JP7187957B2 - Control method, control program, and information processing device - Google Patents

Description

The present invention relates to a control method, a control program, and an information processing device.

It has been proposed to apply a technology of conversation between a person and a robot to a chat service such as LINE (registered trademark) (see Patent Document 1, for example).

WO2017/200078

By the way, there are robots that have a dialogue with a person (hereinafter referred to as a user) according to a predetermined scenario (hereinafter referred to as a talk rule) on a specific topic. When such a robot responds to a user's question based on the talk rules, a question that is less relevant to the intention of the user's question may be selected from the talk rules.

For example, if the user specifies a specific product and asks the robot a question about that product, but the talk rule first defines a question that asks the robot to specify the product to be questioned, the question will be It is selected and notified to the user who asked the question. Therefore, the user makes an answer specifying the product again. In this way, in the case of a robot that interacts with a user according to the talk rule, useless interaction may occur between the user and the robot.

Therefore, in one aspect, it is an object to suppress the execution of question-backing that has a low relationship with the intention of the question.

In one embodiment, the control method receives a question, and receives first information including the content of a reply to the question source of the question and a first answer candidate that is a candidate for the content of the reply to the question; , and the received question , a first similarity is calculated, and a second answer candidate is a candidate for the content of the reply to the question source of the question and the content of the reply to the question. and the received question , a second similarity is calculated, and the difference between the calculated first similarity and the second similarity satisfies the criteria The computer executes a process of controlling whether or not to execute the inquiry back to the question source according to whether or not.

It is possible to suppress the execution of question-backing that has a low relationship with the intention of the question.

FIG. 1 is a diagram for explaining an example of a dialogue system. FIG. 2 shows an example of the hardware configuration of the dialog server. FIG. 3 is an example of a block diagram of a dialogue server and a management server. FIG. 4 is a flow chart showing an example of learning processing of the learning unit. FIG. 5(a) is an example of the incident storage unit. FIG. 5(b) is an example of a trained model. FIG. 6 is a flowchart (part 1) showing an example of interactive processing between an API management unit and an AI unit. FIG. 7 is a flowchart (part 2) showing an example of interaction processing between the API management unit and the AI unit. FIG. 8 is an example of a question screen including an abstract question and its answer. FIG. 9 is a flow chart showing an example of inquiry-back discovery processing of the API management unit. FIG. 10 is an example of the talk rule storage unit. FIG. 11A is a diagram for explaining an example of control. FIG. 11(b) is an example of a question screen including a specific question and its reply. FIGS. 12A and 12B are diagrams for explaining another example of control. FIG. 13 is an example of a question screen including chat and chat response. FIG. 14(a) is an example of the manual storage unit. FIG. 14(b) is an example of a question screen containing manual information.

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

FIG. 1 is a diagram for explaining an example of the dialogue system ST. Dialog system ST includes user terminal 100 , dialog server 200 and management server 300 . The dialogue server 200 is an example of an information processing device. Although FIG. 1 shows a personal computer (PC) as an example of the user terminal 100, it may be a smart device such as a tablet terminal. Note that the components of the dialogue system ST may be changed as appropriate. For example, the components of the dialogue system ST may be the user terminal 100 and the dialogue server 200 . Also, the components of the dialogue system ST may be the dialogue server 200 and the management server 300 .

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 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 information corresponding to the content of the operation. 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 .

In addition, although the details will be described later, when the user operates the input device 110 to input characters, numbers, symbols, etc. in the input fields in the question screen SC, the control device 130 enters the operation contents in the input fields. Display appropriate letters, numbers, symbols, etc. When the user operates the input device 110 to press the send button in the question screen SC, the control device 130 treats the characters, numbers, symbols, etc. in the input fields as user's utterance information (hereinafter simply referred to as utterance). Send to the dialogue server 200 .

User terminal 100 and dialogue server 200 are connected to each other via communication network NW. The communication network NW is, for example, a Local Area Network (LAN). In this embodiment, the operation mode of the interaction server 200 will be described as on-premise. Upon receiving the speech transmitted from the user terminal 100, the dialogue server 200 determines the contents of the speech. When the dialog server 200 determines that the utterance is a question, it replies to the user terminal 100 which is the source of the question. When the dialogue server 200 determines that the utterance is a chat, it returns a chat response to the chat to the user terminal 100 that is the source of the chat. In this way, the dialogue server 200 returns questions and chat responses in response to user's utterances. Therefore, the dialogue server 200 may be called a chatbot. Chatbot is a coined word that combines chat and robot.

