JP5810240B1 - E-mail erroneous transmission determination device, e-mail transmission system, and program - Google Patents

Abstract

An error in a destination address is prevented and erroneous transmission of an electronic mail is prevented with high accuracy. An email mistransmission determination device 30 associates an email body feature vector with an email address in association with an email body feature vector generation unit 35 that generates an email body feature vector related to the content of an email to be transmitted. The mail body feature vector storage unit 36 to be stored, the mail body feature vector stored in association with the address for each stored address, and stored in association with an arbitrary address different from the address. A mail body similarity model generating unit 37 that generates a similarity model to be a determination criterion based on the mail body feature vector, a mail body feature vector relating to the content of the email to be transmitted, and an email to be transmitted Based on the similarity model for the destination of the email, and the trust in the destination of the email to be sent And a, a reliability calculating unit 39 for calculating a. [Selection] Figure 4

Description

  The present invention relates to an erroneous email transmission determination device, an email transmission system, and a program.

Conventionally, a technique for preventing erroneous transmission of an e-mail has been disclosed (see Patent Document 1 and Patent Document 2).
Japanese Patent Application Laid-Open No. 2004-151867 is based on a comparison between the appearance frequency of a predetermined vocabulary in the body of an e-mail and the appearance frequency of the vocabulary in an e-mail body previously transmitted to the same destination as the destination of the e-mail. A technique for controlling whether to send the electronic mail or to hold the transmission is described.

  Further, Patent Document 1 describes a technology that supports the determination of whether or not there is an erroneous transmission based on the politeness of the email text and the presence or absence of a signature attached to the email text.

  Patent Document 2 discloses an analysis unit that analyzes the validity of a destination address of a transmission mail created by a user based on document data of all past transmitted mails stored in a transmitted mail storage unit, and a result thereof. Is described, and a transmission execution means for receiving a transmission instruction of a user who has viewed it and executing transmission of the transmitted mail is described.

  And in patent document 2, each of the destination addresses included in the similar group to which the transmission mail belongs among the similar groups of the transmitted mail configured by the clustering means based on the similarity of the document data as the analysis means, It is described that the similarity with the destination address of the transmitted mail is calculated, and the destination address of the transmitted mail with the highest similarity is proposed as a destination address correction candidate.

Japanese Patent No. 3663844 Japanese Patent No. 4817952

  In the technique of Patent Document 1, in the validity determination of the destination address based on the appearance frequency of the predetermined vocabulary in the e-mail body, the appearance frequency of the predetermined vocabulary in the e-mail body that was simply transmitted to the same destination address in the past as a comparison target. It is only taken into account. As a result, there is a problem that the validity cannot be accurately determined.

  Furthermore, the technique of Patent Document 1 supports the validity determination of the destination address based on the politeness of the e-mail body and the presence or absence of a signature attached to the e-mail body. It only suggests a social relationship with the other party, and is far from sufficient to determine the validity of the destination address.

  Further, the technique of Patent Document 2 extracts a sent mail group having a high similarity to the mail text of the sent mail from all past sent mails, and proposes correction candidates for the destination address based on the extracted mail group. Therefore, a large amount of calculation is required. Furthermore, this technology assumes only mistakes in the destination addresses of outgoing emails that have been sent in the past and the contents of the email texts are similar. It is hard to say that it is directly connected to a simple mistake.

  The present invention has been proposed in view of such circumstances, and an erroneous e-mail transmission determination device, an e-mail transmission system, and a program that can prevent erroneous transmission of an e-mail with high accuracy by preventing an error in a destination address. The purpose is to provide.

An erroneous email transmission determination device according to the present invention includes: feature information generating means for generating feature information relating to the content of an email to be transmitted; feature information relating to the content of a sent email; and a destination of the email. Accumulating means for storing in association, transmission destination candidate selecting means for selecting a transmission destination candidate similar to the transmission destination of the e-mail to be transmitted from among transmission destinations of the transmitted e-mail, and accumulation in the accumulation means For each of the transmission destinations, the feature information stored in the storage unit in association with the transmission destination is stored in the storage unit in association with another transmission destination different from the transmission destination. and the feature information have, based on the number of words included in the e-mail in a word space and dimensionality, the word spatial region characteristic information associated with the destination, to the other destination Been made to separate the words spatial domain response Tagged characteristic information, the characteristic information about the contents of any e-mail, whether belonging to the word spatial region feature information associated with the destination Similarity model generation means for generating a similarity model as a determination criterion, feature information regarding the content of the e-mail to be transmitted generated by the feature information generation means, and the similarity model generation means Based on the similarity model regarding the transmission destination of the transmission target email and the similarity model regarding the transmission destination candidate generated by the similarity model generation unit, the transmission destination of the transmission target email and Reliability calculation means for calculating each reliability of the destination candidate.

