongs to spam and normal email simultaneously with normal email situation out of proportion, the criterion of adding different people is not quite similar, and only consider the identifying of spam to be confused with regard to being enough to from user's feedback information this moment.
Summary of the invention
The object of the present invention is to provide a kind of spam recognition system, it is strict to the balance of sample when spam being discerned processing to be intended to solve prior art, the problem that causes spam erroneous judgement rate to rise.
Another object of the present invention is to provide a kind of spam recognition methods.
The present invention is achieved in that
A kind of method of anti-rubbish mail said method comprising the steps of:
A. handle level by first and the original e-mail sample set is discerned processing, export the final recognition result of non-spam and the first mail sample set according to the first identification principle;
B. by second handle level according to the second identification principle to the described first mail sample set and append the mail sample set and further discern processing, export the final recognition result of non-spam and the second mail sample set;
C. the described second mail sample set is exported as the final recognition result of spam, perhaps continued to input to and export final recognition result of non-spam and the final recognition result of spam after next processing level is discerned processing;
The described first identification principle is used for the mail sample of input is carried out skewed popularity identification, and the described second identification principle is used for described first identification error of handling level is corrected;
It is that the balance of input data is reacted responsive recognizer that the algorithm that adopts is handled in described identification.
Described method further comprises:
D. the identification error of each being handled level feeds back to this processing level and last and handles level, adjusts the identification principle of respective handling level.
The described mail sample set that appends derives from the mail sample set that described original e-mail sample set or last is handled level output.
Mail in the described first mail sample set or the second mail sample set is a spam.
Different processing levels adopts identical or different recognizers.
Each is handled level and all is positioned at server end.
Described first handles level is positioned at server end, and described second handles level is positioned at client;
Described second handles level comprises a plurality of son processing levels, and each son is handled at least one mail account of level correspondence.
A kind of system of anti-rubbish mail, described system comprises:
First order processor is used for according to the first identification principle original e-mail sample set being discerned processing, exports the final recognition result of non-spam and the first mail sample set;
Second level processor, be used for according to the second identification principle to the described first mail sample set and append the mail sample set and further discern processing, export the final recognition result of non-spam and the second mail sample set, and with the described second mail sample set as the output of the final recognition result of spam, perhaps continue to input to next and handle level and discern and handle back output final recognition result of non-spam and the final recognition result of spam;
The described first identification principle is used for the mail sample of input is carried out skewed popularity identification, and the described second identification principle is used for described first identification error of handling level is corrected;
It is that the balance of input data is reacted responsive recognizer that the algorithm that adopts is handled in described identification.
Described first order processor further comprises:
Mail sample storehouse is used to receive the original e-mail sample set;
The mail recognition processing module is used for according to the first identification principle original e-mail sample set in described mail sample storehouse being discerned processing, exports the final recognition result of non-spam and the first mail sample set.
Described second level processor further comprises:
Mail sample storehouse is used to receive the described first mail sample set;
The mail sample appends module, is used for appending corresponding mail sample according to the number of mail of the described first mail sample set;
Append the sample storehouse, be used to receive described mail sample and append the mail sample set that module is appended;
The mail recognition processing module, be used for described first mail sample set and the described mail sample set that appends further being discerned processing according to the second identification principle, export the final recognition result of non-spam and the second mail sample set, and with the described second mail sample set as the output of the final recognition result of spam, perhaps continue to input to next stage and discern and handle back output final recognition result of non-spam and the final recognition result of spam.
The described mail sample set that appends derives from the mail sample set that described original e-mail sample set or last is handled level output.
Mail in the described first mail sample set or the second mail sample set is a spam.
Described first order processor and same recognizer of second level processor adopting or different recognizers.
Described first order processor and second level processor all are positioned at server end.
Described first order processor is positioned at server end, and described second level processor is positioned at client;
Described second level processor comprises a plurality of sub-processors, corresponding at least one mail account of each sub-processor.
The present invention takes that the mail sample is carried out multistage identification and handles, and skewed popularity identification and specific aim correction is combined, and further by feedback mechanism, improved the interception rate of spam, has reduced the False Rate of spam.