A management server 300 is connected to the dialogue server 200 . The dialog server 200 and management server 300 are directly connected by a communication cable CBL. Note that the interaction server 200 and the management server 300 may be indirectly connected via the communication network NW. Thus, the management server 300 is also operated on-premise. The management server 300 manages incidents for an information processing system (not shown). Specifically, the management server 300 manages trouble reports, requests, inquiries, etc. to the information processing system as incidents. The dialogue server 200 learns incidents managed by the management server 300 as learning targets, and generates and manages a learned model. The dialog server 200 uses the trained model to determine whether the utterance is chat. As described above, the operation mode of the dialog server 200 and the management server 300 is described as on-premise, but the operation mode of the dialog server 200 and the management server 300 may be cloud. In this case, the Internet may be used as the communication network NW.

Next, the hardware configuration of the interaction server 200 will be described with reference to FIG. Since the control device 130 and the management server 300 described above have the same hardware configuration as the dialogue server 200, the description thereof will be omitted.

FIG. 2 shows an example of the hardware configuration of the dialogue server 200. As shown in FIG. As shown in FIG. 2, the dialogue server 200 includes at least a central processing unit (CPU) 200A as a hardware processor, a random access memory (RAM) 200B, a read only memory (ROM) 200C, and a network I/F (interface). 200D included. The dialog server 200 may include at least one of a hard disk drive (HDD) 200E, an input I/F 200F, an output I/F 200G, an input/output I/F 200H, and a drive device 200I, as required. 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. A Micro Processing Unit (MPU) may be used as a hardware processor instead of the CPU 200A.

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 Universal Serial Bus (USB) memory and flash memory. The input/output I/F 200H reads programs and data stored in the 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 Compact Disc (CD)-ROM and Digital Versatile Disc (DVD). The drive device 200I reads programs and data recorded on the 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 programs stored in the ROM 200C and HDD 200E are temporarily stored in the RAM 200B described above by the CPU 200A. The program recorded on the portable recording medium 740 is temporarily stored in the RAM 200B by the CPU 200A. By executing the stored programs, the CPU 200A implements various functions described later, and also executes various processes described later. Note that the program may be in accordance with the flow chart described later.

Next, functional configurations of the interaction server 200 and the management server 300 will be described with reference to FIG.

FIG. 3 is an example of a block diagram of the dialogue server 200 and the management server 300. As shown in FIG. FIG. 3 shows main functions of the interaction server 200 and the management server 300 . As shown in FIG. 3, the dialogue server 200 includes a communication section 210, a processing section 220, and a storage section 230 as components. The communication unit 210 can be implemented by the network I/F 200D described above. The processing unit 220 can be realized by the above-described CPU 200A and RAM 200B. Storage unit 230 can be realized by HDD 200E described above. The communication unit 210 and processing unit 220 are functionally connected to each other. The processing unit 220 and the storage unit 230 are functionally connected to each other.

Here, the processing unit 220 includes an API management unit 221 that manages and provides an application programming interface (API) and an AI unit 222 that provides an artificial intelligence (AI) function. The API manager 221 includes a controller 251, a calculator 252, and a talk manager 253 as components. The AI unit 222 includes a learning unit 261, a learned model storage unit 262, and a chat determination unit 263 as components. On the other hand, the storage section 230 includes a talk rule storage section 231 and a manual storage section 232 as components. The control unit 251 cooperates with the talk rule storage unit 231, the manual storage unit 232, the calculation unit 252, the talk management unit 253, and the chat determination unit 263 to execute various processes.

For example, the control unit 251 acquires and accepts an utterance received by the communication unit 210 from the user terminal 100 via the communication network NW. Upon receiving an utterance, control unit 251 determines whether or not the received utterance is a question. If the utterance is not a question, control unit 251 outputs the received utterance to chat determination unit 263 . Chat determination unit 263 determines whether or not the utterance is chat based on the learned model stored in learned model storage unit 262 , and outputs the result to control unit 251 . Control unit 251 determines the type of speech based on the result output from chat determination unit 263 . For example, if the chat determination unit 263 outputs a positive result, the control unit 251 returns a chat response. Conversely, when chat determination unit 263 outputs a negative result, control unit 251 processes the utterance as a question.

