JP2020204711A - Registration system - Google Patents

Abstract

To effectively maintain, by persons, details of answers to questions from customers by a robot which provides interactive service.SOLUTION: The present invention relates to a registration system which registers answers to questions from customers by a robot which provides interactive service, and a robot control device includes: a speech recognition part which recognizes speeches of an interaction between persons; an interaction log database in which speech recognition results by the speech recognition part are registered as an interaction log; an answer database in which combinations of questions and answers to the questions are registered as interaction examples in advance; an interaction generation part which determines propriety of the interaction examples based upon the interaction log registered in the interaction log database and the interaction examples registered in the answer database, and generates correction candidates for interaction examples according to a rule based upon the determination results; and a screen generation part which generates an interaction example correction screen where the correction candidates generated by the interaction generation part are represented. The interaction generation part registers correction candidates selected on the interaction example correction screen in the answer database.SELECTED DRAWING: Figure 3

Description

The present invention relates to a registration system, and is suitable for application to a registration system for registering answers to questions from customers by a robot that provides an interactive service.

In recent years, the functions of voice recognition, dialogue, and translation in robots have been improved, and the cost of robots has been reduced. For this reason, there are increasing opportunities for robots to be used in various stores and facilities to provide dialogue services with humans. Specifically, the dialogue service includes a guidance service that provides voice guidance to a person (customer) to be guided such as a passerby or a visitor, a translation service that performs translation as an intermediary between people, or a translation service. There is an automatic voice response service over the phone.

In order for the robot that provides the dialogue service to answer the question, an example of dialogue that is a combination of the assumed question (Q) and answer (A) (hereinafter referred to as Q & A (Question And Answer)). ) To the robot and record (register) it in the internal response database (hereinafter, the database is abbreviated as DB). The robot searches the questions registered in the answer DB based on the customer's voice recognition result, selects the answer associated with the question with a high match rate, converts the text of the selected answer into a voice file, and uses the speaker. By outputting, it is possible to answer the dialogue. Generally, when registering Q & A in the answer DB, before installing the robot, a user such as an administrator considers the assumed questions according to the role of the robot, the installation location, etc., and then creates a Q & A list to create the answer DB. It is realized by importing to.

In addition, in order for the robot in operation to properly perform its duties, it is necessary to be able to provide appropriate answers to the questions. For example, if the presumed question created in advance and the customer's question do not match, the robot cannot give an appropriate answer. Therefore, by aggregating and analyzing customer questions and robot answer results as a log, it is possible to find out what kind of questions and answers are lacking and expand the Q & A list.

For example, Patent Document 1 discloses a system that can automatically expand the Q & A list registered in the response DB. Specifically, the system disclosed in Patent Document 1 aggregates logs of dialogues between people, and matches a question template with a response template to determine whether the speech recognition result in the dialogue log is a question or a response. Judgment is made, the combination of question and response is registered in the answer DB, and when the robot answers, the closest answer from the Q & A registered in the answer DB is output.

Another service that can be performed by a robot that provides a dialogue service is assisting the dialogue between the customer and the operator. That is, when the robot cannot provide sufficient guidance, the operator may provide appropriate guidance. In such a case, the robot changes its role as an assistant between the customer and the operator and translates the dialogue. By doing so, the operator can be assisted.

Japanese Unexamined Patent Publication No. 2000-339314

By the way, in the dialogue service provided by the robot, the appropriate answer may change due to changes in the situation, so the user such as the administrator confirms and scrutinizes whether the Q & A registered in the answer DB is appropriate. A manual maintenance process is required. For example, if there is a time change such as a change in the installation location of the robot, it may not be possible to provide correct guidance with the previous answers, and it was necessary to manually register a new answer. Also, for example, when a robot assists a dialogue between a customer and an operator, the utterance of the operator, who is originally the respondent, may include a question for confirming the customer's intention. It is difficult to judge whether the voice generated in the above is a question or an answer because it depends on the role of the speaker, and it is necessary to manually examine the contents registered in the answer DB. Furthermore, since dialogue between people includes different characteristics of spoken language for each speaker, there is also the problem that individual answers vary, so manual maintenance is required to eliminate personality. Was preferred.

However, in the case of the system disclosed in Patent Document 1, since Q & A is simply accumulated in the response DB based on the aggregated dialogue log without judging the appropriateness of the response content, there are many inappropriate Q & A. Since it will be registered, there is a problem that it is difficult to carry out effective maintenance, such as the time and effort required for manual Q & A maintenance.

The present invention has been made in consideration of the above points, and an attempt is made to propose a registration system that can support effective manual maintenance of the contents of answers to questions from customers by a robot that provides a dialogue service. Is what you do.

In order to solve such a problem, the present invention is a registration system for registering answers to questions from customers by a robot that provides a dialogue service, a voice recognition unit that recognizes a dialogue between people by voice, and the voice. The dialogue log database in which the voice recognition result by the recognition unit is registered as a dialogue log, the answer database in which the combination of the question and the answer to the question is registered in advance as a dialogue example, and the dialogue log database are registered. A dialogue generation unit that determines the appropriateness of the dialogue example based on the dialogue log and the dialogue example registered in the response database, and generates correction candidates for the dialogue example according to a rule according to the determination result, and the dialogue. A screen generation unit for generating the dialogue example correction screen on which the correction candidate generated by the generation unit is posted is provided, and the dialogue generation unit responds to the correction candidate selected on the dialogue example correction screen. Changed to register in the database.

According to the present invention, it is possible to support the effective implementation of manual maintenance of the contents of answers to questions from customers by a robot that provides a dialogue service.

It is a block diagram which shows the whole structure example of the dialogue registration system of the robot which concerns on one Embodiment of this invention. It is a block diagram which shows the structural example of the control system of the robot shown in FIG. It is a block diagram which shows the structural example of the control system of the robot control apparatus shown in FIG. It is a figure for demonstrating an example of Q & A data. It is a block diagram which shows the structural example of the control system of the operation terminal shown in FIG. It is a flowchart which shows the processing procedure example of the service mode determination processing. It is a flowchart which shows the processing procedure example of the direct dialogue processing. It is a flowchart which shows the processing procedure example of the dialogue mediation processing. It is a figure for demonstrating an example of a dialogue log data. It is a figure which shows the specific example of the Q & A correction screen. It is a flowchart which shows an example of a manual operation procedure in maintenance of Q & A. It is a flowchart which shows the processing procedure example of the correction candidate creation processing. It is a figure which shows the judgment matrix which defined the judgment standard of classification and the automatic generation rule of Q & A. It is a flowchart which shows the processing procedure example of the priority score calculation process.

Hereinafter, an example of the robot dialogue registration system according to the embodiment of the present invention will be described with reference to the drawings, but the present invention is not limited to the following examples.

(1) Configuration Example of Robot Dialogue Registration System 1 FIG. 1 is a block diagram showing an overall configuration example of a robot dialogue registration system according to an embodiment of the present invention. As shown in FIG. 1, the robot dialogue registration system 1 includes a robot 10, a robot control device 20, and an operation terminal 30. The robot 10 and the robot control device 20 and the robot control device 20 and the operation terminal 30 are connected by a communication network such as a wired LAN (Local Area Network) or a wireless LAN.

The robot 10 is a guide robot that can move autonomously, and can move within a predetermined range in the facility where the robot 10 is introduced, for example, in the floor where the robot 10 is installed. Then, the robot 10 provides a service of providing voice guidance to a guide target person (customer) such as a passerby or a visitor in the facility, a service of guiding the customer to a predetermined place by an action such as running or a gesture, and the like. provide. In the present embodiment, as the dialogue service that can be provided by the robot 10, a "guidance service" that provides guidance while interacting with the customer and a "translation service" that supports the dialogue between the customer and the operator by translation are described. To do.

