JP2023008458A - Answer presentation program, answer presentation method and answer presentation device - Google Patents

Abstract

To provide an answer presentation program that improves quality of an answer being presented to a questioner.SOLUTION: An answer presentation program causes a computer to perform processing of: independently acquiring a question and a mail; determining a question category of a question to which a first character belongs on the basis of a specific first character included by a question mark added sentence or an interrogative sentence equivalent when the question includes the question mark or is an interrogative sentence equivalent; determining a first specific word that can identify a commercial transaction object of a question object on the basis of at least one sentence or total sentences of a question; determining a mail category of a mail to which a second character belongs on the basis of a specific second character included by the question mark added sentence or the interrogative sentence equivalent when the mail includes the question mark or is the interrogative sentence equivalent; determining a second specific word that can identify the commercial transaction object being an object in the mail on the basis of at least one sentence or total sentences of the mail; and presenting the mail as an answer candidate for the question on the basis of a first similarity between the question category and the mail category and a second similarity between the first specific word and the second specific word.SELECTED DRAWING: Figure 5

Description

The present invention relates to an answer presentation program, an answer presentation method, and an answer presentation device.

A chatbot is known as a system that automatically responds to inquiries (see Patent Document 1, for example). In addition, when creating FAQ (Frequently Asked Questions) to support automatic responses from dialogue logs, Q candidates (question candidates) and A candidates (answer candidates) to be added to FAQ are selected from a huge amount of dialogue logs. , techniques for efficient extraction using supervised learning have been proposed. As a result, the creation of FAQs is labor-saving (see Patent Document 2, for example).

In addition, there is also known a system in which a question from an inquirer is input as a chat message, and an answerer on the side of the site provider manually inputs an answer to the question as a chat message. In this system, answer candidates are extracted and presented to the respondent using a trained model generated by machine learning using combinations of past questions and answers as teacher data. As a result, the burden on the respondent is reduced while maintaining the quality of the content of the reply (see Patent Document 3, for example).

JP 2019-139746 A JP 2020-071690 A JP 2020-071679 A

However, when machine learning such as supervised learning is used, answer candidates are extracted like a black box, and there is a possibility that high-quality answers that capture the questioner's intentions and questions may not be presented to the questioner.

Therefore, in one aspect, it is an object to provide an answer presentation program, an answer presentation method, and an answer presentation device that improve the quality of the answers presented to the questioner.

In one embodiment, the answer presentation program independently obtains a question and an e-mail, and if the obtained question contains a question mark or if the question corresponds to an interrogative sentence, the question mark is attached. Determine the question category of the question to which the first character belongs based on the first sentence or a specific first character included in the interrogative equivalent, and determine the question category based on at least one sentence or the whole sentence of the acquired question Determining a first specific word that can identify a target commercial transaction object, and adding the question mark if the obtained e-mail contains the question mark or if the e-mail corresponds to the interrogative sentence determining the mail category of the e-mail to which the second character belongs based on the second sentence or a specific second character included in the interrogative sentence equivalent, and based on at least one sentence or the whole text of the acquired e-mail, determining a second specific word capable of identifying the business transaction target addressed by the e-mail, and determining a first similarity between the question category and the mail category; A computer is caused to execute a process of presenting the e-mail as an answer candidate to the question based on the degree of similarity.

It is possible to improve the quality of the answers presented to the questioner.

FIG. 1 is a diagram for explaining an example of a dialogue system. FIG. 2 shows an example of the hardware configuration of the dialogue server. FIG. 3 is an example of a block diagram of a dialog server. FIG. 4 is an example of a scenario DB. FIG. 5(a) is an example of a question screen. FIG. 5(b) is another example of the question screen. FIG. 6 is an example of a mail DB. FIG. 7 is an example of a question DB. FIG. 8 is a flowchart showing an example of dialogue-related processing. FIG. 9(a) is a flowchart illustrating a part of the question category determination process. FIG. 9(b) is a flowchart illustrating the remainder of the question category determination process. FIG. 10 is an example of the category table DB. FIG. 11 is an example of the keyword table DB. FIG. 12 is a flowchart showing an example of mail-related processing. FIG. 13 is an example of an answer DB.

Hereinafter, the form for carrying out this case will be described with reference to the drawings.

As shown in FIG. 1, the dialogue system ST includes a user terminal 100, an administrator terminal 150, and a dialogue server 200. FIG. Dialogue server 200 is an example of an answer presentation device. Although one user terminal 100 and one administrator terminal 150 are shown in FIG. 1, one or both of the user terminal 100 and administrator terminal 150 may be plural. Also, although FIG. 1 shows a PC (Personal Computer) as an example of the user terminal 100, it may be a smart device such as a smart phone or a tablet terminal. The administrator terminal 150 is also similar to the user terminal 100 .

