JP2005149124A - Electronic message filter system and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a mail filter system capable of filtering with high probability and deleting a message associated with a spam and including a spelling-altered word. <P>SOLUTION: The mail filter system 46 includes an object connector 72 providing a connection to a thesaurus 48 for spell checking ; a token look up module 60 reading an e-mail message; and determining, via the connection provided by the object connector 72, whether each token in the e-mail message is defined in the thesaurus 48; and processing modules 62, 64 classifying the e-mail message, by referring to whether the e-mail message is the spam in accordance with the result of token look up. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a mail filter system, and more particularly to a mail filter system that filters out unwanted (spam) email (e-mail) and messages without the need for explicit training.

  E-mail on the Internet is a very convenient communication tool. Sending an e-mail costs little. You can even get in touch with friends on the other side of the globe.

  Unfortunately, e-mail is also a very convenient way to deliver inappropriate commercial e-mail. Such inappropriate commercial e-mail is referred to as “spam”. Today, astronomical numbers of spam are being sent over the network. For those who send spam mails (spammers), it will cost little to send spam mails, so it seems that more and more spam mails will continue to be sent over the network.

  Spam causes enormous economic losses. Global economic losses from spam are expected to exceed $ 20.5 billion in 2003 to over $ 198 billion in 2007. These losses are caused by the loss of bandwidth and CPU time, the loss of security between the network and the targeted computer, and the loss of time and honor of spam recipients.

  Many efforts have been made to prevent or filter spam. The most advanced spam filters are two spam filters based on Bayesian learning and hash blacklists.

  According to the former approach, a message is classified into spam by estimating a probability from a learned spam related probability with respect to a token (word) included in the message. Such classification is estimated according to the following Bayes rule.

例えば、“ｍｏｎｅｙ”という語についてスパムの確率(すなわちＰ（ａ）)が０．６であり、“ｃｏｎｆｉｄｅｎｔｉａｌ”という語について（すなわちＰ（ｂ））が０．８であれば、“ｍｏｎｅｙ”及び“ｃｏｎｆｉｄｅｎｔｉａｌ”の両方の語を含むメッセージの確率（すなわちＰ（ａ，ｂ））は、式（１）により、０．８６となる。

For example, if the spam probability (ie P (a)) is 0.6 for the word “money” and the word “confidential” (ie P (b)) is 0.8, then “money” and The probability of a message including both words “confidental” (that is, P (a, b)) is 0.86 according to equation (1).

  The hash blacklist operates by converting the entire text of an incoming email into a single number (signature) by a process called “hashing”. These numbers (hash values) are simple enough for fast comparisons, but have some useful uniqueness. These are then compared to a list of known spam hashes, and messages matching the “spam hash” are discarded.

  Another example of the spam filter system is disclosed in Patent Document 1. The spam filter system disclosed in Patent Document 1 includes a weighted word list. For each received e-mail, check the weight of the words in the e-mail in a list and calculate the sum. If the total exceeds a predetermined threshold, the e-mail is determined to be spam and placed in a discard folder. Word weights are calculated by “training”. In training, the user manually categorizes received e-mails. If a word is not in this list, a new initial value is assigned to the new word.

JP2003-066734, FIG.

  There is a constantly evolving relationship (arms expansion competition) between spam filter software (spam filter) and the increasingly sophisticated spam generation technology of spam generation software (spam software). As a result, technical problems to be overcome arise. The above three methods for blocking or filtering out spam mail have the following problems.

  The technology used by the Bayesian filter has a common problem with artificial intelligence, where the machine usually cannot understand the meaning of the message. In the case of using a Bayesian filter, the meaning of a message is not grasped, but this is considered as spam based on statistics about the appearance of word combinations. Thus, using currently available spam filters often results in the correct being wrong and the error being correct.