The robot 10 transfers sensor data detected by an internal sensor (camera 131, microphone 132, etc., which will be described later) to the robot control device 20, and the robot control device 20 transfers a person according to the content of the received sensor data. The robot 10 is instructed to guide the robot 10. Then, the robot 10 that has received the instruction gives an answer and guidance by performing an operation such as utterance, motion, or movement through the output device.

The robot control device 20 creates and holds a customer service log based on the sensor data received from the robot 10. In particular, the identification result (voice recognition result) of the voice dialogue between people is held as a dialogue log as an example of the customer service log (dialogue log DB229 described later). Further, the robot control device 20 creates a Q & A correction screen (Q & A correction screen 500 described later) based on the dialogue log.

The operation terminal 30 can access the Q & A correction screen provided by the robot control device 20 to search, edit, and register the Q & A correction candidate.

(1-1) Example of Internal Configuration of Robot 10 FIG. 2 is a block diagram showing a configuration example of the control system of the robot shown in FIG. As shown in FIG. 2, the robot 10 includes a CPU (central processing unit) 110, a storage device 120 for executing various processes under the control of the CPU 110, an input / output device 130, and a communication interface 140.

The CPU 110 is an example of a processor, and by reading and executing a program stored in the storage device 120, for example, each functional unit (drive control unit 121, dialogue control unit 122, The function of the input / output unit 123) is realized.

The storage device 120 is a storage device for storing programs and data, and is composed of a main storage device such as a RAM (Random Access Memory) and an auxiliary storage device such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive). Will be done. FIG. 2 shows a drive control unit 121, an interactive control unit 122, and an input / output unit 123 as an example of a functional configuration in which a function is realized by executing a program stored in the main storage device of the storage device 120. There is.

The drive control unit 121 performs drive control when the robot 10 is autonomously moved. Specifically, for example, the drive control unit 121 is a robot based on an image taken by the camera 131 of the robot 10 when the robot 10 moves autonomously, a position of itself detected by the range sensor 134 by a light wave, and the like. Judge the surrounding situation of 10. Then, the drive control unit 121 autonomously moves the robot 10 while avoiding obstacles such as people and walls, based on the determined contents and preset data regarding the distance to the customer. Further, the drive control unit 121 detects the inclination of the robot 10 based on the detection value of the gyro sensor 133, and controls the robot 10 to move autonomously without falling. If an obstacle is detected based on the information obtained from the camera 131 and / or the range sensor 134 while the robot 10 is moving, the drive control unit 121 stops the movement of the robot 10 or the obstacle. The operation of the robot 10 is controlled so as to avoid the above.

In the present embodiment, the movable range in which the robot 10 can move is limited to a predetermined range (for example, in a facility) determined in advance. That is, the current position of the robot 10 determined by the drive control unit 121 remains within the movable range.

In response to an instruction from the robot control device 20, the dialogue control unit 122 causes the input / output device 130 to perform a dialogue operation with a customer or an operator via voice by using the microphone 132 and the speaker 135.

The input / output unit 123 executes various data input / output operations performed with the input / output device 130, and also performs various data input / output operations performed with the robot control device 20 via the communication interface 140. Execute.

The input / output device 130 includes a camera 131, a microphone 132, a gyro sensor 133, a range sensor 134, a speaker 135, and a drive mechanism 136.

The camera 131 acquires image information such as the face of the customer by photographing the person to be guided (customer). The microphone 132 (an example of the input means, which may be another sound collecting device capable of acquiring voice information) acquires voice information which is information related to voice generated from a customer or an operator. Various data such as image information acquired by the camera 131 and voice information acquired by the microphone 132 are transmitted to the robot control device 20 via the communication interface 140.

The gyro sensor 133 detects the inclination of the robot 10 and the like based on the angular acceleration applied to the robot 10. The gyro sensor 133 transmits the detected tilt angle to the robot control device 20 via the communication interface 140.

The range sensor 134 is a sensor that identifies the position of the robot 10 and detects the surrounding environment of the robot 10 based on the distance measurement using light waves. The range sensor 134 measures the position of the robot 10 and the shape of the surrounding space including obstacles and the like. The range sensor 134 transmits the data obtained by the measurement to the robot control device 20 via the communication interface 140.

The speaker 135 outputs the dialogue voice generated by the dialogue control unit 122, which is necessary for the robot 10 to guide the inside of the facility, to the outside.

The drive mechanism 136 is a mechanism for moving the robot 10 based on an instruction from the drive control unit 121. For example, when the robot 10 is in a form movable by wheels, the drive mechanism 136 is a motor for driving the wheels provided on the robot 10. For example, when the robot 10 has a humanoid form capable of bipedal walking, the drive mechanism 136 is an actuator that drives a member corresponding to a foot provided on the robot 10.

The communication interface 140 is a device for connecting to a network, and is connected to the communication interface 240 of the robot control device 20 (see FIG. 3 described later) via the network.

(1-2) Example of Internal Configuration of Robot Control Device 20 FIG. 3 is a block diagram showing a configuration example of a control system of the robot control device shown in FIG. As shown in FIG. 3, the robot control device 20 includes a CPU 210, a storage device 220, and a communication interface 240.

The CPU 210 is an example of a processor, and by reading and executing a program stored in the storage device 220, for example, each functional unit (speech recognition unit 221, image recognition) shown in the storage device 220 of FIG. 3 The functions of the unit 222, the service control unit 223, the dialogue control unit 224, the translation control unit 225, the screen generation unit 226, and the dialogue generation unit 227) are realized.

The storage device 220 is a storage device for storing programs and data, and is composed of a main storage device such as a RAM and an auxiliary storage device such as an HDD or SSD. FIG. 3 shows, as an example of the functional configuration in which the function is realized by executing the program stored in the main storage device of the storage device 220, the voice recognition unit 221 and the image recognition unit 222, the service control unit 223, and the dialogue control unit 224. , Translation control unit 225, screen generation unit 226, and dialogue generation unit 227 are shown, and answer DB228, dialogue log DB229, and exclusion candidate DB230 are shown as an example of a data group stored in the auxiliary storage device of the storage device 220. Has been done. The various DBs shown in FIG. 3 may be realized by an external storage means capable of communicating with the robot control device 20 (communication interface 240) via a network.

The voice recognition unit 221 performs voice recognition on the voice data transmitted from the robot 10 and received by the communication interface 240, and converts the voice data into text (character string) in the spoken language of the speaker (customer or operator). To carry out. Hereinafter, the text generated by this text conversion may be referred to as "speech text".

As an example of the language setting method used for text conversion by the voice recognition unit 221, when the utterance language of the customer (guidance target person) is set, the utterance content of the customer is received at the start of the service to the customer, and the language is received. By determining which of the plurality of pre-registered languages is closest to, the customer's utterance language can be specified and set. Further, as another example, when setting the utterance language of the operator, the language used by the operator may be set in advance, or the same method as the setting of the utterance language of the customer may be used.

The voice recognition unit 221 analyzes the voice feature amount of the voice data and determines whether the voice feature amount matches the feature amount registered in advance, so that the speaker is a customer or an operator. Can be identified.

The image recognition unit 222 performs person detection by a classifier having facial features based on the image data transmitted from the robot 10 and received by the communication interface 240. The image recognition unit 222 can identify whether the speaker is a customer or an operator by determining whether the feature amount of the face matches the feature amount registered in advance.

The service control unit 223 instructs the robot 10 to serve customers and directs dialogue to the dialogue control unit 224 according to the voice text detected by the voice recognition unit 221 and the person detection result detected by the image recognition unit 222. , Or an instruction to mediate dialogue to the translation control unit 225.