User terminal 100 includes input device 110 , display device 120 , and control device 130 . Both the input device 110 and the display device 120 are connected to the control device 130 . When a user (for example, a questioner) using the dialogue system ST operates the input device 110, the control device 130 controls the display content of the display device 120 based on instructions according to the operation content. For example, when the user operates the input device 110 to access the dialogue server 200 , the control device 130 displays the question screen SC on the display device 120 based on the processing of the dialogue server 200 .

In addition, although the details will be described later, when the user operates the input device 110 to input and transmit a question in sentences such as natural sentences on the question screen SC, the control device 130 asks the input question. Display within the screen SC. Instead of a sentence, a sentence ending with a word such as a noun or a pronoun may be input and transmitted as a question. Control device 130 transmits the input question to dialogue server 200 . When the user operates the input device 110 to select a predetermined selection button appearing on the question screen SC, the control device 130 transmits options corresponding to the selection button to the dialogue server 200 . In addition to the selection buttons, the question screen SC includes character strings that cannot be selected.

The administrator terminal 150 is operated by an administrator who manages the operation of the interaction server 200 . Since the configuration of the administrator terminal 150 is basically the same as that of the user terminal 100, detailed description thereof will be omitted. The administrator operates the administrator terminal 150 to create an FAQ for supporting the automatic response and registers it in the dialogue server 200 . Specifically, guidance questions (for example, questions and answers) and answer candidates to be added to the FAQ are created from a huge amount of dialogue logs between the questioner and the dialogue server 200 and registered in the dialogue server 200 . Further, when the administrator receives an inquiry by e-mail via the dialogue server 200, the administrator replies to the inquiry by e-mail via the dialogue server 200. FIG.

The user terminal 100, administrator terminal 150 and dialogue server 200 are connected to each other via a communication network NW. As the communication network NW, for example, there is a LAN (Local Area Network). In this embodiment, the operation form of the interaction server 200 is described as being on-premise, but the operation form of the interaction server 200 may be cloud. A cloud may be a public cloud or a private cloud. In this case, the Internet may be used as the communication network NW.

When the dialogue server 200 receives a question or an option transmitted from the user terminal 100, it executes various processes. For example, the dialogue server 200 determines questions and options, and returns screen information including answers and guiding questions according to the questions and options to the user terminal 100 . Depending on the option, the dialogue server 200 returns screen information requesting an inquiry by e-mail to the user terminal 100 . This allows the user to make an inquiry by e-mail. In this way, the dialogue server 200 returns guidance questions and answers according to the user's questions and options. Therefore, the dialog system ST or the dialog server 200 may be called a chatbot that can automatically answer questions. Further, the dialog system ST or the dialog server 200 may be called a scenario type chatbot that interacts with a user according to a predetermined scenario regarding a specific topic.

Next, with reference to FIG. 2, the hardware configuration of the dialogue server 200 that executes the answer presentation method will be described. Since the control unit 130 and the control unit of the administrator terminal 150 described above have basically the same hardware configuration as that of the dialogue server 200, the description thereof will be omitted.

As shown in FIG. 2, the dialogue server 200 includes a CPU (Central Processing Unit) 200A as a processor, and RAM (Random Access Memory) 200B and ROM (Read Only Memory) 200C as memories. The dialogue server 200 also includes a network I/F (interface) 200D and a HDD (Hard Disk Drive) 200E. An SSD (Solid State Drive) may be employed instead of the HDD 200E.

Furthermore, the dialog server 200 may include at least one of an input I/F 200F, an output I/F 200G, an input/output I/F 200H, and a drive device 200I. The CPU 200A to the drive device 200I are interconnected by an internal bus 200J. That is, the dialogue server 200 can be implemented by a computer.

An input device 710 is connected to the input I/F 200F. The input device 710 includes, for example, a keyboard and a mouse. Note that the input device 110 described above is similar to the input device 710 . A display device 720 is connected to the output I/F 200G. The display device 720 is, for example, a liquid crystal display. Note that the display device 120 described above is similar to the display device 720 . A semiconductor memory 730 is connected to the input/output I/F 200H. Examples of the semiconductor memory 730 include USB (Universal Serial Bus) memory and flash memory. Input/output I/F 200H reads the answer presentation program stored in semiconductor memory 730 . The input I/F 200F and the input/output I/F 200H are provided with USB ports, for example. The output I/F 200G has, for example, a display port.

A portable recording medium 740 is inserted into the drive device 200I. Examples of the portable recording medium 740 include removable discs such as CD (Compact Disc)-ROM and DVD (Digital Versatile Disc). Drive device 200I reads the answer presentation program recorded in portable recording medium 740 . The network I/F 200D has, for example, a LAN port and a communication circuit. Network I/F 200D is connected to communication network NW.

The answer presentation program stored in the ROM 200C, the HDD 200E, and the semiconductor memory 730 is temporarily stored in the RAM 200B described above by the CPU 200A. The answer presentation program recorded in the portable recording medium 740 is temporarily stored in the RAM 200B by the CPU 200A. As the CPU 200A executes the stored answer presentation program, the CPU 200A realizes various functions described later and also executes various processes described later. It should be noted that the answer presentation program may correspond to the flow chart described later.