  Considering a Bayesian filter, a token that this filter recognizes for the first time should have a neutral effect on the determination of the filter, so the filter should be in the learning (training) state again. Even if such a token is naturally part of a message that is clearly spam, the filter cannot classify the spam as spam. Things get even worse if you receive a larger number of the same spam messages and they contain separate new tokens.

  On the other hand, the obvious solution for spammers is to modify the most important spam-related tokens in spam messages to avoid matching with tokens that are known to have a high probability of spam. For example, assuming the probability that the word “confidental” is part of spam is 0.8, any message containing this would be classified as spam. However, the word “confidental” is intentionally (and randomly) “conf1dental” (uses “1” instead of “i”), “confidentia1” (“1” instead of “l”), “c0nidential” ( Misspelling such as “0” instead of “o”), “confiden_tial” (inserting a blank), etc., naturally, the amount of such a combination becomes enormous and is an important token “confidential”. However, it is replaced with a completely new and unknown token for the Bayesian filter.

  The system disclosed in the prior patent document 1 is the same. If spam software modifies a heavy word, the resulting new word is probably not on the list. As a result, this word is assigned an initial weight that is not very large. Each time a modified token appears, the user must train the system by manually classifying the email.

  It is not a technically serious problem for a spammer to achieve such a modification efficiently. In fact, spammers may run the same Bayes filters and train these filters with the same set of typical spam messages. Spammers are therefore well aware of the approximate value of the probability that the word they are trying to include in the message is related to spam. As a result, words that have a very strong association with spam will be randomized and then broadcast to a large number of potential spam recipients on the network.

  In addition, the modified token can bypass the hash blacklist filter. This is because a new token will generate a new hash value (signature) for content that was previously recognized as spam. Adding more randomly generated meaningless tokens to a spam message is only meant to further degrade the ability of the spam filter to properly classify spam messages using only the message signature. Absent.

  Accordingly, one of the objects of the present invention is to provide a mail filter system capable of filtering and deleting messages containing words related to modified spam with a high probability.

  Another object of the present invention is to provide a mail filter system that can filter and delete messages containing modified spam-related words with high probability without any explicit training.

  An electronic message filter system according to the present invention comprises a connection means for providing a connection to a predetermined dictionary for use in spell checking, a reading means for reading an electronic message, and an electronic read by the reading means. For each token in the message, the means for determining whether the token has a heading in the dictionary, via the connection provided by the connection means, and for the token in the electronic message, obtained by the determination means. And processing means for processing the electronic message according to the result.

  Preferably, the processing means has means for obtaining a ratio of tokens in the electronic message that do not have a heading in the dictionary to all tokens in the electronic message; A determination unit for determining whether or not a predetermined condition is satisfied, and a unit for classifying the electronic message according to the determination of the determination unit.

  The determining means may include means for determining whether or not the rate of tokens without a heading in the dictionary exceeds a predetermined threshold value.

  The means for classifying includes: first classifying means for classifying an electronic message having a token rate exceeding a predetermined threshold as an unnecessary message; and a threshold having a predetermined token rate And a second classification means for classifying a message that does not exceed the normal message as a normal message.

  Preferably, the first classifying means includes means for inserting a predetermined text into an electronic message in which the token rate exceeds a predetermined threshold.

  The electronic message may include a subject part and a body part, and the means for inserting is means for inserting a predetermined text into the subject part of the electronic message where the token rate exceeds a predetermined threshold. May be included.

  Preferably, the first classifying means includes means for sending the electronic message to a predetermined first destination, and the second classifying means is a predetermined number different from the first destination. Means for transmitting to the second destination.

  Preferably, the means for determining includes means for determining, for each word of the electronic message, whether the word has a heading in the dictionary via a connection provided by the connecting means.

  The system further includes means for determining, for each character in the electronic message, whether the character belongs to a word that does not have a heading in the dictionary, and the processing means is in the electronic message for all words in the electronic message. Means for obtaining the rate of words having no heading in the dictionary, means for obtaining the rate of characters contained in words in the electronic message that have no heading, and Means for determining whether the word rate and the character rate satisfy a predetermined condition, and means for classifying the electronic message according to the determination of the determination means.