  An e-mail transmission system according to the present invention includes: an e-mail creating unit that creates an e-mail to be transmitted with a specified destination; and an error regarding the transmission destination of the e-mail to be transmitted created by the e-mail creating unit An e-mail erroneous transmission determination device that determines whether or not there is a high possibility of the e-mail, and the e-mail to be transmitted when the e-mail erroneous transmission determination device determines that there is no high possibility of an error for the destination And transmitting means for deferring transmission of the e-mail to be transmitted when it is determined that there is a high possibility of an error for the transmission destination.

  An e-mail transmission system according to the present invention includes: an e-mail creating unit that creates an e-mail to be transmitted with a specified destination; and an error regarding the transmission destination of the e-mail to be transmitted created by the e-mail creating unit An e-mail erroneous transmission determination device that determines whether or not there is a high possibility, a display unit that displays a destination candidate selected by the selection unit of the e-mail erroneous transmission determination device, and an e-mail erroneous transmission determination The apparatus transmits the e-mail to be transmitted when it is determined by the apparatus that the possibility of error is not high, and the e-mail to be transmitted when it is determined that the possibility of error is high for the destination When one of the transmission destination candidates displayed by the display means is specified while the transmission of the mail is suspended, the transmission destination is designated as the transmission destination candidate. And a transmitting means for transmitting the e-mail Shin target.

The e-mail transmission program according to the present invention is a program for causing a computer to function as each unit of the e-mail erroneous transmission determination device.
An e-mail transmission program according to the present invention is a program for causing a computer to function as each unit of the e-mail transmission system.

  The present invention can prevent an erroneous transmission of an e-mail with high accuracy by preventing an error in a destination address.

It is a figure which shows schematic structure of an electronic mail transmission system. It is a figure which shows the structure of a mail client. It is a flowchart which shows the mail transmission processing routine of a mail client. It is a block diagram which shows the structure of a mail mistransmission determination apparatus. It is a flowchart which shows a mail text feature vector generation routine. It is a figure which shows a noun dictionary. It is a figure which shows a mail text feature vector. It is a flowchart which shows a mail text similarity model production | generation routine. It is a figure which shows an example of the mail text similarity model regarding a predetermined destination address. It is a flowchart which shows an erroneous transmission determination processing routine. It is a figure for demonstrating the calculation method of the reliability using a mail text similarity model.

Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the drawings.
[First embodiment]
FIG. 1 is a diagram showing a schematic configuration of an e-mail transmission system 1.

  The e-mail transmission system 1 includes a mail client 10 that performs mail document creation and transmission processing, a mail error transmission determination device 30 that determines the possibility of erroneous transmission of an email created by the mail client 10, and a mail client 10 And a mail server 50 for transmitting a mail document transmitted from the mail address to a destination address.

FIG. 2 is a diagram showing the configuration of the mail client 10.
The mail client 10 includes an operation unit 11 that is operated by a user, a display unit 12, a communication unit 13 that performs data communication processing with the outside, and a control unit 14 that controls operation of each unit. ing.

The operation unit 11 outputs information according to a user operation, for example, information such as an input mail document, a transmission instruction, and a selection instruction.
The display unit 12 displays information from the operation unit 11, information received by the communication unit 13 from the outside, control information from the control unit 14, and the like. For example, the display unit 12 displays an input mail document, a candidate destination address, and other information necessary for seeking a user's judgment.

  The communication unit 13 communicates with the erroneous mail transmission determination device 30 or the mail server 50 based on the control of the control unit 14. For example, the communication unit 13 transmits authentication data to the mail server 50 and receives authentication data indicating that authentication has been performed from the mail server 50. The communication unit 13 transmits a mail document to be transmitted to the erroneous email transmission determination device 30 and receives a determination result indicating the possibility of erroneous email transmission from the erroneous email transmission determination device 30. Further, the communication unit 13 transmits a mail document to the mail server 50 when the possibility of erroneous mail transmission is small.

The mail client 10 configured as described above transmits a mail document to the mail server 50 according to the following procedure.
FIG. 3 is a flowchart showing a mail transmission processing routine of the mail client 10.