When the control unit 251 receives a question, the calculation unit 252 generates first specific talk information including the content of the question back to the question source and the first answer candidates that are candidates for the content of the response to the question. , to calculate a first degree of similarity with the received question. Similarly, the calculation unit 252 calculates the second specific talk information including the content of the question back to the question source of the question and the second answer candidates that are candidates for the content of the answer to the question, and the received question. A second degree of similarity is calculated. The first answer candidate and the second answer candidate are different. When the calculation unit 252 calculates the first similarity and the second similarity, the control unit 251 determines whether the difference between the first similarity and the second similarity calculated by the calculation unit 252 satisfies the criteria. Accordingly, it controls whether or not to execute the inquiry back to the question source. Details of the functions realized by each component of the API management unit 221 including the control unit 251 and each component of the AI unit 222 including the chat determination unit 263 will be described in detail when the operation of the dialogue server 200 is described. Describe.

Management server 300 includes incident storage unit 310 . Incident storage unit 310 can be realized by HDD 200E described above. Note that the incident storage unit 310 may be included in the storage unit 230 described above. Conversely, at least one of the talk rule storage unit 231 and manual storage unit 232 described above may be included in the management server 300 .

The operation of the dialogue server 200 will be described below.

FIG. 4 is a flow chart showing an example of the learning process of the learning section 261. As shown in FIG. FIG. 5A is an example of the incident storage unit 310. FIG. FIG. 5(b) is an example of a trained model. The learning unit 261 executes learning processing before the user terminal 100 transmits an utterance. The learning unit 261 may dynamically execute the learning process based on the set time, or may execute the learning process based on an operation to activate the learning process.

First, as shown in FIG. 4, the learning unit 261 acquires incident information (step S101). Here, as shown in FIG. 5A, the incident storage unit 310 stores a plurality of pieces of incident information. Each piece of incident information includes an incident ID (denoted as INC ID in FIG. 5A) and an incident as components. The incident ID is identification information that identifies the incident. Incidents are, as described above, trouble reports, requests, inquiries, etc. to the information processing system (not shown). For example, the incident storage unit 310 stores incident information including inquiries about management software α as incidents. The incident storage unit 310 stores incident information including trouble reports regarding the management software β as incidents. The learning unit 261 accesses the incident storage unit 310 and acquires multiple pieces of incident information from the incident storage unit 310 .

When the process of step S101 is completed, the learning unit 261 learns incident information (step S102). More specifically, the learning unit 261 learns incident information by unsupervised learning, which is one of machine learning, and generates a learned model. One-Class Support Vector Machine (OCSVM) will be described as an example of unsupervised learning, but it is not limited to OCSVM, and a known non-hierarchical method such as the k-means method may be employed. As the learning unit 261 learns the incident information, the learning unit 261 generates a learned model as shown in FIG. 5(b). In the learned model, a set of feature points D1 representing incidents of management software α is surrounded by a boundary B1. A set of feature points D2 representing incidents of management software β is surrounded by a boundary B2. Therefore, if there is a feature point Dp that does not belong to either the inside of the boundary B1 or the inside of the boundary B2, the feature point Dp does not correspond to either the incident of the management software α or the incident of the management software β. can be determined to represent

When the process of step S102 is completed, the learning unit 261 saves the learned model (step S103). The learning unit 261 stores the learned model in the learned model storage unit 262 . Thereby, the learned model storage unit 262 stores the learned model. Therefore, when receiving an utterance output from the control unit 251, the chat determination unit 263 can use the learned model stored in the learned model storage unit 262 to determine whether or not the utterance is a chat.

FIG. 6 is a flowchart (part 1) showing an example of interaction processing between the API management unit 221 and the AI unit 222. As shown in FIG. FIG. 7 is a flowchart (No. 2) showing an example of interaction processing between the API management unit 221 and the AI unit 222 . 6 and 7 are continued by corresponding letters "A", "B" and "C".

First, as shown in FIG. 6, the control unit 251 receives an utterance (step S201). More specifically, control unit 251 accepts an utterance received by communication unit 210 . After the processing of step S201 is completed, the control unit 251 checks the dialogue state (step S202) and determines whether or not it is before dialogue (step S203). The control unit 251 manages the state of dialogue with the user terminal 100, and if the control unit 251 receives an utterance while the state of dialogue indicates a state before dialogue, the control unit 251 determines that the state is before dialogue. (Step S203: YES).

