JP6952259B2 - Information processing method, information processing device, and program - Google Patents

Description

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

A dialogue system has been proposed in which a user searches or proposes content (music, product information, etc.) requested by dialogue through dialogue. Such a dialogue system narrows down the content candidates that meet the user's request by gradually acquiring information for narrowing down the content from the user through dialogue, and presents a list of candidates at a certain timing.

For the problem of when to present the list of candidates narrowed down by such dialogue during the dialogue, a method of controlling the presentation timing of the candidate list by reinforcement learning has been proposed (Non-Patent Documents). 1).

Verena Rieser, Oliver Lemon, "Reinforcement Learning for Adaptive Dialogue Systems", Springer Science & Business Media, November 23, 2011

However, in the method of Non-Patent Document 1, whether or not the user is tolerant of the length of time during which the dialogue is repeated, in other words, whether or not the user wants to present a list of candidates at a relatively early stage. Information on the characteristics of is given as an element of the reward function of reinforcement learning. Then, the presentation of the list of candidates is controlled based on the information regarding the above characteristics. Therefore, in order to properly control the timing of presenting the list of candidates, it is necessary to know the information about the above characteristics of the user in advance, and if it is not known, the timing control is not properly performed. There is.

Therefore, it is an object of the present invention to provide an information processing method or the like for more appropriately controlling the timing of presenting a list of candidates even when there is no information on the above characteristics.

The information processing method according to one aspect of the present invention is an information processing method relating to information stored in a storage unit, and retains a history of dialogue including a question to a user and a response by the user to the question. , A mode of appearance of a third response that is neither a first response indicating affirmation to the question and a second response indicating denial to the question in the history of the response included in the history of the dialogue held. However, it is determined whether or not the predetermined condition is satisfied, and when it is determined that the predetermined condition is satisfied, the information stored in the storage unit is presented.

It should be noted that these comprehensive or specific embodiments may be realized in a recording medium such as a system, an apparatus, an integrated circuit, a computer program or a computer-readable CD-ROM, and the system, the apparatus, the integrated circuit, the computer program. And may be realized by any combination of recording media.

The information processing method of the present invention can more appropriately control the timing of presenting a list of candidates.

FIG. 1 is a block diagram showing a configuration of a dialogue system according to an embodiment. FIG. 2 is an explanatory diagram showing an example of an input unit and a display unit according to the embodiment. FIG. 3 is an explanatory diagram showing an example of conversion from spoken text to semantic information by the language comprehension unit according to the embodiment. FIG. 4 is an explanatory diagram showing an example of the dialogue history held in the dialogue history holding unit according to the embodiment. FIG. 5 is an explanatory diagram showing an example of narrowing down information held in the narrowing down information holding unit according to the embodiment. FIG. 6 is an explanatory diagram of a content table showing an example of content stored in the content database according to the embodiment. FIG. 7 is a flow chart showing a processing flow of the dialogue system according to the embodiment. FIG. 8 is a flow chart showing a process related to a determination as to whether or not to present a candidate list in the dialogue system according to the embodiment. FIG. 9 is a first explanatory diagram showing an operation example of the dialogue system according to the embodiment. FIG. 10 is a second explanatory diagram showing an operation example of the dialogue system according to the embodiment. FIG. 11 is an explanatory diagram showing a content table in another application example of the dialogue system according to the embodiment. FIG. 12 is a first explanatory diagram showing an operation example in another application example of the dialogue system according to the embodiment. FIG. 13 is a second explanatory diagram showing an operation example in another application example of the dialogue system according to the embodiment. FIG. 14 is a block diagram showing a configuration of a dialogue system according to a modified example of the embodiment.

