JP7319504B2 - Dialogue system, dialogue method, program, and storage medium - Google Patents

Description

TECHNICAL FIELD Embodiments of the present invention relate to a dialogue system, a dialogue method, a program, and a storage medium.

There is a dialogue system that outputs answers to inquiries from users through dialogue with users. There is a demand for the development of a technology that can more easily construct this dialogue system.

JP 2012-248161 A

Embodiments of the present invention provide a dialog system, dialog method, program, and storage medium that can be constructed more easily.

A dialog system according to an embodiment includes a processing unit. When a user inputs a first sentence including one or more words, the processing unit sets at least part of the one or more words as a first query keyword. The processing unit includes the first query keyword from a question and answer collection including a plurality of combinations of a question and answer pair including a question and an answer to the question and a plurality of question keywords linked to the question and answer pair. Select one or more of the question-answer pairs. When the number of the selected question-and-answer pairs is one, the processing unit outputs a first response indicating the answers included in the selected question-and-answer pair. When the number is equal to or greater than a first threshold, the processing unit extracts a first question keyword different from the first query keyword from the plurality of question keywords associated with one of the selected question-answer pairs. and extracting a second question keyword different from the first query keyword from the plurality of other question keywords associated with another one of the selected question-answer pairs, and extracting the first question keyword and the second Output a second response containing the query keyword. The processing unit uses a plurality of scenario templates each containing blanks when generating the first response and the second response. The first response is generated by filling the blank of one of the plurality of scenario templates with the answer included in the selected question-answer pair. The second response is generated by embedding the first question keyword and the second question keyword in another one of the plurality of scenario templates.

1 is a schematic diagram illustrating an interactive system according to a first embodiment; FIG. It is a table|surface which illustrates a collection of questions and answers. 4 is a table illustrating scenario templates; 4 is a table illustrating scenarios for tasks; 6 is a flowchart illustrating processing of the dialogue system according to the embodiment; FIG. 10 is a schematic diagram illustrating an interactive system according to a modification of the first embodiment; 9 is a flowchart illustrating processing of the dialogue system according to the modification of the first embodiment; 9 is a flowchart illustrating processing of the dialogue system according to the modification of the first embodiment; FIG. 10 is a schematic diagram illustrating an interactive system according to a second embodiment; FIG. 9 is a flowchart illustrating processing of the dialogue system according to the second embodiment; FIG. 10 is a schematic diagram showing the operation of the dialogue system according to the second embodiment; FIG. 11 is a schematic diagram illustrating an interactive system according to a modified example of the second embodiment; FIG. 10 is a schematic diagram for explaining the operation of the dialogue system according to the second embodiment; 1 is a schematic diagram illustrating the configuration of an interactive device according to an embodiment; FIG.

Each embodiment of the present invention will be described below with reference to the drawings.
In the specification and drawings of the present application, elements similar to those already described are denoted by the same reference numerals, and detailed description thereof will be omitted as appropriate.

(First embodiment)
FIG. 1 is a schematic diagram illustrating an interactive system according to the first embodiment.
As shown in FIG. 1, the dialogue system 1 according to the first embodiment includes a processing section 10. As shown in FIG. In the example shown in FIG. 1 , the dialogue system 1 further includes a first storage section 21 and a second storage section 22 .

The processing unit 10 receives a sentence input by the user. The processing unit 10 outputs a response to the input text. Based on the response from the processing unit 10, the user inputs the following text to the processing unit 10. The processing unit 10 outputs a response to the next sentence based on the interaction with the user so far. Thus, dialogue is realized between the user and the dialogue system 1 .

The first storage unit 21 stores a collection of questions and answers. The question-and-answer collection includes a plurality of question-and-answer pairs. Each question-answer pair includes a question and an answer to that question. The second storage unit 22 stores general-purpose scenarios for generating responses output from the processing unit 10 . The processing unit 10 generates a response to the input text while referring to the first storage unit 21 and the second storage unit 22 .

For example, a sentence input by a user includes an inquiry. Here, a sentence input from the user to the processing unit 10 is called a "query sentence". When the processing unit 10 receives the query text, the processing unit 10 refers to the question-and-answer collection in the first storage unit 21 . The processing unit 10 checks whether there is a question related to the query text in the question and answer collection. When the processing unit 10 finds a related question, it outputs a response indicating the answer to the question to the user.

When there are a plurality of related questions, the processing unit 10 outputs a response (answer) for narrowing down the questions to one. The user inputs another query sentence containing information (keywords) for narrowing down the question in response to the response. For example, the input of a query sentence by the user and the response from the processing unit 10 are alternately repeated until the question is narrowed down to one. By answering the responses of the dialogue system, the user can arrive at one answer containing the information the user wants. By using the dialogue system 1, the user can obtain one answer even if the user does not know an appropriate search method.

A scenario is needed to output a response based on a query sentence from a dialog system to a user. There is a method using a concept dictionary to automatically generate scenarios. The concept dictionary includes, for example, attributes, attribute values, and paraphrasing expressions. In a dialogue system using a concept dictionary, when a query sentence is input, the processing unit automatically generates a scenario based on the concept dictionary and the number of answer candidates. Automatic generation of scenarios eliminates the need for humans (eg, system administrators) to create scenarios.
On the other hand, in order to use the concept dictionary, the system administrator needs to assign one or more attributes and attribute values for each attribute to each question included in the question and answer collection. When using a concept dictionary, the time required to create a scenario can be reduced, but it takes time to create the concept dictionary. Also, it depends on the system administrator what to set as an attribute and what to set as an attribute value. You also need to decide how to determine if an attribute value was mentioned in a user query. Therefore, the system administrator is required to have specialized knowledge, which places a heavy burden on the system administrator.
For this reason, it is desired to develop a technology that can more easily construct a dialogue system that allows dialogue with a user.

In the dialogue system 1 according to the embodiment, when a user inputs a query sentence including one or more words, the processing unit 10 sets at least part of the one or more words as a keyword (query keyword). . The processing unit 10 selects a question-and-answer pair including a query keyword from the question-and-answer collection. By using query keywords for selection, advance preparation such as assignment of attributes and attribute values to question and answer collections becomes unnecessary.