Next, the functional configuration of the interaction server 200 will be described with reference to FIG. Note that FIG. 3 shows the essential functions of the interaction server 200 .

As shown in FIG. 3, the dialogue server 200 comprises a storage section 210, a processing section 220, and a communication section 230. FIG. The storage unit 210 can be realized by one or both of the RAM 200B and HDD 200E described above. The processing unit 220 can be realized by the CPU 200A described above. The communication unit 230 can be implemented by the network I/F 200D described above. Therefore, the storage unit 210, the processing unit 220, and the communication unit 230 are connected to each other.

Here, the storage unit 210 includes a scenario DB (Database) 211, a question DB 212, a mail DB 213, a category table DB 214, a keyword table DB 215, and an answer DB 216. On the other hand, the processing section 220 includes a dialogue processing section 221 , an acquisition section 222 , a question category determination section 223 and a first keyword determination section 224 . The processing unit 220 also includes a mail category determination unit 225 , a second keyword determination unit 226 and an answer presentation unit 227 .

For example, the acquisition unit 222 accesses the question DB 212 and acquires question information. Although the details will be described later, the question DB 212 stores question information input to the question screen SC and sent from the user terminal 100 . Also, the acquisition unit 222 accesses the mail DB 213 and acquires the e-mail. Although the details will be described later, the mail DB 213 stores e-mails that contain only inquiries sent by users using e-mails. In other words, the mail DB 213 stores e-mails before replies by the administrator. The mail DB 213 also stores e-mails containing both this inquiry and the administrator's reply to this inquiry. In other words, the mail DB 213 also stores e-mails after responses by the administrator.

Here, the acquisition unit 222 independently acquires the question information and the e-mail. For example, the acquisition unit 222 monitors the question DB 212, and when the dialogue processing unit 221 stores question information in the question DB 212, the acquisition unit 222 immediately acquires the question information. That is, the acquisition unit 222 can acquire question information in real time. On the other hand, the acquisition unit 222 also monitors the mail DB 213, and even if the interaction processing unit 221 stores an e-mail containing only the user's inquiry in the mail DB 213, the acquisition unit 222 waits to acquire the e-mail. When the interactive processing unit 221 stores an e-mail containing both the user's inquiry and the administrator's answer to the inquiry, the acquisition unit 222 acquires the e-mail. That is, unlike acquisition of question information, the acquisition unit 222 cannot acquire an e-mail in real time, and there is a time lag between acquisition of question information.

Therefore, the acquiring unit 222 acquires the question information received by the interactive processing unit 221 during the first period, and also acquires the e-mail received by the interactive processing unit 221 during the second period before the first period. That is, the acquiring unit 222 acquires the e-mails received by the interactive processing unit 221 during the second period past the first period when the interactive processing unit 221 received the question information. Details of the remaining DB elements such as the scenario DB 211 included in the storage unit 210 and the remaining functional units such as the question category determination unit 223 included in the processing unit 220 will be described later.

Next, operations of the interaction server 200 will be described with reference to FIGS. 4 to 13. FIG. Note that the dialogue-related processing shown in FIG. 8 and the mail-related processing shown in FIG. 12 are executed independently and in parallel.

First, when the user operates the input device 110 to access the dialogue server 200, the dialogue processing unit 221 refers to the scenario DB 211 based on the access and acquires the first guidance question information. Here, as shown in FIG. 4, the scenario DB 211 stores a plurality of guiding question information 11, 12, 13, 14 in which the order of guiding questions is defined as scenarios. In FIG. 4, the order of the guiding questions is defined by the guiding ID "Y001" etc. assigned to each of the guiding question information 11, 12, 13 and 14. In FIG. The guidance ID is an identifier that identifies guidance question information. In the present embodiment, the leading question information 11 including the leading question "What is your question..." corresponds to the first leading question information.

When the initial guidance question information is acquired, the dialogue processing unit 221 transmits the screen information of the question screen SC including the guidance question information to the control device 130 . When receiving the screen information, the control device 130 displays the question screen SC on the display device 120 based on the received screen information. As a result, the question screen SC appears on the display device 120, as shown in FIG. 5(a). The question screen SC includes a guide sentence 20 that guides the user on the operation method of the question screen SC, a plurality of selection buttons 21, 22, and 23, an input field 24 for inputting a question, a send button 25, and the like.

When the user operates the input device 110 and presses the selection button 22 with the pointer PT, for example, the control device 130 transmits the option “product system” associated with the selection button 22 to the dialogue server 200 . As a result, the dialogue processing unit 221 acquires the next guiding question information 12 following the guiding question information 11 from the scenario DB 211 and transmits the screen information of the next question screen SC including the guiding question information 12 to the control device 130 .