  Preferably, the determination means includes means for determining whether the word rate exceeds a predetermined threshold and whether the character rate exceeds a predetermined threshold.

  The means for classifying includes: a first classifying means for classifying an electronic message having a word rate and a character rate exceeding respective threshold values as an unnecessary message; and either a word rate or a character rate. May include a second classifying means for classifying an electronic message that does not exceed the corresponding threshold as a normal message.

  Preferably, the first classification means includes means for inserting a predetermined text into an electronic message in which the word rate and the character rate exceed respective threshold values.

  The electronic message may include a subject part and a body part, and the means for inserting is to insert a predetermined text into the subject part of the electronic message where the word rate and the character rate exceed the respective thresholds. Means for doing so may be included.

  Preferably, the means for determining includes, for each word of the electronic message, means for determining via a connection provided by the connection means whether the word has a heading in the dictionary, Further includes means for determining, for each character in the electronic message, whether the character belongs to a word that does not have a heading in the dictionary, the processing means being a word in the electronic message According to the determination of the determination means, a means for obtaining the rate of characters included in those without a headline, a determination unit for determining whether the character rate satisfies a predetermined condition, and an electronic message Means for classifying.

  The determining means may include means for determining whether or not the character rate exceeds a predetermined threshold value.

  The classifying means classifies an electronic message whose character rate exceeds the threshold as an unnecessary message, and classifies an electronic message whose character rate does not exceed the threshold as a normal message. And a second classifying means for performing the processing.

  A second aspect of the present invention relates to a computer-executable program that, when executed on a computer, causes the computer to perform all of the functions described in any of the electronic message filter systems described above.

-Overview-
The mail filter system of the present invention is based on an approach to address advanced spam email messages that seeks to overcome the latest Bayesian and hash blacklist spam filters. This approach is based on the idea that a modified token (word) that has a higher spam probability can be detected using a spell check software agent.

  In the case of a spam message, the amount of tokens and characters that cannot be found in a spell check thesaurus is much larger than in a normal e-mail message. By using a dictionary (thesaurus) provided with an easy-to-operate user interface, which is naturally provided in an office environment with many computers, as proposed here, the spell check software for spam filter according to the present embodiment ( Spam Filtering Spell-Checking Software (FSSCA) uses very common software connectivity, no additional third-party dictionary database for spell-checking, easy customization, etc. It is clear that the benefits can be obtained. The SFSCA according to this embodiment can be used as independent software for spam filters, or it can be used as a plug-in for Bayesian or hash blacklist filters to further enhance their functionality.

  Experience has shown that remodeling tokens related to spam is the most effective way to circumvent the latest spam filters, so you can find these modified tokens and include them in incoming email messages. We discussed how to calculate the rate at which such words exist.

  Remodeled tokens are random, and most importantly, even the same message is different every time it arrives, so what matters are their common features that can be used as a basis for finding them . The system of this embodiment is based on the fact that even though the modified tokens are random, none of them is recognized as a correct word by the spell check routine. The system of this embodiment uses a thesaurus (word) dictionary of a spell check dictionary widely available in office products.

  This approach has the following advantages.

  i) Most desktop computers run one of the widely adopted office software environments.

  ii) The thesaurus in the office software environment can be highly customized, specific terms can be freely added, and special spell check settings can be individually adjusted.

  iii) Since the user normally checks the spelling of the message he / she sends, the risk of making the error correct is further reduced.

iv) The thesaurus is usually implemented in component object technology, and therefore the thesaurus activation by another application (eg, the SFSCA according to the present embodiment) should be supported in a general manner.
-System structure according to the embodiment-
Referring to FIG. 1, the SFSCA 46 of this embodiment includes an e-mail agent 42 having an inbox 44 and a thesaurus 48 used for spell check that can be accessed through an object connection function 50 of the operating system. It is realized as a software agent that operates in an environment defined by. The inbox 44 stores incoming mail that the e-mail agent 42 retrieves from the e-mail server 40.

