JP7124565B2 - Dialogue method, dialogue program and information processing device - Google Patents

Description

The present invention relates to an interactive method, an interactive program, and an information processing apparatus.

In recent years, dialogue systems in which a user and a system have a dialogue using text data are becoming popular. As one of these dialogue systems, there is known one that performs slot-filling dialogue.

In a dialog system that performs a slot-filling dialog, the user is sequentially presented with question sentences for filling in the values of each item prepared in advance, and when the user inputs an answer, this answer is used as the item value Get and perform a specific operation. In some interactive systems, the content of the user's own answer is presented to the user so that the user can confirm whether or not the answer has been conveyed to the interactive system.

JP 2016-71050 A

In the above-described dialogue system, a mechanical method of repeatedly presenting an input answer is used to make the user confirm whether or not the answer has been conveyed to the dialogue system. Therefore, in the dialogue system described above, as the number of times the user answers increases, the user feels uncomfortable or unnatural, which may hinder the continuation of a natural dialogue.

In one aspect, the present invention aims at continuing a natural dialogue with the user.

In one aspect, in a computer-based interaction method, the computer acquires utterance information of a user, and when the acquired utterance information includes an answer to be registered in a storage unit, Then, a plurality of related information relating to the answer is acquired, the related information indicating the attribute of the answer is extracted from the plurality of related information, and the extracted related information is output.

Each of the above procedures may be a functional unit that implements each procedure, a process that implements each procedure, or a computer-readable storage medium that stores a program that causes a computer to execute each process.

Natural dialogue with the user can be continued.

It is a figure explaining the outline|summary of a dialogue system. It is a figure explaining an example of the system configuration|structure of a dialogue system. It is a figure which shows an example of the hardware constitutions of a dialog apparatus. FIG. 4 is a diagram showing an example of a slot table held in a table storage unit; FIG. FIG. 4 is a diagram for explaining functions of an interactive processing unit; It is a figure which shows an example of a related information table. FIG. 10 is a flowchart for explaining processing of a named entity extraction learning unit; FIG. FIG. 10 is a first flow chart for explaining processing of a dialogue processing unit; FIG. FIG. 10 is a second flowchart for explaining processing of the dialogue processing unit; FIG. FIG. 11 is a third flowchart for explaining processing of the dialogue processing unit; FIG.

The present embodiment will be described below with reference to the drawings. FIG. 1 is a diagram for explaining an outline of a dialogue system.

A dialogue system 100 of this embodiment includes a dialogue device 200 and a terminal device 300, and the dialogue device 200 and the terminal device 300 are connected via a network.

In the dialogue system 100 of the present embodiment, the dialogue device 200 is a dialogue device that performs slot-filling dialogue, which is one of algorithms for realizing dialogue with a specific purpose.

Specifically, the interactive device 200 outputs to the terminal device 300 a question sentence for inputting values of items prepared in advance (step S1). When an answer to the question text is input to the terminal device 300 and information indicating the answer is received from the terminal device 300, the interactive device 200 acquires the information indicating the answer as the value of the item included in the question text. , a process corresponding to a specific purpose is executed (step S2).

Further, upon receiving information indicating an answer, the interactive device 200 of the present embodiment acquires related information related to the answer. Then, the dialogue device 200 selects related information indicating attributes of the answer from the acquired related information, and displays it on the terminal device 300 as a response to the answer (step S3). In the following description, a response to an answer may be expressed as a backtrack to an answer, backtracking to an answer, or the like.

In the example of FIG. 1, the dialogue device 200 realizes dialogue with the purpose of "presenting a travel route according to the destination (destination) and the number of people".

The interactive device 200 of the present embodiment generates a question sentence Q1 for prompting the user P of the terminal device 300 to enter an answer to "destination", which is one of the items prepared in advance. Display on the screen 301 .

Note that the screen 301 is an example of a screen displayed on the terminal device 300 when an operation is performed on the terminal device 300 to instruct the start of dialogue with the dialogue device 200 .

