Server, method, and program

Abstract

To provide a server, a method, and a program that prevent incorrect transmission of emails.SOLUTION: In an email incorrect transmission prevention system, a generative AI provision server determines the likelihood of an email being incorrectly transmitted using a machine learning model based on information about the email subject to incorrect transmission determination among emails received from a client terminal by an incorrect transmission prevention server, and instructions for causing the machine learning model to execute the incorrect transmission determination. The incorrect transmission prevention server then transmits information to the client terminal to allow an email sender to confirm a result of the determination before the email is transmitted to a transmission destination.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to a server, a method, and a program. [Background technology]

[0002] A known technique for preventing the erroneous sending of electronic mail (hereinafter also simply referred to as "mail") is to display a confirmation screen for prompting the sender to confirm the legitimacy of the destination address, the legitimacy of the attached file, etc., before sending the email. For example, Patent Document 1 discloses a technique for displaying a confirmation screen for prompting the sender to confirm the legitimacy of the destination address before sending the email, and not allowing the email to be sent until the confirmation result on this confirmation screen is obtained. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2007-87327 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional techniques for preventing erroneous e-mail transmission do not allow for effective confirmation based on the content of the e-mail, so confirmation by the sender can sometimes become a mere formality.

[0005] The present disclosure has been made in consideration of the above points, and aims to prevent erroneous sending of emails. [Means for solving the problem]

[0006] A server according to one aspect of the present disclosure has a judgment unit that judges the possibility that an email has been sent by mistake using a machine learning model based on information regarding the email that is to be judged as being sent by mistake and instructions for causing the machine learning model to perform the judgment on the email as being sent by mistake, and an output unit that outputs information to allow the sender of the email to confirm the result of the judgment made by the judgment unit before the email is sent to the destination. [Effects of the Invention]

[0007] This can prevent emails from being sent to the wrong person. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a diagram illustrating an example of the overall configuration of a system for preventing erroneous email transmission according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram illustrating an example of a functional configuration of a client terminal according to the present embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of the functional configuration of an erroneous transmission prevention server according to the present embodiment. [Figure 4] FIG. 2 is a diagram illustrating an example of the functional configuration of a generation AI providing server according to this embodiment. [Figure 5] FIG. 2 is a sequence diagram showing an example of the operation of the system for preventing erroneous email transmission according to the present embodiment. [Figure 6] FIG. 10 is a diagram showing an example (part 1) of a prompt for determination. [Figure 7] FIG. 10 is a diagram showing an example (part 2) of a prompt for determination. [Figure 8] FIG. 10 is a diagram showing an example (part 3) of a prompt for determination. [Figure 9] FIG. 10 is a diagram showing an example (part 4) of a prompt for determination. [Figure 10] FIG. 10 is a diagram showing an example (part 5) of a prompt for determination. [Figure 11] FIG. 10 is a diagram showing an example (part 6) of a prompt for determination. [Figure 12] FIG. 10 is a diagram showing an example (part 1) of a transmission confirmation screen. [Figure 13] FIG. 10 is a diagram showing an example (part 2) of a transmission confirmation screen. [Figure 14] FIG. 10 is a diagram showing an example (part 3) of a transmission confirmation screen. [Figure 15] FIG. 10 is a diagram showing a modified example of the transmission confirmation screen. DETAILED DESCRIPTION OF THE INVENTION

[0009] An embodiment of the present invention will be described in detail below with reference to the drawings. In the following embodiment, a system 1 for preventing erroneous email transmission will be described. The system 1 can perform an erroneous email transmission determination based on the content of the email to be sent using artificial intelligence (AI), and then display a confirmation screen containing the determination result on the sender's device. The system 1 for preventing erroneous email transmission according to the present embodiment enables effective confirmation based on the content of the email to be sent, thereby preventing confirmation from becoming a mere formality and effectively preventing erroneous email transmission. The confirmation screen is a screen that prompts the sender to confirm the content of the email to be sent (e.g., recipient, body, attachments, etc.). The generative AI refers to a program or a group of modules that receives instructions or questions, such as prompts, executes processing in accordance with the prompts, and generates and outputs various content (e.g., text, images, audio, video, etc.). The prompts may be referred to, for example, as "instructions" or "questionnaires." The generative AI, also known as "generative AI," is generally implemented by a program or a group of modules that includes a machine learning model (especially, a large-scale natural language model (LLM))).

[0010] In the following, it is assumed, as an example, that the sender of the email is a member of a company (e.g., an employee of a company). However, the following embodiment can be similarly applied even if the sender of the email is a person other than a member of a company. Specific examples of a person other than a member of a company include, for example, a school staff member or student, or an organization staff member.

[0011] <Overall configuration example> An example of the overall configuration of an email mis-send prevention system 1 according to this embodiment is shown in Figure 1. As shown in Figure 1, the email mis-send prevention system 1 according to this embodiment includes one or more client terminals 10, a mis-send prevention server 20, and a generation AI provision server 30. Each client terminal 10 and the mis-send prevention server 20 are located within a company's internal system environment E and are communicatively connected via an internal company LAN (Local Area Network) 40. Furthermore, the mis-send prevention server 20 and the generation AI provision server 30 are communicatively connected via an arbitrary communication network 50 such as the Internet.