(Findings underlying the present invention)
A dialogue system has been proposed in which a user searches or proposes content (music, product information, etc.) requested by dialogue through dialogue. Such a dialogue system usually holds a large amount of content that is the subject of a search or suggestion. Then, the dialogue system gradually acquires information for narrowing down the content from the user through the dialogue, thereby narrowing down the content candidates that meet the user's request. The narrowed down candidates are presented to the user as a list of candidates on the screen, for example, when the number of the candidates falls below a certain number. The user can confirm the presented candidates and select a desired candidate.

Here, regarding the timing of presenting candidates, it is not always appropriate to use the fact that the number of candidates is less than a certain number as a criterion. This is because if the number of dialogues repeated before the candidate is presented is greater than the number intended by the user, the user's dissatisfaction with the difficulty in presenting the result of the search or proposal increases. On the other hand, if the number of repeated dialogues before the candidates are presented is less than the number intended by the user, the user's dissatisfaction with the failure to sufficiently narrow down the candidates by including many conditions increases. .. In this way, it is considered that the timing (how early or late) the user wants to browse the narrowed-down list of candidates in the process of dialogue differs depending on the user.

The control method for presenting the narrowing-down result by the above-mentioned conventional technique is the error rate of voice recognition due to noise (that is, the probability that the actual content of the user's utterance does not match the content of the utterance recognized by the dialogue system) and the narrowing-down result. A reward function calculated from three parameters, such as the size of the screen for displaying the screen and the user's tolerance for the number of times the dialogue is repeated, is used. Then, the control method repeats the determination of whether to ask a question or present a list so that the reward calculated by the reward function is maximized in the dialogue by reinforcement learning. That is, in the control method, the threshold value of a predetermined number of candidates is not set and controlled as to whether or not to present the result, and the situation in which the dialogue set by the above three parameters is performed is rewarded. By comprehensively evaluating as a function, the timing to display the list of narrowing results is dynamically determined.

However, in this conventional technique, the timing is controlled on the premise that the parameter of tolerance for the number of repetitions of the user's dialogue is known. Therefore, there is a problem that the timing cannot be controlled appropriately when this parameter is not known. In other words, in this prior art, there is a problem that it is necessary to know parameters such as tolerance of the user in advance in order to properly control the timing of presenting the list of candidates.

Therefore, it is an object of the present invention to provide an information processing method or the like for more appropriately controlling the timing of presenting a list of candidates even when there is no information on the above characteristics.

In order to solve such a problem, the information processing method according to one aspect of the present invention is an information processing method relating to information stored in a storage unit, and is a question to a user and a question to the user by the user. The history of the dialogue including the response is retained, and in the history of the response included in the history of the dialogue, the first response indicating affirmation to the question and the second response indicating denial to the question It is determined whether or not the mode of appearance of the third response, which is neither of them, satisfies the predetermined condition, and when it is determined that the predetermined condition is satisfied, the information stored in the storage unit is presented. ..

According to the above aspect, the information processing method controls the timing of presenting a list of candidates from the content of the user's utterance in the dialogue with the user. The third response, which is neither an affirmative response nor a negative response, is also considered to be an expression of the user's intention not to provide clear narrowing conditions. Therefore, the mode of appearance of the third response can be a criterion for determining whether or not the user wants to see the list of content candidates. By doing this, even if you do not know whether you want to see the list of content candidates narrowed down in advance (that is, the above "tolerance") quickly, you can estimate the user's intention from the content of the user's utterance. Therefore, it is possible to control the timing of presenting the narrowed list of content candidates.

This information processing method can be applied to an application of a dialogue system that narrows down and presents a large number of contents according to the user's intention by dialogue. In particular, it can be used in an application in which a user presents a candidate at an appropriate timing from an interactive system and lets the user select it, especially when the information for narrowing down the content is not so clearly verbalized.

For example, the predetermined condition is that the third response appears a predetermined number of times in succession in the history of the response.

According to the above aspect, the information processing method realizes the information processing method more concretely by using the condition that the third response appears consecutively a predetermined number of times, and the timing of presenting the list of candidates is more appropriate. Can be controlled.

