JP4780684B2 - Mass mail detection method and mail server - Google Patents

Description

  The present invention relates to spam processing of electronic mail, and in particular, unsolicited advertisements included in electronic mail delivered via electronic mail servers by operators operating large-scale electronic mail servers such as mobile phones and ISPs. The present invention relates to a mass mail detection method suitable for detecting a junk mail such as a mail server equipped with a mass mail detection method.

  With the spread of electronic mails, annoying mails that use electronic mails as a means of transport have increased and have become a social problem. Conventionally, as a means for preventing such annoying mail, there is a method of preparing a mechanism for detecting annoying mail in a terminal used for reception by an electronic mail recipient and automatically deleting the annoying mail by the mechanism. It was general.

  For example, SpamAssassin is a rule-based software, and bogofilter is a machine-learning software, both of which are mainly used as an effective mechanism among PC users. These softwares are shown in Non-Patent Documents 1 and 2, respectively.

http://www.au.spamassassin.org/presentations/SAGE_IE_2002/ http://bogofilter.sourceforge.net/bogofilter_man.html

  The above-described prior art is based on the premise that the recipient of electronic mail uses a receiving terminal having a certain level of information processing capability such as a PC, and a receiving terminal having a relatively low ability such as a mobile phone. Not suitable for. In order to support a relatively low-capacity receiving terminal such as a mobile phone, it is desirable to provide a mail server on the provider side with means for detecting mass mail.

  However, the above-described conventional technique has a problem in that a processing speed is low and a large-scale facility is required for use in a server. In addition, it is difficult to create mass mail detection rules and machine learning results that are common to a large number of users, and the maintenance costs for dealing with new types of spam are enormous. .

  SUMMARY OF THE INVENTION An object of the present invention is to provide a mass-mail detection method that does not require prior rule creation or learning and that operates at high speed, and a mail server equipped with the mass-mail detection method in order to solve the above-described problems. is there.

In order to achieve the above-described object, the present invention performs electronic mail pre-processing in accordance with electronic mail collection means for collecting electronic mail to be delivered and specifications when the electronic mail is displayed to a recipient. Pre-processing means, feature quantity conversion means for converting the pre-processed electronic mail into feature quantities, and mass mail detection means for detecting mass mail using the converted feature quantities, wherein the feature quantity conversion means is A partial character string is extracted from the body of the electronic mail, and a collection of hash values calculated from the partial character string is used as a feature amount.The mass mail detecting means includes a feature amount database and a pointer to the feature amount database. It is determined whether or not the hash value of the entered feature value is entered in the pointer. If it is entered, the feature value is used using the pointer. Accesses the database by comparing the hash value of the electronic mail that has already been registered in the feature database, to determine the similarity between the electronic mail of the new e-mail and registered, the hash value of more than a certain number of The matching electronic mail is determined as a similar electronic mail, and the similar electronic mail is determined as a mass mail when a predetermined number of similar electronic mails are detected.

  According to the present invention, an operator operating a large-scale electronic mail server such as a mobile phone or an ISP detects an unsolicited mail such as an unsolicited advertisement contained in the electronic mail delivered via the electronic mail server. Therefore, it is possible to provide a mass mail detection method suitable for the above.

1 is a block diagram showing a system configuration including the present invention. It is a flowchart which shows an example of the process procedure of an electronic mail collection means. It is explanatory drawing which shows an example of the process of a feature-value conversion means. It is a flowchart which shows an example of the process procedure of a feature-value conversion means. A Directed Mapped Cache, which is an example of a data structure used by mass mail detection means, is shown. It is a flowchart which shows an example of a process of a massmail detection means. It is a flowchart which shows one specific example of step S30 of FIG. It is a flowchart which shows one specific example of step S35 and S38 of FIG. It is explanatory drawing of the feature-value of the mail 1, 2, 3, 4 extracted sequentially. It is explanatory drawing which shows the data example of Directed Mapped Cache with respect to the mail 1. FIG. It is explanatory drawing which shows the data example of Directed Mapped Cache with respect to the mails 1 and 2. FIG. It is explanatory drawing which shows the data example of Directed Mapped Cache with respect to mail 1, 2, and 3. FIG. It is explanatory drawing which shows the data example of Directed Mapped Cache with respect to mail 1, 2, 3, 4. It is a block diagram which shows the principal part of other embodiment of this invention.

  Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing an example of a system configuration to which the present invention is applied.

  In the figure, 1 is, for example, a mail server group of a telecommunications carrier (provider), 2 is the Internet, 3 is a switching hub used for data distribution between them, and the mail server group 1 is connected to a user terminal 4. . Further, the mass mail detection device 5 of the present invention is connected to the switching hub 3.

  In the present embodiment, mass mail is detected using the mass mail detection device 5 from electronic mail distributed between the mail server group 1 and the Internet 2 using the SMTP protocol.

  The mass mail detection device 5 includes an electronic mail collection means 51, a feature amount conversion means 52 and a mass mail detection means 53. The electronic mail collecting means 51 collects electronic mail to be delivered, and may be an appropriate computer program. The feature quantity conversion means 52 converts the electronic mail collected by the electronic mail collection means 51 into a feature quantity, and may be an appropriate computer program. The mass mail detection means 53 detects mass mail using the converted feature quantity, and may be an appropriate computer program. Reference numeral 55 denotes a detection result mass mail.

  Next, the operation of this embodiment will be described. The electronic mail collecting means 51 analyzes the electronic mail delivery protocol flowing on the network, and extracts the electronic mail text from the electronic mail traffic flowing on the network. Next, the feature amount conversion means 52 calculates, for example, several hash values from the electronic mail text, and sets it as the feature amount of the mail. Finally, the past electronic mail stored by the mass mail detecting means 53 and the newly received electronic mail are compared using the feature amount, the similarity is determined according to a specific criterion, and they are similar. In this case, it is determined as a mass mail candidate (similar mail), and when a certain number of similar mails are detected, it is determined as a mass mail.

  FIG. 2 is a flowchart showing an example of the processing procedure of the electronic mail collecting means 51. A plurality of mails are delivered in parallel between the mail server group 1 and the Internet 2. Therefore, in step S10, the electronic mail collection means 51 receives a TCP packet by tapping with the switching hub 3. The received TCP packet is a mixture of a plurality of electronic mail information. In step S15, the electronic mail collection means 51 determines the type of packet. That is, it is determined whether the packet is a mail, if it is a mail, whether it is a new mail packet, a processing mail packet, or a processing mail end packet.

  If the received packet is for a new mail, the process proceeds to step S11 to initialize a new mail storage area. On the other hand, if the received packet is a packet that means the end of the mail being processed, the process proceeds to step S13, the mail text being processed is transmitted to the feature amount conversion means 52, and then the process proceeds to step S14 to be processed. Discard / release the mail storage area. If the received packet is a mail packet being processed other than the end, the process proceeds to step S12, and the contents of the mail included in the TCP packet are recorded in the mail storage area being processed. If it is determined in step S15 that the packet is other than mail, the process ends without performing any processing. In FIG. 2, the end (end) is described for the sake of convenience, but it is apparent that the processes in steps S10 to S14 are continuously repeated without ending the actual process.

  FIG. 3 is an explanatory diagram showing an example of processing of the feature quantity conversion means 52, and FIG. 4 is a flowchart showing an example of processing procedures of the feature quantity conversion means 52.

  In the present embodiment, a set of hash values of a character string (for example, 4 characters) having a predetermined length L is used as the feature amount of the mail text. Specifically, according to the procedure of FIG. 4, first, in step S21, L characters are extracted sequentially from the beginning of the mail body 100, and the hash value is calculated. Next, the process proceeds to step S22, where the calculated hash values are sorted, and the first N (for example, 100) values are transmitted to the mass mail detection means 53 as feature quantities.