When the control unit 251 determines that it is before the dialogue, the control unit 251 switches the dialogue state from the state before the dialogue to the state during the dialogue, and determines whether or not the utterance is a question (step S204). Here, the questions are broadly divided into two types of questions. For example, a question such as "Can I ask you something?" is a question that does not specifically specify the target of the question. Such questions are hereinafter referred to as abstract questions. On the other hand, the question "Please tell me about the installation of the management software α." is a question specifically specifying the subject of the question. Such questions are hereinafter referred to as specific questions. In addition, depending on the utterance, there is also a case of chatting such as "It's hot and humid today." The control unit 251 analyzes the received utterance by morphological analysis. For example, if the utterance contains a specific symbol such as "?" The unit 251 determines that the utterance is a question (step S204: YES). Conversely, if the speech does not include such a specific symbol or a specific character string, the control unit 251 determines that the speech is not a question (step S204: NO). Even if the control unit 251 determines that the utterance is not a question, it does not determine that the utterance is chatting at this stage, and reserves the judgment.

When the control unit 251 determines that the utterance is a question, the API management unit 221 executes inquiry-back discovery processing (step S205). Although the details will be described later, the return inquiry discovery process is a process of specifying a return inquiry corresponding to a question. When the API management unit 221 identifies the reply to the question corresponding to the question, the communication unit 210 returns the reply to the inquiry identified by the API management unit 221 to the user terminal 100 (step S206).

A plurality of answer candidates including a first answer candidate and a second answer candidate are associated with the return question. Therefore, when the control device 130 receives the reply from the dialogue server 200, the control device 130 displays on the display device 120 the question screen SC including the contents of the reply and the talk information having a plurality of answer candidates for the reply. Thereby, as shown in FIG. 8, the display device 120 displays the question screen SC. The question screen SC includes talk information, an input field BX for inputting speech, and a send button BT for sending speech by pressing the pointer Pt. In the question screen SC shown in FIG. 8, the content of the question in response to the user's abstract question "What is your product?" It is shown.

Here, with reference to FIG. 9, the inquiry-back discovery process described above will be described in detail.

FIG. 9 is a flow chart showing an example of inquiry-back discovery processing of the API management unit 221 . In the process of step S204 shown in FIG. 6, when the control unit 251 determines that the utterance is a question, the control unit 251 accepts the question (step S301). Upon receiving a question, control unit 251 determines whether or not an answer candidate has been found from the question (step S302). Answer candidates are predetermined. Answer candidates include character strings corresponding to the content of the inquiry, such as "management software α", "management software β", and "installation". Therefore, in the case of the abstract question described above, since these character strings do not exist in the utterance, the control unit 251 determines that no answer candidates have been found (step S302: NO). Conversely, in the case of the above-described specific question, since these character strings appear in the utterance, the control unit 251 determines that answer candidates have been found (step S302: YES).

When the control unit 251 determines that no answer candidate has been found, the control unit 251 specifies the first question-back (step S303). More specifically, the control unit 251 refers to the talk rule storage unit 231 to identify the first question-back. As shown in FIG. 10, the talk rule storage unit 231 stores a plurality of pieces of talk information 10, 20, 30, 35, 40, 50, 55 that define the order of questions and answers. In FIG. 10, the order of questions is defined by the character string "Q1" included in the talk information 10, the character string "Q2" included in the talk information 20, and the like. Therefore, in this embodiment, the talk information 10 containing the content of the inquiry "What is your product?" corresponds to the first inquiry.

When the control unit 251 identifies the first return question, the control unit 251 executes the return question (step S304) and terminates the return inquiry process. Accordingly, the communication unit 210 executes the process of step S206 (see FIG. 6), and the control device 130 displays the question screen SC shown in FIG.

On the other hand, when the control unit 251 determines that an answer candidate has been found, the control unit 251 identifies the corresponding question-back (step S305). More specifically, the control unit 251 refers to the talk rule storage unit 231 to specify the question-back corresponding to the found answer candidate. For example, if the specific question includes answer candidates such as "management software α" and "installation", the query corresponding to these answer candidates is specified. Therefore, referring to the talk rule storage unit 231 shown in FIG. 10, the control unit 251 identifies the talk information 10 including the content of the return inquiry "What is your product?" as the corresponding return inquiry. In addition, the control unit 251 identifies the talk information 40 including the content of the reply "What is the operation?" as the corresponding reply. Furthermore, the control unit 251 identifies the talk information 50, 55 including the content of the inquiry "What is the detailed operation?" as the corresponding inquiry.