For example, the predetermined condition is that the third response is accumulated and appears a predetermined number of times in the history of the response.

According to the above aspect, the information processing method realizes the information processing method more concretely by using the condition that the third response is accumulated and appears a predetermined number of times, and the timing of presenting the list of candidates is more appropriate. Can be controlled.

For example, at the time of the presentation, at least a part of the information stored in the storage unit is presented.

According to the above aspect, the information processing method allows the user to browse the narrowed list of content candidates based on the intention estimation. In this way, by estimating the user's intention from the content of the user's utterance, it is possible to control the timing of presenting the narrowed list of content candidates.

For example, at the time of the presentation, a question to the user as to whether or not to present at least a part of the information stored in the storage unit is presented.

According to the above aspect, the information processing method can present a question as to whether or not to allow the user to browse the narrowed list of content candidates based on the intention estimation. In the timing control based on the intention estimation, the user may try to present the list of content candidates at a timing different from the timing actually intended. Therefore, by confirming the actual intention of the user by the above question, it is possible to suppress the presentation of the list of content candidates contrary to the intention of the user.

For example, at the time of the presentation, when the user obtains an affirmative response to the question, at least a part of the information stored in the storage unit is presented.

According to the above aspect, the information processing method presents the list of content candidates and allows the user to browse after confirming that the presentation of the list of content candidates matches the actual intention of the user. can.

For example, in the information processing method, a process of narrowing down the information stored in the storage unit is further performed by using the response included in the dialogue, and at the time of the presentation, after the process is performed. The presentation regarding the information stored in the storage unit is made.

According to the above aspect, the information processing method can narrow down the information stored in the storage unit by the interface of dialogue and present the narrowed down information to the user. As a result, the user can search for information based on sequential information input by dialogue, and can receive proposals from the dialogue system, unlike searching for information by a search engine or the like.

For example, in the information processing method, when it is determined that the history of the response includes only one or more of the third response, the information processing method is stored in the storage unit using predetermined information. The processing for narrowing down the information to be stored is performed, and at the time of the presentation, the presentation regarding the information stored in the storage unit is performed after the processing is performed.

According to the above aspect, when there is no narrowing condition, the information processing method can avoid displaying all the information stored in the storage unit, that is, the information that has not been narrowed down at all as a list.

For example, the question is a closed question for causing the user to make any of the first response, the second response, and the third response.

According to the above aspect, the information processing method can make a more specific search and suggestion based on a closed question that provides options and makes an answer selected, and a dialogue composed of the answers.

For example, in the storage unit, a plurality of pieces of information indicating a cooking menu are stored as the information, and the dialogue includes a question regarding the search of the cooking menu to the user and the user in response to the question. Includes response.

According to the above aspect, the information processing method can provide the user with a search or suggestion of the cooking menu by using the narrowing down of the information indicating the cooking menu.

For example, in the storage unit, a plurality of information indicating the writer is stored as the information, and the dialogue includes a question regarding the search for the writer to the user and the response by the user to the question. including.

According to the above aspect, the information processing method can provide the user with a search or suggestion of a writer by using the narrowing down of information indicating the writer.