Further, for example, when the selection button 23 is pressed, the control device 130 displays a predetermined screen (not shown) on the display device 120 for supporting creation of an e-mail and transmission/reception of the e-mail. This predetermined screen contains the mail address of the user terminal 100 representing the source of the e-mail. This predetermined screen also includes the mail address of the dialogue server 200 as the destination of the electronic mail. This allows the user to send an inquiry to the interaction server 200 by e-mail. When the user operates input device 110 to instruct transmission of an e-mail containing an inquiry, control device 130 transmits the e-mail to dialogue server 200 . As a result, the interactive processing unit 221 receives the e-mail. Upon receiving the e-mail, the interactive processing unit 221 stores the e-mail in the mail DB 213 and transfers the received e-mail to the administrator terminal 150 . This allows administrators to create answers to queries.

When the administrator operates the input device of the administrator terminal 150 to instruct transmission of an e-mail containing the reply, the control section of the administrator terminal 150 transmits this e-mail to the interaction server 200 . As a result, the interactive processing unit 221 receives the e-mail. Upon receiving the e-mail, interactive processing unit 221 stores the e-mail in mail DB 213 . Thereby, as shown in FIG. 6, the mail DB 213 stores the e-mail 31 including both the inquiry and the reply. In addition to the e-mail 31, the e-mail DB 213 stores a plurality of e-mails 32 and 33 each containing a single inquiry. When the interactive processing unit 221 receives the e-mail containing the reply, it transmits the received e-mail to the e-mail address of the user terminal 100 . This allows the user to confirm the answer to the inquiry.

By the way, as shown in FIG. 5A, when the user operates the input device 110 to input a question in the input field 24 and presses the send button 25, the control device 130 sends the question input in the input field 24. The question information included is transmitted to the dialogue server 200 . In this embodiment, the control device 130 transmits question information including the question "How much is the A product?" Upon receiving the question information, the dialogue processing unit 221 adds a question ID for identifying the question to the received question information, and stores it in the question DB 212 together with the date and time of reception. Thereby, as shown in FIG. 7, question DB212 memorize|stores question information.

As described above, the acquisition unit 222 monitors the question DB 212, and when the dialogue processing unit 221 stores question information in the question DB 212, acquires the question as shown in FIG. 8 (step S1). For example, as shown in FIG. 7, if a plurality of pieces of question information with different reception dates are stored in the question DB 212, the acquisition unit 222 acquires the question with the latest reception date. In this embodiment, the acquisition unit 222 acquires the question "How much is product A?"

When the acquisition unit 222 acquires a question, as shown in FIG. 8, the question category determination unit 223 executes question category determination processing (step S2). In the question category determination process, for example, when a question acquired by the acquisition unit 222 contains a question mark, a question category representing the category of the question to which the character belongs is determined based on a specific character included in the sentence with the question mark. It is a process to In the question category determination process, even if the question does not contain question marks, if the question corresponds to an interrogative sentence, the question category is determined based on the specific characters of the interrogative sentence equivalent.

Details of the question category determination process will now be described with reference to FIGS. 9 and 10. FIG. When the question category determining section 223 starts executing the question category determining process, first, as shown in FIG. 9A, it determines whether or not the question contains a question mark (step S21). The question mark is specifically a question mark "?". When the acquisition unit 222 acquires the question “How much is product A?”, the question has a question mark at the end, so the question category determination unit 223 determines that the question includes a question mark (step S21: YES). As a result, the question category determination unit 223 identifies sentences with question marks (step S22). Therefore, the question category determination unit 223 identifies the text "How much is the A product?". Note that the sentences determined by the question category determination unit 223 may include question marks.

After identifying the sentence, the question category determination unit 223 identifies characters in the sentence (step S23). More specifically, the question category determination unit 223 refers to the category table DB 214 to specify characters in the text. As shown in FIG. 10, the category table DB 214 stores a category table, which is an example of category information in which a plurality of characters and categories to which the plurality of characters belong are associated in advance. Several characters and categories are pre-associated. The question category determination unit 223 acquires a plurality of (or all) characters such as the character "how much" and the character "performance" registered in the category table, compares the acquired plurality of characters with sentences one by one, and finds a match. Identify characters. For example, when specifying the text "How much is product A?", the question category determining unit 223 specifies the characters "how much".

After identifying the characters, the question category determination unit 223 determines the question category (step S24) and terminates the process. More specifically, the question category determination unit 223 refers to the category table DB 214 and determines the category to which the specified character belongs as the question category. For example, when the character "how much" is specified, as shown in FIG. 10, the question category determining unit 223 determines the category "cost" to which the character "how much" belongs as the question category.

On the other hand, if it is determined that the question does not contain a question mark (step S21: NO), the question category determination unit 223 determines whether the question corresponds to an interrogative sentence (step S25). For example, when the acquisition unit 222 acquires a question “cost of product A” that does not contain a question mark, it can be estimated that the question corresponds to an interrogative sentence because the question is a non-verbal statement. In such a case, the question category determination unit 223 determines that the question corresponds to an interrogative sentence (step S25: YES). For example, when the acquisition unit 222 acquires a question “I would like to know the price of product A” that does not contain a question mark, since the word “I want to know” is written at the end of the question to ask for resolution of the question, it corresponds to an interrogative sentence. Then it can be estimated. Similarly, in such a case, the question category determination unit 223 determines that the question corresponds to an interrogative sentence. In addition to the word "want to know", the word "want" or the word "tell me" may be used.