When an answer sentence (speech sentence) A1 is entered in the input field 302 on the screen 301, the terminal device 300 of the present embodiment displays the answer sentence A1 in association with the question sentence Q1, and displays information indicating the answer sentence A1. to the interactive device 200 .

In the following description, information indicating a question sentence may be referred to as question information, and information indicating an answer sentence (utterance sentence) may be referred to as utterance information. Further, in the following description, a character string that is included in the utterance information and that is the value of an item included in the question may be called an answer or a character string that becomes an answer.

When the dialogue device 200 of the present embodiment receives speech information indicating the answer sentence A1 from the terminal device 300, it extracts a character string indicating the value of the item "destination" from the speech information as an answer. In the example of FIG. 1, the character string (answer) that is the value of the item "destination" is "Chicago", so the character string "Chicago" is extracted as the answer.

After extracting an answer from the utterance information, the dialogue device 200 of the present embodiment searches external information sources to obtain information related to the answer, and selects relevant information indicating attributes of the answer from the obtained answer information. do.

Then, the interactive device 200 causes the terminal device 300 to display the backtracking sentence including the selected related information.

Further, after displaying the backtracking sentence, the interactive device 200 causes the terminal device 300 to display a question sentence for inputting the value of the next item.

In the example of FIG. 1, the item included in the question Q1 is "destination", and the attribute of this item is "place". Therefore, the attribute of the answer that is the value of the item "destination" is also "location".

Therefore, dialogue device 200 selects related information R1 in which "Chicago" indicates "place" from among the related information acquired based on the answer "Chicago", and generates backtrack H1 including related information R1. to display it on the screen 301 of the terminal device 300 .

In the example of FIG. 1, the text data "Chicago is famous for being the birthplace of Mr. Bob XX" is displayed as the backtrack H1, and the related information R1 included therein is "Chicago is , famous for the birthplace of Bob Davidson” indicates a place called “Chicago”. Also, in the backtrack H1, it can be seen that the related information R1 includes the character string "Chicago" as an answer.

Also, the backtracking sentence H1 is displayed as a response to the reply sentence A1, and it can be seen that the backtracking sentence H1 is associated with the reply sentence A1.

In this embodiment, after displaying the backtracking sentence H1 on the screen 301, the dialogue device 200 causes the screen 301 of the terminal device 300 to display the question sentence Q2 for inputting the value of the item "companion".

In this way, the dialogue device 200 of the present embodiment selects related information indicating attributes of an answer from related information related to an answer, and uses the selected related information as a back-and-forth to confirm that the answer has been received. It is displayed on the terminal device 300 as a sentence.

In other words, the dialogue device 200 displays a question sentence for the user, and when receiving an input of an utterance sentence including an answer to the question sentence, a response sentence (backhand sentence) including the answer and related information related to the answer is displayed. It is generated and displayed in correspondence with the utterance sentence. In the following description, information indicating a response sentence (backhand sentence) may be expressed as response information.

Therefore, according to the present embodiment, the backtracking sentence output from the interactive device 200 differs according to the attribute of the answer, and a wide variety of backtracking sentences can be output in response to the user's answer.

Further, according to the present embodiment, since the reply and related information selected based on the attributes of the reply are included in the backtracking sentence, the degree of understanding of the contents of the reply by the interactive device 200 can be presented to the user. can be done. Therefore, in this embodiment, it is possible to allow the user to confirm whether the interactive device 200 correctly understands the answer.

For these reasons, according to the present embodiment, it is possible to contribute to the continuation of natural dialogue with the user.

The dialogue system 100 of this embodiment will be further described below. FIG. 2 is a diagram illustrating an example of the system configuration of the dialogue system.

The dialogue device 200 of the dialogue system 100 of this embodiment has a dialogue processing section 210 , a named entity extraction learning section 220 , a model storage section 230 and a table storage section 240 .