In step S1, when the mail client 10 transmits authentication data to the mail server 50 and receives authentication data indicating that the mail server 50 has been authenticated, the process proceeds to step S2.
In step S2, the mail client 10 transmits the mail document scheduled to be transmitted to the mail error transmission determination device 30, and proceeds to step S3.

  In step S3, the mail client 10 stands by until a determination result indicating the possibility of erroneous transmission is received from the erroneous mail transmission determination device 30, and when the determination result is received, the process proceeds to step S4. Note that the erroneous email transmission determination device 30 can also transmit a candidate destination address in place of the destination address when transmitting a determination result indicating that there is a high possibility of erroneous transmission to the email client 10. .

  In step S4, the mail client 10 analyzes the determination result to determine whether there is a high possibility of erroneous transmission. If the determination is affirmative (if the possibility of erroneous transmission is large), the process proceeds to step S5. In the case of determination, the process proceeds to step S10.

  In step S5, the mail client 10 proceeds to step S10 when the forced transmission operation is performed by the user, and proceeds to step S6 when the forced transmission operation is not performed by the user.

  In step S6, the mail client 10 determines whether there is a candidate destination address transmitted from the erroneous mail transmission determination device 30, and if there is a candidate destination address, the process proceeds to step S7, and if there is no candidate destination address, step S9. Proceed to

  In steps S7 and S8, the mail client 10 displays the candidate destination address and waits. When the candidate destination address is selected, the mail client 10 returns to step S5.

  In step S9, since the mail client 10 is not instructed to forcibly send the mail document and there is no candidate destination address, the mail client 10 discards the mail document and ends this routine.

  In step S10, the mail client 10 transmits a mail document to the mail server 50 because the possibility of erroneous transmission is small or there is an instruction for forced transmission, and this routine ends. Then, the mail server 50 transmits the mail document transmitted from the mail client 10 to the destination address according to a predetermined protocol.

FIG. 4 is a block diagram illustrating a configuration of the erroneous mail transmission determination device 30.
The mail mistransmission determination device 30 includes a mail document analysis unit 31 that analyzes a mail document transmitted from the mail client 10, an address book 32 that is a storage unit that stores a destination address of the transmitted mail document, and an address book 32. A candidate destination address selection unit 33 for selecting a candidate destination address in place of the destination address from the address stored in the address, a noun dictionary 34 which is a storage unit for storing a noun, and a mail body feature vector generation for generating a feature vector of a mail document Part 35.

  The user can also select an address to be stored in the address book 32. That is, the addresses stored in the address book 32 may be all the destination addresses of the sent mail document, or may be one selected by the user among all the destination addresses.

  The erroneous email transmission determination device 30 further includes an email body feature vector storage unit 36 that stores email body feature vectors, an email body similarity model generation unit 37 that generates an email body similarity model, and an email body similarity model. A mail body similarity model accumulating unit 38 for accumulating, a reliability calculating unit 39 for calculating the reliability, and a destination address validity determining unit 40 for determining the validity of the destination address are provided.

FIG. 5 is a flowchart showing a mail text feature vector generation routine.
In step S11, the mail document analysis unit 31 extracts a mail text from the mail document, further extracts a destination address from the mail document, and proceeds to step S12. If the extracted destination address is not stored in the address book 32, the destination address is written in the address book 32.

  However, when the destination address has predetermined characteristics, for example, when the destination address has the same domain as the sender's domain, the destination address may not be written in the address book 32. In other words, for example, only e-mails sent to a domain different from the sender's domain may be determined for erroneous transmission.

In step S12, the mail document analysis unit 31 performs morphological analysis on the mail text, and extracts all nouns included in the mail text using a noun dictionary. The nouns extracted here are preferably, for example, savory connection nouns, general nouns, proper nouns, and the like. This is because the appearance frequency of each of the S-variant connection noun, the general noun, and the proper noun varies greatly depending on the contents of the mail text. Further, the mail document analysis unit 31 may extract a word even if it is a part of speech other than a noun as long as it is a word that characterizes the content of the mail text.
In step S13, the mail document analysis unit 31 selects the first noun from all the extracted nouns, and proceeds to step S14.

  In step S14, the mail document analysis unit 31 determines whether the noun being selected has been registered in the noun dictionary 34. If it has not been registered, the process proceeds to step S15. If it has been registered, the process proceeds to step S16. move on.

  FIG. 6 is a diagram showing the noun dictionary 34. In the noun dictionary 34, nouns and correspondence indexes that are one-to-one correspondences with the nouns are registered. The correspondence index is a value for specifying a noun, and in this embodiment, the corresponding index is the order in which the noun is registered (serial number starting from 0).