The service control unit 223 can determine and set the "service mode" that the robot control device 20 has as an internal state. When the service mode is set to a predetermined mode by the service control unit 223, the robot control device 20 controls the robot 10 based on the mode, and the robot 10 can provide a service according to the mode to the customer. it can.

In the present embodiment, "guidance service mode", "translation service mode", and "Q & A correction mode" will be described as specific examples of the internal state of the robot control device 20. Among these, as specific examples of the service mode related to the dialogue service, the "guidance service mode" set when the robot 10 provides the guidance service and the "translation" set when the robot 10 provides the translation service. "Service mode" is prepared. In addition, the "Q & A correction mode" is a mode set when performing Q & A maintenance manually (user such as an administrator). Although the details will be described later, in the Q & A correction mode, "Q & A correction candidates" are displayed on the Q & A correction screen referenced from the operation terminal 30, and the Q & A correction candidates selected by the user can be registered in the answer DB 228.

The details will be described later with reference to FIG. 6, but when the service control unit 223 determines that the service mode is the "guidance service mode", the service control unit 224 directly talks with the customer based on the voice text to the dialogue control unit 224. Instruct the implementation of (answer). On the other hand, when the service control unit 223 determines that the service mode is the "translation service mode", the service control unit 225 tells the translation control unit 225 the voice text and the speaker determination result based on the result of identifying whether the speaker is a customer or an operator. And instruct the implementation of dialogue mediation that mediates the dialogue between the customer and the operator by translation.

The dialogue control unit 224 searches the answer DB 228 based on the voice text, and outputs the answer according to the search result. Specifically, the dialogue control unit 224 searches for whether a hypothetical question (Q) similar to or matching the voice text is registered in the answer DB 228, and if there is a corresponding hypothetical question (Q), if there is a corresponding hypothetical question (Q), The answer (A) registered in association with the assumed question (Q) in the answer DB 228 is output, and if there is no corresponding assumed question (Q), a predetermined answer prepared in advance is output. The answer output by the dialogue control unit 224 is transferred to the robot 10 after performing voice synthesis and converting it into a voice file. As a result, the robot 10 can answer the customer's question by outputting the transferred audio file to the speaker 135.

The translation control unit 225 performs mutual translation between the customer and the operator based on the voice text and the speaker determination result. That is, the translation control unit 225 translates the customer's utterance language into the operator's utterance language and outputs it, and translates the operator's utterance language into the customer's utterance language and outputs it. Specifically, for example, when the customer speaks in English and the operator speaks in Japanese, the translation control unit 225 translates the customer's utterance result from English to Japanese, performs voice synthesis, and creates a voice file. After conversion, it is transferred to the robot 10. As a result, the robot 10 can translate the customer's utterance content and convey it to the operator by outputting the transferred audio file to the speaker 135. The utterance contents and translation results of the customer and the operator are stored in the dialogue log DB229.

The screen generation unit 226 generates a screen to be displayed on a predetermined display device (in this example, the display 331 of the operation terminal 30 is used, but other display devices may be provided). Specifically, for example, when the robot control device 20 receives a predetermined access request (Q & A correction request) from the operation terminal 30, the screen generation unit 226 can register and correct the Q & A managed by the response DB 228. Output the correction screen. The Q & A correction screen is provided by, for example, a GUI (Graphical User Interface), and the detailed contents of the screen will be described later with reference to FIG.

The dialogue generation unit 227 automatically generates and outputs Q & A correction candidates in the "Q & A correction mode" in which a user such as an administrator can correct the Q & A through an operation from the Q & A correction screen. When the dialogue generation unit 227 receives a search request on the Q & A correction screen by the user's operation on the operation terminal 30, the dialogue generation unit 227 accesses the response DB 228, the dialogue log DB 229, and the exclusion candidate DB 230 to search each DB. , Automatically create correction candidates for Q & A to be added or corrected, and calculate the priority score for each correction candidate. Then, the result generated (created and calculated) by the dialogue generation unit 227 is displayed on the Q & A correction screen as a search result.

The answer DB 228 holds a Q & A for the robot 10 to answer in the dialogue service. Specifically, the combination of the expected question and the answer is registered in the answer DB 228 (Q & A data). As described above, the response DB 228 accepts searches from the dialogue control unit 224 and the dialogue generation unit 227.

FIG. 4 is a diagram for explaining an example of Q & A data. The Q & A data 410 illustrated in FIG. 4 is information stored in the answer DB 228, in which the question (Q) 411 in which the expected question is recorded and the answer corresponding to the question in the line (record) are recorded. Answer (A) 412, and edit time 413 in which the time stamp when any of the question (Q) 411 and answer (A) 412 of the record was last edited (registered, modified, etc.) is recorded. Is configured with. By holding the Q & A data 410 including the question (Q) 411 and the answer (A) 412, the answer DB 228 can register the expected combination of Q & A in line (record) units. Further, the edit time 413 is referred to in the priority score calculation process described later.

Specifically, in the case of Fig. 4, when looking at the Q & A on the first line of the data line, in response to the Japanese question (Q) 411, "Where is the restroom?", "Go straight here and turn left. The Japanese answer (A) 412, "If you proceed, there is." Is registered. Also, looking at the Q & A on the second line of the data line, the Q & A in English is registered with the same content as the first line. As described above, in the answer DB 228, although the combination of individual Q & A is registered in the same language, as a whole, the Q & A in a plurality of kinds of languages that can be translated by the translation control unit 225 can be registered.

The dialogue log DB 229 holds a log of customer service performed by the robot 10, specifically, a log that aggregates the results of dialogue in the guidance service and the results of translation in the translation service. As described above, the dialogue log DB 229 accepts a search from the dialogue generation unit 227.

The exclusion candidate DB 230 holds information about the Q & A set to be excluded in the Q & A modification mode. Specifically, the exclusion candidate DB 230 holds, for example, the Q (question) of the Q & A for the Q & A generated by the dialogue generation unit 227 and set as "exclusion" in the search result displayed on the Q & A correction screen. .. As described above, the exclusion candidate DB 230 accepts the search from the dialogue generation unit 227.

The communication interface 240 is a device for connecting to a network, and is connected to the communication interface 340 of the operation terminal 30 (see FIG. 5 described later) via the network.

(1-3) Example of Internal Configuration of Operation Terminal 30 FIG. 5 is a block diagram showing a configuration example of a control system of the operation terminal shown in FIG. As shown in FIG. 5, the operation terminal 30 is, for example, a general-purpose PC, and includes a CPU 310, a storage device 320 that executes various processes under the control of the CPU 310, an input / output device 330, and a communication interface 340. To be equipped.

The CPU 310 is an example of a processor, and by reading and executing a program stored in the storage device 320, for example, the functions of each function unit (browser 321 and input / output unit 322) shown in the storage device 320. To realize.

The storage device 320 is a storage device for storing programs and data, and is composed of a main storage device such as a RAM and an auxiliary storage device such as an HDD or SSD. FIG. 5 shows a browser 321 and an input / output unit 322 as an example of a functional configuration in which a function is realized by executing a program stored in the main storage device of the storage device 320.

The browser 321 accesses the robot control device 20 via the communication interface 340, receives the information on the Q & A correction screen output from the screen generator 226, and based on the information, Q & A on the display 331 of the input / output device 330. Display the correction screen. On the Q & A correction screen displayed by the browser 321, Q & A registration and correction can be performed by operating the displayed items with the keyboard 332 and the mouse 333.

The communication interface 340 is a device for connecting to a network, and is connected to the communication interface 240 of the robot control device 20 via the network.