When the question category determination unit 223 determines that the question corresponds to an interrogative sentence, it identifies characters in the interrogative sentence (step S26). More specifically, the question category determination unit 223 acquires a plurality of (or all) characters such as the characters “expense” and the character “price” registered in the category table, and determines each character of the acquired plurality of characters as an interrogative sentence. Contrast to identify matching characters. For example, when the question "Cost of product A" is acquired, the question category determination unit 223 identifies the characters "cost". When the question "I would like to know the price of product A" is acquired, the question category determination unit 223 specifies the characters "price". When the question category determining unit 223 identifies the characters in the interrogative sentence, it determines the category "cost" to which the identified characters belong as the question category by the process of step S24.

If it is determined that the question does not correspond to an interrogative sentence (step S25: NO), the dialog processor 221 presents a leading question as shown in FIG. 9B (step S27). More specifically, the dialogue processing unit 221 acquires guidance question information corresponding to the question from the scenario DB 211 and transmits screen information of the question screen SC including the guidance question information to the control device 130 . As a result, a new question screen (not shown) appears on the display device 120 . For example, when the user operates the input device 110 to press a selection button, the control device 130 transmits options associated with the selection button to the dialogue server 200 as answers to the guiding questions. As a result, the interactive processing unit 221 obtains an answer (step S28).

When the dialogue processing unit 221 acquires the answer, the question category determination unit 223 determines whether or not the question category can be specified (step S29). More specifically, the question category determination unit 223 compares the answer acquired by the interactive processing unit 221 with a plurality of characters registered in the category table one by one, and determines whether or not a matching character can be specified. If the matching characters can be specified, the question category can be determined based on the specified characters, so the question category determination unit 223 determines that the question category can be specified (step S29: YES). In this case, as shown in FIG. 9(a), the question category determination unit 223 determines the question category by the process of step S24, and ends the process.

On the other hand, if the matching character cannot be identified, the question category determination unit 223 determines that the question category cannot be identified (step S29: NO), and determines whether the number of attempts to identify the question category is a predetermined number or more. (step S30). The predetermined number of times represents a threshold for this determination, and can be several times such as 3 to 5 times, for example. If the number of attempts to specify the question category is less than the predetermined number (step S30: NO), the question category determining section 223 executes the process of step S27. As a result, the user terminal 100 is presented with the next leading question following the new leading question.

On the other hand, if the number of attempts to identify the question category is equal to or greater than the predetermined number (step S30: YES), the question category determination unit 223 notifies the administrator terminal 150 of a category identification error (step S31), and FIG. End the process as shown in a). This makes it possible to investigate the reason why the administrator could not specify the question category. In this way, the question category of the question can be determined by the question category determining section 223 executing the question category determining process in cooperation with the interactive processing section 221 .

Returning to FIG. 8, after the question category determination unit 223 finishes executing the question category determination process, the first keyword determination unit 224 determines a question-based first keyword (step S3). Question-based means question-based. The processing order of the processing in step S2 and the processing in step S3 may be reversed, or may be performed in parallel. More specifically, the first keyword determination unit 224 refers to the keyword table DB 215 and determines a first keyword that can specify the product targeted by the question, based on at least one sentence or the entire sentence of the question. A product is an example of a commercial transaction object. A first keyword is an example of a first specific word. Instead of products, services may be the object of commercial transactions.

As shown in FIG. 11, the keyword table DB 215 stores a keyword table in which a plurality of words related to products and services are associated with keywords that can specify the products and services. A keyword table is an example of specific word information. For example, multiple words related to HCI (Hyper-Converged Infrastructure) (such as storage and computing) are registered in the keyword table, and keywords that can identify HCI-related products (such as ○○ integrated system, etc.) are registered in these multiple words. ) may be associated and registered. Note that multiple words and keywords are associated in advance.

The keyword table is commonly used in both cases of specifying a question-based first keyword and specifying a mail-based second keyword, which will be described later. From the keyword table, if a word corresponds to one of a plurality of words, a keyword associated with that word can be determined. In this embodiment, the acquisition unit 222 acquires the question “How much is product A?” Then, the first keyword "A product" is determined from the question.

By the way, as described above, since the dialogue-related process and the mail-related process are executed independently and in parallel, the acquisition unit 222 also monitors the mail DB 213 . Acquisition unit 222 monitors mail DB 213 , and executes mail-related processing when interaction processing unit 221 stores an e-mail containing an administrator's response in mail DB 213 .

Details of the mail-related processing will be described with reference to FIG. 12 and the like. As described above, when the interactive processing unit 221 stores the e-mail containing the manager's response in the mail DB 213, the acquisition unit 222 acquires the e-mail (step S41). As shown in FIG. 6 , if the e-mails 31 each containing the administrator's reply are stored in the mail DB 213 , the obtaining unit 222 obtains the e-mails 31 .