  For example, as shown in FIG. 3, if the length L is 4 and the mail text 100 is “text of mail text”, “mail book” (101 in FIG. 3), “il text” , Hash values 201 to 202 such as “le text” and “text of text” (102 in FIG. 3) are calculated (step S21 in FIG. 4). Next, after the hash values 201 to 202 are sorted, the first N pieces are transmitted as the feature quantity 200 to the mass mail detection means 53 (step S22 in FIG. 4). The hash value can be represented by a 64-bit integer, for example.

  FIG. 5 shows an example of the data structure used by the mass mail detection means 53. Reference numeral 300 denotes a data structure Directed Map Cache (hereinafter referred to as DMC), that is, a management map cache method for preferentially storing electronic mail that is frequently delivered in the feature amount database 310. The DMC 300 includes a feature quantity database 310 and a pointer 311 to the feature quantity database 310. For each electronic mail, the feature quantity database 310 stores the feature quantity (hash value 1 to N), the number of similar mails of the corresponding electronic mail, and the number of pointers referring to the corresponding mail in the entry of the pointer 311 (the number of DMC references). ) Is a data structure on a computer. Each of the pointers 311 can be expressed by 64 bits, for example.

  When the electronic mail collecting means 51 extracts the electronic mail, the feature amount converting means 52 calculates the feature amount 200 (see FIG. 3) from the electronic mail text, and finally the mass mail detecting means 53 follows the procedure illustrated in FIG. Similar mails that are delivered in large quantities are detected as mass mail. Specifically, the feature quantity 200 calculated by the feature quantity conversion means 52 for one electronic mail has N hash values (N is a positive integer), but the mass mail detection means 53 is assigned to each mail. On the other hand, the processing from step S31 to step S41 is repeated at most N times according to the procedure of FIG.

  In step S <b> 30, based on the feature amount 200, it is determined whether there is already a mail similar to the electronic mail collected by the electronic mail collection means 51. A specific example of this process will be described with reference to the flowchart of FIG.

  In step S301, the number m indicating the number of the feature quantity 200 is set to 1, and in step S302, the m-th hash value in the feature quantity 200 of the new mail is extracted. In step S303, it is determined whether or not the hash value is entered in the pointer 311. If this determination is affirmative, the process proceeds to step S304 to determine the similarity to the entry in the feature amount database 310 referenced from the current pointer 311. For example, if there is 80% similarity, it is determined that the mail is similar, and if it is less than 80%, it is determined that the mail is dissimilar. In step S305, it is determined whether or not m = N is established. If the determination is negative, the process proceeds to step S306, and 1 is added to m. Next, returning to step S302, the second hash value is extracted. Hereinafter, similarly, the above-described processing is repeatedly performed, and when the determination in step S305 becomes affirmative, the processing in step S30 ends.

  The similarity determination in step S304 uses, for example, the number of new electronic mail hash values 200 (see FIG. 3) and the hash values in the feature amount database 310 that match. For example, when the number N of feature quantities is 100 and the similarity threshold is 80%, it is determined that the mails are similar if 80 hash values match. It is preferable to sort the hash values in advance in order to speed up the number of coincidence measurement processes. In FIG. 7, the similarity is determined for N hash values. However, it is not always necessary to determine the similarity for N hash values, and the similarity may be determined using fewer hash values.

  Returning to FIG. 6, the description is continued. In step S31, m is set to m = 1 again, and in step S32, the m-th hash value in the feature amount 200 is extracted. Next, the process proceeds to step S33, where it is determined whether or not the m-th hash value is a hash value of similar mail. If this determination is negative, that is, a new electronic mail (non-similar mail), the similar mail is not stored in the feature quantity database 310, so the process proceeds to step S 34, and the feature quantity of the corresponding mail is the feature quantity database 310. Is registered as a new entry. Specifically, the feature quantity 200 (see FIG. 3) of the new electronic mail is stored as hash value 1 to hash value N of the feature quantity database 310. In step S35, the contents of the DMC 300 are updated with the vector value of the feature amount.