In addition to the terminal device 300, the dialogue device 200 of the present embodiment is also connected to, for example, an information source 400 and a named entity tagged corpus 500 via a network.

The information source 400 is, for example, a group of databases constructed on a network and stores various information. The information source 400 is referenced by the interactive processor 210 .

The named entity-tagged corpus 500 is a set of documents in which a structure indicating the range of proper nouns such as names of people, places, organizations, and products is given in advance as metadata.

In the example of FIG. 2, the information source 400 and the named entity tagged corpus 500 are provided outside the dialogue system 100, but the present invention is not limited to this. Information source 400 and named entity tagged corpus 500 may be included in dialogue system 100 . Also, the information source 400 and the named entity tagged corpus 500 may each be provided in the dialogue device 200 .

The interaction processing unit 210 of this embodiment executes interaction processing for realizing interaction with the user of the terminal device 300 . Specifically, the dialogue processing unit 210 refers to the slot table 241 held by the table storage unit 240 to generate question information.

Further, the dialogue processing unit 210 of this embodiment, based on the result obtained by inputting the related information related to the answer to the named entity extraction model 231 held in the model storage unit 230, , selects relevant information that indicates the attributes of the answer and generates response information. Details of the interactive processing unit 210 will be described later.

The named entity extraction learning unit 220 generates a named entity extraction model 231 using the named entity tagged corpus 500 and stores it in the model storage unit 230 as pre-processing before executing the dialogue processing by the dialogue processing unit 210 . The named entity extraction model 231 is used when the dialogue processing unit 210 generates response information.

The named entity extraction learning unit 220 of this embodiment generates the named entity extraction model 231 by learning only the context information included in the named entity tagged corpus 500 .

When related information is input, the named entity extraction model 231 of this embodiment outputs a score indicating the probability that the related information indicates the attribute of the answer based on the context of the related information. The details of the processing of the named entity extraction learning unit 220 will be described later.

The model storage unit 230 is a storage area for holding named entity extraction models 231 . The table storage unit 240 is a storage area for holding the slot table 241 referenced by the interactive processing unit 210 .

Next, the hardware configuration of the interactive device 200 of this embodiment will be described with reference to FIG. FIG. 3 is a diagram showing an example of the hardware configuration of the interactive device.

The interactive device 200 of this embodiment includes an input device 21, an output device 22, a drive device 23, an auxiliary storage device 24, a memory device 25, an arithmetic processing device 26, and an interface device 27, which are connected to each other via a bus B. It is an information processing device.

The input device 21 is a device for inputting various kinds of information, and is realized by, for example, a keyboard or pointing device. The output device 22 is for outputting various kinds of information, and is realized by, for example, a display. The interface device 27 includes a LAN card and the like, and is used for connecting to a network.

The dialogue program that implements the dialogue processing unit 210 and the named entity extraction learning unit 220 of this embodiment is at least part of various programs that control the dialogue device 200 . The interactive program is provided, for example, by distribution on the storage medium 28, download from a network, or the like. The storage medium 28 in which the dialog program is recorded is a storage medium for optically, electrically or magnetically recording information such as a CD-ROM, a flexible disk, a magneto-optical disk, etc., or a storage medium for storing information such as a ROM, a flash memory, etc. Various types of storage media can be used, such as semiconductor memories that record electrically.

Further, the dialogue program is installed in the auxiliary storage device 24 from the storage medium 28 via the drive device 23 when the storage medium 28 recording the dialogue program is set in the drive device 23 . A dialog program downloaded from the network is installed in the auxiliary storage device 24 via the interface device 27 .

The auxiliary storage device 24 stores the installed dialogue program, as well as necessary files and data such as the named entity extraction model 231 and the slot table 241 described above. The memory device 25 reads and stores the dialogue program from the auxiliary storage device 24 when the dialogue device 200 is activated. The arithmetic processing unit 26 implements various processes described later in accordance with the dialogue program stored in the memory unit 25 .