[0012] The client terminal 10 is a terminal of any type used by a sender or receiver of email (hereinafter also referred to as a "user"). The client terminal 10 creates and sends emails (including forwarding and replying to emails) and receives emails in response to user operations. The client terminal 10 is realized, for example, by a PC (personal computer). However, the client terminal 10 may also be realized, for example, by a smartphone, tablet terminal, wearable device, etc.

[0013] The erroneous transmission prevention server 20 receives an email sent from the client terminal 10 and sends a request for erroneous transmission determination (hereinafter also referred to as an "erroneous transmission determination request") according to the content of the email to the generation AI provision server 30. Here, the erroneous transmission determination request includes, for example, a prompt for erroneous transmission determination (hereinafter also referred to as a "determination prompt") and information about the email that is the target of erroneous transmission determination (hereinafter also referred to as "email information"). Furthermore, when the erroneous transmission prevention server 20 receives the result of the erroneous transmission determination in response to the erroneous transmission determination request (hereinafter also referred to as the "erroneous transmission determination result"), it causes the client terminal 10 to display a confirmation screen (hereinafter also referred to as a "transmission confirmation screen") that includes the erroneous transmission determination result. Furthermore, the erroneous transmission prevention server 20 checks (checks) the transmission confirmation screen displayed on the client terminal 10, and if the sender instructs permission to send, it sends (forwards) the email received from the client terminal 10 to its destination. The email information includes, for example, the body of the email that is the target of erroneous transmission determination, destinations (To, Cc, Bcc), attachments, etc. The erroneous transmission prevention server 20 is realized by, for example, one or more computers.

[0014] When the generation AI provision server 30 receives the erroneous transmission determination request sent from the erroneous transmission prevention server 20, it performs an erroneous transmission determination using the generation AI based on the determination prompt and email information included in the erroneous transmission determination request. The generation AI provision server 30 also transmits the erroneous transmission determination result for that erroneous transmission determination to the erroneous transmission prevention server 20. The generation AI provision server 30 is realized, for example, by one or more computers.

[0015] The overall configuration of the email mis-sending prevention system 1 shown in Figure 1 is an example and is not limited to this. For example, the internal system environment E is not limited to the internal system environment of a certain company, but may be a system environment of a school or organization, or may be a system environment between multiple companies. Furthermore, the generation AI provision server 30 may exist within the internal system environment E, or the mis-sending prevention server 20 and the generation AI provision server 30 may be configured as a single unit. Furthermore, the person managing the internal system environment E may be different from the person managing the generation AI provision server 30 (i.e., the mis-sending prevention server 20 may send a mis-sending determination request to the generation AI provision server 30 managed and operated by an external service vendor, etc.).

[0016] <Example of functional configuration> <Client terminal 10> An example of the functional configuration of the client terminal 10 according to this embodiment is shown in Fig. 2. As shown in Fig. 2, the client terminal 10 according to this embodiment has a communication unit 101, an email function unit 102, and a UI control unit 103. These units are realized, for example, by processing executed by a computing device such as a CPU (Central Processing Unit) by one or more programs installed in the client terminal 10. The client terminal 10 according to this embodiment also has a storage unit 104. The storage unit 104 is realized, for example, by a storage area of ​​a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory.

[0017] The communication unit 101 performs various data communications, including sending and receiving emails. The email function unit 102 creates emails in response to operations accepted by the UI control unit 103. The UI control unit 103 accepts user operations using input devices such as a keyboard, mouse, or touch panel, and displays various screens (e.g., a sending confirmation screen) on a display device such as a display. The storage unit 104 stores various data (e.g., email sending and receiving history, files attached to those emails, etc.).

[0018] <<Wrong Transmission Prevention Server 20>> An example of the functional configuration of the erroneous transmission prevention server 20 according to this embodiment is shown in Fig. 3. As shown in Fig. 3, the erroneous transmission prevention server 20 according to this embodiment includes a communication unit 201, a target email judgment unit 202, a prompt acquisition unit 203, and a judgment request creation unit 204. Each of these units is realized, for example, by a processing unit such as a CPU executing one or more programs installed in the erroneous transmission prevention server 20. The erroneous transmission prevention server 20 according to this embodiment also includes a memory unit 205. The memory unit 205 is realized, for example, by a storage area of ​​a storage device such as an HDD, SSD, or flash memory.

[0019] The communication unit 201 performs various data communications, including sending and receiving emails, sending requests for erroneous transmission determination, and receiving erroneous transmission determination results, and cancels the transmission of emails for which a cancellation instruction has been issued. The target email determination unit 202 determines whether an email received by the communication unit 201 is subject to erroneous transmission determination. When the target email determination unit 202 determines that the email is subject to erroneous transmission determination, the prompt acquisition unit 203 acquires a determination prompt stored in the memory unit 205. The determination request creation unit 204 creates an erroneous transmission determination request that includes the determination prompt acquired by the prompt acquisition unit 203 and email information about the email that is subject to erroneous transmission determination. The memory unit 205 stores various data that includes at least the determination prompt.

[0020] <Generation AI provision server 30> An example of the functional configuration of the generation AI provision server 30 according to this embodiment is shown in Figure 4. As shown in Figure 4, the generation AI provision server 30 according to this embodiment has a communication unit 301 and an erroneous transmission determination unit 302. Each of these units is realized, for example, by a process in which one or more programs installed in the generation AI provision server 30 are executed by a computing device such as a CPU. The generation AI provision server 30 according to this embodiment also has a memory unit 303. The memory unit 303 is realized, for example, by a memory area of ​​a storage device such as an HDD, SSD, or flash memory.