Furthermore, when the number of selected question-answer pairs is one, the processing unit 10 outputs the first response indicating the answer of the one question-answer pair. For example, the first response includes the questions and answers included in the selected question-answer pair. The first response may contain the URL of the web page containing the response. The user can confirm the answer to the inquiry by referring to the first response.
The processing unit 10 outputs the second response when the number of selected question-answer pairs is two or more. The second response is a question keyword extracted from a plurality of keywords (question keywords) included in one of the selected question-answer pairs and another plurality of keywords included in another one of the selected question-answer pairs. and another question keyword extracted from the question keyword. These question keywords contained in the second response are different from the query keywords. By referring to the second response, the user can know what kind of keyword should be included in the next query text in order to narrow down the question-and-answer pairs.
Also, a scenario corresponding to the number of question-and-answer pairs selected as described above can be automatically generated using a general-purpose dictionary. Therefore, it is possible to reduce the burden required for scenario creation on the system administrator.

For example, the user inputs to the processing unit 10 another query text containing any of the question keywords included in the second response. The processing unit 10 sets at least part of one or more words included in the other query sentence as another query keyword. The processing unit 10 further selects a part of the selected two or more question-answer pairs using the other query keyword. As a result, when the number of selected question-and-answer pairs is one, the processing unit 10 outputs the first response. When the number of selected question-and-answer pairs is two or more, the processing unit 10 outputs the second response again.

According to the embodiment, for example, dialogue with a user about a question and answer collection can be realized without using a concept dictionary. Further, according to the embodiment, when two or more question-answer pairs are selected, the responses including question keywords are displayed. This allows the user to know what kind of keyword a query sentence should be input to the dialogue system 1 . As a result, it becomes easier for the user to reach the desired answer. The convenience of the dialogue system 1 can be improved.

The processing unit 10 includes a central processing unit including electrical circuits (processing circuits). The 1st memory|storage part 21 and the 2nd memory|storage part 22 each contain a memory|storage device. The storage device includes, for example, a hard disk drive (HDD), a network attached storage (NAS), an embedded multimedia card (eMMC), a solid state drive (SSD), and/or a solid state hybrid drive (SSHD). One storage device may function as both the first storage section 21 and the second storage section 22 . The processing unit 10, the first storage unit 21, and the second storage unit 22 may be included in one computer, or may be connected to each other by wire, wirelessly, or via a network.

The dialogue system 1 according to the embodiment will be specifically described.
FIG. 2 is a table illustrating a collection of questions and answers.
FIG. 3 is a table illustrating scenario templates.
FIG. 4 is a table illustrating scenarios for tasks.

The first storage unit 21 stores, for example, a collection of questions and answers 110 shown in FIG. Question-and-answer collection 110 includes a plurality of question-and-answer pairs 110a-110K. The question-answer pairs 110a-110K include IDs 111a-111K that identify the question-answer pairs, questions 112a-112K, and answers 113a-113K to the questions, respectively.

Question keywords 114a-114K are associated with the question-answer pairs 110a-110K, respectively. The question keywords 114a to 114K are included in the question and answer collection 110, for example. Alternatively, the question keywords 114a to 114K may be stored in the first storage unit 21 or other storage units separately from the question and answer collection 110. FIG.

Question keywords 114a-114K are extracted from questions 112a-112K, respectively. A plurality of question keywords may be extracted from one question. For example, a processing device external to the dialogue system 1 extracts question keywords 114a-114K from the questions 112a-112K. Alternatively, question keywords 114a-114K may be assigned to questions 112a-112K by a system administrator.

The question-and-answer collection may include information other than multiple question-and-answer pairs and multiple question keywords. For example, the question and answer collection may include questions for which corresponding answers are not set. A question-and-answer collection may include a combination of one or more questions and one or more answers, where one question and one answer are not paired. For example, the question-and-answer collection includes a combination of two questions and one answer set as a common answer to the two questions. In these cases, the processing unit 10 executes dialogue with the user, for example, based only on a plurality of question-answer pairs. The processing unit 10 does not use information other than the plurality of question-and-answer pairs for dialogue with the user.
In the question and answer collection, one answer may include multiple conditions and detailed answers to each condition. In this case, multiple conditions and multiple detailed answers can be collectively regarded as one answer.

The second storage unit 22 stores, for example, the general-purpose model 120 shown in FIG. Generic model 120 includes a plurality of scenario templates 121a-121c. Scenario templates 121a-121c are used as the basis for task scenarios. The task scenario contains specific information and is used as a response output from the processing unit 10 . Numbers of answer candidates 122a-122c are set in scenario templates 121a-121c, respectively. A scenario template to be used is determined according to the number of answer candidates.

Scenario templates 121a-121c are described generically so as not to depend on user queries. Specifically, each of the scenario templates 121a-121c includes <blank>. As will be described later, questions, answers, or question keywords are applied to <blank> when task scenarios are generated.

The general-purpose model 120 may include a standard scenario that does not include blanks, like the scenario template 121d. Standard scenarios correspond to greetings, thanks, apologies, and the like.

The processing unit 10 includes, for example, a reception unit 11, a setting unit 12, a selection unit 13, and a generation unit 14.
The accepting unit 11 accepts a query sentence input by a user. A query sentence is input by, for example, a user's operation of a keyboard, touch panel, or the like. Alternatively, the query sentence may be generated based on the user's utterance. For example, a microphone records the user's speech and recognizes the voice. A query sentence is generated by speech recognition.

The setting unit 12 sets a query keyword based on the query text. For example, the setting unit 12 sets the entire query sentence including one or more words as one query keyword. The selection unit 13 selects one or more question-and-answer pairs including a query keyword from the question-and-answer collection. The generation unit 14 generates a task scenario according to the number of selected question-and-answer pairs. The generation unit 14 outputs the generated task scenario to an external device such as a monitor, speaker, or printer. These devices output task scenarios in a way that the user can perceive them.

As a specific example, let Ns be 3 in the general-purpose model 120 shown in FIG. Ns is set to switch responses from the dialogue system 1 according to the number of question-answer pairs selected when a plurality of question-answer pairs are selected. For example, the user inputs "balance confirmation" to the dialogue system 1. FIG. The reception unit 11 receives the query text. The setting unit 12 uses the query text as a query keyword. That is, "balance confirmation" is the query keyword. The selection unit 13 refers to the question-and-answer collection in the first storage unit 21 and selects a question-and-answer pair including the query keyword. The answers included in the selected question-answer pair are candidates for answers to the user's inquiry.