(2) Dialogue Service The details of the processing performed when the robot 10 provides the dialogue service (guidance service, translation service) will be described below.

(2-1) Service Mode Determination Process FIG. 6 is a flowchart showing an example of a processing procedure of the service mode determination process. The service mode determination process shown in FIG. 6 is a process of determining the dialogue service provided by the robot 10 through a dialogue between the robot 10 and a person (customer), and is a person detection based on an image taken by the camera 131 in the robot 10. Or, when interrupt detection is performed based on the voice information acquired by the microphone 132, the following sequence is executed through a dialogue with a person (customer).

According to FIG. 6, first, the robot 10 performs a process of asking the customer for a service decision (step S11). Specifically, for example, the service control unit 223 of the robot control device 20 instructs the dialogue control unit 122 of the robot 10 to execute a processing for prompting the customer to select a service, and the dialogue control unit 122 receives the instruction. By using the speaker 135 of the robot 10 to speak such as "is it a guidance service? Is it a translation service?", The customer is encouraged to select a service. The instruction for prompting the customer to select a service in step S12 is not limited to voice, and the service options may be displayed on a touch display or the like so that the customer can touch and select the service.

In the next step S12, the service control unit 223 determines whether or not the service selected by the customer in step S11 is a guidance service. If it is determined that the selected service is a guidance service (YES in step S12), the process proceeds to step S13, and if it is determined that the selected service is not a guidance service, that is, it is a translation service (step). NO) in S12, the process proceeds to step S14.

In step S13, the service control unit 223 sets the internal state of the robot control device 20 to the "guidance service mode", and then the robot 10 executes the guidance service to the customer. The robot control device 20 in the guidance service mode presents the options of the guidance service that it can implement, and thereafter, hears the customer's request through dialogue with the customer and performs customer service using the Q & A registered in the answer DB 228. To do.

On the other hand, in step S14, the service control unit 223 sets the internal state of the robot control device 20 to the "translation service mode", and then performs the translation service to the customer by the robot 10. Here, when the translation service is implemented, the robot 10 supports the dialogue between the operator who manages the facility in which the robot 10 is installed and the customer by translation. Therefore, when the translation service mode is set, the translation service is provided via the communication interface 240 of the robot control device 20 (specifically, using some notification means such as telephone, mail, or lamp display). It is preferable to notify the operator to that effect.

In step S15 following step S14, the service control unit 223 performs a setting process (customer identification setting) for identifying the customer and the operator who are speakers. The customer identification setting will be described in detail. When the translation service is provided, mutual translation processing is performed between the customer's utterance language and the operator's utterance language. Therefore, the robot control device 20 determines whether the speaker is the customer or the operator, and each of them. It is necessary to be able to identify in which language the speaker speaks. Therefore, in the customer identification process, the service control unit 223 first outputs a predetermined voice from the robot 10 to conduct a hearing with the customer, and based on the utterance of the customer who responded, the utterance language and direction of the speaker (customer). And record. Further, the service control unit 223 also conducts a hearing with the operator, identifies and records the utterance language and direction of the speaker (operator) based on the utterance of the operator who responded. Here, the "speaker direction" means, for example, the relative direction (or position) of each speaker with respect to the robot 10, and by setting the relative direction of each speaker. , It becomes possible to identify whether the speaker is a customer or an operator in the subsequent dialogue. It should be noted that the present embodiment is not limited to identifying the speaker based on the direction (position), but is performed by another method, for example, analyzing the characteristics of the speaker's voice. You may make a setting that can identify. Further, the instruction to hear the customer and the operator in step S15 is not limited to the one by voice, and the service options may be displayed on the touch display or the like so that the customer can touch and select.

Further, in step S15, by setting the operator information in advance, the hearing process with the operator can be skipped. Specifically, the operator registers the spoken language, facial features, voice features, and the like in the robot control device 20 in advance, so that the robot control device 20 can be transferred to the robot 10 by the operator's image or voice. It is possible to identify which of the two people (customer, operator) facing each other is the operator.

As described above, when the process of step S13 is completed, the robot control device 20 is ready to start the guidance service. Therefore, after that, the robot 10 can carry out the guidance service while interacting with the customer. The flow of processing in the guidance service will be described later with reference to FIG. 7. Further, when the process of step S15 is completed, the robot control device 20 is ready to start the translation service. Therefore, after that, the robot 10 can implement a translation service that supports the dialogue between the customer and the operator by mutual translation. .. The flow of processing in the translation service will be described later with reference to FIG.

(2-2) Guidance Service FIG. 7 is a flowchart showing an example of a processing procedure for direct dialogue processing. The direct dialogue process shown in FIG. 7 is a process in which the robot 10 executes a guidance service through a dialogue with a person (customer) when the service mode is set to the guidance service mode.

According to FIG. 7, first, when the robot 10 receives the customer's voice, the voice recognition unit 221 of the robot control device 20 performs text conversion on the voice data transmitted from the robot 10 to generate voice text. (Step S21).

Next, the dialogue control unit 224 removes fillers such as "um ..." and "uh ..." from the voice text generated in step S21, and then answers DB228 based on the voice text after removal. The question registered in is searched, and the answer associated with the corresponding question is acquired (step S22). The voice recognition unit 221 may remove the filler from the voice text. Further, since the voice text after removing the filler may contain variations (fluctuations) such as Japanese wording, the dialogue control unit 224 further performs morphological analysis on the voice text after removing the filler. Therefore, it is preferable to carry out the search in step S22 for the question in the answer DB 228 in which the morphological analysis has been similarly performed. In the above description, the dialogue control unit 224 performs filler removal and morphological analysis, but other functional units of the robot control device 20 (for example, voice recognition unit 221) may perform the filler removal and morphological analysis.

Next, in step S23, the dialogue control unit 224 determines whether or not a valid answer has been obtained by the search in step S22. If a valid answer can be obtained (YES in step S23), the process proceeds to step S24, and if a valid answer cannot be obtained (NO in step S23), the process proceeds to step S25.

In step S24, the dialogue control unit 224 voice-synthesizes the answer obtained in step S22, converts it into a voice file, transfers it to the robot 10, and reproduces it from the speaker 135. By performing such processing, the robot 10 can utter an appropriate answer to the customer's question.

On the other hand, in the case of step S25, since a valid answer has not been obtained in step S22, the dialogue control unit 224 has a predetermined hearing content (fixed) such as "Can you say it again?" Answer) is voice-synthesized, converted into a voice file, transferred to the robot 10, and reproduced from the speaker 135. By performing such processing, the robot 10 can utter a request for re-questioning to the customer. The content of the hearing output in step S25 is not limited to the above example, and the robot indicates that it cannot answer this question, for example, "I don't know" or "Please change the question". It may be the one that makes 10 speak.

Then, after the processing of step S24 or step S25, the process returns to step S21, and when the robot 10 receives the next voice by the customer, the processing after step S21 is repeated again to continue the dialogue between the customer and the robot 10. be able to.

As described above, in the direct dialogue process, the answer to the customer's question is determined based on the search of the Q & A registered in the answer DB228. Therefore, in order for the robot 10 to provide an effective customer service (guidance service) to the customer, Q & A corresponding to various questions (all questions) is registered in the answer DB 228 of the robot control device 20. It is required to be.

(2-3) Translation service FIG. 8 is a flowchart showing an example of a processing procedure for dialogue mediation processing. The dialogue mediation process shown in FIG. 8 is a process in which the robot 10 executes a translation service that mediates a dialogue between a person (customer) and a person (operator) by translation when the service mode is set to the translation service mode. is there.