Description of drawings
Fig. 1 is the implementation method schematic diagram of spam recognition training provided by the invention;
Fig. 2 is the structure chart of the system of anti-rubbish mail provided by the invention;
Fig. 3 is the realization schematic diagram of among the present invention anti-rubbish mail being handled when expanding to user level.
Embodiment
In order to make purpose of the present invention, technical scheme and advantage clearer,, the present invention is further elaborated below in conjunction with drawings and Examples.Should be appreciated that specific embodiment described herein only in order to explanation the present invention, and be not used in qualification the present invention.
The present invention takes the method for multistep treatment to the identification of spam, upper level is handled and is carried out the identification of skewed popularity, the next stage processor is then corrected targetedly at the recognition result that upper level is handled, simultaneously, further the error feedback that the corresponding levels are produced is handled to the corresponding levels and upper level, participates in Classification and Identification training process next time.The present invention is not limited to the balance of mail sample, can handle flexibly, has improved the interception rate of spam, has reduced False Rate.
Need to prove, the processing method that processing at different levels is adopted among the present invention is necessary for because the characteristics of core calculations formula or technical limitations are caused to the highstrung method of input data balancing, as sorting techniques such as Bayesian class, decision tree class, rules, or K-means (K-average), KNN (k-nearestneighbors method, the most contiguous algorithm) clustering method such as, or insensitive but just can bring the method for balance sensitivity through simple modification to the input data balancing originally.When adopting identical algorithms, input data set at different levels needs different with the ratio of spam and normal email; When adopting algorithms of different, allow input data set at different levels identical under the limiting case.After increasing the error feedback processing on this basis again, the system with anti-rubbish mail of directed deflection mistake and the corrigendum of orientation deflection is just set up.
Fig. 1 shows the realization principle of spam recognition training provided by the invention, is example with the two-stage, and certainly, the present invention is not limited in two-stage, and a plurality of processing ranks can be set as required.
In the present invention, processing at different levels are generally all adopted with a kind of algorithm, by having a mind to change the input mail of each grade processing and the parameter of training method changes its recognition result.The situation that other certain one-levels are handled or certain what processing makes mistakes is easily only handled in each grade processing, and therefore, the recognition result of any one-level can be set up a corrigendum level in principle.
Aspect application deployment, the present invention is not strict with the deployed position of processing at different levels, according to the different application demand mail server side can be set all, can partly be arranged on server end yet, and part is arranged on client, as Foxmail, and Outlook etc.
As one embodiment of the present of invention, the target that upper level is handled specifically can be by relaxing the criterion of identification of spam for obtaining good non-spam recognition result, and the interception rate that improves spam realizes.Certainly, target also can be for obtaining good spam recognition result, can selecting flexibly.
Upper level is handled the processing mode of when training the input mail being gathered:
A. with the mail that all can obtain, promptly the original e-mail sample set is as input;
B. will be wherein all mails that are considered to have the spam tendency all treat as spam;
C. the mail that will wherein appear at the set of non-spam and spam is simultaneously treated as spam.
The mail of exporting through the upper level processing has been divided into two parts: a part is reliable non-spam, enters final result; A part is a spam in addition, is taken as the non-spam of part that spam is treated and wherein may comprise, and this a part of mail is sent to next stage and handles and further to discern and judge.
Next stage is handled the spam recognition result of being handled the high error rate that sends over by upper level is corrected targetedly, identifies non-spam wherein again.
Next stage is handled the mail set of input when training and is made up of two parts: a part is that upper level is handled the spam that sends over; Another part is that next stage handle to adopt certain strategy to extract the part mail from the mail sample set of processing is discerned in original e-mail sample set or upper level input, as appending the sample storehouse, if for example present sample set that need further handle major part in theory all is misjudged spam, then needing to add a certain proportion of real spam does corresponding, this ratio can be adjusted (experience setting/recurrence testing setup) as required flexibly, and the target of adjustment is that final result of determination can be accepted.Two parts mail combines the training data set of handling as next stage.
The algorithm of next stage processing also can be selected flexibly, but the present invention's general employing in concrete the application caught up with the identical algorithm of one-level processing.