The generator 14 counts the number of selected question-answer pairs. The number of question-answer pairs selected corresponds to the number of answer candidates. Of the questions 112a, 112b, 112c, and 112K of the question and answer collection shown in FIG. 2, only the question 112a contains the question keyword of "balance check". Therefore, the number of question-answer pairs selected is one. The generation unit 14 refers to the scenario template 121a corresponding to the answer candidate number of one in the general-purpose model 120 . The generating unit 14 substitutes <blank 1> with the question 112a “I want to check the balance of my deposit account”. The generation unit 14 substitutes the answer 113a, "to check the balance...", into <blank 2>. As a result, as shown in FIG. 4, a task scenario 131 of "Your question is 'I want to check the balance of my bank account.' 'How to check the balance...'" is generated.

As another example, the user inputs “application” to the dialogue system 1 . The setting unit 12 uses “application” as a query keyword. In the question and answer collection shown in FIG. 2, two questions, question 112b and question 112K, include "application". The number of question-answer pairs selected is two. The generation unit 14 refers to the scenario template 121b corresponding to the number of answer candidates of two in the general-purpose model 120 . The generating unit 14 substitutes <blank 3a> with the question 112b “I want to apply for a savings account”. The generation unit 14 substitutes the question 112K “I want to apply for a loan” in <blank 3b>. As a result, a task scenario 132 of "Do you want to apply for a bank account?" or "Apply for a loan?" is generated.

As yet another example, the user inputs "deposit account" to the dialog system 1. In the question and answer collection shown in FIG. 2, three questions 112a, 112b, and 112c include "deposit account." The number of question-answer pairs selected is three. The generation unit 14 refers to the scenario template 121c corresponding to the number of answer candidates of 3 in the general-purpose model 120 . The generation unit 14 also extracts question keywords different from the query keywords from the question keywords 114a to 114c corresponding to the questions 112a to 112c, respectively. The generation unit 14 substitutes the question keyword “balance check” of the question 112a into <blank 4a>. The generation unit 14 substitutes the question keyword “application” of the question 112b into <blank 4b>. The generating unit 14 substitutes the question keyword “cancellation” of the question 112c into <blank 4c>. As a result, a task scenario 133 of "Do you have any additional keywords? About 'balance check', 'application', 'cancellation', etc." is generated.

When the number of selected question-answer pairs is 0, the generation unit 14 refers to the scenario template 121d corresponding to the number of answer candidates of 0 in the general-purpose model 120 . The scenario template 121d is a standard scenario that does not contain blanks. The generation unit 14 directly uses the scenario template 121d as a task scenario.

In this way, the processing unit 10 automatically generates a task scenario based on the query text input by the user, the question and answer collection 110, and the general-purpose model 120. FIG. The generator 14 outputs the generated task scenario as a response to the user's inquiry.

FIG. 5 is a flowchart illustrating processing of the dialogue system according to the embodiment.
The reception unit 11 receives a query text input by a user (step S101). The setting unit 12 sets at least part of the query text as a query keyword (first query keyword) (step S102). The selection unit 13 selects a question-answer pair including the first query keyword (step S103). The generator 14 counts the number N of the selected question-answer pairs (step S104).

The generator 14 determines whether N is 0 (step S105). When N is 0, the generator 14 outputs a fixed response (step S106). A canned response corresponds to a canned scenario. When N is not 0, the generator 14 determines whether N is 1 (step S107). When N is 1, the generator 14 generates a first response (step S108). The first response indicates the answer included in the selected question-answer pair. The first response is, for example, the task scenario 131 shown in FIG.

When N is not 1, the generator 14 determines whether N is greater than or equal to Ns (step S109). When N is equal to or greater than Ns, the generator 14 selects a question keyword (first question keyword) different from the first query keyword for one of the selected question-answer pairs. The generating unit 14 selects another question keyword (second question keyword) different from the first query keyword for another one of the selected question-answer pairs. The first question keyword and the second question keyword are different from each other. The generator 14 generates a second response including the first question keyword and the second question keyword (step S110). The second response is, for example, the task scenario 133 shown in FIG.

When N is less than Ns, the generator 14 generates a third response including N questions included in N question-answer pairs (step S111). For example, the generation unit 14 generates a third response including a first question included in one of the N question-answer pairs and a second question included in another one of the N question-answer pairs. do. The third response is, for example, the task scenario 132 shown in FIG. After generating any of the first to third responses, the generator 14 outputs the generated responses (step S112).

After outputting the response, the processing unit 10 determines whether the dialogue has ended (step S113). For example, the processing unit 10 measures the elapsed time after outputting the response. If the elapsed time exceeds the threshold without the user inputting another query text, the processing unit 10 determines that the dialogue has ended. A sentence indicating that the dialogue has ended may be input from the user to the dialogue system 1 . When the processing unit 10 receives the text, it determines that the dialogue has ended. If the reception unit 11 receives the next query text before these termination conditions are satisfied, the processing unit 10 executes the above-described processing again.

When the generation unit 14 outputs the response to the monitor, the generation unit 14 may output the question, the keyword, etc. on the graphical user interface (GUI) displayed on the monitor so as to be selectable. For example, when the generation unit 14 generates the second response, it is output so that the first question keyword and the second question keyword can be selected on the GUI. When the user clicks the first question keyword or the second question keyword on the GUI, the clicked keyword is input to the processing unit 10 as the next query sentence. For example, when the generator 14 generates the third response, it is output so that the first question and the second question can be selected on the GUI. When the user clicks the first question or the second question on the GUI, the clicked question is input to the processing unit 10 as the next query text.

According to the examples shown in FIGS. 2 to 5, when multiple question-answer pairs are selected, the second response or the third response is generated according to the number of question-answer pairs selected. The second response includes multiple question keywords different from the query keyword. Therefore, the user can know what keywords should be included in the next query text. The third response includes multiple questions included in the selected question-answer pair. Therefore, the user can directly confirm the questions included in the selected question-answer pair. These responses make it easier for users to get to the answers they seek.

(Modification)
FIG. 6 is a schematic diagram illustrating an interactive system according to a modification of the first embodiment;
In the interactive system 1a according to the modification, a task model is generated in advance before the user makes an inquiry. The task model includes question keywords included in the question-and-answer collection, the number of question-and-answer pairs including each question keyword, and a plurality of task scenarios.

A task model is generated by the processing unit 30 . For example, the dialogue system 1a includes the processing unit 30. FIG. The processing unit 30 stores the generated task model in the third storage unit 23 . The processing unit 30 includes a central processing unit including electrical circuitry. A single device may function as both processing unit 10 and processing unit 30 . The 3rd memory|storage part 23 contains a memory|storage device. Storage devices include, for example, at least one of HDD, NAS, eMMC, SSD, and SSHD. A single storage device may function as the first storage section 21 to the third storage section 23 .