According to FIG. 8, when the robot 10 detects the customer's voice, it assumes that there is a voice interrupt and transfers the voice data to the voice recognition unit 221 of the robot control device 20 (step S31). Then, in the robot control device 20, the service control unit 223 identifies the speaker (customer or operator) of the received voice data and its utterance language based on the customer identification setting set in step S15 of FIG. Specifically, the service control unit 223 can identify whether the speaker is a customer or an operator based on the direction in which the voice data is uttered (the direction of the speaker).

Next, in step S32, the voice recognition unit 221 (which may be the service control unit 223) determines the identification result of step S31. If it is determined that the speaker is a customer (YES in step S32), the process proceeds to step S33, and if it is determined that the speaker is not a customer, that is, an operator (NO in step S32), step S35. Proceed to.

When the process proceeds to step S33, the voice recognition unit 221 generates voice text by performing voice recognition (text conversion) on the received voice data of the customer in the spoken language of the customer. Further, the translation control unit 225 translates the generated voice text into the utterance language of the operator (if the utterance language of the customer and the operator match, the translation does not have to be performed).

Then, in the next step S34, for example, the service control unit 223 holds the voice text generated in step S33 and the translation result thereof in the data as a customer question (Q). When the process of step S34 is completed, the process proceeds to step S38 described later.

On the other hand, when the process proceeds to step S35, the voice recognition unit 221 generates voice text by performing voice recognition (text conversion) on the received voice data of the operator in the spoken language of the operator. Further, the translation control unit 225 translates the generated voice text into the utterance language of the speaker (if the utterance language of the customer and the operator match, the translation does not have to be performed).

Then, in the next step S36, for example, the service control unit 223 holds the voice text generated in step S35 and the translation result thereof in the data as the operator's answer (A).

Further, in the next step S37, for example, the service control unit 223 determines the voice text and translation result of the customer question (Q) held in the previous step S34 and the operator's answer (A) held in step S36. A Q & A is created by combining the voice text and the translation result, and registered in the dialogue log DB229 (dialogue log data). After registering in the dialogue log DB 229, the service control unit 223 deletes each of the held data and proceeds to step S38.

Explaining the creation of the Q & A in step S37 in more detail, the customer's question (Q) and the operator's answer (A) may have different utterance languages. Therefore, at this time, the customer's question before translation (that is, the question) (Q) voice text) and the translated operator's answer (that is, the translation result of answer (A)) are combined into one combination, and the translated customer's question (that is, the translation result of question (Q)) The operator's answer before translation (that is, the voice text of answer (A)) is used as one combination. In step S37, a Q & A in the same language is created as described above and registered in the dialogue log DB229.

FIG. 9 is a diagram for explaining an example of dialogue log data. In the case of FIG. 9, the dialogue log data stored in the dialogue log DB 229 is divided into tables according to the language. Specifically, FIG. 9A exemplifies the dialogue log data 420 in which the dialogue log is collected in Japanese, and FIG. 9B exemplifies the dialogue log data 430 in which the dialogue log is collected in English. Has been done.

The dialogue log data 420 includes a log date and time 421 in which the time stamp for collecting the log of the record is recorded, a question (Q) 422 in which the customer's question is recorded, and an answer (A) 423 in which the operator's answer is recorded. And are configured with. Since the data structure (log date and time 431, question (Q) 432, answer (A) 433) in the dialogue log data 430 is the same as that of the dialogue log data 420, the description thereof will be omitted.

The dialogue log DB229 stores the dialogue log data 420 and 430 for each language as described above, extracts a combination of questions and answers from the dialogue contents between the customer and the operator, and further translates them into the same language. It can be retained as a dialogue log.

Then, as described above, after the processing of step S34 or step S37, the processing of step S38 is performed. In step S38, the dialogue control unit 224 is the translation result translated immediately before (specifically, if it is via step S34, it is the translation result of the question (Q) by step S33, and if it is via step S37, it is the translation result. The text of the answer (A) translated in step S35) is voice-synthesized, converted into a voice file, transferred to the robot 10, and reproduced from the speaker 135. By performing such processing, the robot 10 can translate the utterance content of one speaker who has spoken immediately before into the utterance language of the other speaker and speak.

Then, after the processing of step S38, the process returns to step S31, and when the robot 10 detects the next voice interrupt, the processing after step S31 is repeated again to support the dialogue between the customer and the operator by mutual translation. Can be done.

Although detailed description is omitted, in the dialogue mediation process as well as in the case of the direct dialogue process, the dialogue control unit 224 (or the voice recognition unit 221 etc.) responds to the voice text (or its translation result). It is preferable to remove the filler and perform morphological analysis. By removing the filler and performing the morphological analysis, it is possible to register the Q & A excluding unnecessary utterance contents and fluctuations in the dialogue log DB229.

(3) Q & A correction mode The processing in the Q & A correction mode will be described in detail below. As described above, the "Q & A correction mode" is a mode that is set to the internal state of the robot control device 20 when the Q & A maintenance is performed manually (user such as an administrator) in the robot dialogue registration system 1. The user can access the GUI (Q & A correction screen) provided by the robot control device 20 from the operation terminal 30.

FIG. 10 is a diagram showing a specific example of the Q & A correction screen. Further, FIG. 11 is a flowchart showing an example of a manual operation procedure in Q & A maintenance.

The Q & A correction screen 500 shown in FIG. 10 includes a search condition input field 510, a search button 520, a search result display field 530, and a batch reflection button 540. The procedure shown in FIG. 11 will be described with reference to the Q & A correction screen 500.

According to FIG. 11, first, the user operates the operation terminal 30 to input the search condition for narrowing down the Q & A to be maintained in the search condition input field 510 of the Q & A correction screen 500, and the input of the search condition is completed. After that, by clicking the search button 520, the robot control device 20 is requested to output the Q & A correction candidate (step S41). More specifically, when inputting search conditions, the user specifies a range of log dates, a language, and the like.

When the output of the Q & A correction candidate is requested in step S41, the dialogue generation unit 227 and the screen generation unit 226 of the robot control device 20 execute the correction candidate creation process. The details will be described later with reference to FIG. 12 and the like, but in the correction candidate creation process, the dialogue generation unit 227 uses the dialogue log stored in the dialogue log DB 229 and the answer stored in the response DB 228 based on the response DB 228. The appropriateness (appropriate degree) of the answer stored in is determined, correction candidates for Q & A are generated according to the rules according to the judgment result, the priority of each correction candidate is calculated, and then output. Then, the screen generation unit 226 displays the information regarding the Q & A correction candidates output from the dialogue generation unit 227 in the search result display field 530 of the Q & A correction screen 500 in descending order of priority.

Here, as shown in FIG. 10, the search result display field 530 has display items of question 531, answer 532, answer 533, classification 534, reflection check box 535, and exclusion check box 536. Of these, question 531 and answer 532 are questions and answers automatically generated by shaping the dialogue log, and answer 533 is the answer of the dialogue log held in the dialogue log DB229. Therefore, by comparing the answer 532 and the answer 533, it is possible to confirm how the dialogue generation unit 227 corrected the answer in the correction candidate creation process. Further, in the classification 534, the classification result of the appropriateness of the corresponding correction candidate (see step S55 of FIG. 12 described later) is displayed.

After step S41, the user confirms the display contents (specifically, questions 531 to 534) of the search result display field 530 of the Q & A correction screen 500, operates the operation terminal 30 at the user's discretion, and reflects the display contents. Check the check box 535 or the exclusion check box 536 (step S42). Specifically, when the user determines that the automatically generated question 531 is a question to be answered by the robot 10 (it is an answer target), the user checks the reflection check box 535. On the other hand, when the user determines that the question 531 is a question that the robot 10 does not need to answer (it is not an answer target), the user checks the exclusion check box 536. If the user wants to withhold the judgment, he / she may not check any of the check boxes.