When systematic training, said process may need repeated multiple times, in the parameter of adjusting algorithms at different levels and after appending the content in sample storehouse, after dropping between a designation area, comprehensive interception rate, False Rate can stop learning process, model is preserved the participation practical application.
When concrete the application, also several process of feedbacks need be set at the wrong identification result of each grade processing, error result is turned back to different local (corresponding levels and upper levels) respectively participate in following training.For multistage spam recognition system, the later processing module of two-stage only need be handled upper level or what is gone up and wrong situation occur, and the set of its input sample generally can reduce step by step.
In sum, the present invention program has following distinguishing feature:
The first, input data sets at different levels close difference, and this reaches by having a mind in identifyings at different levels even painstakingly making the disequilibrium of importing the mail sample.
The second,, adopt the strategy of certain final recognition objective of deflection, or be partial to the recognition accuracy of non-spam or be partial to the recognition accuracy of spam to corresponding input data at different levels.
Three, can adopt identical or different method in the processing at different levels, mail is discerned.
Like this, the output result of all processing handled (serial or parallel) after, then the uneven inferior position of input sample can be become advantage.
Fig. 2 shows the structure of the system of anti-rubbish mail provided by the invention, for the ease of understanding, is example with the bi-level treatment.
The mail in input mail sample storehouse 101 is the original e-mail sample set, mail recognition processing module 102 is discerned spam by above-mentioned spam recognizer, as one embodiment of the present of invention, mail recognition processing module 102 loosens the spam criterion of identification, improve the interception rate of spam, from obtaining good non-spam recognition result.
The non-spam recognition result of output is as final mail recognition result after 102 identifications of mail recognition processing module are handled, and the spam recognition result of output continues to be input to the 200 correction property identifications of second level processor to be handled.
The spam recognition result that back output is handled in 102 identifications of mail recognition processing module is input to mail sample storehouse 201.The mail sample appends module 202 and is used for being input to according to the number of mail of the spam recognition result of first order processor 100 output is appended corresponding proportion from original e-mail sample or other mail samples mail sample and appends sample storehouse 203.Mail recognition processing module 204 is further discerned processing according to the principle of the spam recognition result correction property identification that first order processor 100 is exported to mail sample storehouse 201 and the mail that appends sample storehouse 203, further improve the interception rate of spam, export non-spam recognition result and spam recognition result respectively as final recognition result.Certainly, also can adopt more multi-level processor, for example third level processor, fourth stage processor ... the time, with the spam recognition result correction property processing of second level processor 200 outputs, with the recognition effect of further raising spam.At this moment, third level processor, fourth stage processor ... structure identical with the structure of second level processor 200.
Fig. 3 shows the present invention realization principle when expanding to user level is handled in spam identification, can the anti-rubbish module of final stage that meet its characteristic most be set at each user, realizing the requirement of customer personalized service, can be effectively the preference of different mail be realized anti-rubbish mail effect more accurately at different user.
In the present invention, other processor of any one grade can be a processor in logic, allows the sub-processor of a plurality of high correlations of encapsulation, corresponding at least one email account of each sub-processor.In the present embodiment, regard the integral body of all unique user aspects as second level processor exactly.
If continue to use traditional Bayesian algorithm as anti-spam technologies, present embodiment is then based on the system of a two-stage Bayesian anti-rubbish mail.Real fine rule represents that output needs next stage to continue the result who handles in each step among the figure; The result of dotted line representative output can fully trust, as final result; Real thick line is represented the wrong identification result is fed back.Identifying to spam among this embodiment is consistent with said process, repeats no more.
What need supplementary notes is, the mistake that second level processor recognizes can be passed through user, mailing system operating personnel, and perhaps spam gatherer/probe feeds back.The mistake of finding in the processor of the second level in the present embodiment is fed back to the corresponding levels and upper level respectively, participates in Classification and Identification model training process next time.Generally speaking, error sample is according to the place that different higher levels' significance level decision is returned, and the wrong feedback that only makes progress.
The above only is preferred embodiment of the present invention, not in order to restriction the present invention, all any modifications of being done within the spirit and principles in the present invention, is equal to and replaces and improvement etc., all should be included within protection scope of the present invention.