[0021] The communication unit 301 performs various data communications, including receiving an erroneous transmission determination request and transmitting an erroneous transmission determination result. The erroneous transmission determination unit 302 performs an erroneous transmission determination using a generation AI based on the determination prompt and email information included in the erroneous transmission determination request received by the communication unit 301. The storage unit 303 stores various data (e.g., trained parameters of a machine learning model that realizes the generation AI). Note that a machine learning model whose parameters have been trained may be called a "trained machine learning model" or the like.

[0022] <Example of operation> An example of the operation of the system 1 for preventing erroneous email transmission according to this embodiment will be described below with reference to Fig. 5. In the following, it is assumed that, for example, a user performs an operation to create an email with an attachment on an email creation screen displayed by the UI control unit 103 (hereinafter also referred to as an "email creation operation"), and that the email creation operation is accepted by the UI control unit 103.

[0023] The email function unit 102 of the client terminal 10 creates an email with an attachment (step S101) in response to an email creation operation accepted by the UI control unit 103. Hereinafter, the email created in this step will also be referred to as an "email to be sent."

[0024] The communication unit 101 of the client terminal 10 transmits the email to be sent created in step S101 above to the erroneous transmission prevention server 20 (step S102).

[0025] The communication unit 201 of the erroneous transmission prevention server 20 receives the email to be sent sent in step S102 above (step S103).

[0026] The target email determination unit 202 of the erroneous transmission prevention server 20 determines whether the target email received in step S103 above is a target for erroneous transmission detection (step S104). For example, the target email determination unit 202 may determine that the target email is a target for erroneous transmission detection if it satisfies a predetermined condition, and may determine that the target email is not a target for erroneous transmission detection if it does not. Various conditions can be used as the condition, such as "the sender address of the target email is not included in the list of email addresses that are not subject to erroneous transmission detection." However, this is just one example, and the condition may also be, for example, "the sender address of the target email is included in the list of email addresses that are subject to erroneous transmission detection."

[0027] If it is determined in step S104 above that the email to be sent is subject to erroneous transmission determination, the prompt acquisition unit 203 of the erroneous transmission prevention server 20 acquires the determination prompt stored in the storage unit 205 (step S105).

[0028] The determination request creation unit 204 of the erroneous transmission prevention server 20 creates an erroneous transmission determination request that includes the determination prompt acquired in step S105 and the email information of the email to be sent (step S106). The email information includes, for example, the body of the email, attachments, destination (To), destination (Cc), destination (Bcc), etc.

[0029] The communication unit 201 of the erroneous transmission prevention server 20 transmits the erroneous transmission determination request created in the above step S106 to the generation AI providing server 30 (step S107).

[0030] The communication unit 301 of the generation AI provision server 30 receives the erroneous transmission determination request sent in the above step S107 (step S108).

[0031] The erroneous transmission judgment unit 302 of the generation AI providing server 30 performs an erroneous transmission judgment for the email to be sent based on the judgment prompt and email information included in the erroneous transmission judgment request received in step S108 above (step S109). That is, the erroneous transmission judgment unit 302 inputs the judgment prompt and at least one of the following information: the body, attachment, destination (To), destination (Cc), and destination (Bcc), and uses the generation AI to judge the possibility of erroneous transmission for the email to be sent. However, the generation AI may also input other necessary information (e.g., upper limit values, etc., as described below). Specific examples of erroneous transmission judgment and judgment prompts for making that judgment will be described later.

[0032] The communication unit 301 of the generation AI provision server 30 transmits the erroneous transmission determination result indicating the result of the erroneous transmission determination in the above step S109 to the erroneous transmission prevention server 20 (step S110).

[0033] The communication unit 201 of the erroneous transmission prevention server 20 receives the erroneous transmission determination result transmitted in step S110 (step S111).

[0034] The communication unit 201 of the erroneous transmission prevention server 20 transmits the erroneous transmission determination result received in step S111 above to the client terminal 10 that is the sender of the email to be sent (step S112). At this time, the communication unit 201 of the erroneous transmission prevention server 20 may also transmit display information for the transmission confirmation screen (e.g., information such as HTML (HyperText Markup Language) that defines the transmission confirmation screen) to the client terminal 10.

[0035] The communication unit 101 of the client terminal 10 receives the erroneous transmission determination result transmitted in the above step S112 (step S113).

[0036] The UI control unit 103 of the client terminal 10 displays a transmission confirmation screen including the erroneous transmission determination result received in step S113 above (step S114). Hereinafter, the transmission confirmation screen includes display components such as buttons for the user to instruct permission to send or cancellation of sending of the email to be sent, and it is assumed that an operation to instruct permission to send (hereinafter also referred to as a "transmission permission instruction operation") or an operation to instruct cancellation of sending (hereinafter also referred to as a "transmission cancellation instruction operation") has been performed. Specific examples of the transmission confirmation screen will be described later.

[0037] The UI control unit 103 of the client terminal 10 accepts an operation to instruct permission to send or an operation to instruct cancellation of sending on the sending confirmation screen displayed in the above step S114 (step S115).