Next, the user confirms the automatically generated Q & A (question 531 and answer 532) for the record for which the reflection check box 535 is checked in step S42, and if the wording or the content of the Q & A is not appropriate, the search result. After editing the Q & A (question 531 and answer 532) by inputting text in the corresponding field of the display field 530, click the batch reflection button 540 (step S43).

When the batch reflection button 540 is selected (clicked) in step S43, the dialogue generation unit 227 of the robot control device 20 asks Q & A (question 531) of the record in which the reflection check box 535 is checked in the search result display field 530. Answer 532) is registered in the answer DB 228, and Q (question 531) of the record in which the exclusion check box 536 is checked in the search result display field 530 is registered in the exclusion candidate DB 230. When registering a Q & A in the answer DB 228 based on the selection of reflection, if the same question (Q) exists in the answer DB 228, update it with the Q & A (question 531 and answer 532) selected for reflection. Is preferable.

As described above, by executing the steps S41 to S43, the Q & A determined to be reflected by the user is newly registered in the answer DB 228, so that the robot 10 gives an appropriate answer to the customer in the guidance service. You will be able to do it.

Further, the Q (question) registered in the exclusion candidate DB 230 is not output as a Q & A correction candidate because it is not selected as a correction candidate in the subsequent correction candidate creation process, but the change in the environment of the robot 10 with the passage of time is taken into consideration. Then, after a certain period of time has passed, it is preferable that the correction candidate is selected again. In this case, the robot control device 20 (for example, the dialogue generation unit 227) rotates the data registered in the exclusion candidate DB 230 at a predetermined time (for example, a date), and after the data is registered in the exclusion candidate DB 230. When the predetermined period has passed, the Q (question) data may be deleted. Further, the process of deleting the Q (question) data from the exclusion candidate DB 230 may be instructed by the user.

(3-1) Correction candidate creation process Next, the correction candidate creation process will be described in detail.

FIG. 12 is a flowchart showing a processing procedure example of the correction candidate creation process. As described in step S41 of FIG. 11, in the manual Q & A maintenance (Q & A correction mode), when the search condition input field 510 of the Q & A correction screen 500 is input and the search button 520 is clicked, the robot control device 20 (dialogue generation unit 227, screen generation unit 226) executes the correction candidate creation process.

According to FIG. 12, first, when the dialogue generation unit 227 receives the output request of the Q & A correction candidate (see step S41 in FIG. 11) from the operation terminal 30, the dialogue log is entered in the search condition input field 510 with the search condition. A search for DB229 is performed (step S51).

The search of the dialogue log DB229 in step S51 is a process of sequentially extracting one question (Q), which will be described in detail below. First, the dialogue generation unit 227 refers to one line (record) in order from the top, for example, with respect to the dialogue log data stored in the dialogue log DB 229, and extracts the question (Q) recorded in the reference record (FIG. 9). Then, the processes of steps S52 to S56 are performed by using the extracted question (Q) as a question searched from the dialogue log DB229. After that, it is determined in step S57 whether the search for all the dialogue logs is completed. At this time, if the record for which the question (Q) has not been extracted remains in the dialogue log DB229, NO in step S57. Is determined, the next one record (for example, the record of the next line) is referred to, the question (Q) recorded in this reference record is extracted, and the processes of steps S52 to S56 are performed. By repeating the loop process in this way, the question (Q) is finally searched for all the records registered in the dialogue log DB229, and YES is determined in step S57.

After searching the question (Q) from the dialogue log DB 229 in step S51, the dialogue generation unit 227 searches the exclusion candidate DB 230 for the question (Q) searched from the dialogue log DB 229 (step S52).

Then, in the search in step S52, the dialogue generation unit 227 determines whether or not the question (Q) searched in step S51 matches the question registered in the exclusion candidate DB 230 (step S53). If it is determined in step S53 that they match (YES in step S53), the question registered in the exclusion candidate DB 230 is not the target of the Q & A correction candidate, so the process of steps S54 to S56 is skipped and the process proceeds to step S57. .. On the other hand, if it is determined in step S53 that they do not match (NO in step S53), the Q & A correction candidate may be the target, so the process proceeds to step S54.

In step S54, the dialogue generation unit 227 searches the Q & A registered in the answer DB 228 using the question (Q) searched from the dialogue log DB 229 as a key. Further, the dialogue generation unit 227 performs a classification process of classifying the corresponding Q & A into categories according to its appropriateness based on the search results in steps S51 and S54 (step S55).

Here, the classification process performed in step S55 will be described. In the following explanation, the question searched from the dialogue log DB229 is referred to as "Q1", the answer corresponding to Q1 is referred to as "A1", and the question searched from the answer DB228 using "Q1" as a key is referred to as "Q2". , The answer corresponding to Q2 is referred to as "A2". A1 is registered in the dialogue log DB229, and A2 is registered in the answer DB228.

When classifying the Q & A registered in the answer DB 228 in step S55, the dialogue generation unit 227 first performs morphological analysis on Q1, Q2, A1 and A2, and divides each sentence into clauses.

Next, the dialogue generation unit 227 calculates "F1" indicating the degree of agreement between Q1 and Q2 and "F2" indicating the degree of agreement between A1 and A2 based on the result of the morphological analysis. The degree of agreement F1 is calculated by, for example, the following equation 1. The larger the calculated F value, the higher the similarity between Q1 and Q2, and the smaller the F value, the lower the similarity between Q1 and Q2. The degree of coincidence F2 can be similarly calculated using Equation 1 by replacing Q1 and Q2 with A1 and A2.

Finally, the dialogue generation unit 227 classifies the Q & A based on the combination of F1 and F2 calculated using Equation 1. The criteria for classification will be described later with reference to FIG.

Returning to the description of FIG. After performing the classification process in step S55, the dialogue generation unit 227 automatically generates a Q & A based on the classification result (step S56). The procedure (rule) for automatically generating the Q & A will be described later with reference to FIG.

Next, in step S57, as described above in step S51, the dialogue generation unit 227 determines whether or not the search has been completed for all the dialogue logs, and if it is determined to be complete (YES in step S57), step. Proceed to S58. On the other hand, if it is determined that an unsearched dialogue log remains in the dialogue log DB 229 during the search process (NO in step S57), the dialogue generation unit 227 asks the next dialogue log (question (Q)). The search is performed, the process returns to step S52, and the process is repeated.

In step S58, the dialogue generation unit 227 calculates a score (priority score) indicating the priority of correction for each Q & A automatically generated in step S56, and interacts in descending order of the calculated priority score. The results (search results) obtained by executing the processes of steps S54 to S56 are sorted for each log. The procedure for calculating the priority score (priority score calculation process) will be described later with reference to FIG.

Next, the dialogue generation unit 227 outputs the search results sorted in step S58, and the screen generation unit 226 displays the output information in the search result display field 530 of the Q & A correction screen 500 (step S59). .. As a result, in the search result display column 530, search results (information regarding correction candidates for Q & A) are displayed in descending order of priority for correction.

(3-2) Classification Judgment Criteria and Q & A Automatic Generation Rule FIG. 13 is a diagram showing a judgment matrix in which classification judgment criteria and Q & A automatic generation rules are defined. As shown in FIG. 13, in the determination matrix 440, rules for Q & A classification and automatic generation are defined according to the combination of the F1 similarity and the F2 similarity. Here, the degree of similarity with F1 is a division of the degree of agreement F1 with reference to a predetermined threshold value, and indicates the degree of similarity as a question between Q1 and Q2. Similarly, the F2 similarity degree is a division of the degree of agreement F2 based on a predetermined threshold value, and indicates the degree of similarity as an answer between A1 and A2. In this example, three types of F1 similarity and F2 similarity are used in descending order of similarity: "large", "medium", and "small".