When the processing of step S305 is completed, the calculation unit 252 calculates a plurality of degrees of similarity (step S306). More specifically, as shown in FIG. 11(a), the calculation unit 252 calculates the first specific talk information 11 including the specified return question and the first answer candidate to the return question, and the first Calculate the similarity. In the present embodiment, since the first answer candidate “management software α” to the specified return question and the character string “management software α” included in the specific question are common, the calculation unit 252 calculates the similarity “0.3”, for example. is calculated as the first degree of similarity. Similarly, the calculation unit 252 calculates a second degree of similarity between the second specific talk information 12 including the specified return question and second answer candidates to the return question and the specific question. In the present embodiment, since the second answer candidate “management software β” to the specified return question and the character string “management software α” included in the specific question are not common, the calculation unit 252 calculates the degree of similarity “0”, for example. 2 similarity. Note that the calculation unit 252 calculates the first similarity and the second similarity using at least one of the cosine similarity, the Jaccard coefficient, and the Dice coefficient.

In addition, as shown in FIG. 12(a), the calculation unit 252 calculates the first similarity between the first specific talk information 41 including the specified return question and the first answer candidate to the return question and the first similarity to the specific question. Calculate In the present embodiment, the first answer candidate “install” to the specified return question and the character string “install” included in the specific question are common. Calculated as similarity. Similarly, the calculation unit 252 calculates a second degree of similarity between the second specific talk information 42 including the specified return question and second answer candidates to the return question and the specific question. In the present embodiment, since the second answer candidate “setting” for the specified question-back and the character string “install” included in the specific question are not common, the calculation unit 252 may set the similarity “0” to the second similarity Calculate as Further, as shown in FIG. 12(b), the calculation unit 252 calculates the first similarity between the first specific talk information 51 including the specified return question and the first answer candidate to the return question and the first similarity to the specific question. Calculate In the present embodiment, since the first answer candidate to the specified return question "installation on the operation server" and the character string "installation" included in the specific question are common, the calculation unit 252 calculates the degree of similarity "0.22", for example. is calculated as the first degree of similarity. Similarly, the calculation unit 252 calculates a second degree of similarity between the second specific talk information 52 including the specified return question and second answer candidates to the return question and the specific question. In the present embodiment, since the second answer candidate to the specified question is "installation on mail server" and the character string "installation" included in the specific question is common, the calculation unit 252 calculates the degree of similarity "0.23", for example. is calculated as the second degree of similarity. Note that the character string "install" is omitted from the first answer candidate and the second answer candidate shown in FIG. 12(b). As described above, the calculation unit 252 calculates the first degree of similarity and the second degree of similarity for a plurality of pieces of talk information 10, 40, and 50. Calculate to In this way, the calculation unit 252 calculates the first similarity and the second similarity for each specified return question. If there is a third answer candidate, the calculation unit 252 also calculates a third similarity using the same technique.

After the process of step S306 is completed, the calculation unit 252 calculates the difference obtained by subtracting the smaller one of the first similarity and the second similarity from the larger one of the first similarity and the second similarity. (Step S307). Therefore, as shown in FIG. 11A, in the case of the first similarity having a similarity of "0.3" and the second similarity having a similarity of "0", the calculation unit 252 calculates the difference as "0.3". 3” is calculated. The case of FIG. 12(a) is similar to the case of FIG. 11(a). On the other hand, as shown in FIG. 12B, when the first similarity is "0.22" and the second similarity is "0.23", the calculation unit 252 calculates the difference " 0.01" is calculated.

After the process of step S307 is completed, the control unit 251 then determines whether or not the difference is greater than the reference value (step S308). In other words, the processing of step S308 is such that the control unit 251 determines whether or not the difference satisfies the criterion. The reference value may be appropriately determined depending on the design. In this embodiment, the reference value "0.1" is adopted. Therefore, since the difference "0.3" is larger than the reference value "0.1", the control unit 251 determines that the difference is larger than the reference value (step S308: YES). Conversely, since the difference "0.01" is smaller than the reference value "0.1", the control unit 251 determines that the difference is smaller than the reference value (step S308: NO).

When the control unit 251 determines that the difference is larger than the reference value, the talk management unit 253 selects the larger degree of similarity between the first similarity degree and the second similarity degree from which the difference “0.3” is calculated. The specific talk information corresponding to the person is managed (step S309). Therefore, in the case of the first specific talk information 11 and the second specific talk information 12 shown in FIG. 11(a), the talk management unit 253 manages the first specific talk information 11 shown in FIG. 11(a). . Similarly, for the first specific talk information 41 and the second specific talk information 42 shown in FIG. 12(a), the talk manager 253 manages the first specific talk information 41 shown in FIG. 12(a). do. That is, the talk management unit 253 manages the information including the content of the question that is presumed to require no answer from the user based on the specific question as a skip target of the question. When the talk management unit 253 manages the skip target, the control unit 251 selects the next talk information of the skip target, which is earlier in the order, as the return target of the inquiry. That is, when the talk management unit 253 manages the first specific talk information 11 shown in FIG. 11(a) and the first specific talk information 41 shown in FIG. 12(a) as skip targets, the control unit 251 The talk information 20 (see FIG. 10) containing "Q2" is selected as a reply target of the inquiry.