Claims (15)
1, a kind of method of anti-rubbish mail is characterized in that, said method comprising the steps of:
A. handle level by first and the original e-mail sample set is discerned processing, export the final recognition result of non-spam and the first mail sample set according to the first identification principle;
B. by second handle level according to the second identification principle to the described first mail sample set and append the mail sample set and further discern processing, export the final recognition result of non-spam and the second mail sample set;
C. the described second mail sample set is exported as the final recognition result of spam, perhaps continued to input to and export final recognition result of non-spam and the final recognition result of spam after next processing level is discerned processing;
The described first identification principle is used for the mail sample of input is carried out skewed popularity identification, and the described second identification principle is used for described first identification error of handling level is corrected;
It is that the balance of input data is reacted responsive recognizer that the algorithm that adopts is handled in described identification.
2, the method for anti-rubbish mail as claimed in claim 1 is characterized in that, described method further comprises:
D. the identification error of each being handled level feeds back to this processing level and last and handles level, adjusts the identification principle of respective handling level.
3, the method for anti-rubbish mail as claimed in claim 1 is characterized in that, the described mail sample set that appends derives from the mail sample set that described original e-mail sample set or last is handled level output.
4, the method for anti-rubbish mail as claimed in claim 1 is characterized in that, the mail in the described first mail sample set or the second mail sample set is a spam.
5, the method for anti-rubbish mail as claimed in claim 1 is characterized in that, different processing levels adopts identical or different recognizers.
6, the method for anti-rubbish mail as claimed in claim 1 is characterized in that, each is handled level and all is positioned at server end.
7, the method for anti-rubbish mail as claimed in claim 1 is characterized in that, described first handles level is positioned at server end, and described second handles level is positioned at client;
Described second handles level comprises a plurality of son processing levels, and each son is handled at least one mail account of level correspondence.
8, a kind of system of anti-rubbish mail is characterized in that, described system comprises:
First order processor is used for according to the first identification principle original e-mail sample set being discerned processing, exports the final recognition result of non-spam and the first mail sample set;
Second level processor, be used for according to the second identification principle to the described first mail sample set and append the mail sample set and further discern processing, export the final recognition result of non-spam and the second mail sample set, and with the described second mail sample set as the output of the final recognition result of spam, perhaps continue to input to next and handle level and discern and handle back output final recognition result of non-spam and the final recognition result of spam;
The described first identification principle is used for the mail sample of input is carried out skewed popularity identification, and the described second identification principle is used for described first identification error of handling level is corrected;
It is that the balance of input data is reacted responsive recognizer that the algorithm that adopts is handled in described identification.
9, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, described first order processor further comprises:
Mail sample storehouse is used to receive the original e-mail sample set;
The mail recognition processing module is used for according to the first identification principle original e-mail sample set in described mail sample storehouse being discerned processing, exports the final recognition result of non-spam and the first mail sample set.
10, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, described second level processor further comprises:
Mail sample storehouse is used to receive the described first mail sample set;
The mail sample appends module, is used for appending corresponding mail sample according to the number of mail of the described first mail sample set;
Append the sample storehouse, be used to receive described mail sample and append the mail sample set that module is appended;
The mail recognition processing module, be used for described first mail sample set and the described mail sample set that appends further being discerned processing according to the second identification principle, export the final recognition result of non-spam and the second mail sample set, and with the described second mail sample set as the output of the final recognition result of spam, perhaps continue to input to next stage and discern and handle back output final recognition result of non-spam and the final recognition result of spam.
11, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, the described mail sample set that appends derives from the mail sample set that described original e-mail sample set or last is handled level output.
12, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, the mail in the described first mail sample set or the second mail sample set is a spam.
13, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, described first order processor and same recognizer of second level processor adopting or different recognizers.
14, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, described first order processor and second level processor all are positioned at server end.