In this example, there are five categories of Q & A classification destinations: "Answer error", "Answer error possibility", "No answer", "Synonyms / insufficient wording", and "Answer possession". The Q & A automatic generation rule is also determined depending on which category it is classified into. Each category will be described in detail below.

As the first combination, when the F1 similarity is large and the F2 similarity is small (F1 similarity (large) and F2 similarity (small)), it is classified into the category of "answer error". This is a situation in which the operator's answer A1 and the answer A2 registered in the answer DB 228 are different from each other even though the customer's question Q1 and the assumed question Q2 registered in the answer DB 228 have a high degree of similarity (match). At this time, since there is a high possibility that the answer A2 registered in the answer DB 228 is incorrect, it is classified into the category of "answer error".

When the category is "answer error", there is a high possibility that the answer A1 by the operator is more correct than the answer A2 registered in the answer DB228. In addition, Q2, which is manually determined in advance and registered in the answer DB 228, is shaped as a sentence rather than Q1 of the customer question, which is the result of voice recognition. Therefore, Q & A correction candidates are automatically generated using Q2 and A1 (Q = Q2, A = A1).

As the second combination, when the F1 similarity is large and the F2 similarity is medium (F1 similarity (large) and F2 similarity (medium)), it is classified into the category of "possible answer error". In the case of the second combination, as in the case of the first combination, the answer A2 registered in the answer DB 228 may be incorrect. Therefore, when the category is "possible answer error", Q & A correction candidates are automatically generated using Q2 and A1 as in the case of the first combination "answer error" (Q = Q2, A = A1). ).

As a third combination, when the F1 similarity is small and the F2 similarity is also small (F1 similarity (small) and F2 similarity (small)), it is classified into the category of "no answer". This means that the assumed question Q2 corresponding to the question Q1 by the customer is not registered in the answer DB 228. Therefore, when the category is "no answer", the correction candidate of Q & A is automatically generated by using the question Q1 by the customer and the answer A1 by the operator (Q = Q1, A = A1).

As a fourth combination, when the F1 similarity is medium and the F2 similarity is large (F1 similarity (medium) and F2 similarity (large)), it is classified into the category of "synonyms / insufficient wording". This is because the answer A1 by the operator and the answer A2 registered in the answer DB228 have a high degree of similarity (match), while the question Q1 by the customer and the assumed question Q2 registered in the answer DB228 are slightly similar. The situation is that there is. In this case, since it is determined that the correspondence of synonyms and phrases is insufficient in the assumed question Q2 registered in the answer DB 228, it is classified into the category of "synonyms / insufficient phrases". Then, when the category is "synonyms / insufficient wording", the correction candidate of Q & A is automatically generated by using the question Q1 by the customer and the answer A2 registered in the answer DB228 (Q = Q1, A = A2). ..

As a fifth combination, when the F1 similarity is large and the F2 similarity is also large (F1 similarity (large) and F2 similarity (large)), the conversation between the customer and the operator (Q1, A1) has already been answered DB228. Since it means that it is registered in (Q2, A2), it is classified into the category of "answer possession". If the category is "Owned answer", it is not necessary to automatically generate correction candidates for Q & A.

It should be noted that the processing commonly performed when automatically generating correction candidates for Q & A is supplemented. When the question Q1 by the customer and the answer A1 by the operator are used, since Q1 and A1 are voice recognition results, the dialogue generation unit 227 (may be another functional unit of the robot control device 20) is "um ... , "Ah ..." and other fillers are removed to carry out the generation. In addition, since there is a possibility that the wording of A1 may vary depending on the person, the dialogue generation unit 227 morphologically analyzes the sentence, shapes it, and then outputs the sentence. For example, if an imperative form such as "go" is detected in the inflected form of a verb, it is converted to a honorific expression such as "please come".

As described above, as described in detail with reference to the determination matrix 440 of FIG. 13, the “Q & A correction candidates” automatically generated by the rules according to the categories classified by the dialogue generation unit 227 are registered in the answer DB 228. A modified version of the Q & A (specifically, automatically created with Q = Q1) and a new Q & A not registered in the answer DB228 (specifically, automatically created with Q = Q2). And are included. In the present embodiment, such "Q & A correction candidates" are displayed in the search result display field 530, and are registered in the response DB 228 after the user selects the reflection. Therefore, the "Q & A modification candidate" in the present embodiment is a candidate for modification to the Q & A registered in the answer DB 228 or a new registration in the answer DB 228.

Further, in the dialogue registration system 1 according to the present embodiment, as described above, when the dialogue between people is performed in different languages in the translation service mode, the utterances are mutually translated by the translation control unit 225. , The translation result is stored in the dialogue log DB229 in a unified language, and the Q & A unified in the same language is also registered in the answer DB 228. Therefore, the dialogue is based on the information held in the dialogue log DB 229 and the answer DB 228. The "Q & A correction candidates" automatically generated by the generation unit 227 are also Q & A unified in the same language. That is, since the Q & A correction candidates unified in the same language are also displayed in the search result display field 530 of the Q & A correction screen 500, an appropriate Q & A combination can be presented and the maintenance user reflects the correction candidates. It is possible to make it easier to determine whether or not to do so.

(3-3) Priority Score Calculation Process FIG. 14 is a flowchart showing a processing procedure example of the priority score calculation process. The priority score calculation process shown in FIG. 14 is a process (step S57) for calculating a priority score indicating the priority of correction for the correction candidates of the Q & A automatically generated in step S56 of FIG. 12, and is a dialogue. It is executed by the generation unit 227.

According to FIG. 14, in steps S61 to S69, the dialogue generation unit 227 values the parameter Af indicating the priority regarding the classification result according to the classification result of the Q & A registered in the answer DB 228 (step S55 in FIG. 12). To set. The set values α1 to α5 of Af are preset constants, but have a relationship according to the priority of modification for each classification. Specifically, when the priority of correction is "Answer error> No answer> Possibility of answer error> Synonyms / Insufficient wording> Answer possession", the relationship of α1> α2> α3> α4> α5 Has sex.

After the Af is set, the dialogue generation unit 227 refers to the answer DB 228, and among the Q & A data stored in the answer DB 228, the registration date and time of the Q & A corresponding to the Q & A to be calculated (at the edit time 413 in FIG. 4). (Equivalent) is acquired (step S70).

Next, the dialogue generation unit 227 calculates the value of the parameter Tf indicating the priority with respect to the elapsed time (step S71). Specifically, the dialogue generation unit 227 calculates the number of elapsed days from the difference between the registration date and time acquired in step S70 and the current date and time, and calculates the value of Tf by multiplying the number of elapsed days by a predetermined coefficient β1. Here, the reason for calculating Tf by "elapsed days x β1" is that the Q & A registered in the past (the longer the elapsed days), the more the facilities to be guided are changed, and the situation changes. This is because the correct answer is likely to have changed, and the correction should be prioritized.

Next, the dialogue generation unit 227 acquires the number of times the same question is asked by the customer in the designated search period as the parameter N indicating the multiplicity of questions (step S72). The reason for acquiring the parameter N as described above is that it can be expected that the guidance service by the robot 10 will be effective by preparing the answer as soon as the Q & A is asked many times. is there.

なお、式２によって算出された優先度スコアＭｆは、Ｍｆの値が大きいほど、回答ＤＢ２２８に現在登録されているＱ＆Ａを修正する必要性が高いことを意味する。 Finally, the dialogue generation unit 227 calculates the final priority score Mf using Af, Tf, and N calculated or acquired in steps S61 to S72 (step S73). Specifically, the priority score Mf is calculated by the following equation 2.