In the dialogue system 1 a , the processing section 30 includes an extracting section 31 , an extracting section 32 and a generating section 33 . The processing unit 10 includes a reception unit 11 , a setting unit 12 and a selection unit 13 .

FIG. 7 is a flowchart illustrating processing of the dialogue system according to the modification of the first embodiment;
The flow chart of FIG. 7 illustrates the process of creating a model for a task.

The retrieval unit 31 refers to the question-and-answer collection in the first storage unit 21 and retrieves one question-and-answer pair from a plurality of question-and-answer pairs (step S121). The extraction unit 32 extracts the question keyword registered for the one question-answer pair (step S122). The generation unit 33 determines whether the extracted question keyword is registered in the task model (step S123).

When not registered, the generation unit 33 refers to the general-purpose model in the second storage unit 22 . The generation unit 33 searches for a scenario template corresponding to one answer candidate. The generator 33 substitutes the questions and answers of the selected question-answer pair into a plurality of blanks of the scenario template to generate a task scenario (step S124). The generation unit 33 associates the extracted question keyword, the number of question-and-answer pairs including the question keyword, and the generated task scenario, and adds them to the task model in the third storage unit 23 (step S125). .

In step S123, when the extracted question keyword is registered in the task model, the generator 33 updates the task model (step S126). Specifically, the generation unit 33 adds 1 to the number of question-answer pairs including the extracted question keyword in the task model. The generator 33 compares the number of question-and-answer pairs after adding 1 with the threshold value Ns.

When the number of question-answer pairs is less than Ns, the generator 33 searches for scenario templates corresponding to the number of answer candidates equal to or greater than Ns and less than Ns. The generation unit 33 substitutes a plurality of questions of a plurality of question-answer pairs including the extracted question keyword into a plurality of blanks of the scenario template to generate a task scenario.

When the number of question-answer pairs is Ns or more, the generator 33 extracts another question keyword from each question included in the question-answer pair containing the extracted question keyword. The generation unit 33 searches for scenario templates corresponding to the number of answer candidates equal to or greater than Ns. The generation unit 33 substitutes the plurality of different extracted question keywords into the plurality of blanks of the scenario template to generate a task scenario. The generation unit 33 registers the generated task scenario and the number of updated question-answer pairs in the task model.

After step S125 or S126, the processing unit 30 determines whether or not all question-answer pairs have been extracted from the question-and-answer collection (step S127). If there is a question-and-answer pair that has not been extracted, the processing unit 30 executes step S121 again. In step S121, the extractor 31 extracts another one of the plurality of question-answer pairs. Thereafter, another question keyword is extracted from another one of the plurality of question-answer pairs. The processing of steps S121 to S126 is repeated until all question-and-answer pairs included in the question-and-answer collection are extracted.

FIG. 8 is a flowchart illustrating processing of the dialogue system according to the modification of the first embodiment;
The flowchart of FIG. 8 illustrates the process of responding to query sentences.

The reception unit 11 receives a query sentence input by the user (step S141). The setting unit 12 sets the query text as a query keyword (step S142). The selection unit 13 selects a response based on the query keyword (step S143). Specifically, the selection unit 13 accesses the third storage unit 23 . The selection unit 13 determines whether the task model includes a query keyword.

When a query keyword is included in the task model, the selection unit 13 selects a task scenario associated with the query keyword from the task model. When the query keyword is not included in the task model, the selection unit 13 selects, as the task scenario, a scenario template (standard scenario) corresponding to the number of answer candidates of 0 in the general-purpose model of the second storage unit 22. . A scenario template corresponding to 0 answer candidates may be registered in the task model in advance as a task scenario. The selection unit 13 outputs the selected response (step S144). The processing unit 10 determines whether the dialogue has ended (step S145).

According to the interactive system 1a according to the modification, a task model including a plurality of task scenarios is generated before the processing unit 10 receives a query sentence. Therefore, the processing unit 10 does not need to generate a task scenario based on the query text after receiving the query text. For example, the processing load on the processing unit 10 after receiving the query text can be reduced.

(Second embodiment)
FIG. 9 is a schematic diagram illustrating an interactive system according to the second embodiment.
In the dialogue system 2 according to the second embodiment, the processing unit 10 refers to the fourth storage unit 24 in addition to the first storage unit 21 and the second storage unit 22 when generating a response. The dialogue system 2 may also include a fourth storage unit 24 . The fourth storage unit 24 stores models or rules for extracting keywords from sentences.

The dialogue system 2 further includes a processing section 40 . The processing unit 40 includes, for example, an extraction unit 41 and an extraction unit 42 . The retrieval unit 41 accesses the first storage unit 21 and retrieves one question-answer pair from the question-and-answer collection. The extraction unit 42 extracts question keywords from the question-answer pair according to the model or rules stored in the fourth storage unit 24 . The extraction unit 42 stores the extracted question keyword in the first storage unit 21, for example. The processing unit 40 includes a central processing unit including electrical circuitry. A single device may function as both processing unit 10 and processing unit 40 . The fourth storage unit 24 includes a storage device. Storage devices include, for example, at least one of HDD, NAS, eMMC, SSD, and SSHD. A single storage device may function as the first storage section 21 , the second storage section 22 and the fourth storage section 24 .

When the receiving unit 11 receives a query sentence including one or more words, the setting unit 12 extracts a part of one or more words according to the model or rule stored in the fourth storage unit 24. The setting unit 12 sets the extracted words as query keywords. That is, the setting unit 12 extracts query keywords from the query text. Multiple query keywords may be extracted from the query sentence.

For example, keyword extraction is based on the part of speech of each word that makes up a sentence. In this case, the fourth storage unit 24 stores morphological analysis models or morphological analysis rules. The setting unit 12 divides the sentence into a plurality of words by morphologically analyzing the sentence, and estimates the part of speech of each word. The setting unit 12 sets a word of a specific part of speech as a keyword. For example, nouns and verbs are used as specific parts of speech. For a compound noun, which is one of nouns, the setting unit 12 may register each noun in the compound noun as a keyword, register the entire compound noun as one keyword, or combine them. You can register with The setting unit 12 may further extract words (unknown words) that cannot be classified into any part of speech as keywords.

FIG. 10 is a flowchart illustrating processing of the dialogue system according to the second embodiment.
FIG. 10 shows processing when the processing unit 40 extracts question keywords from the question-and-answer collection. The processing shown in FIG. 10 is executed, for example, when a user inputs a query sentence.