  FIG. 2 is a detailed block diagram of the SFSCA 46. Referring to FIG. 2, the SFSCA 46 provides an object connector 72 that provides connection to the thesaurus 48, and all the unread e-mail messages stored in the inbox 44 via the object connector 72 and the object connection function 50. A token lookup module 60 that looks for a token in the thesaurus 48 and determines whether the token is listed in the thesaurus, and a rate of words not listed in the thesaurus 48 based on the output of the token lookup module 60 And a calculation module 62 for calculating. The calculation module 62 also calculates the ratio of characters in words that are not defined in the thesaurus in the message to characters in words that are listed in the thesaurus 48 in the message.

The SFSCA is further connected to the calculation module 62 and determines whether the e-mail is spam based on the work memory 70 for storing the rate calculated by the calculation module 62 and the rate stored in the work memory 70. Determination module 64, and two memories 66 and 68 for storing a predetermined threshold rate of undefined words (TH ₁ ) and a threshold rate of characters (TH ₂ ). The threshold rates stored in memories 66 and 68 can be manually modified.

  Further, the determination module 64 has a function of inserting a specific keyword (for example, “spam”) as a mark in the header of the e-mail message determined to be spam, so that the e-mail agent 42 can read the message. Filtering can be performed easily. Modules 62, 64 and memories 66, 68, 70 process each email according to the result of the token lookup by token lookup module 60. That is, modules 62 and 64 classify each e-mail as spam (unwanted e-mail message) or regular message.

  The determination module 64 of this embodiment has a threshold at which the rate of undefined words exceeds the threshold stored in the memory 66 and the rate of characters in the undefined words is stored in the memory 68. If it exceeds, it is determined that the e-mail is spam.

The threshold can be determined according to the user's environment. In this embodiment, if the rate of undefined words exceeds 20% and the rate of characters in the undefined words exceeds 20%, it is determined that the e-mail message is spam. If the rate of undefined words does not exceed 20%, or the rate of characters in the undefined words does not exceed 20%, it is determined that the e-mail is not spam.
-Operation-
The SFSCA 46 according to this embodiment operates as follows. The e-mail agent 42 retrieves messages from the e-mail server 40 and stores them in the inbox 44. At a pre-scheduled timing or in response to a manual trigger signal, the SFSCA 46 reads unread e-mail in the inbox 44.

  Referring to FIG. 2, for each unread e-mail message, token lookup module 60 searches the thesaurus 48 for each word to see if the word is defined in the thesaurus 48. The token lookup module 60 outputs the number of undefined words to the calculation module 62. The calculation module 62 counts the number of words and characters in the email message and, in response to the output of the token lookup module 60, the number of characters in the undefined words, the rate of undefined words and the undefined words. Calculate the percentage of characters in the middle.

  The determination module 64 reads the calculated rate from the work memory 70. Decision module 64 compares the calculated rate of undefined words with a threshold value stored in memory 66. If the calculated rate exceeds the threshold, the calculated rate of characters in the undefined word is compared to the threshold stored in memory 68. If the calculated rate exceeds the threshold, the determination module 64 determines that the email message is spam, marks the subject of the email message “spam”, and sets the subject The e-mail message marked with is stored in the inbox 44. The token lookup module 60 then reads another unread message in the inbox. If none of the calculated rates exceeds the corresponding threshold, the decision module simply reads another unread message in the inbox.

  The above operation is repeated until there are no unread e-mail messages in the inbox 44.

  FIG. 3 shows an example of an e-mail message. According to the SFSCA 46 of this embodiment, the rate of undefined words is 3%, and the rate of characters in undefined words is 5%. Therefore, it is determined that this e-mail message is not spam. It appears that the sender did not spell check the message considered here, but still the rate of undefined words and the rate of characters in undefined words is relatively small.