Next, the slot table 241 of this embodiment will be described with reference to FIG. FIG. 4 is a diagram showing an example of a slot table held in a table storage unit.

The slot table 241 of this embodiment is prepared in advance for each interactive system 100 and held in the table storage unit 240 .

The slot table 241 of this embodiment has slot names, types, and slot values as information items.

The value of the item “slot name” indicates the name of the item included in the question information transmitted from the interactive device 200 to the terminal device 300 . In other words, the value of the item “slot name” is an item included in the question text displayed on the terminal device 300, and is an item for the user of the terminal device 300 to input a value.

The value of the item "type" indicates the attribute of the corresponding slot name. In other words, the value of the item "type" indicates the attribute of the item included in the question (attribute of the answer included in the utterance).

The value of the item "slot value" indicates the value of the item included in the question information extracted from the utterance information. In other words, the value of the item "slot value" is the answer included in the utterance sentence to the question sentence.

In the example of FIG. 4, the slot name "destination" is associated with the type "place". Therefore, if the item included in the question text is "destination", the attribute of the slot value "Chicago" that is the answer to the question text is "location", and the slot value "Chicago" is treated as a character string indicating the location. will be

Further, in the example of FIG. 4, the slot name "period" is associated with the type "date". Therefore, when the item included in the question is "period", the slot value that is the answer to the question is treated as a character string indicating the date.

Next, with reference to FIG. 5, functions of the interactive processing unit 210 of this embodiment will be described. FIG. 5 is a diagram for explaining functions of the interactive processing unit.

The dialogue processing unit 210 of this embodiment has a question generation unit 211 , an utterance reception unit 212 , an answer extraction unit 213 , an answer registration unit 214 , a backtrack generation unit 215 and an output unit 216 .

When receiving an instruction to start a dialogue from the terminal device 300, the question text generation unit 211 initializes the value of the item "slot value" in the slot table 241, and includes the value of the item "slot name" as an item to be answered. Generate a question sentence. In other words, the question sentence generation unit 211 generates question information including the value of the item “slot name” of the slot table 241 .

The speech accepting unit 212 accepts input of speech information received from the terminal device 300 . The answer extractor 213 extracts an answer to the question from the speech information received by the speech receiver 212 . In other words, the answer extraction unit 213 extracts the slot value from the utterance sentence indicated by the utterance information. Specifically, the answer extraction unit 213 analyzes the utterance information and extracts words of the type (attribute) associated with the slot value from the words included in the utterance information as slot values.

The answer registration unit 214 stores the slot values extracted by the answer extraction unit 213 in the slot table 241 .

The backtracking sentence generation unit 215 generates response information indicating a backtracking sentence for the answer from the speech information received by the speech reception unit 212 and the information source 400 . Details of the backtracking sentence generation unit 215 will be described later.

The output unit 216 outputs the question information generated by the question generation unit 211 and the response information generated by the backtrack generation unit 215 to the terminal device 300 .

The backtracking sentence generation unit 215 will be described below. The backtrack generation unit 215 of this embodiment has a related information acquisition unit 251 , a score acquisition unit 252 , and a backtrack selection unit 253 .

The related information acquisition unit 251 acquires related information related to the answer included in the utterance information. In addition, the related information of this embodiment includes various information related to the answer.

Specifically, the related information acquisition unit 251 searches the information source 400 based on the slot value (answer) extracted by the answer extraction unit 213 . Then, the related information acquisition unit 251 holds the information extracted as the search result in the related information table 254 as related information related to the answer. In other words, the related information table 254 of the present embodiment is created each time response information (return sentence) is generated.

In this embodiment, the related information stored in the related information table 254 is a backtrack candidate that can be part of a backtrack sentence. Also, the related information stored in the related information table 254 includes the same character string as the slot value (answer). In the following description, the same character string as the slot value included in the related information may be expressed as the same character string in the related information. Details of the related information table 254 will be described later.