[0038] The communication unit 101 of the client terminal 10 transmits a transmission control instruction to the erroneous transmission prevention server 20 (step S116). Here, if a transmission permission instruction operation is accepted in the above step S115, the transmission control instruction includes information indicating transmission permission. On the other hand, if a transmission cancellation instruction operation is accepted in the above step S115, the transmission control instruction includes information indicating transmission cancellation.

[0039] The communication unit 201 of the erroneous transmission prevention server 20 receives the transmission control instruction transmitted in step S116 above (step S117).

[0040] If the information included in the transmission control instruction received in step S117 indicates permission to send, the communication unit 201 of the erroneous transmission prevention server 20 sends the email to be sent to the destination address (step S118). That is, the communication unit 201 of the erroneous transmission prevention server 20 sends the email to be sent to the email addresses specified in the destination (To), Cc, and Bcc fields of the email to be sent.

[0041] On the other hand, if the information included in the transmission control instruction received in step S117 indicates that transmission should be stopped, the communication unit 201 of the erroneous transmission prevention server 20 stops sending the email to be sent (step S119). Note that the email whose transmission has been stopped is, for example, discarded.

[0042] If it is not determined in step S104 above that the email to be sent is subject to erroneous transmission judgment (i.e., the email to be sent is not subject to erroneous transmission judgment), the communication unit 201 of the erroneous transmission prevention server 20 sends the email to the destination address (step S120). That is, the communication unit 201 of the erroneous transmission prevention server 20 sends the email to the email addresses specified in the destination (To), Cc, and Bcc fields of the email to be sent.

[0043] <Examples of erroneous transmission detection and prompts for that detection> The following describes a specific example of the erroneous transmission determination in step S109 of FIG. 5 and a prompt for making the determination.

[0044] Example of erroneous transmission detection (part 1) In Example 1 of erroneous transmission judgment, the body of the email and attachments are each classified into six predetermined categories: "Quotation," "Contract (Order / Order)," "Inquiry," "Meeting Request," "Notice," and "Other," and then the possibility of erroneous transmission (risk level) is judged on a three-level scale based on the classification results and the basis for the classification.

[0045] Specifically, the possibility of erroneous transmission is determined by the following steps 1-1 and 1-2.

[0046] Step 1-1: The body of the email is sorted into one of six categories, and the attachments are also sorted into one of six categories. At this time, the basis for the decision to sort into one of the six categories is also calculated. When sorting attachments, the text contained in the attachments is also sorted.

[0047] Step 1-2: Based on the classification of the email body and the basis for that classification, and the classification of the attachment and the basis for that classification, the possibility of erroneous transmission (risk level) is determined on a three-level scale: "high," "medium," or "low." In other words, if the classifications of the email body and the attachment are different, the risk level is "high." If the classifications of the email body and the attachment are the same but the basis for the judgment cannot be clearly indicated, the risk level is "medium." If the classifications of the email body and the attachment are the same and the basis for the judgment can be clearly indicated, the risk level is "low."

[0048] FIG. 6 shows an example of a prompt for determining the possibility of erroneous transmission according to steps 1-1 and 1-2 above. The prompt for determination 1100 shown in FIG. 6 includes a "#instruction" definition 1110, a "#constraint" definition 1120, a "#category and its knowledge" definition 1130, a "#determination rule" definition 1140, a "#input" definition 1150, and a "#output" definition 1160. By inputting the prompt for determination 1100 shown in FIG. 6 and the body, attachments, and destination (To) of the email to be sent, the generation AI can perform the erroneous transmission determination shown in steps 1-1 and 1-2 above. When the prompt for determination 1100 shown in FIG. 6 is input, the classification of the body is output as "text_category," the classification of the attachment as "file_text_category," and the risk level as "level." In addition to these, the reason for the judgment result is output as "description," the recipient's company name and name written in the body of the email are output as "recipient," and a summary of the body of the email is output as "summary."

[0049] Example of erroneous transmission detection (part 2) In Example 2 of the mis-sending judgment, the destination indicated by the email address specified in the To field is compared with the destination included in the body of the email, and the possibility of mis-sending (risk level) is judged in two stages based on the comparison results.

[0050] Specifically, the possibility of erroneous transmission is determined by the following steps 2-1 and 2-2.

[0051] Step 2-1: Extract the "company name" and "name" from the body of the email as the first recipient.

[0052] Step 2-2: After extracting the email address specified in the destination (To) as the second destination, the first destination and the second destination are compared, and the possibility of erroneous transmission (risk level) is determined based on the comparison results. That is, if the first destination and the second destination are different, the risk level is set to "incorrect destination," and if the first destination and the second destination are the same, the risk level is set to "not incorrect destination."

[0053] FIG. 7 shows an example of a prompt for determining the possibility of erroneous transmission according to steps 2-1 and 2-2 above. The prompt for determination 1200 shown in FIG. 7 includes a "#instruction" definition 1210, a "#constraint" definition 1220, a "#determination rule" definition 1230, a "#input" definition 1240, and a "#output" definition 1250. By inputting the prompt for determination 1200 shown in FIG. 7 and the body, attachments, and destination (To) of the email to be sent, the generation AI can perform the erroneous transmission determination shown in steps 2-1 and 2-2 above. When the prompt for determination 1200 shown in FIG. 7 is used as input, the risk level is output as "level." Additionally, the reason for the determination result is output as "description," the recipient's company name and name written in the body as "recipient," and a summary of the body as "summary."