  If the determination in step S33 is affirmative, that is, if there is a similar mail, the process proceeds to step S37. In step S37, 1 is added to the number of similar mails (see FIG. 5) already stored in the feature amount database 310. In step S38, the contents of the DMC 300 are updated with the vector values stored in the feature amount database 310. The step S38 is the same process as the step S35, and a specific example thereof will be described later with reference to FIG.

  In step S39, it is determined whether or not the number of similar mails is greater than or equal to a predetermined value S. If it is greater than or equal to S, the process proceeds to step S40 and the corresponding mail is determined to be spam. On the other hand, when the determination in step S39 is negative, the process proceeds to step S36. In step S36, it is determined whether or not m = N. If this determination is negative, the process proceeds to step S41, and 1 is added to m. Then, the operation from step S32 is repeated again.

  FIG. 8 illustrates the procedure of the DMC update process in steps S35 and S38 in FIG. The mass mail detecting means 53 first determines in step S351 when processing each electronic mail. That is, it is determined whether or not the hash value refers to an old entry in the feature amount database 310 from the current pointer 311. When this determination is negative, the process proceeds to step S352, where the corresponding entry of the pointer 311 is set so as to point to the entry of the new feature quantity database 310, and 1 is added to the number of references in the feature quantity database.

  On the other hand, when the determination in step S351 is affirmative, that is, when the hash value refers to an old entry in the feature amount database 310 from the current pointer 311, the process proceeds to step S353, where the hash value is an entry of its own. It is determined whether or not it is referenced. That is, it is determined whether or not the hash value is included in the similar mail. If this determination is affirmative, no processing is performed and the processing of FIG.

  When the determination in step S353 is negative, that is, when the hash value is not included in the similar mail, the process proceeds to step S354, and the DMC of the old entry in the feature amount database 310 referenced from the current pointer 311 is obtained. Subtract 1 from the number of references. Next, the process proceeds to step S355, where it is determined whether or not the DMC referenced number is zero. When this determination is affirmative, the process proceeds to step S356, and the previous mail entry whose DMC reference count is 0 is deleted from the feature amount database 310. When the determination in step S355 is negative, the process proceeds to step S352, where the corresponding entry of the pointer 311 is set so as to point to the entry of the new feature quantity database 310, and the number of reference in the feature quantity database is incremented by one.

  According to the above processing, mails with many similar mails are frequently started from step S38 in FIG. 6 and the update process in FIG. 8 (more specifically, step S352) moves, so the DMC reference count becomes zero. Although it is difficult, if there is no similar mail, the number of DMC referenced is reduced with the lapse of time by overwriting the data hit by the hash value, and finally deleted in step S356.

  Next, specific examples of the operations shown in FIGS. 6 to 8 will be described with reference to FIGS. 9 to 13. Now, it is assumed that new mails are collected in the order of mails 1, 2, 3, and 4 as shown in FIG. 9 through the Internet, and the feature amount of the mail (feature amount 200 in FIG. 3). However, for mail 1, hash values h1, h2, h3, and h4, for mail 2, hash values h2, h3, h6, and h7, for mail 3, hash values h4, h8, h9, h0, and for hash 4 are hash values. Assume that h1, h2, h3, and h0. Here, if the criterion for determining similar mail is 75% or more, Mail 4 is similar to Mail 1. This determination is performed by the process of FIG. Here, in order to simplify the description, it is assumed that each mail has four feature amounts.

  First, when the mail 1 is extracted via the Internet, the determination in step S33 in FIG. 6 is negative, and the processing in steps S34 and S35 is performed. As a result of the process in step S34, h1 to h4 are registered in the hash values 1 to 4 in FIG. 10B in the feature amount database 310, and the pointer 311 of the DMC 300 becomes as shown in FIG. The number of DMC referenced is 4.