The score acquisition unit 252 acquires a score associated with the related information based on the result of applying the related information stored in the related information table 254 to the named entity extraction model 231 and the type of the slot value.

The score acquired by the score acquisition unit 252 is a value that indicates the probability that the same character string in the related information is treated as a character string of the same type as the slot value. The score acquired for each piece of related information by the score acquisition unit 252 is held in association with the related information in the related information table 254 .

The backtracking sentence selection unit 253 selects the related information with the highest score acquired by the score acquisition unit 252 as the related information to be included in the backtracking sentence, and generates response information indicating the backtracking sentence.

The related information table 254 of this embodiment will be described below with reference to FIG. FIG. 6 is a diagram showing an example of a related information table.

In the related information table 254 of this embodiment, related information and scores are associated with each other. In the following description, in the related information table 254, information in which related information is associated with a score will be referred to as scored related information.

In the example of FIG. 6, the slot name (item) included in the question sentence is "destination", the slot value (answer) included in the utterance sentence is "Chicago", and the type (attribute) associated with the slot name is ) is “location”, an example of the related information table 254 is shown.

In the related information table 254, in the scored related information 61, the score "0.07" is associated with the related information "the best song in Chicago".

That is, in the related information included in the scored related information 61, the probability that the same character string "Chicago" in this related information is treated as the same attribute "location" as the slot value "Chicago" is 7%. indicates that there is

Similarly, in the scored related information 62, the score "0.75" is associated with the related information, and in the scored related information 63, the score "0.91" is associated with the related information. attached.

Therefore, among the related information stored in the related information table 254, the related information ``Chicago is famous for being the birthplace of Mr. It can be seen that the probability that it is treated as a character string indicating is the highest.

Next, the processing of the named entity extraction learning unit 220 of this embodiment will be described with reference to FIG. FIG. 7 is a flowchart for explaining the processing of the named entity extraction learning unit. The processing in FIG. 7 is performed as pre-processing for the processing by the interactive processing unit 210 .

The named entity extraction learning unit 220 of the present embodiment learns context information from the named entity tagged corpus 500, generates a named entity extraction model 231 (step S701), and stores the named entity extraction model 231 in the model storage unit 230. is stored (step S702), and the process ends.

Here, generation of the named entity extraction model 231 will be described.

The named entity extraction learning unit 220 of the present embodiment generates named entity extraction models by machine learning using context information of document data included in the named entity tagged corpus 500, which is a collection of documents to which tags have been added in advance, as learning data. 231 is generated.

The context information is information indicating in what context each word is used, and is information indicating a character string within an arbitrary range before and after the word, a co-occurrence relationship between words, and the like.

The context information will be specifically described below. For example, as learning data included in the named entity tagged corpus 500, document data "I'm going to <place name>Kawasaki</place name> today" and "I borrowed a book from <person's name>Kawasaki</person's name>". Suppose there exists document data of

The contextual information of this learning data is "words that appear before 'niyuku' are likely to be place names" and "words that appear before 'you' are likely to be person names."

The named entity extraction learning unit 220 of the present embodiment generates the named entity extraction model 231 by performing learning using only such context information as feature amounts.

In the present embodiment, by generating the named entity extraction model 231 in this way, when the related information is applied to the named entity extraction model 231, the target word (answer) is, for example, a place name. The probability of being a person's name can be output as a score.

Next, the processing of the interactive processing unit 210 of this embodiment will be described with reference to FIGS. 8 to 10. FIG. FIG. 8 is a first flowchart for explaining the processing of the dialogue processing section.

FIG. 8 shows the initial operation of the interaction processing unit 210 when interaction with the user of the terminal device 300 is started.

When receiving a dialogue start instruction from the terminal device 300, the dialogue processing unit 210 of the present embodiment initializes the slot values of the slot table 241 by the question sentence generation unit 211 (step S801). Specifically, the question sentence generating unit 211 empties the value of the item “slot value” in the slot table 241 .