  FIG. 4 shows another example of an e-mail message. According to the SFSCA 46 of this embodiment, the rate of undefined words is calculated to be 46%, and the rate of characters in undefined words is calculated to be 45%. Therefore, this e-mail message is determined to be spam.

  As described above, the SFSCA 46 of this embodiment can determine whether an e-mail is spam using a dictionary that can be easily used in a normal office environment. It is not necessary to further use a dictionary database for third-party spell checking. However, this embodiment does not exclude the use of such an additional dictionary. This embodiment can itself determine spam email messages (with the aid of a spell check dictionary) and can therefore be used as independent software. When combined with a conventional Bayes list filter or hash black list filter or both, the SFSCA 46 of this embodiment can further improve their spam checking capabilities.

  It should be emphasized that this embodiment does not require an explicit “training” phase. If a user knows that a frequently used word is not defined in the spell check dictionary, he / she will add the word to the spell check dictionary without being conscious of improving the spam filtering capability of the SFSCA 46. Will do.

In the above embodiment, the ratio of undefined words and the ratio of characters in undefined words are both calculated. However, the present invention is not limited to such an embodiment. For example, only the rate of undefined words or only the rate of characters in undefined words may be used.
-Example of implementation by computer-
The above-described SFSCA 46 can be realized as software executed on ordinary computer hardware. FIG. 5 is an external view of such a normal computer system 120, and FIG. 6 shows the system 120 in a functional block diagram.

  Referring to FIG. 5, the computer system 120 includes a computer 140 having an FD (flexible disk) drive 152 and a CD-ROM (compact disk read only memory) drive 150, a keyboard 146, a mouse 148, and a monitor 142. Including.

  Referring to the block diagram of FIG. 6, the computer 140 further includes a CPU (central processing unit) 156, a bus 166 connected to the CPU 156, the CD-ROM drive 150, and the FD drive 152, and a program such as a bootup program. A read-only memory (ROM) 158 for storing, a random access memory (RAM) 160 connected to the CPU 160 for storing application program instructions, system programs, data and the like to be executed, and a hard disk drive 154 are included.

  Computer 140 further includes a network adapter board 174 for providing computer 140 with a connection to a local area network (LAN) 184. A mail server (not shown) is connected to the computer system 120 via the LAN 176.

  The thesaurus 48 shown in FIGS. 1 and 2 is stored in the hard disk drive 154, and when the program is executed, the thesaurus 48 is read from the hard disk drive 154 and stored in the RAM 160.

  FIG. 7 shows a control structure of a program for causing the computer 120 to perform mail filtering according to this embodiment. This program is stored in the CD-ROM 180 or FD 182, inserted into the CD-ROM drive 150 or FD drive 152, and further transferred to the hard disk drive 154. Alternatively, the program may be transmitted to the computer 140 via the LAN 176 and stored in the hard disk drive 154. The program is loaded into the RAM 164 at the time of execution. The program may be directly loaded into the RAM 164 via the CD-ROM 180, the FD 182 or the LAN 176.

  The program described below includes a plurality of instructions that cause the computer 140 to operate as the SFSCA 46. Some of the basic functions required to perform this mail filtering function are operating system (OS) running on the computer 140, third party programs or modules such as a toolbox installed on the computer 140. Since the program is provided, the program may not necessarily include all the basic functions necessary to realize the function of the SFSCA 46 according to this embodiment. The program only needs to include the portion of the instructions that performs mail filtering by calling the appropriate function or “tool” in a controlled manner so that the desired result is obtained. How the computer system 120 operates is well known and will not be described in detail here.