When the obtaining unit 222 obtains the e-mail, the mail category determination unit 225 determines whether or not the mail category can be specified (step S42). More specifically, the mail category determination unit 225 compares the e-mail acquired by the acquisition unit 222 with a plurality of characters such as the characters “approximate” and the characters “spec” registered in the category table one by one, and determines the matching characters. Determine whether identification is possible. If the matching characters can be specified, the mail category can be determined based on the specified characters, so the mail category determining unit 225 determines that the mail category can be specified (step S42: YES). For example, when the acquisition unit 222 acquires the e-mail 31 (see FIG. 6), the e-mail 31 includes the characters "approximate" registered in the category table. Therefore, in this case, the mail category determining unit 225 determines that the mail category can be specified.

When the mail category can be identified, the mail category determining unit 225 determines the mail category (step S43). More specifically, the mail category determining unit 225 refers to the category table DB 214 and determines the category to which the specified character belongs as the mail category. For example, when the characters "approximate" are identified, the mail category determination unit 225 determines the category "cost" to which the characters "approximate" belong as the mail category, as shown in FIG. After determining the mail category, the mail category determining unit 225 associates the mail category with the e-mail (step S44), and terminates the process. As a result, e-mails can be searched using the mail category as a key item.

If it is determined that the mail category cannot be specified (step S42: NO), the mail category determining unit 225 notifies the administrator terminal 150 of a category specifying error (step S45), and terminates the process. This makes it possible to investigate the reason why the administrator could not specify the mail category.

Returning to FIG. 8, when the first keyword determination unit 224 determines the first keyword in the process of step S3, the answer presentation unit 227 calculates the first similarity between the question category and the mail category (step S4). For example, the answer presentation unit 227 uses the cosine similarity to calculate the first similarity between the question category and the mail category. If the question category completely matches the mail category, the answer presenting unit 227 calculates the first degree of similarity "1.0". On the other hand, if the question category and the mail category do not match at all, the answer presenting unit 227 calculates the first degree of similarity "0.0". Therefore, the answer presenting unit 227 calculates a numerical value within a range from the first similarity "0.0" to the first similarity "1.0" as the first similarity based on the question category and the mail category. Instead of the cosine similarity, a known technique for calculating the similarity between character strings, such as the Levenshtein distance, may be employed to calculate the first similarity between the question category and the mail category.

After calculating the first degree of similarity, the answer presentation unit 227 searches for e-mails based on the first degree of similarity (step S5), and determines whether or not the search has been narrowed down (step S6). As mentioned above, e-mails have mail categories associated with them (see FIG. 12). The answer presenting unit 227 searches for e-mails and narrows down e-mails associated with mail categories whose first degree of similarity is greater than or equal to the threshold similarity. For example, by adopting a threshold similarity greater than the threshold similarity "0.5", it is possible to narrow down the e-mails associated with the mail category that matches or has a high degree of similarity with the question category.

For example, if no e-mail associated with a mail category whose first similarity is greater than or equal to the threshold similarity is found (that is, if the search result is 0), the answer presenting unit 227 determines that the e-mails could not be narrowed down. (step S6: NO). In this case, the answer presentation unit 227 terminates the dialogue-related processing. Before ending the dialogue-related processing, the answer presenting unit 227 presents to the administrator terminal 150 a question of the question category from which the first similarity is calculated, and creates and adds an answer to the question. You can request it.

On the other hand, if an e-mail associated with a mail category whose first degree of similarity is greater than or equal to the threshold similarity can be found (that is, if the search result is one or more), the answer presenting unit 227 determines that the e-mails have been narrowed down. (Step S6: YES). However, the number of search results may be enormous, for example, 1000 cases. In this case, the answer presenting unit 227 weights (specifically, assigns scores to) the e-mails according to a predetermined standard, and narrows down the e-mails to the top 10 e-mails with the highest weights (that is, the highest scores). Predetermined criteria include, for example, the frequency of responses by e-mail, the update date of e-mails, and the level of response evaluation, which will be described later.

When the e-mails are narrowed down, the second keyword determination unit 226 determines mail-based second keywords from the e-mails that have been narrowed down (step S7). Mail-based means email-based. More specifically, the second keyword determining unit 226 refers to the keyword table DB 215 and determines a second keyword that can identify the product targeted by the e-mail based on the full text of the e-mail. The second keyword determining unit 226 may determine the second keyword based on at least one sentence of the e-mail. For example, when the acquisition unit 222 acquires the e-mail 31 (see FIG. 6), the second keyword determination unit 226 compares a plurality of words registered in the keyword table DB 215 with the entire text in the e-mail 31 word by word. Then, the second keyword “A product” is determined from the e-mail 31 .