  The correspondence index is not limited to the order of registration as long as it is a value that can uniquely identify a noun in the noun dictionary 34, for example, UUID (Universally Unique IDentifier), or a hash value (character string of the noun). It may be a numerical value derived from a hash function as an input). When the UUID or the hash value is used, the corresponding index is globally unique. As a result, a plurality of noun dictionaries 34 can be combined into one noun dictionary 34, or a single noun dictionary 34 can be shared by a plurality of mail client users.

  In step S15, the mail document analysis unit 31 generates a new correspondence index for the noun because the noun being selected is not registered in the noun dictionary 34, and is generated by associating the noun with the noun. The corresponding index is paired and registered in the noun dictionary 34, and the process proceeds to step S16.

  In step S16, the mail document analysis unit 31 extracts the correspondence index of the currently selected noun from the noun dictionary 34, and proceeds to step S17.

  In step S17, the mail document analysis unit 31 increments the appearance frequency counter of the extracted corresponding index, and proceeds to step S18. Therefore, every time one noun is extracted from the mail text, the appearance frequency counter of the noun (corresponding index) increases by one.

  In step S18, the mail document analysis unit 31 determines whether there is an unselected noun among all the nouns extracted in step S12. If there is an unselected noun, the process proceeds to step S19, where an unselected noun. If not, the process proceeds to step S20.

  In step S19, the mail document analysis unit 31 selects the next noun from unselected nouns, and returns to step S14. Then, for all nouns extracted in step S12, the processing from step S14 to step S19 is repeated until the appearance frequency counter of the corresponding index is incremented.

  In step S <b> 20, the mail document analysis unit 31 outputs a pair of correspondence indexes and appearance frequency counters for all nouns included in the mail body to the mail body feature vector generation unit 35. The mail body feature vector generation unit 35 generates a mail body feature vector by arranging the corresponding indexes and appearance frequency counters output from the mail document analysis unit 31 in a predetermined order.

  FIG. 7 is a diagram showing a mail text feature vector. As shown in the figure, the mail body feature vector is a pair of correspondence indexes and appearance frequency counters arranged in a predetermined order. The mail body feature vector generated by the mail body feature vector generation unit 35 is stored in the mail body feature vector storage unit 36 in association with the destination address.

  As described above, the mail body feature vectors of all mail documents transmitted to the mail erroneous transmission determination device 30 are sequentially stored in the mail body feature vector storage unit 36. Then, this routine ends.

  Note that the mail body feature vector storage unit 36 does not continue to store the mail body feature vectors of all the mail documents that have been sent, but stores only the mail body feature vectors transmitted from the present time to a predetermined period before. Also good. In other words, the mail text feature vector storage unit 36 may delete the mail text feature vector that is older than a predetermined period from the current time.

  Further, the mail body feature vector storage unit 36 may store only the mail body feature vector of the mail document having each address in the address book 32 as the destination address, instead of all the mail documents that have been sent.

FIG. 8 is a flowchart showing a mail text similarity model generation routine. This routine is batch-processed at regular intervals, for example, every 24 hours.
In step S21, the mail text similarity model generation unit 37 selects the first address from the address book 32, and proceeds to step S22. The first address corresponds to, for example, the destination address first registered in the address book 32, but is not limited to this, and may be any address registered in the address book 32.

  However, the addresses selected from the address book 32 may be limited to addresses related to the mail text similarity model to be updated, instead of all registered addresses. The email body similarity model to be updated corresponds to the email body similarity model for the address that is not reflected even though a new email is sent to the address related to the email body similarity model To do.

  In step S22, the mail body similarity model generation unit 37 extracts all mail body feature vectors having the selected address as the destination address from the mail body feature vector storage unit 36, and proceeds to step S23.

  In step S23, the mail body similarity model generation unit 37 randomly extracts a mail body feature vector having a destination address other than the currently selected address from the mail body feature vector accumulation unit 36, and proceeds to step S24. . The number of mail text feature vectors extracted in this step is the same as the number of mail text feature vectors extracted in step S22.

  In step S23, the mail body feature vector having an address other than the currently selected address as the destination address is extracted “randomly” because the mail body feature vector group has an address other than the currently selected address as the destination address. This is for retaining the characteristics (variation tendency) as a group of all the mail body feature vectors. In addition, the number of “same” number of the mail text feature vectors extracted from step S22 is the contribution from the mail text feature vector group having the selected address as the destination address with respect to the mail text similarity model. This is because the contribution from the mail body feature vector group with the address other than the selected address as the destination address is made equal.