Subsequently, the dialogue processing unit 210, the question text generating unit 211 refers to the slot table 241, and generates a question text including the slot value corresponding to the empty slot value (step S802).

Subsequently, the dialogue processing unit 210 outputs the question sentence by the output unit 216, displays it on the terminal device 300 (step S803), and ends the initial operation.

FIG. 9 is a second flowchart for explaining the processing of the dialogue processing section. FIG. 9 shows the process of responding to the user's answer.

In the dialogue processing unit 210 of the present embodiment, when input of speech information is received by the speech reception unit 212 (step S901), the answer extraction unit 213 extracts slot values (answers) from the speech information (step S902). .

Subsequently, the dialogue processing unit 210 determines whether or not the slot value is extracted from the utterance information by the answer registration unit 214 (step S903). In other words, the answer registration unit 214 determines whether or not the utterance information includes an answer to be registered in the slot table 241 of the table storage unit 240 .

If the slot value is not extracted in step S903, the process proceeds to step S906, which will be described later.

Note that the case where the slot value has not been extracted is, for example, the case where the user has not entered an answer to the question. Specifically, for example, when the question is "Who is your companion?"

If slot values have been extracted in step S903, the interactive processing unit 210 causes the answer registration unit 214 to register the extracted slot values in the slot table 241 (step S904).

Subsequently, the dialogue processing unit 210 generates a backtrack by the backtracking sentence generation unit 215 (step S905). Details of step S905 will be described later.

Subsequently, the dialogue processing unit 210 causes the question text generation unit 211 to generate a question text including the slot name corresponding to the slot value with an empty value (step S906).

Subsequently, the dialogue processing unit 210 outputs the generated backtracking sentence and question sentence to the terminal device 300 by the output unit 216 (step S907), and ends the process.

Incidentally, in the example of FIG. 9, the backtracking sentence for the answer and the next question sentence are respectively generated, and then both the backtracking sentence and the next question sentence are output, but the present invention is not limited to this.

The back-and-forth sentence and the next question sentence may be output separately. Specifically, for example, after the backtracking sentence is generated, only the backtracking sentence is output to the terminal device 300, and after the backtracking sentence is output, the next question sentence is generated and output to the terminal device 300. good.

In this case, on the terminal device 300, as shown in the screen 301 (see FIG. 1), the backtracking sentence H1 and the next question sentence Q2 do not have to be displayed in one balloon. Sentence Q2 may be displayed in separate balloons.

Also, in the present embodiment, the backtracking sentence and the question sentence are output in succession, but the present invention is not limited to this. In the present embodiment, for example, after outputting the backtracking sentence, after receiving a response from the user of the terminal device 300 indicating that the content of the response has been confirmed, the next question sentence may be generated and output.

FIG. 10 is a third flowchart for explaining the processing of the dialogue processing section. FIG. 10 shows the details of the processing of the backtracking sentence generation unit 215 in step S915.

The backtracking sentence generation unit 215 of the present embodiment acquires the related information of the answer by the related information acquisition unit 251 (step S1001). Specifically, the backtracking sentence generation unit 215 searches the information source 400 using the slot value registered in the slot table 241 as a search key, acquires the search result as a group of related information, and saves each of them in the related information table 254 . Store related information.

Subsequently, the backtracking sentence generation unit 215 applies the named entity extraction model 231 to each related information stored in the related information table 254 by the score acquisition unit 252, and acquires a score for each related information (step S1002).

Acquisition of a score for each piece of related information by the score acquisition unit 252 will be described below. Below, the case of acquiring the score of each related information stored in the related information table 254 shown in FIG. 6 will be described.

In the related information table 254 of FIG. 6, related information acquired using the extracted slot value "Chicago" as a search key is "the best songs in Chicago", "Chicago is synonymous with Bob XX", "Chicago is Bob・Famous for Mr. ○○'s hometown" is stored.