After the second keyword determination unit 226 determines the second keyword, the answer presentation unit 227 calculates a second similarity between the first keyword and the second keyword (step S8). As in the case of calculating the first degree of similarity between the question category and the mail category, the answer presenting unit 227 uses cosine similarity or the like to calculate the second degree of similarity between the first keyword and the second keyword. Therefore, based on the first keyword and the second keyword, the answer presenting unit 227 calculates a numerical value in the range of the second similarity "0.0" to the second similarity "1.0" as the second similarity. .

After calculating the second degree of similarity, the answer presentation unit 227 further searches for e-mails based on the second degree of similarity (step S9), and determines whether there is any e-mail (step S10). More specifically, the answer presenting unit 227 searches a plurality of e-mails that have already been narrowed down, and searches for e-mails that contain words associated with the second keyword whose second similarity is equal to or greater than the threshold similarity. For example, by adopting a threshold similarity greater than the threshold similarity of "0.5", it is possible to identify e-mails that include words that match the first keyword or are associated with a second keyword that has a high degree of similarity. In this embodiment, since the first keyword "A product" is determined, the e-mail 31 containing the word "A product" associated with the second keyword "A product" that matches the first keyword "A product" can be specified. Therefore, answer presentation unit 227 determines that there is an e-mail.

If there is no e-mail (step S10: NO), the answer presentation unit 227 terminates the dialogue-related processing. On the other hand, if there is an e-mail (step S10: YES), the answer presenting unit 227 presents the e-mail as an answer candidate to the user terminal 100 (step S11), and terminates the dialogue-related processing. More specifically, the answer presenting unit 227 transmits the extracted information obtained by extracting the entire text or a part of the text from the specified e-mail as an answer candidate to the control device 130 . When receiving the extracted information, the control device 130 updates the question screen SC displayed on the display device 120 based on the received extracted information.

Thereby, the question screen SC is updated on the display device 120 . For example, the answer presenting unit 227 identifies the e-mail 31 (see FIG. 6), and transmits extracted information obtained by extracting a part of the text from the e-mail 31, "Product A is 3,500,000 yen.", as an answer candidate. In that case, as shown in FIG. 5(b), the answer candidate appears in the question screen SC. When the answer presenting unit 227 extracts a part of the text from the e-mail, for example, if the part containing the person's name, the part containing the greeting, and the part containing the quotation marks are removed from the entire text, the remaining part is taken as the part of the text. good. In the case of the e-mail 31, the portions including the person's name "Mr. Taro Yamada" and "Responsible for company XX: Hanako Fuji" are removed. Similarly, the part containing the greeting "... thank you." is removed, and the part containing the quotation mark ">" is also removed.

As shown in FIG. 5(b), in the question screen SC, in addition to the answer candidates, the answer source "Answer DB" and the answer source "Mail DB" of the answer candidates appear. In this embodiment, since the answer candidates are extracted from the mail DB 213, the answer presenting unit 227 strikes out the answer source "answer DB" and underlines the answer source "mail DB". . Thereby, the user operating the user terminal 100 can visually recognize the source of the answer candidate.

Even if one of the answer sources is crossed out and the other answer source is not underlined, and either one of the two answer sources appears, the user operating the user terminal 100 cannot select the answer candidates. The source can be visually recognized. On the other hand, these two answer sources may not be included in the question screen SC. As a result, the user does not need to be aware of the source of the answer candidates.

Further, as shown in FIG. 5(b), answer evaluations may appear on the question screen SC in addition to answer candidates and answer sources. The answer evaluation is a questionnaire as to whether or not presentation of answer candidates by dialogue server 200 was desirable for the user. When the affirmative button 41 is pressed, the control device 130 transmits affirmative information to the dialogue server 200 . When the negative button 42 is pressed, the control device 130 transmits negative information to the dialogue server 200 .

In the dialogue server 200, for example, the answer presenting unit 227 receives affirmative information or negative information, and associates the received affirmative information or negative information with the e-mails that are answer candidates. The answer presenting unit 227 may present an e-mail associated with affirmative information or negative information to the administrator terminal 150 . As a result, the administrator can consider various measures for improving the quality of the answer presented to the questioner.

In addition, in the process of step S1 described above, when the acquisition unit 222 acquires a question and the acquisition unit 222 can extract an answer corresponding to the acquired question from the answer DB 216, the question category determination unit 223, etc. You may skip the process of S11. As shown in FIG. 13, the answer DB 216 stores, in addition to answers corresponding to questions, answer information including answer IDs for identifying answers, and a plurality of transition states obtained by combining guidance IDs and options, in association with each other. ing.

For example, when the option "first option" is selected from among a plurality of options in the guidance question information 11 with the guidance ID "Y001", the information transitions to the guidance question information 12 with the guidance ID "Y002" (see FIG. 4). Then, when the option "first option" is selected from among the multiple options in the guidance question information 12, the answer presenting unit 227 selects the answer with the answer ID "A001" associated with these transition states as an answer candidate. Presented on the terminal 100 . In this case, on the question screen SC shown in FIG. 5(b), since the answer candidate is extracted from the answer DB 216, the answer source "answer DB" is underlined and the answer source "email DB" is struck through. be.