The priority score Mf calculated by Equation 2 means that the larger the value of Mf, the higher the need to correct the Q & A currently registered in the answer DB 228.

As described above, the robot dialogue registration system 1 according to the present embodiment automatically generates Q & A correction candidates to the answer DB 228 regarding the Q & A corresponding to the search condition specified by the user in the Q & A correction mode. Since it can be displayed in the search result display field 530 of the Q & A correction screen 500 in order from the candidate that needs to be corrected at an early stage, it is possible to support the effective implementation of Q & A maintenance by humans (users). ..

Further, in the robot dialogue registration system 1 according to the present embodiment, Q & A correction candidates are automatically generated in the Q & A correction mode and registered in the answer DB 228 by the user's selection, so that the domain is not expected in advance. Q & A can be expanded and the quality of dialogue service can be improved. In addition, it is possible to reduce the workload of the user who performs maintenance manually.

Further, in the robot dialogue registration system 1 according to the present embodiment, as described with reference to the determination matrix 440 of FIG. 13, the Q & A classification destination is the category of "synonymous word / insufficient wording" in the correction candidate creation process. In this case, by automatically generating Q & A correction candidates using the customer's question Q1 and the answer A2 registered in the answer DB228, it is possible to solve the problem of personality such as different wording depending on the person. it can. In addition, when using voice recognition results such as question Q1 by the customer and answer A1 by the operator in the generation of correction candidates for Q & A, by performing shaping using filler removal and morphological analysis, further personality is obtained from the generated Q & A. Can be excluded. As a result, the effect of reducing the burden on the user in manual maintenance can be expected.

Further, in the robot dialogue registration system 1 according to the present embodiment, in the Q & A correction mode, the user checks the exclusion check box 536 among the Q & A correction candidates displayed in the search result display field 530 of the Q & A correction screen 500. The Q & A that has been performed can be excluded from the subsequent search targets. As a result, in the subsequent manual maintenance, correction candidates for Q & A that do not need to be corrected are not displayed, and the efficiency of maintenance by the user can be improved.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations. Further, it is possible to add / delete / replace a part of the configuration of the embodiment with another configuration. For example, in the above embodiment, the robot dialogue registration system 1 has been described as being composed of the robot 10, the robot control device 20, and the operation terminal 30, but the configuration of the robot dialogue registration system according to the present invention is The robot control device 20 is not limited to this, and the robot control device 20 is provided with a function of displaying to the user or a function of receiving input from the user in place of the operation terminal 30, and the robot 10 is equipped with the robot control device 20 and the operation terminal 30. The main configurations of the robot interactive registration system, such as being equipped with functions, may be integrated into one or two devices.

Further, the scenes and situations to which the present invention can be applied are not limited to the above-described embodiments. For example, in the above embodiment, the dialogue registration system 1 has been described in the situation of a translation service in which the robot 10 supports the dialogue between a person (customer) and a person (operator) by translation. It is not limited to this, and can be applied to other scenes such as growing the Q & A of the chatbot to be used together based on the voice recognition result of the guidance service by the customer and the operator at the call center or the like. In the case of applying to the above call center situation, since the call source is the customer and the call recipient is the operator, the process of identifying the customer shown in step S15 of FIG. 6 does not need to be executed. ..

Further, each of the above configurations, functions, processing units, processing means and the like may be realized by hardware by designing a part or all of them by, for example, an integrated circuit. Further, each of the above configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function can be placed in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

Further, in the drawings, the control lines and information lines are shown as necessary for explanation, and not all the control lines and information lines are necessarily shown in the product. In practice it may be considered that almost all configurations are interconnected.

1 Dialogue registration system 10 Robot 20 Robot control device 30 Operation terminal 110, 210, 310 CPU
120, 220, 320 Storage device 121 Drive control unit 122 Dialogue control unit 123 Input / output unit 130, 330 Input / output device 131 Camera 132 Microphone 133 Gyro sensor 134 Range sensor 135 Speaker 136 Drive mechanism 140, 240, 340 Communication interface 221 Voice Recognition unit 222 Image recognition unit 223 Service control unit 224 Dialogue control unit 225 Translation control unit 226 Screen generation unit 227 Dialogue generation unit 228 Answer DB
229 Dialogue log DB
230 Exclusion candidate DB
321 Browser 322 Input / output section 331 Display 332 Keyboard 333 Mouse 410 Q & A data 420,430 Dialogue log data 440 Judgment matrix 500 Q & A correction screen 510 Search condition input field 520 Search button 530 Search result display field 540 Batch reflection button

Claims (7)

A registration system that registers answers to customer questions by robots that provide dialogue services.
A voice recognition unit that recognizes human-to-human dialogue,
A dialogue log database in which the voice recognition result by the voice recognition unit is registered as a dialogue log, and
An answer database in which a combination of the question and the answer to the question is registered in advance as an example of dialogue,
The appropriateness of the dialogue example is determined based on the dialogue log registered in the dialogue log database and the dialogue example registered in the response database, and correction candidates of the dialogue example are generated according to the rules according to the determination result. Dialog generator and
A screen generation unit for generating the dialogue example correction screen on which the correction candidates generated by the dialogue generation unit are posted is provided.
The dialogue generation unit
The dialogue example A robot dialogue registration system characterized in that the correction candidates selected on the correction screen are registered in the answer database. The dialogue generation unit
A category according to the appropriateness of the dialogue example based on the degree of agreement between the question in the dialogue log and the question in the dialogue example and the degree of agreement between the answer in the dialogue log and the answer in the dialogue example. Classified into
For each of the generated modification candidates, the category to which the corresponding dialogue example is classified, the elapsed time since the dialogue example was registered in the answer database, and the question in the dialogue example are asked by the customer. Calculate the modification priority based on the number of times
The robot dialogue registration system according to claim 1, wherein each of the correction candidates is posted on the dialogue example correction screen in an order according to the calculated priority. The voice recognition unit
It is characterized in that questions and answers are identified from voice data of human-to-human dialogue, converted into texts, and the converted texts are shaped by removing fillers or performing morphological analysis. The robot dialogue registration system according to claim 1. The dialogue generation unit
When generating the correction candidate, when the answer of the correction candidate is generated from the answer of the dialogue log according to the rule, the answer of the dialogue log obtained by performing a predetermined shaping process is regarded as the answer of the correction candidate. ,
The screen generator
The robot dialogue registration system according to claim 1, wherein the Q & A correction screen on which the correction candidates are posted is generated in a mode including display of answers before and after the shaping process. The dialogue generation unit
Based on the degree of agreement between the question in the dialogue log and the question in the dialogue example, and the degree of agreement between the answer in the dialogue log and the answer in the dialogue example, the dialogue example was adapted to the appropriateness. Categorize and
The screen generator
The robot dialogue registration system according to claim 1, wherein the category of the classification destination of each of the modification candidates is associated with the modification candidate and posted on the dialogue example modification screen. When the dialogue is performed in a different language, the voice recognition result of the dialogue is translated by the voice recognition unit to unify the language, and the translation result is stored in the dialogue log database as the dialogue log. With a translation control unit
The dialogue generation unit
The correction candidate is generated based on the dialogue log and the dialogue example unified in the same language.
The screen generator
The robot dialogue registration system according to claim 1, wherein the correction candidate of the unified language is posted on the dialogue example correction screen. The user can edit the correction candidate displayed on the dialogue example correction screen on the dialogue example correction screen.
The dialogue generation unit
The robot dialogue registration system according to claim 1, wherein when the edited modification candidate is selected on the dialogue example modification screen, the edited modification candidate is registered in the answer database.

Images