When the related information "Chicago's best song" is input to the named entity extraction model 231, the named entity extraction model 231 obtains a score indicating the probability that the same character string "Chicago" is treated as a person's name, and the same character string " and a score indicating the probability that "Chicago" is treated as a place.

In other words, the named entity extraction model 231 obtains a score indicating the probability that the attribute of the same character string "Chicago" is "person's name" and the attribute of the same character string "Chicago" is " It outputs a score that indicates the probability of being "location".

Here, the score indicating the probability that the same character string "Chicago" has the attribute "person's name" is 0.82, and the score indicating the probability that the same character string "Chicago" has the attribute "location" is 0.07. Suppose it was

In addition, the named entity extraction model 231 has a score of 0.32 indicating the probability that the attribute of the same character string "Chicago" is "person's name" when the related information "Chicago, Bob Mr. ○○" is input. , a score of 0.75 indicating the probability that the attribute of the same character string "Chicago" is "location" is output. Also, the named entity extraction model 231 outputs a score of 0.48 indicating the probability that the attribute of the character string "Bob XX" is "place".

Also, the named entity extraction model 231, when the related information "Chicago, Bob XX" and "Chicago is famous for being the birthplace of Bob XX" are input, the same character string "Chicago ' has the attribute "person's name", and the score 0.91 indicates the probability that the same character string "Chicago" has the attribute "location".

In this way, the named entity extraction model 231 of this embodiment outputs a score indicating the probability that a word included in related information is treated as a word with a certain attribute.

When the score for each attribute of the same character string is output for each piece of related information, the score acquisition unit 252 of the present embodiment calculates the score corresponding to the attribute “location” of the slot value “Chicago” for each piece of related information. Acquire as a score associated with related information.

Specifically, the score acquisition unit 252 sets the score “0.07” indicating the probability that the same character string “Chicago” has the attribute “place” as the score associated with the related information “Chicago's best song”. Acquired and stored in the related information table 254 in association with the related information "the best songs in Chicago".

In addition, the score acquisition unit 252 obtains a score “0. 75”.

It should be noted that the character string "Bob XX" has the attribute "place", but is excluded because the character string "Bob XX" is not the same character string.

Similarly, the score acquisition unit 252 indicates the probability that the attribute of the same character string "Chicago" is "location" as the score associated with the related information "Chicago is famous for being the hometown of Mr. Bob XX". Get a score of 0.91.

Following the processing of the score acquisition unit 252 in step S1002, the backpaste generation unit 215 selects the related information with the highest score acquired in step S1002 by the backpaste selection unit 253 (step S1003). That is, the backtracking sentence selection unit 253 of the present embodiment is an example of an extraction unit that extracts related information indicating attributes of answers from a plurality of pieces of related information acquired by the related information acquisition unit 251 .

Here, the related information "Chicago is famous for being the hometown of Mr. Bob XX" associated with the score of 0.91 is selected.

Subsequently, the backtracking sentence generating unit 215 generates a backtracking sentence including the selected related information (step S1004), and ends the process.

In this embodiment, with the above processing, it is possible to present related information of the category implied by the user's answer as a back-and-forth sentence. Therefore, according to this embodiment, it is possible to indicate to the user that the system understands the meaning of the user's utterance, and to reduce the unnaturalness of the dialogue.