  That is, in step S23, the mail body feature vector is extracted “randomly” and “same” as in step S22, so that the mail body similarity can be determined with high accuracy. This is to generate a model. However, in step S23, it is not essential to extract the mail body feature vector “randomly” and “same” as in step S22. That is, even if such a condition is not satisfied, a mail text similarity model necessary for mail mistransmission determination can be generated.

  In step S24, the mail text similarity model generation unit 37 uses the mail text feature vector group extracted in steps S22 and S23 to generate a mail text similarity model related to the selected address. The mail text similarity model is generated using, for example, SVM (Support Vector Machine).

In SVM, assuming a word space whose dimension is the number of all nouns used in the email body of all emails, the email body feature vector extracted from the email body of one email is Represented as one data point in the word space. Within the word space, the data points represented by all the mail body feature vector groups extracted at step S22 as a destination address and the mails extracted at step S23 as the destination address. If an optimum hyperplane that separates the data points represented by the text feature vector group can be generated, it is possible to classify any data point in the word space by the hyperplane.

  Therefore, the mail text similarity model generation unit 37 generates a mail text similarity model using SVM as the hyperplane. Then, the mail text similarity model generation unit 37 stores the mail text similarity model in the mail text similarity model storage unit 38, and proceeds to step S25.

  In step S25, the mail text similarity model generation unit 37 determines whether there is an unselected address among the addresses to be selected among the addresses registered in the address book 32, and the unselected address is determined. If there is, the process proceeds to step S26, and if there is no unselected address, this routine is terminated.

  In step S26, the mail text similarity model generation unit 37 selects an unselected address and returns to step S22. Then, by repeating the processing from step S22 to step S26, the mail text similarity model generation unit 37 calculates the mail text similarity model for all the addresses to be selected from the addresses registered in the address book 32. It is generated and stored in the mail text similarity model storage unit 38. As a result, the latest mail text similarity model for all addresses registered in the address book 32 is stored in the mail text similarity model storage unit 38.

  FIG. 9 is a diagram illustrating an example of a mail text similarity model relating to a predetermined address (address A). Here, it is assumed that the noun included in the mail text is a two-dimensional word space having only “XYZ company” and “Saito”. The mail text similarity model is expressed as a broken line in FIG. In FIG. 9, white circles are data points represented by the mail body feature vector group of the transmitted mail body with the address A as the destination address. A black circle is a data point represented by a mail body feature vector group of a sent mail body having an address other than the address A as a destination address. The mail body similarity model for the address addressed to A is a hyperplane (in this case, a straight line) that separates the area where the white circle mark exists from the area where the black circle mark exists with the highest probability in the two-dimensional word space. Expressed.

FIG. 10 is a flowchart showing an erroneous transmission determination processing routine.
In step S31, the reliability calculation unit 39 calculates the reliability for the destination address of the planned outgoing mail. Specifically, the reliability calculation unit 39 includes the mail body feature vector of the destination address generated by the mail body feature vector generation unit 35 and the mail body relating to the destination address stored in the mail body similarity model storage unit 38. The reliability of the destination address is calculated based on the similarity model.

  FIG. 11 is a diagram for explaining a reliability calculation method using the mail text similarity model. White and black triangles are the mail body feature vectors of the two mail documents to be transmitted. The reliability corresponds to the distance from the mail text similarity model (broken line) to the mail text feature vector (for example, each triangle mark) for which the reliability is calculated. In other words, the longer the distance is, the higher the reliability that the mail body feature vector belongs to the region.

  In FIG. 11, the white triangle mark is in the back of the area of the white circle mark, and the distance from the mail text similarity model is long. The black triangle mark is within the area of the black circle mark, but exists near the mail text similarity model, and the distance from the mail text similarity model is short. Therefore, the mail body having the mail body feature vector corresponding to the white triangle mark can be estimated as a transmission mail with the address addressed to A as the destination address, and its reliability is high. On the other hand, although it can be estimated that the mail body having the mail body feature vector corresponding to the black triangle mark is not a transmission mail with the address addressed to A as the destination address, its reliability is low.

  In step S <b> 32, the candidate destination address selection unit 33 selects one or more candidate destination addresses to replace the destination address extracted by the mail document analysis unit 31 from all the addresses registered in the address book 32. . The candidate destination address selection unit 33 selects a candidate destination address that is a character address similar to the destination address from the addresses in the address book 32 using, for example, the Levenshtein method or the n-gram method.