[0054] In step 2-1 above, "company name" and "name" are extracted from the text, but this is just one example, and information other than "company name" and "name" may be extracted as long as it is information that represents the destination. For example, "company name," "department name," and "name" may be extracted, or "company name," "department name," "section name," and "name" may be extracted.

[0055] Example of erroneous transmission detection (part 3) In the example of erroneous sending judgment (part 3), it is judged whether the destination included in the body of the email is included in the attached file, and based on the judgment result, the possibility of erroneous sending (risk level) is judged in two stages.

[0056] Specifically, the possibility of erroneous transmission is determined by the following steps 3-1 and 3-2.

[0057] Step 3-1: Extract the "company name," "department name," and "name" from the text.

[0058] Step 3-2: If all of the "company name," "department name," and "name" extracted in step 3-1 above are present in the text included in the attachment, the risk level is set to "no incorrect destination," otherwise the risk level is set to "incorrect destination."

[0059] FIG. 8 shows an example of a prompt for judgment when judging the possibility of erroneous transmission according to steps 3-1 and 3-2 above. The prompt for judgment 1300 shown in FIG. 8 includes a "#instruction" definition 1310, a "#constraint" definition 1320, a "#judgment rule" definition 1330, a "#input" definition 1340, and a "#output" definition 1350. By inputting the prompt for judgment 1300 shown in FIG. 8 and the body and attachment of the email to be sent, the generation AI can make the judgment of erroneous transmission shown in steps 3-1 and 3-2 above. When the prompt for judgment 1300 shown in FIG. 8 is used as input, the risk level is output as "level." In addition, the reason for the judgment result is output as "description," the recipient's company name and name written in the body as "recipient," and a summary of the body as "summary."

[0060] In step 3-1 above, "company name," "department name," and "name" are extracted from the text, but this is just one example, and information other than "company name," "department name," and "name" may be extracted as long as it is information that represents the destination. For example, "company name" and "name" may be extracted, or "company name," "department name," "section name," and "name" may be extracted.

[0061] Furthermore, in step 3-1 above, it is determined whether all of the "company name," "department name," and "name" are present in the text included in the attached file, but this is just one example, and it may be determined whether all or part of the "company name," "department name," and "name" are present in the text included in the attached file according to conditions set or selected by the user. Here, the conditions may be, for example, "all of the "company name," "department name," and "name" are present in the text included in the attached file, excluding differences in spelling (e.g., katakana spelling, romanized spelling, etc.)," ​​"at least the "company name" and "name" are present in the text included in the attached file," "at least the "company name" and "department name" are present in the text included in the attached file," or other conditions.

[0062] Example of erroneous transmission detection (part 4) In Example 4 of Mis-Sending Detection, an address book that associates email addresses with names and their affiliations (company names, department names, etc.) and job titles is used to compare the address indicated by the email address specified in the To field with the address included in the body of the email, and based on the comparison results, the possibility of mis-sending (risk level) is determined in two stages.

[0063] Specifically, the possibility of erroneous transmission is determined by the following steps 4-1 and 4-2.

[0064] Step 4-1: Extract the "company name," "department name," and "name" from the body of the email as the first recipient.

[0065] Step 4-2: Using the address book, extract the "company name," "department name," and "name" from the email address specified in the "To" field as secondary recipients. Note that the generation AI can obtain or reference the address book using a technology or function called Retrieval Augmented Generation (RAG).

[0066] Step 4-3: If the first and second destinations are the same, the risk level is set to "no incorrect destination," and if the first and second destinations are different, the risk level is set to "incorrect destination."

[0067] If multiple email addresses are specified in the destination (To) field, step 4-2 above is executed for each email address according to a predetermined priority, with the upper limit set or selected by the user or the like as the upper limit. For example, the priority may be higher for email addresses whose "name" extracted from the body of the email is associated with the address book, and lower for email addresses that do not, or the higher the job title associated with the email address in the address book, the higher the priority.

[0068] FIG. 9 shows an example of a prompt for determining the possibility of erroneous transmission according to steps 4-1 to 4-3 above. The prompt for determination 1400 shown in FIG. 9 includes a "#instruction" definition 1410, a "#constraint" definition 1420, a "#determination rule" definition 1430, a "#input" definition 1440, and a "#output" definition 1450. By inputting the prompt for determination 1400 shown in FIG. 9, the body and destination (To) of the email to be sent, and an upper limit, the generation AI can perform the erroneous transmission determination shown in steps 4-1 to 4-3 above. When the prompt for determination 1400 shown in FIG. 9 is used as input, the risk level is output as "level." Additionally, the reason for the determination result is output as "description," the recipient's company name and name written in the body as "recipient," and a summary of the body as "summary."

[0069] In step 4-1 above, "company name," "department name," and "name" are extracted from the text, but this is just one example, and information other than "company name," "department name," and "name" may be extracted as long as it is information that represents the destination. For example, "company name" and "name" may be extracted, or "company name," "department name," "section name," and "name" may be extracted.

[0070] Example of erroneous transmission detection (part 5) In Example 5 of the erroneous transmission judgment, the body of the email and the attachment are each classified into six predetermined categories, and it is determined whether the destination included in the body of the email is included in the attachment. The possibility of erroneous transmission (risk level) is judged based on the classification and judgment results.