After the process of step S309 is completed, the control unit 251 then determines the magnitudes of the first similarity and the second similarity (step S310). When the control unit 251 determines that the first degree of similarity is greater than the second degree of similarity (step S310: YES), it selects new talk information using the first answer candidates (step S311). More specifically, the control unit 251 refers to the talk rule storage unit 231, and selects the talk information transitioned in the case of the first answer candidate as new talk information for the user who is the question source. Therefore, in the case of the first answer candidate "management software α", the control section 251 selects the talk information 30 including the character string "Q3" regarding the version of the first answer candidate "management software α". When the process of step S311 is completed, the control unit 251 ends the inquiry-back discovery process.

Accordingly, when the control unit 251 executes the process of step S206 (see FIG. 6), the control device 130 displays the question screen SC shown in FIG. 11B on the display device 120. FIG. That is, when the user inputs a specific question on the question screen SC, the appearance of the talk information 10 including the character string "Q1" is suppressed and the content of the talk information 20 including the character string "Q2" is displayed on the question screen SC. Appear. That is, the dialogue server 200 determines that the answer to the talk information 10 containing the character string "Q1" has been obtained from the content of the specific question, and requests the user to answer the talk information 20 containing the character string "Q2".

On the other hand, when the control unit 251 determines that the second degree of similarity is greater than the first degree of similarity (step S310: NO), it selects new talk information using the second answer candidate (step S312). . More specifically, the control unit 251 refers to the talk rule storage unit 231 and selects the talk information to which the second answer candidate transitions as new talk information for the user who is the question source. Therefore, in the case of the second answer candidate "management software β", the control unit 251 selects talk information containing the character string "Q3" regarding the version of the second answer candidate "management software β". When the processing of step S312 is completed, the control unit 251 terminates the query return discovery processing.

If the control unit 251 determines in the process of step S308 that the difference is smaller than the reference value, the control unit 251 executes the process of step S304. That is, as shown in FIG. 12(b), it is impossible to narrow down clear answer candidates from the specific question to the content of the question return of the character string "Q5". to end the inquiry-back discovery process.

Returning to FIG. 6, in the process of step S204, when the control unit 251 determines that the utterance is not a question, the chat determination unit 263 determines whether or not the utterance is a chat (step S207). More specifically, when chat determination unit 263 receives an utterance output from control unit 251, it determines whether or not the received utterance is chat based on the learned model. For example, if the feature point corresponding to the utterance does not belong to either the inside of the boundary B1 or the inside of the boundary B2 (see FIG. 5(b)), it is determined that the received utterance is chat (step S207: YES), Output a positive result. As a result, the control unit 251 returns a chat response (step S208). Therefore, as shown in FIG. 13, the display device 120 displays a question screen SC including a chat response to the user's chat, "It's hot today, isn't it raining?"

On the other hand, if chat determination unit 263 determines that the received utterance is not chat (step S207: NO), it outputs a negative result. As a result, the API management unit 221 executes inquiry-back discovery processing (step S209). For example, when the content of the utterance is "Management software α is easy to use", even if there is a high possibility that it corresponds to chatting, the content includes "Management software α". feature points may belong inside the boundary B1. In such a case, the API management unit 221 executes inquiry-back discovery processing, and the control unit 251 accepts this content as a question and executes subsequent processing.

Here, in the process of step S203 described above, if the control unit 251 receives an utterance while the dialogue state indicates a state indicating that the dialogue is in progress, the control unit 251 determines that the dialogue is in progress (step S203: NO). ). In this case, as shown in FIG. 7, the control unit 251 determines whether the utterance is a number (step S301). For example, as shown in FIG. 11(b), when the control unit 251 determines that the utterance is numeric (step S301: YES), it refers to the talk rule storage unit 231 to determine whether or not there is the next question. is determined (step S302). In this embodiment, as shown in FIG. 10, talk information 20 containing the character string "Q2" asking about the operating system (OS) is stored next to talk information 30 containing the character string "Q3". Therefore, the control unit 251 determines that there is the next question (step S302: YES), and transitions to the next question (step S303). Then, the control unit 251 identifies the current query (step S304), and executes the process of step S206. As a result, control device 130 displays talk information 30 including the character string “Q3” on display device 120 .