The retrieval unit 41 retrieves one of a plurality of question-answer pairs from the question-and-answer collection (step S201). The extraction unit 42 extracts a question keyword from the question of the extracted question-answer pair (step S202). The extraction unit 42 registers the extracted question keyword in the first storage unit 21 (step 203). The retrieval unit 41 determines whether all question-and-answer pairs included in the question-and-answer collection have been retrieved (step S204). If there is a question-and-answer pair that has not been extracted, the extractor 41 executes step S201 again. In step S201, the retrieving unit 41 retrieves another one of the plurality of question-answer pairs from the question-and-answer collection. Thereafter, another question keyword is extracted from another one of the plurality of question-answer pairs. The processing of steps S201 to S203 is repeated until question keywords are extracted from all question-answer pairs included in the question-and-answer collection.

The processing unit 10 uses the question keyword extracted from the question-and-answer collection to execute the same process as in the flowchart shown in FIG. For example, in step S102, the setting unit 12 extracts a part of the query sentence input by the user and sets it as one or more query keywords. In step S103, the selection unit 13 selects a question-answer pair including at least one of the extracted one or more query keywords.

According to the second embodiment, a part of the query text is extracted and set as the query keyword. According to this method, a question-answer pair including a question corresponding to a user's inquiry is more likely to be selected than when the entire query text is set as one query keyword. This makes it easier for the user to reach the desired answer. The convenience of the dialogue system 2 can be improved.

Further, according to the second embodiment, when the user inputs a query sentence, a question keyword is extracted from the question-and-answer collection. Therefore, even when the question-and-answer collection is frequently updated, question-and-answer pairs are selected based on the question keywords included in the latest question-and-answer collection. For example, responses can be output using question keywords included in the latest collection of questions and answers. This makes it easier for the user to reach the desired answer.

In step S103, the selection unit 13 may select a question-answer pair containing a question keyword similar to the query keyword in addition to a question-answer pair containing the same question keyword as the query keyword. The selection unit 13 calculates the degree of similarity between each of the one or more query keywords and each of the multiple question keywords extracted from the question-and-answer collection. When the degree of similarity between one query keyword and one question keyword is equal to or greater than a preset threshold, the one query keyword is determined to be a paraphrased expression of the one question keyword. The selection unit 13 determines that a question-and-answer pair including a question keyword similar to the query keyword substantially includes the query keyword, and selects the question-and-answer pair.

For example, the reading of a certain question keyword is the same as the reading of a certain query keyword. As an example, at least some of the query keywords are written in one of hiragana and katakana, and at least some of the query keywords are written in the other of hiragana and katakana. In this case, the question keyword is a paraphrase of the query keyword. As another example, the question keyword and the query keyword contain the same kanji characters but different okurigana characters. Again, the question keyword is a paraphrase of the query keyword.
Even when the question keyword and the query keyword contain the same kanji characters and the okurigana characters are different due to the difference in the part of speech, the question keyword is determined to be a paraphrased expression of the query keyword. For example, when the question keyword is "transfer" and the query keyword is "transfer", the question keyword is determined to be a paraphrased expression of the query keyword.

An edit distance, for example, is used to determine the degree of similarity. The selection unit 13 uses the edit distance between keywords as the degree of similarity.
The degree of similarity may be determined using a pre-created paraphrasing dictionary. The paraphrasing dictionary includes a set of multiple keywords. Each set contains multiple keywords. A plurality of keywords included in each set are paraphrase expressions that are similar to each other. For example, the selection unit 13 sets the degree of similarity between keywords included in one set to 1, and sets the degree of similarity between other keywords to 0.
Similarity may be determined using a pre-trained distributed representation model. The distributed representation model describes the distributed representation of words. The selection unit 13 determines the degree of similarity based on the cosine similarity or distance between the vectors of each keyword.

By selecting a question-answer pair containing a question keyword similar to the query keyword, it becomes easier to select a question-answer pair containing a question corresponding to the user's inquiry. This makes it easier for the user to reach the desired answer.

FIG. 11 is a schematic diagram showing the operation of the dialogue system according to the second embodiment.
FIG. 11 shows an interaction between the user US and the interaction system 2. FIG. For example, when the dialogue system 2 detects the user US, it outputs the sentence TX1 "Ask a question." The user US enters the text TX2 “About bank account” into the dialog system 2 . The dialogue system 2 decomposes the sentence TX2 (query sentence) into a plurality of words and estimates the part of speech of each word. The dialogue system 2 extracts "deposit account" as a query keyword from the sentence TX2 based on the part of speech. The dialog system 2 selects a question-and-answer pair in which "deposit account" is registered as a question keyword from the question-and-answer collection.

As an example, the dialogue system 2 includes a question-and-answer pair containing the question keywords "deposit account" and "application", a question-and-answer pair containing the question keywords "deposit account" and "cancellation", and "deposit account" and " A question-and-answer pair including the question keyword of "balance confirmation" and three question-and-answer pairs are selected. Based on this selection result, the dialogue system 2 outputs the text TX3 "Do you have any additional keywords? Regarding application, cancellation, balance confirmation, etc.". Text TX3 corresponds to the second response.

In sentence TX3, the word corresponding to the question keyword may be output so as to be distinguishable from other words. For example, when a response output from the dialogue system 2 is displayed on a monitor, words corresponding to the question keyword are displayed bolder, underlined or colored than other words. Words corresponding to the question keyword may be displayed as clickable. When the user US clicks on a word corresponding to the question keyword, the next sentence containing the question keyword may be automatically input to the dialogue system 2 . When the response output from the dialogue system 2 is output as voice, the words corresponding to the question keyword are read aloud at a higher volume or more slowly than other words.

The user US inputs the text TX4 “application” into the dialogue system 2 based on the text TX3. The dialogue system 2 sets "application" as a query keyword based on the text TX4. The dialogue system 2 further selects a part of the previously selected question-answer pair using the query keyword based on the text TX4. For example, the query keyword "application" based on sentence TX4 is similar to the question keyword "application." The dialogue system 2 selects a question-answer pair containing the question keywords "deposit account" and "application" from the above three question-answer pairs. Since the number of question-and-answer pairs has been narrowed down to one, the dialogue system 2 displays the sentence TX5 indicating the answers included in the question-and-answer pair: "Your question is 'I want to apply for a savings account.' …” is output. Text TX5 corresponds to the first response.

The extraction unit 42 may determine a score representing the importance of each question keyword when extracting question keywords from the question-and-answer collection. The selection unit 13 selects a question-answer pair based on the score.