[0071] Specifically, the possibility of erroneous transmission is determined by the following steps 5-1 to 5-5.

[0072] Step 5-1: As in step 1-1 above, the body of the email and any attachments are sorted into one of six categories.

[0073] Step 5-2: As in step 3-1 above, extract the "company name," "department name," and "name" from the text.

[0074] Step 5-3: If the classification result of the body of the email differs from that of the attachment, the risk level is set to "Incorrect attachment." If the classification result of the body of the email differs from that of the attachment, the risk level is set to "No incorrect attachment." Furthermore, if none of the "company name," "department name," or "name" extracted in step 5-2 above is present in the text included in the attachment, the risk level is set to "Incorrect attachment."

[0075] Step 5-4: If all of the "company name," "department name," and "name" extracted in step 5-2 above are present in the text included in the attachment, add "no incorrect destination" to the risk level; otherwise, add "incorrect destination" to the risk level.

[0076] Step 5-5: If the classification result of the body of the email and the classification result of the attachment are the same, and the classification is "well-known" or "other," and the "company name" extracted from the body of the email is present in the text included in the attachment, add "no incorrect destination" to the risk level; otherwise, add "incorrect destination" to the risk level.

[0077] FIG. 10 shows an example of a prompt for determination of the possibility of erroneous transmission according to steps 5-1 to 5-5 above. The prompt for determination 1500 shown in FIG. 10 includes a "#instruction" definition 1510, a "#constraint" definition 1520, a "#category and its knowledge" definition 1530, a "#determination rule" definition 1540, a "#input" definition 1550, and a "#output" definition 1560. By inputting the prompt for determination 1500 shown in FIG. 10 and the body and attachment of the email to be sent, the generation AI can perform the erroneous transmission determination shown in steps 5-1 to 5-5 above. When the prompt for determination 1500 shown in FIG. 10 is input, the classification of the body is output as "text_category," the classification of the attachment as "file_text_category," and the risk level as "level." In addition to these, the reason for the judgment result is output as "description," the recipient's company name and name written in the body of the email are output as "recipient," and a summary of the body of the email is output as "summary."

[0078] In step 5-2 above, "company name," "department name," and "name" are extracted from the text, but this is just one example, and information other than "company name," "department name," and "name" may be extracted as long as it is information that represents the destination. For example, "company name" and "name" may be extracted, or "company name," "department name," "section name," and "name" may be extracted.

[0079] Example of erroneous transmission detection (part 6) In Example 6 of Mis-Sent Judgment, the possibility of mis-sent email (risk level) is judged in two stages by determining whether the format of the body of the email, the destination (To) address, and the attached file violates the rules stated in documents such as company regulations and team regulations.

[0080] Specifically, the possibility of erroneous transmission is determined by the following steps 6-1 and 6-2.

[0081] Step 6-1: Determine whether the format of the body, destination (To), and attachments violates the rules. Note that using a technology or function called Search Augmented Generation (RAG), the generation AI can obtain or refer to the document in which the rules are written.

[0082] Step 6-2: If the format of the body of the email, the destination (To) address, and the attachments do not violate the rules, the risk level is set to "No format violation," otherwise the risk level is set to "Format violation."

[0083] FIG. 11 shows an example of a prompt for determining the possibility of erroneous transmission according to steps 6-1 and 6-2 above. The prompt for determination 1600 shown in FIG. 11 includes a "#instruction" definition 1610, a "#constraint" definition 1620, a "#determination rule" definition 1630, a "#input" definition 1640, and a "#output" definition 1650. By inputting the prompt for determination 1600 shown in FIG. 11 and the body of the email to be sent, attachments, destination (To), destination (Cc), and destination (Bcc), the generation AI can perform the erroneous transmission determination shown in steps 6-1 and 6-2 above. When the prompt for determination 1600 shown in FIG. 11 is used as input, the risk level (determination result) is output as "level." In addition, the reason for the determination result is output as "description," the recipient's company name and name written in the body as "recipient," and a summary of the body as "summary."

[0084] In the above steps 6-1 to 6-2, company regulations, team regulations, etc. are assumed as rules, but this is just an example, and the rules are not limited to company regulations, team regulations, etc. For example, rules may be based on customer information managed by CRM (Customer Relationship Management), past approval history, workflow, information stored in an internal database, etc.

[0085] <Example of the sending confirmation screen> A specific example of the transmission confirmation screen displayed in step S114 of FIG. 5 will now be described.

[0086] Example of a sending confirmation screen (part 1) An example of a transmission confirmation screen (part 1) is shown in Fig. 12. The transmission confirmation screen 2100 shown in Fig. 12 includes display fields 2110, 2120, and 2130. Display field 2110 displays information such as the subject, sender, and sender of the email to be sent, and display field 2120 displays information that allows the user to confirm the legitimacy of the destination address using existing erroneous transmission prevention technology (e.g., the erroneous transmission prevention technology described in Patent Document 1). Meanwhile, display field 2130 displays the erroneous transmission determination result received in step S113 of Fig. 5. In the example shown in Fig. 12, the "level" and "description" included in the erroneous transmission determination result of the above erroneous transmission determination example (part 1) are displayed in display field 2131 as the "possibility of erroneous transmission" and "reason," respectively, and the "recipient" and "summary" included in the erroneous transmission determination result are displayed in display field 2132 as the "analysis result of the body of the email." This allows the user to know the possibility that the email to be sent was sent in error and the reason for this, the name of the recipient's company and name, and a summary of the email body.