  Referring to FIG. 7, the program that implements the SFSCA 46 of this embodiment initializes the required resources in step 200 after starting. For example, the program connects to the thesaurus 48 resident in the RAM 160 at this step. Next, in step 202, the system attempts to read the next unread e-mail in inbox 44. In step 204, it is determined whether there is an unread e-mail. If there is no unread e-mail message, the program performs a cleanup process at step 240 and ends. Otherwise, in step 206, the variables CNT_C and CNT_W are set to zero. The variable CNT_C is for counting the number of characters in the undefined word. CNT_W is for counting the number of undefined words.

  At step 208, the total number of words in the email message is confirmed. Similarly, at step 210, the total number of characters is confirmed.

  For all words in the e-mail message, it is determined in step 220 whether the word is in the thesaurus 48. If not, 1 is added to CNT_W. Similarly, for all characters in the e-mail message, it is determined in step 222 whether the characters are for words in the thesaurus 48. If not, 1 is added to CNT_C.

  In step 224, the rate of undefined words (RATIO_W) and the rate of characters in undefined words (RATIO_C) are calculated using the number of words and characters calculated in steps 208 and 210 and the variables CNT_C and CNT_W.

In step 226, the calculated RATIO_W whether more than a threshold value TH ₁ is determined. If RATIO_W> TH ₁ , control proceeds to step 228, otherwise returns to step 202.

In step 228, the calculated RATIO_C whether more than a threshold value TH ₂ is determined. If RATIO_C> TH ₂ , control proceeds to step 230, otherwise returns to step 202.

  At step 230, the subject of the e-mail is marked with “SPAM” and control returns to step 202.

  When the program configured as described above is executed on the computer system 120, it causes the computer system 120 to execute all the functions of the SFSCA 46 shown in FIGS.

  If the e-mail message is determined to be an inappropriate message, the determination module 64 of this embodiment adds a specific keyword to the header portion. The present invention is not limited to such an implementation. For example, the SFSCA may send unnecessary e-mail messages to a predetermined destination such as a discard folder and send other mails to the normal inbox of the e-mail agent.

  The calculation module 62 shown in FIG. 2 calculates the rate of tokens not in the thesaurus 48, and the determination module 64 determines that the email is spam if the rate exceeds a threshold. However, the present invention is not limited to such an implementation. The calculation module may calculate the rate of tokens in the thesaurus 48 and the determination module may determine that the email is spam if the rate is below a threshold. These are equivalent to each other. Similarly, any reasonable criteria may be employed in determining whether an e-mail is spam.

  In this embodiment, an email is classified as spam if the rate of undefined words and the rate of characters in undefined words exceed their respective thresholds. The present invention is not limited to this embodiment. An email may be classified as spam if either the rate of undefined words or the rate of characters in undefined words exceeds a corresponding threshold.

  Further, the message is not necessarily limited to an e-mail message. It may be an instant messenger message or any other form of electronic message.

  The above-described embodiments are merely examples and should not be construed as limiting. The scope of the present invention is indicated by each claim in the claims after taking into account the description of the detailed description of the invention, and all modifications within the meaning and scope equivalent to the wording described therein are intended. Including.

1 is a diagram showing an overall configuration of an e-mail system including an SFSCA 46 according to an embodiment of the present invention. It is a detailed block diagram of SFSCA46 concerning one embodiment of this invention. It is a figure which shows the example of the e-mail which is not spam. It is a figure which shows the example of spam e-mail. 1 is an external view of a computer system that implements an SFSCA 46 according to an embodiment of the present invention. It is a functional block diagram of a computer system. It is a flowchart which shows the structure of the program which implement | achieves SFSCA46 which concerns on one embodiment of this invention.