The score of a question keyword is set higher as the number of question-answer pairs containing that question keyword is smaller. Keywords with high scores are not often included in the question-and-answer collection, and are useful for selecting question-and-answer pairs. The setting unit 12 determines the score of each question keyword by using the appearance frequency in the question and answer collection, the reverse document frequency in the question and answer collection, or a method such as bm25. For example, the setting unit 12 calculates the ratio of the number of question-and-answer pairs including the question keyword to the total number of question-and-answer pairs included in the question-and-answer collection. The setting unit 12 sets the reciprocal of the ratio (reverse document frequency) as a score. Alternatively, the setting unit 12 may set a numerical value based on the ratio as the score. For example, the setting unit 12 may convert the reciprocal of the ratio using a logarithmic function and use the obtained numerical value as the score.

As a specific example, assume that the question-and-answer collection consists of four question-and-answer pairs 110a, 110b, 110c, and 110K, as shown in FIG. In this question-and-answer collection, there are three question-and-answer pairs including "deposit account". The "savings account" score is set to 4/3, which is the reciprocal of 3/4. Similarly, the score for "Balance Check" is set to 4/1, which is the reciprocal of 1/4. The score for "Sign Up" is set to 4/2, which is the reciprocal of 2/4.

After selecting the question-and-answer pairs including the query keyword, the selection unit 13 refers to the score of the question keyword corresponding to the query keyword for each of the selected question-and-answer pairs. A question-answer pair containing a question keyword corresponding to the query keyword is given a score set for the question keyword. The selection unit 13 calculates the total score for each question-and-answer pair. The selection unit 13 selects a question-answer pair whose total score exceeds a threshold (second threshold) from the question-answer pairs including the query keyword. The generator 14 generates a response according to the number of question-answer pairs whose total score exceeds the threshold.

For example, the setting unit 12 sets part of the query text input by the user as the first query keyword, and sets another part of the query text as the second query keyword. The selection unit 13 gives the score set to the question keyword to one or more question-answer pairs including the question keyword corresponding to the first query keyword. The selection unit 13 gives the score set for the other question keyword to one or more question-answer pairs including another question keyword corresponding to the second query keyword. The selection unit 13 selects one or more question-and-answer pairs whose score exceeds the threshold.

By selecting question-and-answer pairs based on scores, the number of question-and-answer pairs can be narrowed down. In addition, selection based on scores can prevent the answer desired by the user from being deviated from the selection. Therefore, it becomes easier for the user to reach the desired answer.

The processing unit 10 may determine whether the query sentence is a negative sentence. For example, when setting a query keyword, the setting unit 12 determines whether the query sentence is a negative sentence. When the setting unit 12 determines that the query sentence is a negative sentence, the selection unit 13 selects question-answer pairs that do not include the query keyword.

For example, after the dialog system 3 outputs the text TX3, the user inputs a query text, "Not cancellation and balance confirmation." The setting unit 12 extracts "cancellation" and "balance check" from this query text as query keywords. Furthermore, the setting unit 12 determines that this query sentence is a negative sentence. Based on this determination result, the selection unit 13 selects a question-answer pair that does not include the question keywords "cancellation" and "balance confirmation" from the three question-answer pairs that were previously selected. As a result, the dialogue system 2 outputs the text TX5.

In order to determine whether the query sentence is a negative sentence, the setting unit 12 refers to, for example, rules created in advance. As an example, a rule may be stated such as "A query sentence is a negative sentence when the query sentence contains 'not', 'not', or 'other than'." The setting unit 12 determines whether the query sentence is a negative sentence by determining whether the query sentence satisfies the conditions described in the rule.
The setting unit 12 may refer to a model created in advance. For example, the model is an artificial neural network model or a recurrent neural network model. The model is trained in advance using training data. The learning data includes multiple pairs of data. Each pair of data includes a sentence and information indicating that the sentence is negative. The model is trained to output information indicating a negative sentence when a negative sentence is input.

(Modification)
FIG. 12 is a schematic diagram illustrating an interactive system according to a modification of the second embodiment.
In the dialog system 2a according to the modification, a task model is generated in advance before the user makes an inquiry. A task model is generated by the processing unit 40 . For example, the dialogue system 2a includes a processing unit 40. FIG. The processing unit 40 stores the generated task model in the third storage unit 23 .

In the dialogue system 2 a , the processing section 40 includes an extracting section 41 , an extracting section 42 and a generating section 43 . The processing unit 10 includes a reception unit 11 , a setting unit 12 and a selection unit 13 .

The retrieval unit 41 accesses the first storage unit 21 and retrieves one question-answer pair from the question-and-answer collection. The extraction unit 42 extracts question keywords from the question-answer pair according to the model or rules stored in the fourth storage unit 24 . The extraction unit 42 stores the extracted question keyword in the first storage unit 21, for example. The generator 43 generates a task scenario based on the extracted question keyword, and updates the task model. The processing by the generation unit 43 is the same as steps S123 to S126 in the flowchart shown in FIG.

The processing unit 10 uses the task model to output a response to the query text input by the user. The processing of the processing unit 10 in the dialogue system 2a is the same as the flowchart shown in FIG.

According to the interactive system 2a according to the modification, similarly to the interactive system 1a, a task model including a plurality of task scenarios is generated before the processing unit 10 receives a query sentence. Therefore, for example, the processing load on the processing unit 10 after receiving the query text can be reduced.

In the dialogue system 2a, the extraction unit 42 may determine a score representing the importance of each question keyword when extracting question keywords from the question and answer collection. The selection unit 13 selects a task scenario from the task model based on the score.

As an example, the setting unit 12 extracts two query keywords (a first query keyword and a second query keyword) from the query text input by the user. The task model includes a task scenario including first question keywords corresponding to the first query keywords, and a task scenario including second question keywords corresponding to the second query keywords. The task model does not include a task scenario that includes both the first question keyword and the second question keyword.
In this example, when selecting one task scenario, the selection unit 13 compares the score of the first question keyword and the score of the second question keyword. The selection unit 13 selects a task scenario including a question keyword with a higher score. The selection unit 13 outputs the selected task scenario as a response.

FIG. 13 is a schematic diagram for explaining the operation of the dialogue system according to the second embodiment.
The upper part of FIG. 13 illustrates the interaction between the user and the interaction system. The middle part of FIG. 13 illustrates a collection of questions and answers. The lower part of FIG. 13 illustrates the task model.