The technology disclosed herein can take forms such as those described below.
(Appendix 1)
A computer-based method of interaction, wherein the computer comprises:
Get the user's utterance information,
when the acquired utterance information includes an answer to be registered in a storage unit, acquiring a plurality of related information related to the answer based on the utterance information;
extracting relevant information indicating an attribute of the answer from the plurality of relevant information;
An interactive method that outputs relevant extracted information.
(Appendix 2)
generating response information indicating a back-and-forth sentence containing the relevant information;
The interaction method according to appendix 1, wherein the response information is output in association with the utterance information.
(Appendix 3)
Generating a named entity extraction model using contextual information contained in a set of documents as learning data,
obtaining a score indicating a probability that the related information is related information indicating an attribute of the answer for each of the plurality of related information, using the plurality of related information as inputs for the named entity extraction model;
3. The interactive method according to appendix 1 or 2, wherein the relevant information with the highest score is extracted as the relevant information indicating the attribute of the answer.
(Appendix 4)
4. The dialogue method according to any one of Appendices 1 to 3, wherein the related information includes a character string similar to the character string indicating the answer.
(Appendix 5)
A computer-based method of interaction, wherein the computer comprises:
Display a question sentence for the user on the terminal device,
A dialogue method comprising: receiving an input of an utterance sentence containing an answer to the question sentence, and causing the terminal device to display an utterance sentence containing the answer and related information related to the answer in association with the utterance sentence.
(Appendix 6)
Get the user's utterance information,
when the acquired utterance information includes an answer to be registered in a storage unit, acquiring a plurality of related information related to the answer based on the utterance information;
extracting relevant information indicating an attribute of the answer from the plurality of relevant information;
An interactive program that causes a computer to execute a process that outputs relevant information that has been extracted.
(Appendix 7)
an utterance reception unit that acquires utterance information of a user;
a related information acquiring unit for acquiring a plurality of related information related to an answer based on the utterance information when the acquired utterance information includes an answer to be registered in a storage unit;
an extraction unit for extracting relevant information indicating attributes of the answer from among the plurality of relevant information;
and an output unit that outputs the extracted related information.

The invention is not limited to the specifically disclosed embodiments, and various modifications and changes are possible without departing from the scope of the claims.

REFERENCE SIGNS LIST 100 dialogue system 200 dialogue device 210 dialogue processing unit 211 question generation unit 212 utterance reception unit 213 answer extraction unit 214 answer registration unit 215 backtrack generation unit 216 output unit 220 named entity extraction learning unit 230 model storage unit 231 named entity extraction model 240 Table storage unit 241 Slot table 251 Related information acquisition unit 252 Score acquisition unit 253 Return sentence selection unit 300 Terminal device 400 Information source
Corpus with 500 named entity tags

Claims (6)

A computer-based method of interaction, wherein the computer comprises:
Get the user's utterance information,
when the acquired utterance information includes an answer to be registered in a storage unit, acquiring a plurality of related information related to the answer based on the utterance information;
extracting relevant information indicating an attribute of the answer from the plurality of relevant information;
An interactive method that outputs relevant extracted information. generating response information indicating a back-and-forth sentence containing the relevant information;
2. The interaction method according to claim 1, wherein said response information is output in association with said utterance information. Generating a named entity extraction model using contextual information contained in a set of documents as learning data,
obtaining a score indicating a probability that the related information is related information indicating an attribute of the answer for each of the plurality of related information, using the plurality of related information as inputs for the named entity extraction model;
3. The dialogue method according to claim 1, wherein the relevant information with the highest score is extracted as the relevant information indicating the attribute of the answer. A computer-based method of interaction, wherein the computer comprises:
Display a question sentence for the user on the terminal device,
Receiving an input of an utterance sentence including an answer to the question sentence, and backtracking including the answer and related information indicating an attribute of the answer among related information indicating a plurality of types of attributes acquired based on the answer. A dialog method, wherein a sentence is associated with the uttered sentence and displayed on the terminal device. Get the user's utterance information,
when the acquired utterance information includes an answer to be registered in a storage unit, acquiring a plurality of related information related to the answer based on the utterance information;
extracting relevant information indicating an attribute of the answer from the plurality of relevant information;
An interactive program that causes a computer to execute a process that outputs relevant information that has been extracted. an utterance reception unit that acquires utterance information of a user;
a related information acquiring unit for acquiring a plurality of related information related to an answer based on the utterance information when the acquired utterance information includes an answer to be registered in a storage unit;
an extraction unit for extracting relevant information indicating attributes of the answer from among the plurality of relevant information;
and an output unit that outputs the extracted related information.

Images