  Next, when the mail 2 is extracted, the determination in step S33 is negative, so the process proceeds to steps S34 and S35. The feature database 310 registers h2, h3, h6, and h7 in the hash values 1 to 4 of the mail 2 in FIG. 11B by the process of step S34, and the pointer 311 of the DMC 300 is updated by the process of step S35. ), The number of DMC references in Mail 1 is 2, and the number of DMC references in Mail 2 is 4.

  Subsequently, when the mail 3 is extracted, the determination in step S33 is negative, so that the processes in steps S34 and S35 are performed. As shown in FIG. 12B, h4, h8, h9, h0 are registered in the hash values 1 to 4 of the mail 3 by the process of step S34, and the pointer of the DMC 300 is obtained by the process of step S35. As shown in (a), the DMC referenced numbers of the mails 1, 2, and 3 are 1, 4, and 4, respectively.

  Further, when the mail 4 is extracted, the mail 4 is similar to the registered mail 1, so that the determination in step S33 is affirmative and the processes in steps S37 and S38 are performed. By the process of step S37, 1 is added to the number of similar mails of mail 1 in the feature quantity database 310, and the result is as shown in FIG. Further, by the processing of step S38, the pointer of the DMC 300 becomes as shown in FIG. 5A, and the DMC reference numbers of the mails 1, 2, and 3 are 4, 2, and 3, respectively.

  That is, when a similar mail arrives, 1 is added to the number of similar mails of mail 1 in step S37. Next, in the process of step S38, that is, the process of FIG. 8, the hash value h1 indicates that the pointer 31 points to its own mail 1, so the determination in step S353 is affirmative and the process of FIG. Since the pointers of the next hash values h2 and h3 both point to mail 2, the determination in step S353 is negative, and the process proceeds to step S354 and subsequent steps. Then, 1 is subtracted from the DMC referenced number of Mail 2 in Step S354, and 1 is added to the DMC referenced number of Mail 1 in Step S352. The next hash value h0 is similarly processed.

  As described above, a mail having many similar mails is frequently started from step S38 in FIG. 6 and the update process in FIG. 8 is started, and the number of DMC referenced is increased. On the other hand, as is apparent from Mail 2, the number of DMC references decreases with the passage of time by overwriting the data that has been hit by the hash value if there is no similar mail.

  In addition, this invention is not limited to the said embodiment, It is also possible to deform | transform as follows. In the above embodiment, the electronic mail collecting means 51 collects the electronic mail flowing on the network by tapping the switching hub 3, but the mail server software is changed so that the mail server selects the mail to be delivered. You may make it transmit to the feature-value conversion means 52 directly. The mail delivery protocol is assumed to be SMTP, but other delivery forms such as WWW mail using HTTP may be used.

  In addition, when the mail server sends the mail to be delivered to the feature amount conversion means 52, the mail that has already been determined to be spam is sent with a mark as spam, and using the information, the mass mail detection means 53 Mails similar to those marked with may be immediately identified as spam. In addition, the mail server may be configured to include up to the feature amount conversion unit 52, and the feature amount converted by the feature amount conversion unit 52 may be transmitted to the mass mail detection unit 53 via a network.

  In the above embodiment, the mass mail detection means 53 uses the DMC 300 (see FIG. 5) as a mechanism for preferentially storing electronic mails that are frequently delivered to the storage area. You can use another mechanism such as. When using an LRU, specifically, an LRU list for managing entries in the feature quantity database 310 is created, and the entry in the feature quantity database 310 to be processed in step S37 (see FIG. 6) is moved to the head of the LRU. The process is included in step S37. In addition, the storage area required when creating a new entry in the feature amount database 310 in step S34 is secured by discarding the last entry of the LRU, and the new entry is added to the head of the LRU.