The question-and-answer collection consists of six question-and-answer pairs shown in FIG. 13, for example. Each question includes one of the question keywords "internet banking", "overseas", and "remittance". The question and answer collection does not include questions that include all of "internet banking", "overseas", and "remittance". In this case, the processing unit 40 generates a task scenario corresponding to the question keyword "internet banking", a task scenario corresponding to the question keyword "overseas", and a task scenario corresponding to the question keyword "remittance". , to generate

When question keywords are extracted from the question-and-answer collection to generate a task scenario, the extraction unit 42 determines the score of each question keyword. In the example of FIG. 13, there are two questions (answer candidates) including "internet banking". There are three questions that include "overseas." There is one question that includes "remittance". The score is the number of questions containing each query keyword for the number of question-answer pairs (6) selected. "Internet Banking" scores 6/2. The score for "overseas" is 6/3. "Remittance" scores 6/1.

The user US, as shown in FIG. 13, inputs into the dialogue system 2a, "Can you remit money overseas by internet banking?" The setting unit 12 extracts "internet banking", "overseas", and "remittance" from this query text as query keywords. The selection unit 13 extracts from the task model a task scenario generated corresponding to one of the question keywords "internet banking", "overseas", and "remittance".

The selection unit 13 outputs one of a plurality of task scenarios as a response. In order to select one task scenario from a plurality of task scenarios, the selection unit 13 compares scores of question keywords corresponding to each task scenario. As a result of the score comparison, it is determined that the score of "remittance" is the highest. The selection unit 13 outputs a task scenario corresponding to "remittance" as a response.

FIG. 14 is a schematic diagram illustrating the configuration of the interactive device according to the embodiment.
The dialogue system according to each of the embodiments described above is implemented by, for example, the dialogue device 3 shown in FIG. The dialogue device 3 is implemented by, for example, a hardware configuration similar to that of a general computer (information processing device). The interactive device 3 includes a CPU (Central Processing Unit) 91, an input section 92, an output section 93, a ROM (Read Only Memory) 94, a RAM (Random Access Memory) 95, a storage section 96, and a communication device 97. , and a bus 98 . Each section is connected by a bus 98 .

The CPU 91 cooperates with various programs pre-stored in the ROM 94 or the storage unit 96 to execute various processes, and controls the operation of each unit constituting the interactive device 3 in an integrated manner. CPU91 uses the predetermined area|region of RAM95 as a work area in a process. The CPU 91 cooperates with a program pre-stored in the ROM 94 or storage unit 96 to implement an input unit 92, an output unit 93, a communication device 97, and the like.

The input unit 92 includes, for example, at least one of a keyboard, a microphone, and a touch panel. The input unit 92 receives information input by the user as an instruction signal, and outputs the instruction signal to the CPU 91 . The output unit 93 includes, for example, at least one of a monitor, a speaker, and a printer. The output unit 93 outputs various information based on the signal output from the CPU 91 .

The ROM 94 non-rewritably stores programs and various setting information used for controlling the interactive device 3 . The RAM 95 is a volatile storage medium such as SDRAM (Synchronous Dynamic Random Access Memory). The RAM 95 functions as a work area for the CPU 91 . Specifically, it functions as a buffer or the like that temporarily stores various variables, parameters, and the like used by the interactive device 3 .

The storage unit 96 is a rewritable recording device such as a semiconductor storage medium such as a flash memory, or a magnetically or optically recordable storage medium. The storage unit 96 stores programs used for controlling the interactive device 3, various setting information, and the like. Storage unit 96 functions as first storage unit 21 to fourth storage unit 24 . A communication device 97 is used to communicate with an external device to transmit and receive information.

According to each of the embodiments described above, it is possible to more easily construct a dialogue system that allows dialogue with a user. According to each embodiment, it is possible to provide a dialogue system, a dialogue method, a program, and a storage medium that make it easier for the user to reach the desired answer.

The various data processing described above is executed based on, for example, a program (software). For example, a computer stores this program and reads out this program to process the various information described above.

The processing of the various information described above can be performed by using a program that can be executed by a computer, such as a magnetic disk (flexible disk, hard disk, etc.), an optical disk (CD-ROM, CD-R, CD-RW, DVD-ROM, DVD±R). , DVD±RW, etc.), a semiconductor memory, or other recording media.

For example, information recorded on a recording medium can be read by a computer (or embedded system). Any recording format (storage format) can be used in the recording medium. For example, a computer reads a program from a recording medium and causes a CPU to execute instructions written in the program based on the program. Acquisition (or reading) of a program in a computer may be performed through a network.

At least part of the processing of the above information may be implemented in various software running on a computer based on a program installed from a recording medium to the computer (or embedded system). This software includes, for example, an OS (operating system). This software may include, for example, middleware that operates on a network.

A recording medium according to an embodiment stores a program that can cause a computer to process the various types of information described above. The recording medium according to the embodiment also includes a recording medium in which a program is stored after being downloaded via a LAN or the Internet. The above processing may be performed based on a plurality of recording media.

A computer according to embodiments includes one or more devices (eg, personal computers, etc.). A computer according to an embodiment may include multiple devices connected by a network.

The embodiments of the present invention have been described above with reference to specific examples. However, embodiments of the invention are not limited to these specific examples. For example, with respect to the specific configuration of each element such as a processing unit and a storage unit included in the dialogue system, those skilled in the art can implement the present invention in the same manner and obtain the same effect by appropriately selecting from a range known to those skilled in the art. are included within the scope of the invention as far as possible.

Any combination of two or more elements of each specific example within the technically possible range is also included in the scope of the present invention as long as it includes the gist of the present invention.

In addition, based on the dialogue system, dialogue method, program, and storage medium described above as embodiments of the present invention, all dialogue systems, dialogue methods, programs, and memories that can be implemented by those skilled in the art by appropriately modifying their designs. A medium also belongs to the scope of the present invention as long as it includes the gist of the present invention.

In addition, within the scope of the idea of the present invention, those skilled in the art can conceive various modifications and modifications, and it is understood that these modifications and modifications also belong to the scope of the present invention. .

While several embodiments of the invention have been described, these embodiments have been presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and equivalents thereof.