  The candidate destination address selection unit 33 compares the local part (user name) excluding the domain of the destination address with the local part of each address registered in the address book using, for example, the n-gram method, For each of the 32 addresses, the address similarity (0 to 100%), which is an index similar to the destination address, is calculated. Then, the candidate destination address selection unit 33 selects an address having an address similarity equal to or higher than a threshold value (for example, 50%) among the address similarities obtained for each address in the address book 32 as a candidate destination address.

  Note that the candidate destination address selection unit 33 may compare the destination address with each address in the address book 32, including not only the local part of the mail address but also the comment part of the mail address.

For example, in the case of “Taro Shinjuku” <shinjuku.tarou@example.co.jp>, the comment part is “Taro Shinjuku”. The candidate destination address selection unit 33 can select a candidate destination address having a higher similarity to the destination address by using not only the local part of the address included in the address book 32 but also the comment part. .
Then, the reliability calculation unit 39 selects an arbitrary candidate destination address from one or more candidate destination addresses selected by the candidate destination address selection unit 33, and proceeds to step S33.

  In step S33, the reliability calculation unit 39 calculates the reliability for the selected candidate destination address. Specifically, the reliability calculation unit 39 first reads the mail text similarity model related to the selected candidate destination address from the mail text similarity model storage unit 38. Next, the reliability calculation unit 39 receives the email body feature vector of the email to be transmitted generated by the email body feature vector generation unit 35 and the email body similarity model regarding the read candidate destination address being selected. Based on this, the reliability of the selected candidate destination address is calculated, and the process proceeds to step S34.

  In step S34, the destination address validity determination unit 40 determines whether or not the “reliability with respect to the candidate destination address” being selected is greater than the “reliability with respect to the destination address”. If the determination is affirmative, the process proceeds to step S35. If the determination is negative, the process proceeds to step S36. Even if the destination address validity determination unit 40 makes a negative determination, the destination address validity determination unit 40 may proceed to step S35 if the reliability of the destination address is lower than a predetermined threshold.

  In step S35, the destination address validity determination unit 40 determines that there is a possibility of erroneous transmission for the mail document to be transmitted, and sets the selected candidate destination address as the candidate destination address to be presented to the mail client 10. Then, the process proceeds to step S36. Note that the destination address validity determination unit 40 may set the candidate destination address when a negative determination is made in step S34 and the reliability of the destination address is lower than a predetermined threshold.

  In step S36, the destination address validity determination unit 40 determines whether there is an unselected candidate destination address. If there is an unselected candidate destination address, the process proceeds to step S37, and if there is no unselected candidate destination address. Advances to step S38.

  In step S37, the destination address validity determination unit 40 selects an unselected candidate destination address and returns to step S33. Then, by repeatedly executing steps S33 to S37, all combinations of “reliability with respect to destination address” and “reliability with respect to candidate destination address” are respectively compared, and the process proceeds to step S38.

  In step S38, the destination address validity determination unit 40 transmits the determination result to the mail client 10. Specifically, the destination address validity determination unit 40 determines that the possibility of erroneous transmission is low when the “reliability with respect to the destination address” is higher than all the “reliabilities with respect to candidate destination addresses”. Is sent to the mail client 10.

  In addition, when there is a “reliability with respect to the candidate destination address” higher than the “reliability with respect to the destination address”, the destination address validity determination unit 40, together with the determination result that the possibility of erroneous transmission is high, in step S35. The candidate destination address set in (1) is transmitted to the mail client 10 as a candidate destination address to be presented to the mail client 10. Further, when there are a plurality of candidate destination addresses, the destination address validity determination unit 40 transmits the priority order (reliability order) of the candidate destination addresses to the mail client 10 together with the plurality of candidate destination addresses. Also good.

As a result, when the mail client 10 transmits the mail document to the mail error transmission determination device 30, the mail client 10 receives a determination result indicating the possibility of erroneous transmission.
Further, the mail client 10 can receive from the mail erroneous transmission determination device 30 a candidate destination address that replaces the original destination address, together with a determination result that has a high possibility of erroneous transmission. Further, the mail client 10 can receive the priority order of candidate destination addresses when there are a plurality of candidate destination addresses.