15, the system of anti-rubbish mail as claimed in claim 8 is characterized in that, described first order processor is positioned at server end, and described second level processor is positioned at client;
Described second level processor comprises a plurality of sub-processors, corresponding at least one mail account of each sub-processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2006100339800A CN100561988C (en) | 2006-02-23 | 2006-02-23 | A kind of method and system of anti-rubbish mail |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2006100339800A CN100561988C (en) | 2006-02-23 | 2006-02-23 | A kind of method and system of anti-rubbish mail |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101026593A CN101026593A (en) | 2007-08-29 |
| CN100561988C true CN100561988C (en) | 2009-11-18 |
Family
ID=38744497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2006100339800A Expired - Fee Related CN100561988C (en) | 2006-02-23 | 2006-02-23 | A kind of method and system of anti-rubbish mail |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100561988C (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101150756B (en) * | 2007-11-08 | 2010-05-19 | 电子科技大学 | A spam filtering method |
| CN101610461B (en) * | 2008-06-17 | 2012-12-12 | 朗讯科技公司 | Anti-spam system and method, and communication network |
| CN101388859B (en) * | 2008-09-16 | 2010-09-01 | 王玉冰 | System and method preventing junk mail |
| CN101917420B (en) * | 2010-08-04 | 2014-12-03 | 安徽天虹数码技术有限公司 | Behavior filtering method of job network behavior fire wall |
| CN104243501B (en) * | 2014-10-14 | 2017-04-12 | 四川神琥科技有限公司 | Filtering and intercepting method for junk mail |
| CN106021299B (en) * | 2016-05-03 | 2020-07-10 | Tcl科技集团股份有限公司 | Text dimension reduction feature vector determination method and device |
| CN111291272A (en) * | 2018-12-06 | 2020-06-16 | 阿里巴巴集团控股有限公司 | File identification method and device and electronic equipment |
| CN112803803B (en) * | 2021-01-29 | 2022-04-22 | 中国兵器工业集团第二一四研究所苏州研发中心 | Flexible multi-state switch control method and system based on fuzzy logic PI controller |
-
2006
- 2006-02-23 CN CNB2006100339800A patent/CN100561988C/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN101026593A (en) | 2007-08-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100561988C (en) | A kind of method and system of anti-rubbish mail | |
| CN106453033B (en) | Multi-level process for sorting mailings based on Mail Contents | |
| US8572184B1 (en) | Systems and methods for dynamically integrating heterogeneous anti-spam filters | |
| US7590694B2 (en) | System for determining degrees of similarity in email message information | |
| US7930351B2 (en) | Identifying undesired email messages having attachments | |
| CN103441924B (en) | A kind of rubbish mail filtering method based on short text and device | |
| Del Castillo et al. | A multistrategy approach for digital text categorization from imbalanced documents | |
| CN103166830B (en) | A kind of Spam Filtering System of intelligent selection training sample and method | |
| CN101604322B (en) | Decision level text automatic classified fusion method | |
| US20100077051A1 (en) | Phonetic Filtering of Undesired Email Messages | |
| CN101330473A (en) | Method and apparatus for filtrating network rubbish information supported by multiple protocols | |
| CN105871887B (en) | Client-based individual electronic mail filtering system and filter method | |
| CN102404249B (en) | Method and device for filtering junk emails based on coordinated training | |
| Tuteja et al. | Email Spam filtering using BPNN classification algorithm | |
| CN110442568A (en) | Acquisition methods and device, storage medium, the electronic device of field label | |
| CN102214320A (en) | Neural network training method and junk mail filtering method using same | |
| CN108090503B (en) | Online adjustment method and device for multiple classifiers, storage medium and electronic equipment | |
| CN101299729B (en) | Method for judging rubbish mail based on topological action | |
| CN109800852A (en) | A kind of multi-modal spam filtering method | |
| CN106156163A (en) | File classification method and device | |
| CN108446214A (en) | Test case evolution generation method based on DBN | |
| CN110505144A (en) | Process for sorting mailings, device, equipment and storage medium | |
| CN111414973A (en) | Classification framework based on generating extremely unbalanced data for a countermeasure network | |
| CN105337842B (en) | A kind of rubbish mail filtering method unrelated with content | |
| CN1991879A (en) | Filtration method of junk mail |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C57 | Notification of unclear or unknown address | ||
| DD01 | Delivery of document by public notice |
Addressee: Chen Yan Document name: Notice of application for publication of patent for invention and entry into the substantive examination procedure |
|
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20091118 Termination date: 20130223 |