Explanation of symbols

40 e-mail server, 42 e-mail agent, 44 inbox, 46 SFSCA (spell check software agent for spam filter), 48 thesaurus, 60 token lookup module, 62 calculation module, 64 decision module, 66 and 68 threshold Value memory, 70 work memory

Claims (17)

Connection means for providing a connection to a predetermined dictionary for use in spell checking;
Reading means for reading electronic messages;
Means for determining, for each token in the electronic message read by the reading means, via a connection provided by the connecting means whether the token has a heading in the dictionary;
An electronic message filter system comprising: processing means for processing the electronic message according to a result obtained by the means for determining for tokens in the electronic message. The processing means is
Means for obtaining a ratio of tokens in the electronic message that do not have a heading in the dictionary to all tokens in the electronic message;
A determination means for determining whether a rate of tokens without a heading in the dictionary satisfies a predetermined condition;
Means for classifying the electronic message according to a determination of a determination means. The system of claim 2, wherein the determining means includes means for determining whether the rate of tokens without headings in the dictionary exceeds a predetermined threshold. The means for classifying comprises
First classifying means for classifying an electronic message having a token rate exceeding the predetermined threshold as an unnecessary message;
The system according to claim 3, further comprising: a second classifying unit for classifying a message whose token rate does not exceed the predetermined threshold as a normal message. 5. The system of claim 4, wherein the first classifying means includes means for inserting a predetermined text into an electronic message in which the token rate exceeds the predetermined threshold. The electronic message includes a subject part and a body part,
6. The system of claim 5, wherein the means for inserting includes means for inserting a predetermined text into a subject portion of an electronic message where the rate of the token exceeds a predetermined threshold. . The first classifying means includes means for sending the electronic message to a predetermined first destination;
5. The system of claim 4, wherein the second classifying means includes means for sending the electronic message to a predetermined second destination that is different from the first destination. The means for determining includes, for each word of the electronic message, means for determining via the connection provided by the connecting means whether the word has a heading in the dictionary; The system of claim 1. For each character of the electronic message further comprising means for determining whether the character belongs to a word without a heading in the dictionary;
The processing means is
Means for obtaining a rate of words in the electronic message that do not have a heading in the dictionary with respect to all words in the electronic message;
Means for obtaining a rate of characters contained in words in the electronic message that do not have a heading in the dictionary;
Determining means for determining whether the word rate and the character rate satisfy a predetermined condition;
9. A system according to claim 8, comprising means for classifying the electronic message according to a determination of the determination means. 10. The means of claim 9, wherein the determining means includes means for determining whether the word rate exceeds the predetermined threshold and the character rate exceeds a predetermined threshold. The described system. The means for classifying comprises
First classification means for classifying an electronic message in which the word rate and the character rate exceed respective threshold values as an unnecessary message;
11. A system according to claim 10, comprising: a second classifying means for classifying as an ordinary message an electronic message in which either the word rate or the character rate does not exceed a corresponding threshold. 12. The system of claim 11, wherein the first classification means includes means for inserting predetermined text into an electronic message in which the word rate and the character rate exceed respective thresholds. . The electronic message includes a subject part and a body part,
The means for inserting includes means for inserting predetermined text into a subject portion of an electronic message in which the word rate and the character rate exceed respective thresholds. The described system. The means for determining includes, for each word of the electronic message, means for determining via a connection provided by the connecting means whether the word has a heading in the dictionary;
The system further includes means for determining, for each character in the electronic message, whether the character belongs to a word without a heading in the dictionary;
The processing means is
Means for obtaining a rate of characters contained in words in the electronic message that do not have a heading in the dictionary;
Determining means for determining whether the character rate satisfies a predetermined condition;
Means for classifying the electronic message according to a determination of the determining means. The system of claim 14, wherein the determining means includes means for determining whether the rate of the characters exceeds a predetermined threshold. The means for classifying comprises
First classification means for classifying an electronic message having a character rate exceeding a threshold as an unnecessary message;
16. A system according to claim 15, comprising: a second classifying means for classifying an electronic message whose character rate does not exceed a threshold value as a normal message. A computer-executable program that, when executed on a computer, causes the computer to perform all of the functions recited in any one of claims 1 to 16.

Images