Although the preferred embodiments of the present invention have been described in detail above, the present invention is not limited to the specific embodiments, and various modifications can be made within the spirit and scope of the present invention described in the claims.・Changes are possible. The various servers described above may be realized by physical servers or by virtual servers.

In the above-described embodiment, the case where the question category and the mail category match, and the first keyword and the second keyword match has been described. It may be similar. Also, the question category and the mail category may be similar and the first keyword and the second keyword may be the same, or the question category and the mail category may be similar and the first keyword and the second keyword are similar. There may be. In the case of similarity, all or part of the text of the e-mail with the highest first similarity and highest second similarity may be presented to the user terminal 100 as an answer candidate. Note that the number of presentations may be one, or may be two or more.

ST dialogue system 100 user terminal 150 administrator terminal 200 dialogue server 210 storage unit 220 processing unit 221 dialogue processing unit 222 acquisition unit 223 question category determination unit 224 first keyword determination unit 225 mail category determination unit 226 second keyword determination unit 227 answer Presentation unit 230 Communication unit

Claims (8)

Get questions and emails independently,
If the obtained question contains a question mark or if the question corresponds to an interrogative sentence, the first determining the question category of said question to which the character belongs;
Based on at least one sentence or the entire sentence of the obtained question, determining a first specific word that can specify a commercial transaction target in the question,
If the obtained e-mail contains the question mark or if the e-mail corresponds to the interrogative sentence, based on the second sentence with the question mark or a specific second character included in the interrogative sentence equivalent , determining the mail category of said email to which said second character belongs;
Based on at least one sentence or the entire text of the acquired e-mail, determining a second specific word capable of specifying the commercial transaction target in the e-mail,
presenting the e-mail as an answer candidate to the question based on a first degree of similarity between the question category and the mail category and a second degree of similarity between the first specific word and the second specific word;
An answer presentation program for causing a computer to execute processing. The process of determining the question category determines the question category based on the first character and category information in which a plurality of characters and categories to which the plurality of characters belong are associated in advance,
the process of determining the mail category determines the mail category based on the second character and the category information;
The answer presentation program according to claim 1, characterized by: The process of determining the first specific word includes, in advance, a first word included in at least one sentence or the entire sentence of the acquired question, a plurality of words related to the business transaction object, and a specific word capable of specifying the business transaction object. Determine the first specific word based on the specific word information associated with
In the process of determining the second specific word, the second specific word is determined based on the second word included in at least one sentence or the entire sentence of the acquired e-mail and the specific word information.
3. The answer presentation program according to claim 1 or 2, characterized by: In the process of determining the question category, if the question category cannot be determined, presenting a guiding question that can determine the question category, and determining the question category based on the answer to the presented guiding question.
4. The answer presentation program according to any one of claims 1 to 3, characterized by: The process of independently obtaining the question and the e-mail includes obtaining the question received during a first period and obtaining the e-mail received during a second period prior to the first period.
5. The answer presentation program according to any one of claims 1 to 4, characterized by: The answer presentation program is a program used for a scenario-type chatbot that interacts with the questioner according to a predetermined scenario about a specific topic.
6. The answer presentation program according to any one of claims 1 to 5, characterized by: Get questions and emails independently,
If the obtained question contains a question mark or if the question corresponds to an interrogative sentence, the first determining the question category of said question to which the character belongs;
Based on at least one sentence or the entire sentence of the obtained question, determining a first specific word that can specify a commercial transaction target in the question,
If the obtained e-mail contains the question mark or if the e-mail corresponds to the interrogative sentence, based on the second sentence with the question mark or a specific second character included in the interrogative sentence equivalent , determining the mail category of said email to which said second character belongs;
Based on at least one sentence or the entire text of the acquired e-mail, determining a second specific word capable of specifying the commercial transaction target in the e-mail,
presenting the e-mail as an answer candidate to the question based on a first degree of similarity between the question category and the mail category and a second degree of similarity between the first specific word and the second specific word;
An answer presentation method in which processing is performed by a computer. Get questions and emails independently,
If the obtained question contains a question mark or if the question corresponds to an interrogative sentence, the first determining the question category of said question to which the character belongs;
Based on at least one sentence or the entire sentence of the obtained question, determining a first specific word that can specify a commercial transaction target in the question,
If the obtained e-mail contains the question mark or if the e-mail corresponds to the interrogative sentence, based on the second sentence with the question mark or a specific second character included in the interrogative sentence equivalent , determining the mail category of said email to which said second character belongs;
Based on at least one sentence or the entire text of the acquired e-mail, determining a second specific word capable of specifying the commercial transaction target in the e-mail,
presenting the e-mail as an answer candidate to the question based on a first degree of similarity between the question category and the mail category and a second degree of similarity between the first specific word and the second specific word;
An answer presentation device comprising a processing unit.

Images