[0087] 12 includes a transmission permission button 2140 and a transmission cancellation button 2150. The user can perform a transmission permission instruction operation by pressing the transmission permission button 2140, and can perform a transmission cancellation instruction operation by pressing the transmission cancellation button 2150.

[0088] Example of sending confirmation screen (part 2) An example of a transmission confirmation screen (part 2) is shown in Fig. 13. The transmission confirmation screen 2200 shown in Fig. 13 includes display fields 2210, 2220, and 2230. Display field 2210 displays information such as the subject, sender, and sender of the email to be sent, and display field 2220 displays information that allows the user to confirm the legitimacy of the destination address using existing erroneous transmission prevention technology (e.g., the erroneous transmission prevention technology described in Patent Document 1). Meanwhile, display field 2230 displays the erroneous transmission determination result received in step S113 of Fig. 5. In the example shown in Fig. 13, the "level" and "description" included in the erroneous transmission determination result of the above erroneous transmission determination example (part 2) are displayed in display field 2231 as the "possibility of erroneous transmission" and "reason," respectively, and the "recipient" and "summary" included in the erroneous transmission determination result are displayed in display field 2232 as the "analysis result of the body of the email." This allows the user to know the possibility that the email to be sent was sent in error and the reason for this, the name of the recipient's company and name, and a summary of the email body.

[0089] 13 includes a transmission permission button 2240 and a transmission cancellation button 2250. The user can perform a transmission permission instruction operation by pressing the transmission permission button 2240, and can perform a transmission cancellation instruction operation by pressing the transmission cancellation button 2250.

[0090] Example of sending confirmation screen (part 3) An example of a transmission confirmation screen (part 3) is shown in Fig. 14. The transmission confirmation screen 2300 shown in Fig. 14 is a transmission confirmation screen when an erroneous transmission determination is made according to the above-mentioned erroneous transmission determination example (part 3), and includes display fields 2310, 2320, and 2330. Display field 2310 displays information indicating whether or not there is a possibility of erroneous transmission based on the "level" included in the erroneous transmission determination result according to the erroneous transmission determination example (part 3). In the example shown in Fig. 14, information indicating that the destination of the body of the email and the destination of the attached file are different is displayed in display field 2310. Furthermore, display field 2320 displays information about the body of the email, such as the text contained in the body of the email, and display field 2330 displays information about the attached file.

[0091] When information indicating that the destination of the body of the email differs from the destination of the attached file is displayed in display field 2310, display field 2320 is highlighted by display components 2340 and 2350, and similarly display field 2330 is highlighted by display components 2360 and 2370. This allows the user to easily know which display field to check when the destination of the body of the email differs from the destination of the attached file.

[0092] 14 includes a transmission permission button 2380 and a transmission cancellation button 2390. The user can perform a transmission permission instruction operation by pressing the transmission permission button 2380, and can perform a transmission cancellation instruction operation by pressing the transmission cancellation button 2390.

[0093] <Modification> Modifications of the above embodiment will be described below. Note that two or more of the following modifications may be combined as appropriate, as long as they are not mutually contradictory.

[0094] Variation 1 In step S112 of Fig. 5, the erroneous transmission determination result was sent to the client terminal 10. However, for example, the erroneous transmission determination result may also be sent to a server that provides a chat application service used by the user of that client terminal 10. This allows the user to know the erroneous transmission determination result on the chat application service. At this time, the user may also be able to perform an operation to instruct permission to send or an operation to instruct cancellation of sending on the chat application service. Furthermore, in addition to this, the actual email transmission result (sent / cancelled / discarded, etc.) may be notified to the user by any means (e.g., chat application service, etc.).

[0095] Variation 2 In the above variant example 1, the erroneous transmission prevention server 20 transmitted the erroneous transmission judgment result to the server providing the chat application service, but for example, the generation AI providing server 30 may transmit the erroneous transmission judgment result to the server providing the chat application service in step S110 of Figure 5.

[0096] Variation 3 14, display field 2320 is highlighted by display component 2340 and display component 2350, and display field 2330 is highlighted by display component 2360 and display component 2370, but the highlighting method is not limited to this. Various highlighting methods may be used, such as enlarging the text, turning it red, blinking, or displaying it in dynamic animation.

[0097] Variation 4 If a message such as an instruction or warning to the sender based on the determination of erroneous transmission is present in display field 2320 or display field 2330 of transmission confirmation screen 2300 shown in FIG. 14 , the message may be highlighted. Specific examples of such messages include, for example, "Open a new window and check the main text" as an instruction to prompt the sender to check the email body, and "Open a new window and check the attached file" as an instruction to prompt the sender to check the attached file. Furthermore, the message is not necessarily limited to a message, and instructions or warnings to the sender based on the determination of erroneous transmission may also be highlighted in some manner. Specific examples of such highlighting include, for example, highlighting the need to check the destination when prompting the sender to check the destination.

[0098] Variation 5 All or part of the erroneous transmission determination result included on the transmission confirmation screen may be read aloud by synthetic voice, etc. In particular, the "possibility of erroneous transmission" and "reason" may be read aloud by synthetic voice, etc.