Reference Signs List 1, 1a, 2, 2a dialogue system, 3 dialogue device, 10 processing unit, 11 reception unit, 12 setting unit, 13 selection unit, 14 generation unit, 21 first storage unit, 22 second storage unit, 23 third storage Unit 24 Fourth Storage Unit 30 Processing Unit 31 Extraction Unit 32 Extraction Unit 33 Generation Unit 40 Processing Unit 41 Extraction Unit 42 Extraction Unit 43 Generation Unit 91 CPU 92 Input Unit 93 Output Unit , 94 ROM, 95 RAM, 96 storage unit, 97 communication device, 98 bus, 110 question and answer collection, 110a to 110K question and answer pair, 111a to 111K ID, 112a to 112K question, 113a to 113K answer, 114a to 114K question keyword , 120 general-purpose model, 121a-121d scenario template, 122a-122d number of answer candidates, 131-133 task scenario, Ns threshold, TX1-TX6 text, US user

Claims (9)

when a user inputs a first sentence containing one or more words, setting at least part of the one or more words as a first query keyword;
From a question-and-answer collection containing a plurality of combinations of a question-and-answer pair including a question and an answer to the question, and a plurality of question keywords linked to the question-and-answer pair, one or more of the above including the first query keyword Select a question-and-answer pair,
when the number of the selected question-and-answer pairs is 1, outputting a first response indicating the answers included in the selected question-and-answer pair;
When the number is equal to or greater than the first threshold, a first question keyword different from the first query keyword is extracted from the plurality of question keywords associated with one of the selected question-answer pairs, and the selected A second question keyword different from the first query keyword is extracted from the plurality of other question keywords associated with another one of the question-answer pairs, and a second question keyword including the first question keyword and the second question keyword is extracted. 2 outputs a response,
A dialog system comprising a processing unit,
The processing unit uses a plurality of scenario templates each containing a blank when generating the first response and the second response,
The first response is generated by embedding the answer included in the selected question-answer pair in the blank of one of the plurality of scenario templates;
The dialogue system, wherein the second response is generated by embedding the first question keyword and the second question keyword in another one of the plurality of scenario templates. After outputting the second response, the processing unit receives a second sentence including one of the first question keyword and the second question keyword, and then receives the one of the first question keyword and the second question keyword. 2. The dialogue system according to claim 1, which outputs a response indicating said answer contained in said question-answer pair containing . The first threshold is 3 or more,
3. The dialog according to claim 1, wherein, when the number is equal to or greater than 2 and less than the first threshold, the processing unit outputs a third response including the plurality of questions respectively included in the selected question-answer pair. system. The processing unit, in selecting the question-and-answer pair,
The plurality of first similar keywords similar to the first query keyword are selected based on the degree of similarity between the first query keyword and a plurality of question keywords included in the plurality of questions of the question and answer collection. from the question keywords of
Selecting one or more question-answer pairs including at least one of the first query keyword and the first similar keyword from the question-and-answer collection;
Dialog system according to any one of claims 1-3. The processing unit, in setting the first query keyword,
dividing the first sentence into the one or more words;
estimating a part of speech for each of the one or more words;
setting one of a verb, a noun, and an unknown word contained in the one or more words as the first query keyword;
Dialog system according to any one of claims 1-4. when a user inputs a first sentence including a plurality of words, setting part of the plurality of words as a first query keyword and setting another part of the plurality of words as a second query keyword;
A question-and-answer collection containing a plurality of combinations of a question-and-answer pair including a question and an answer to the question and a plurality of question-keywords linked to the question-and-answer pair is stored, and a score is assigned to each of the plurality of question-keywords. referring to the set first storage unit,
giving a score set to the one of the plurality of question keywords to one or more question-answer pairs including one of the plurality of question keywords corresponding to the first query keyword, and the second query assigning a score set to another one of the plurality of question keywords to one or more question-answer pairs including another one of the plurality of question keywords corresponding to the keyword;
Select one or more question-answer pairs whose assigned scores exceed the second threshold;
When the number of the selected question-and-answer pairs is 1, outputting a first response indicating the answers included in the selected question-and-answer pairs,
When the number is equal to or greater than the first threshold, a first question keyword different from the first query keyword is extracted from the plurality of question keywords associated with one of the selected question-answer pairs, and the selected A second question keyword different from the first query keyword is extracted from the plurality of other question keywords associated with another one of the question-answer pairs, and a second question keyword including the first question keyword and the second question keyword is extracted. 2 outputs a response,
A dialog system comprising a processing unit,
The processing unit uses a plurality of scenario templates each containing a blank when generating the first response and the second response,
The first response is generated by embedding the answer included in the selected question-answer pair in the blank of one of the plurality of scenario templates;
The dialogue system, wherein the second response is generated by embedding the first question keyword and the second question keyword in another one of the plurality of scenario templates. to the processing unit,
when a first sentence including one or more words is input by a user, setting at least part of the one or more words as a first query keyword;
From a question-and-answer collection containing a plurality of combinations of a question-and-answer pair including a question and an answer to the question, and a plurality of question keywords linked to the question-and-answer pair, one or more of the above including the first query keyword Select a question-and-answer pair,
when the number of the selected question-and-answer pairs is one, outputting a first response indicating the answers included in the selected question-and-answer pairs;
If the number is equal to or greater than the first threshold, a first question keyword different from the first query keyword is extracted from the plurality of question keywords associated with one of the selected question-answer pairs, and the selected A second question keyword different from the first query keyword is extracted from the plurality of other question keywords associated with another one of the question-answer pairs, and a second question keyword including the first question keyword and the second question keyword is extracted. 2 to output a response,
A dialogue method comprising:
causing the processing unit to use a plurality of scenario templates each containing a blank when generating the first response and the second response;
The first response is generated by embedding the answer included in the selected question-answer pair in the blank of one of the plurality of scenario templates;
The interactive method, wherein the second response is generated by embedding the first question keyword and the second question keyword in the blank of another one of the plurality of scenario templates. to the processing unit,
when a first sentence including one or more words is input by a user, setting at least part of the one or more words as a first query keyword;
From a question-and-answer collection containing a plurality of combinations of a question-and-answer pair including a question and an answer to the question, and a plurality of question keywords linked to the question-and-answer pair, one or more of the above including the first query keyword Select a question-and-answer pair,
when the number of the selected question-and-answer pairs is one, outputting a first response indicating the answers included in the selected question-and-answer pairs;
If the number is equal to or greater than the first threshold, a first question keyword different from the first query keyword is extracted from the plurality of question keywords associated with one of the selected question-answer pairs, and the selected A second question keyword different from the first query keyword is extracted from the plurality of other question keywords associated with another one of the question-answer pairs, and a second question keyword including the first question keyword and the second question keyword is extracted. 2 to output a response,
a program,
causing the processing unit to use a plurality of scenario templates each containing a blank when generating the first response and the second response;
The first response is generated by embedding the answer included in the selected question-answer pair in the blank of one of the plurality of scenario templates;
The program, wherein the second response is generated by embedding the first question keyword and the second question keyword in another one of the plurality of scenario templates. A storage medium storing the program according to claim 8 .

Images