The mail client 10 transmits the e-mail to be transmitted via the mail server 50 when the possibility of erroneous transmission is low, and the e-mail to be transmitted when the possibility of erroneous transmission is high. Suspend sending
When the mail client 10 suspends transmission of the e-mail, the mail client 10 can present the candidate destination address in place of the current destination address to the user to alert the user. In addition, when there are a plurality of candidate destination addresses, the mail client 10 presents them to the user from the candidate destination addresses with the highest priority (for example, the candidate destination address with the higher priority is higher in the drop-down list). You can also

In addition, this invention is not limited to embodiment mentioned above, It can apply also to what was changed in the design within the range of the matter described in the claim.
For example, in FIG. 1, the erroneous email transmission determination device 30 is configured independently of the email client 10 or the email server 50, but may be configured to be incorporated in the email client 10 or the email server 50.

  In the embodiment described above, destination addresses are extracted from all mail documents transmitted to the erroneous mail transmission determination device 30 shown in FIG. 4, and all the extracted destination addresses are registered in the address book 32. However, the present invention is not limited to this example, and only the destination address of the electronic mail transmitted through the result of the erroneous mail transmission determination may be registered in the address book 32.

  In addition, the mail document analysis unit 31 shown in FIG. 4 performs morphological analysis on only the mail body. If the text information is included in the e-mail, the mail document analysis unit 31 performs morphological analysis including the mail header and other information. Also good.

  The mail text similarity model is not limited to SVM, and is not particularly limited as long as it statistically indicates the certainty that the mail text is related to a specific destination address. As the mail text similarity model, for example, a naive Bayes, a decision tree (CART), a maximum entropy (ME) model, or the like may be used.

  The mail body similarity model of the present embodiment is sequentially updated from the beginning based on the mail body feature vectors of all mail documents transmitted to the mail erroneous transmission determination device 30. However, the mail text similarity model may reflect, for example, a difference from the mail text feature vector of today's mail document based on the one generated at a predetermined point in the past.

Further, the destination address validity determination unit 40 illustrated in FIG. 4 may be provided in the mail client 10 instead of the mail erroneous transmission determination device 30. In this case, the reliability calculation unit 39 of the erroneous mail transmission determination device 30 may transmit the reliability for the destination address and the candidate destination address to the mail client 10.
On the other hand, the control unit 14 of the mail client 10 shown in FIG. 2 has the function of the destination address validity determination unit 40 as described above, and uses each reliability transmitted from the mail erroneous transmission determination device 30. Thus, the processes of steps S34 to S37 shown in FIG. 10 are repeatedly executed. In this way, the mail client 10 may make a final determination as to whether or not the mail is erroneously transmitted.

[Second Embodiment]
The erroneous email transmission determination device 30 according to the second embodiment of the present invention determines whether an email to be transmitted is erroneously transmitted without generating a candidate destination address. In this case, in the erroneous transmission determination processing routine shown in FIG. 10, the candidate destination address selection processing in steps S32 and steps S33, S36, and S37 are omitted. Further, in step S34, unlike the first embodiment, the following processing is executed.

  In step S34, the destination address validity determination unit 40 determines whether or not the “reliability with respect to the destination address” is greater than a predetermined threshold value. If the determination is affirmative, the destination address validity determination unit 40 determines that the possibility of erroneous transmission is low. In the case of negative determination, it is determined that there is a high possibility of erroneous transmission, and this determination result is transmitted to the mail client 10.

The mail client 10 uses the determination result by the erroneous mail transmission determination device 30 to transmit the electronic mail when the possibility of erroneous transmission of the electronic mail is low, and when the possibility of erroneous transmission of the electronic mail is high. Can suspend the transmission of the e-mail.
The destination address validity determination unit 40 may transmit not only the determination result but also the reliability to the mail client 10. The mail client 10 may ask for the final determination of mail transmission by presenting the reliability to the user, for example.

  The mail client 10, the mail erroneous transmission determination device 30, and the mail server 50 described above are not limited to a hardware configuration, and may be a general-purpose computer in which a program for executing the processing described above is installed.

  Further, the destination address validity determination unit 40 illustrated in FIG. 4 may be provided in the mail client 10 instead of the mail erroneous transmission determination device 30. In this case, the reliability calculation unit 39 of the erroneous mail transmission determination device 30 may transmit the calculated reliability to the mail client 10.

  On the other hand, the control unit 14 of the mail client 10 shown in FIG. 2 has the function of the destination address validity determination unit 40 as described above, and whether or not the reliability of the destination address is greater than a predetermined threshold value. Determine. When the reliability is greater than a predetermined threshold, the control unit 14 determines that the possibility of erroneous transmission is low, and controls the communication unit 13 to transmit an e-mail. In addition, when the reliability is equal to or lower than a predetermined threshold, the control unit 14 determines that there is a high possibility of erroneous transmission of an electronic mail and suspends transmission of the electronic mail. In this way, the mail client 10 may make a final determination as to whether or not the mail is erroneously transmitted.