As shown in FIG. 7, when the control unit 251 determines that the utterance is not a number (step S301: NO), it executes the process of step S304. That is, in the present embodiment, when an utterance other than a number is entered in the entry field BX and sent, the same question is repeatedly asked. That is, in the case shown in FIG. 8, the control device 130 displays the talk information 10 including the character string "Q1" on the display device 120 again.

In the process of step S302, when the control unit 251 determines that there is no next question (step S302: NO), or when the process of step S209 shown in FIG. 6 is completed, the control unit 251 searches manual information ( step S210). That is, in this embodiment, when the control unit 251 refers to the talk rule storage unit 231 and confirms that there is no talk information following the talk information 50 containing the character string "Q5", the control unit 251 searches manual information. .

Here, the control unit 251 refers to the manual storage unit 232 to search for manual information. The manual storage unit 232 stores a plurality of pieces of manual information with different manual titles, as shown in FIG. 14(a). The control unit 251 refers to the manual storage unit 232 and searches for manual information corresponding to the official answer specified by the utterance transmitted from the control device 130 and the answer to the specific talk information managed by the talk management unit 253. to identify. When the process of step S210 is completed, the control unit 251 returns the specified manual information (step S211), and ends the process. Thereby, as shown in FIG. 14B, the control device 130 displays the question screen SC including the manual information on the display device 120. FIG. When the user specifies the manual title he/she is looking for from among the manual titles included in the question screen SC and designates it, the control device 130 displays the operation manual associated with the specified manual title on the display device 120. do.

As described above, according to this embodiment, the interaction server 200 includes the processing unit 220 . In particular, the processing unit 220 includes an API management unit 221 , a control unit 251 and a calculation unit 252 . Control unit 251 accepts questions. The calculation unit 252 calculates the first specific talk information including the content of the reply to the question source of the question and the first answer candidate that is a candidate of the content of the reply to the question, and the first similarity of the received question. Calculate degrees. In addition, the second similarity between the second specific talk information including the content of the reply to the question source of the question and the second answer candidate that is the candidate of the content of the reply to the question and the second similarity with the received question is calculated. calculate. The control unit 251 controls whether or not to answer the question, depending on whether the difference between the first degree of similarity and the second degree of similarity calculated by the calculation unit 252 satisfies a criterion. As a result, it is possible to suppress the execution of question-backing that has a low relationship with the intention of the question.

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.

Note that the following notes are further disclosed with respect to the above description.
(Appendix 1) A question is received, and first information including the content of a reply to the question source of the question and a first answer candidate that is a candidate for the content of the reply to the question, and the received question. second information including second answer candidates that are candidates for the content of the reply to the question and the content of the reply to the question, and received calculating a second degree of similarity with the question, and executing the question back to the question source according to whether a difference between the calculated first degree of similarity and the second degree of similarity satisfies a criterion; A control method characterized in that a computer executes processing for controlling whether or not to perform.
(Supplementary Note 2) Control according to Supplementary Note 1, characterized in that when the difference is smaller than a reference value, the execution of the inquiry is performed, and when the difference is larger than the reference value, the execution of the inquiry is suppressed. Method.
(Appendix 3) When the difference is greater than the reference value and the first degree of similarity is greater than the second degree of similarity, the first answer candidate is used to 3. The control method according to appendix 1 or 2, further comprising: selecting content of a new question-back to the question source from candidates of content of the question-back.
(Appendix 4) Receiving an utterance, determining whether or not the received utterance includes a specific symbol or a specific character string representing a question, and determining whether the utterance includes the specific symbol or the specific character string 4. The control method according to any one of appendices 1 to 3, wherein the question is accepted when it is determined that
(Appendix 5) receiving an utterance, determining whether or not the utterance is chat based on the received utterance and a trained model generated based on a learning object, and determining that the utterance is not chat 5. The control method according to any one of appendices 1 to 4, wherein the question is received when the
(Appendix 6) The control method according to any one of Appendices 1 to 5, characterized in that the content of the inquiry is stipulated in the order of the inquiry.
(Appendix 7) A question is received, and first information including the content of a reply to the question source of the question and a first answer candidate that is a candidate for the content of the reply to the question; and the received question. second information including second answer candidates that are candidates for the content of the reply to the question and the content of the reply to the question, and received calculating a second degree of similarity with the question, and executing the question back to the question source according to whether a difference between the calculated first degree of similarity and the second degree of similarity satisfies a criterion; A control program characterized by causing a computer to execute a process that controls whether to perform or not.
(Appendix 8) A question is received, and first information including the content of a reply to the question source of the question and a first answer candidate that is a candidate for the content of the reply to the question; and the received question. second information including second answer candidates that are candidates for the content of the reply to the question and the content of the reply to the question, and received calculating a second degree of similarity with the question, and executing the question back to the question source according to whether a difference between the calculated first degree of similarity and the second degree of similarity satisfies a criterion; An information processing apparatus, comprising: a processing unit that executes processing for controlling whether or not to perform processing.
(Supplementary note 9) The processing unit executes the question return when the difference is smaller than a reference value, and suppresses execution of the question return when the difference is larger than the reference value. 9. The information processing device according to 8.
(Supplementary note 10) When the difference is greater than the reference value and the first similarity is greater than the second similarity, the processing unit uses the first answer candidate to store the 10. The information processing apparatus according to appendix 8 or 9, wherein a new question-back content to the question source is selected from among a plurality of question-back content candidates stored in the .
(Supplementary Note 11) The processing unit receives an utterance, determines whether the received utterance includes a specific symbol or a specific character string representing a question, and determines whether the utterance includes the specific symbol or the specific character string. 11. The information processing apparatus according to any one of appendices 8 to 10, wherein the question is accepted when it is determined that the character string is included.
(Supplementary Note 12) The processing unit receives an utterance, determines whether the utterance is chat based on the received utterance and a trained model generated based on a learning target, and determines whether the utterance is chat. 12. The information processing apparatus according to any one of appendices 8 to 11, wherein the question is accepted when it is determined that the question is not a chat.
(Appendix 13) The information processing apparatus according to any one of Appendices 8 to 12, characterized in that the content of the inquiry is stipulated in the order of the inquiry.