[0099] Variation 6 Because the generation AI's determination of erroneous transmission may require a certain amount of processing time, for example, the transmission confirmation screen 2400 shown in Fig. 15 may be displayed before the transmission confirmation screen 2300 shown in Fig. 14 is displayed. The transmission confirmation screen 2400 shown in Fig. 15 includes a display field 2410 that displays a message indicating that the generation AI is determining whether the transmission has been erroneous.

[0100] At this time, a cancel button for canceling the erroneous transmission determination by the generation AI may be displayed in display field 2410, allowing the user to cancel the erroneous transmission determination at any time. Also, for example, if a processing time of more than a predetermined time has elapsed during the erroneous transmission determination by the generation AI, a message indicating that the determination has failed may be displayed in display field 2410.

[0101] Variation 7 In the above embodiment, whether or not the email to be sent is subject to the erroneous transmission determination is determined in step S104 of Fig. 5, but this determination does not have to be performed. In other words, this determination may not be performed and all emails to be sent may be subject to the erroneous transmission determination.

[0102] Variation 8 The generation AI may be retrained (tuned) using information fed back from a user who confirmed the erroneous transmission judgment result (such as "possibility of erroneous transmission," "reason," and "analysis result of the text") as learning data. Specifically, a user who confirmed the erroneous transmission judgment result may determine whether the erroneous transmission judgment result is correct or incorrect, and the correctness result may be sent as feedback data from the client terminal 10 to the generation AI provision server 30. This allows the generation AI provision server 30 to retrain (tune) the generation AI using the feedback data and the email information corresponding to that feedback data as learning data.

[0103] However, the feedback data is not limited to the correct or incorrect result, but may be, for example, the content of the correction if the user corrects the erroneous transmission judgment result, or the item that is incorrect in the erroneous transmission judgment result if the user specifies that item.

[0104] Variation 9 In the above examples of erroneous transmission determination (2) and erroneous transmission determination (4), the address written in the body of the email is compared with the address indicated by the email address specified in the destination (To) field or with the address extracted from that email address using an address book, but the comparison may be made without directly using information stored in the email address or address book (personal information such as "company name," "department name," and "name").

[0105] For example, instead of an email address or personal information stored in an address book in association with that email address, public information (e.g., domain owner information, etc.) or email address naming conventions may be estimated or referenced, and the company name, department name, job title, etc. corresponding to the email address may be estimated or identified, and compared with the address stated in the body of the text.

[0106] Alternatively, for example, some or all of the personal information stored in the address book may be deleted or masked, and the generation AI may then use search expansion generation to obtain the deleted or masked personal information. In addition, for example, the address described in the body of the email may be compared with the address indicated by the email address specified in the To field or with the address extracted from that email address using an address book by means other than the generation AI, and the comparison results may be input into the generation AI to determine the possibility of erroneous transmission.

[0107] <Summary> As described above, the system 1 for preventing erroneous email transmission according to the present embodiment can use a generation AI to determine whether an email has been sent by mistake, based on the content of the email to be sent, and then display a transmission confirmation screen containing the result of the determination on the client terminal 10 used by the sender. Therefore, the system 1 for preventing erroneous email transmission according to the present embodiment can perform effective confirmation according to the content of the email to be sent, preventing confirmation from becoming a mere formality and is expected to effectively prevent erroneous email transmission.

[0108] The present invention is not limited to the above-described specifically disclosed embodiments, and various modifications, changes, and combinations with known technologies are possible without departing from the scope of the claims. [Explanation of symbols]

[0109] 1. Email Mis-Sending Prevention System 10 Client Terminal 20. Server for preventing erroneous transmission 30 Generation AI provision server 40 Internal LAN 50 Communication Network 101 Communications Department 102 Email function section 103 UI control section 104 Storage section 201 Communications Department 202 Target email determination unit 203 Prompt Acquisition Unit 204 Judgment Request Creation Unit 205 Storage section 301 Communications Department 302 False transmission detection unit 303 Storage section

Claims

1. a determination unit that determines the possibility that the email has been sent by mistake using a machine learning model based on information about the email that is to be determined to be sent by mistake and instructions for causing the machine learning model to perform the determination of the email being sent by mistake; an output unit that outputs information for allowing a sender of the email to confirm the result of the determination made by the determination unit before the email is sent to a destination; A server having:

2. The machine learning model is a trained machine learning model that realizes generative AI, The determination unit The server according to claim 1 , wherein the server receives information about the email as input and determines the likelihood that the email was sent by mistake using the machine learning model in accordance with the instructions.

3. The server described in claim 1 or 2, wherein the information about the email includes at least one of a string representing the body of the email, a string contained in an attachment attached to the email, an email address specified in the To field of the email, an email address specified in the Cc field of the email, and an email address specified in the Bcc field of the email.

4. a determination procedure for determining the possibility that the email has been sent by mistake using a machine learning model based on information about the email to be determined as having been sent by mistake and instructions for causing the machine learning model to perform the determination of whether the email has been sent by mistake; an output step of outputting information for allowing a sender of the email to confirm the result of the determination step before the email is sent to a destination; How a computer runs.

5. a determination procedure for determining the possibility that the email has been sent by mistake using a machine learning model based on information about the email to be determined as having been sent by mistake and instructions for causing the machine learning model to perform the determination of whether the email has been sent by mistake; an output step of outputting information for allowing a sender of the email to confirm the result of the determination step before the email is sent to a destination; A program that causes a computer to execute the following.

Images