DESCRIPTION OF SYMBOLS 10 Mail client 30 Mail mistransmission determination apparatus 31 Mail document analysis part 32 Address book 33 Candidate destination address selection part 34 Noun dictionary 35 Mail text feature vector generation part 36 Mail text feature vector storage part 37 Mail text similarity model generation part 38 Mail Text similarity model storage unit 39 Reliability calculation unit 40 Destination address validity determination unit 50 Mail server

Claims (9)

Feature information generating means for generating feature information relating to the content of the e-mail to be transmitted;
Storage means for storing the feature information related to the content of the transmitted electronic mail and the transmission destination of the electronic mail in association with each other;
A transmission destination candidate selection means for selecting a transmission destination candidate similar to the transmission destination of the e-mail to be transmitted from among transmission destinations of the transmitted e-mail;
For each transmission destination stored in the storage unit, the feature information stored in the storage unit in association with the transmission destination is associated with another transmission destination different from the transmission destination. Based on the feature information stored in the means, in a word space having the number of words included in the e-mail as a dimensional number, the word space region of the feature information associated with the destination, and the other Is separated from the word space area of the feature information associated with the transmission destination, and the feature information related to the contents of any e-mail is stored in the word space area of the feature information associated with the transmission destination. a similarity model generating means for generating a similarity model to which the genus either criterion,
Feature information related to the content of the transmission target email generated by the feature information generation unit, the similarity model regarding the transmission destination of the transmission target email generated by the similarity model generation unit, and the similarity Reliability calculation means for calculating each reliability of the transmission destination and destination candidates of the transmission target email based on the similarity model related to the transmission destination candidates generated by the degree model generation means;
An e-mail erroneous transmission determination device comprising: It further includes an erroneous transmission determination unit that determines whether or not there is a high possibility of an error for the transmission destination of the email to be transmitted by comparing the reliability calculated by the reliability calculation unit.
The e-mail erroneous transmission determination device according to claim 1 . When the erroneous transmission determination means determines that there is a high possibility of an error for the transmission destination of the email to be transmitted, the transmission destination having a higher reliability than the reliability of the transmission destination of the email to be transmitted A selection means for selecting candidates is further provided.
The electronic mail erroneous transmission determination device according to claim 2 . The transmission destination candidate selection means selects the transmission destination candidate using a local part obtained by removing a domain part from the transmission destination.
The electronic mail erroneous transmission determination device according to any one of claims 1 to 3 . The transmission destination candidate selection means selects the transmission destination candidate by further using a comment part added to the transmission destination.
Email wrong transmission determining apparatus according to any one of claims 1 to 4. An e-mail creation means for creating an e-mail to be sent with a specified destination;
The e-mail erroneous transmission determination device according to claim 2 , wherein it is determined whether or not there is a high possibility of an error with respect to a transmission destination of the e-mail to be transmitted generated by the e-mail generation unit,
When the e-mail erroneous transmission determination device determines that the possibility of an error is not high for the transmission destination, the e-mail to be transmitted is transmitted, and when it is determined that the error is high for the transmission destination A transmission means for deferring transmission of the e-mail to be transmitted;
E-mail transmission system equipped with. An e-mail creation means for creating an e-mail to be sent with a specified destination;
The e-mail erroneous transmission determination device according to claim 3 , wherein it is determined whether or not there is a high possibility of an error with respect to a transmission destination of the e-mail to be transmitted generated by the e-mail generation unit,
Display means for displaying the destination candidates selected by the selection means of the e-mail erroneous transmission determination device;
When the e-mail erroneous transmission determination device determines that the possibility of an error is not high for the transmission destination, the e-mail to be transmitted is transmitted, and when it is determined that the error is high for the transmission destination The transmission target electronic mail is suspended, and when one of the transmission destination candidates displayed by the display means is designated during the suspension of transmission, the transmission target electronic mail is designated as the designated transmission destination candidate. A transmission means for sending an email;
E-mail transmission system equipped with. A program for causing a computer to function as each unit of the erroneous e-mail transmission determination apparatus according to any one of claims 1 to 5 . The program for functioning a computer as each means of the electronic mail transmission system of Claim 6 or Claim 7 .