  In the above embodiment, the preprocessing of the feature amount conversion unit 52 has not been described. However, a preprocessing unit may be provided between the electronic mail collection unit 51 and the feature amount conversion unit 52 of FIG. good. This pre-processing means may be a means for extracting a character string, and an e-mail address, a telephone number, etc. may be extracted by this pre-processing means. Also, any other means for preprocessing may be used. This pre-processing may be performed according to the specifications when the electronic mail is displayed to the recipient. In this pre-processing, for example, a character displayed in the first part (for example, the first two pages) of the receiver's terminal may be selected as a processing target. The specifications assumed at this time include HTML and MIME processing that affect the display, but other specifications are also acceptable. Also, when displaying, uppercase and lowercase letters, full-width characters and half-width characters are treated as similar characters, and the same feature amount is calculated (for example, all double-byte uppercase characters are converted into half-width lowercase characters in advance) ) It doesn't matter.

  In the above embodiment, the hash value of the character string included in the text of the electronic mail is used as the feature quantity. However, other feature quantities such as bigram and word appearance frequency may be used.

  FIG. 14 is a block diagram showing a configuration of another embodiment of the present invention, which is a configuration example in which the present invention is incorporated in a mail server. In FIG. 14, the same reference numerals are given to the same or equivalent parts as in FIG. 1, and the description overlapping with FIG. 1 is omitted.

  Assuming that the mail server group 1 in FIG. 14 (a) is composed of a plurality of mail servers 1a, 1b, 1c, etc., this embodiment is configured to collect electronic mail as shown in FIG. 14 (b). The feature 51, the feature amount conversion means 52, the mass mail detection means 53, and the mail processing means 57 are characterized in that they are incorporated in each mail server 1a, 1b, 1c.

  In the present embodiment, the result of detection as to whether the mail is mass mail is sent to the mail processing means 57. The processing performed by the mail processing means 57 performs deletion of the mass mail, display of the mass mail in the mail title section, and the like based on the mass mail detection result. It may also be a mass mail notification to the mail server operator.

  As is clear from the above description, according to the present invention, prior rule creation and learning are unnecessary. In addition, similar mail is detected simply by comparing feature quantities of electronic mail, and when the number of similar mail reaches a certain number, it is determined as mass mail, so that mass mail detection operation can be performed at high speed.

DESCRIPTION OF SYMBOLS 1 ... Mail server group, 2 ... Internet, 3 ... Switching hub, 5 ... Mass-mail detection apparatus, 51 ... Electronic mail collection means, 52 ... Feature-value conversion means, 53 ... Mass mail detection means, 55... Mass mail detection result, 300... Directed Mapped Cache, 310... Feature amount database, 311.

Claims (7)

Electronic mail collecting means for collecting electronic mail to be delivered;
Pre-processing means for pre-processing the electronic mail according to a specification when the electronic mail is displayed to the recipient;
A feature amount conversion means for converting the preprocessed electronic mail into a feature amount;
Comprising mass mail detecting means for detecting mass mail using the converted feature quantity;
The feature amount conversion means extracts a partial character string from the body of the electronic mail, uses a set of hash values calculated from the partial character string as a feature amount,
The mass mail detection means includes a feature quantity database and a pointer to the feature quantity database, and determines whether or not a hash value of the entered feature quantity is entered in the pointer. Determines the similarity between a new electronic mail and a registered electronic mail by accessing the feature database using the pointer and comparing it with the hash value of the electronic mail already registered in the feature database. A mass-mail detection method characterized in that an electronic mail having a hash value equal to or greater than a predetermined number is determined as a similar electronic mail, and the similar electronic mail is determined as a mass mail when a predetermined number of the similar electronic mails are detected. The mass mail detection method according to claim 1, wherein the preprocessing unit gives priority to a character displayed in an initial part. The pretreatment means, a character-like double-byte characters and single-byte characters corresponding thereto, Masumeiru detection method according to claim 1, wherein the handle with the same characters該似was characters. The pretreatment means, Masumeiru detection method of claim 1, a character resembling the corresponding lowercase to uppercase and this, and wherein the handle and the same character a該似was characters. The mass mail detection method according to claim 1, wherein the preprocessing means includes HTML processing. The mass mail detection method according to claim 1, wherein the preprocessing means includes MIME processing. Mail server with a Masumeiru detection method according to any one of claims 1 to 6.

Images