ST dialogue system 100 user terminal 200 dialogue server 210 communication unit 220 processing unit 251 control unit 252 calculation unit 230 storage unit 300 management server

Claims (8)

take questions,
A first information that includes the content of the question back to the question source of the question and a first answer candidate that is a candidate for the content of the answer to the question, and the first information that is the degree of similarity between the received question and the first information. Calculate the similarity of
A degree of similarity between the received question and second information including the content of the reply to the question source of the question and a second answer candidate that is a candidate for the content of the reply to the question. Calculate the similarity of 2,
Depending on whether the difference between the calculated first degree of similarity and the second degree of similarity satisfies a criterion, controlling whether or not to execute the inquiry back to the question source,
A control method characterized in that the processing is executed by a computer. If the difference is smaller than a reference value, the execution of the inquiry is performed, and if the difference is larger than the reference value, the execution of the inquiry is suppressed.
2. The control method according to claim 1, characterized by: When the difference is greater than the reference value and the first degree of similarity is greater than the second degree of similarity, the content of a plurality of questions and answers stored in a storage unit using the first answer candidate. select the contents of a new question back to the question source from among the candidates of
3. The control method according to claim 1 or 2, characterized in that: accept utterances,
Determining whether the received utterance contains a specific symbol or a specific character string representing a question,
Accepting the question when it is determined that the specific symbol or the specific character string is included in the utterance;
4. The control method according to any one of claims 1 to 3, characterized in that: accept utterances,
Determining whether the utterance is chat based on the received utterance and a trained model generated based on the learning target,
Accepting the question when it is determined that the utterance is not the chat;
5. The control method according to any one of claims 1 to 4, characterized in that: In the content of the inquiry, the order of inquiry is stipulated,
The control method according to any one of claims 1 to 5, characterized in that: take questions,
A first information that includes the content of the question back to the question source of the question and a first answer candidate that is a candidate for the content of the answer to the question, and the first information that is the degree of similarity between the received question and the first information. Calculate the similarity of
A degree of similarity between the received question and second information including the content of the reply to the question source of the question and a second answer candidate that is a candidate for the content of the reply to the question. Calculate the similarity of 2,
Depending on whether the difference between the calculated first degree of similarity and the second degree of similarity satisfies a criterion, controlling whether or not to execute the inquiry back to the question source,
A control program that causes a computer to execute processing. take questions,
A first information that includes the content of the question back to the question source of the question and a first answer candidate that is a candidate for the content of the answer to the question, and the first information that is the degree of similarity between the received question and the first information. Calculate the similarity of
A degree of similarity between the received question and second information including the content of the reply to the question source of the question and a second answer candidate that is a candidate for the content of the reply to the question. Calculate the similarity of 2,
Depending on whether the difference between the calculated first degree of similarity and the second degree of similarity satisfies a criterion, controlling whether or not to execute the inquiry back to the question source,
An information processing apparatus comprising a processing unit that executes processing.

Images