Further, the information processing method according to one aspect of the present invention is an information processing method relating to a menu preferred by the user using a dialogue with the user in a chat format via the user's terminal, and is provided in the terminal. A first question asking a request for the menu is displayed in a balloon type on the display, and the database associates a plurality of menus with classification information indicating the classification of each of the plurality of menus. The classification information includes hot / cold of the menu, the genre of the menu, the category of the menu, the cooking method of the menu, and the ingredients used in the menu, and the first question is asked in the database for the menu. Asks whether to request the included first classification, receives the first answer to the first question input by the user through the terminal, and in the first answer, a phrase affirming the first classification. If it is determined that is included, a second question is asked to the display asking whether to request the second classification corresponding to the first menu corresponding to the first classification based on the database. In the second classification, unlike the first classification, the second answer to the second question input by the user is received through the terminal, and the second classification is displayed in the second answer. If it is determined that neither the phrase affirming the above nor the phrase denying the second classification is included, the display is requested to request the third classification corresponding to the first menu based on the database. The third question to be asked is displayed in a balloon type, and unlike the first category and the second category, the third category receives the third answer to the third question input by the user through the terminal. (A) If it is determined that the third answer contains words and phrases that affirm the third category, the second menu corresponding to the third category of the first menu is supported. The fourth question asking whether or not to request the four classifications is displayed in a balloon type, and the fourth classification is different from any of the first classification, the second classification, and the third classification, and (b) the first classification. (Iii) If it is determined that the answer contains a phrase that denies the third category, whether or not to request the fifth category corresponding to the third menu that does not correspond to the third category among the first menus. The fifth question asking is displayed in a balloon type, and the fifth classification is the first minute. Unlike any of the categories, the second category, the third category, and the fourth category, (c) the third answer includes a phrase that affirms the third category and a phrase that denies the third category. If it is determined that the question is not displayed, a fifth question asking whether the information indicating the first menu may be displayed is displayed on the display in a balloon type, and the user input the information through the terminal. When the fifth answer to the fifth question is received and it is determined that the fifth answer contains a phrase affirming that the information indicating the first menu is displayed, the first menu is displayed on the display. Display information indicating.

According to the above aspect, the same effect as that of the above information processing method is obtained.

Further, the information processing device according to one aspect of the present invention is an information processing device for performing information processing on information stored in a storage unit, and includes a question to a user and a response by the user to the question. In the dialogue history holding unit that holds the history of the dialogue including the above, the first response indicating the affirmation of the question in the history of the response included in the history of the dialogue being held, and the first response indicating the denial of the question. The mode of appearance of the third response, which is neither of the two responses, is stored in the dialogue management unit for determining whether or not the predetermined condition is satisfied, and in the storage unit when it is determined that the predetermined condition is satisfied. It is provided with a presentation unit for presenting the above-mentioned information.

According to the above aspect, the same effect as that of the above information processing method is obtained.

Further, the program according to one aspect of the present invention is a program for causing a computer to execute the above information processing method.

According to the above aspect, the same effect as that of the above information processing method is obtained.

It should be noted that these comprehensive or specific embodiments may be realized in a recording medium such as a system, an apparatus, an integrated circuit, a computer program or a computer-readable CD-ROM, and the system, the apparatus, the integrated circuit, the computer program. Alternatively, it may be realized by any combination of recording media.

Hereinafter, embodiments will be specifically described with reference to the drawings.

It should be noted that all of the embodiments described below show comprehensive or specific examples. The numerical values, shapes, materials, components, arrangement positions and connection forms of the components, steps, the order of steps, etc. shown in the following embodiments are examples, and are not intended to limit the present invention. Further, among the components in the following embodiments, the components not described in the independent claims indicating the highest level concept are described as arbitrary components.

(Embodiment)
In the present embodiment, the dialogue system 1 that more appropriately controls the timing of presenting the list of candidates will be described. The dialogue system 1 corresponds to an information processing device. Further, the control method of the dialogue system 1 is also referred to as a narrowing-down result presentation control method or an information processing method.

FIG. 1 is a block diagram showing an example of the configuration of the dialogue system 1 according to the present embodiment.

As shown in FIG. 1, the dialogue system 1 according to the present embodiment includes an input unit 101, a display unit 102, a language understanding unit 103, a dialogue management unit 104, a response generation unit 105, and a content database 106. A narrowing down information holding unit 107, a narrowing down processing unit 108, a dialogue history holding unit 109, a user intention estimation unit 110, and a candidate list display generation unit 111 are provided.

The dialogue system 1 is a system that narrows down the contents stored in the content database 106 by dialogue with the user and presents the narrowed down contents to the user. This dialogue is, for example, a dialogue with a user in a chat format, and the dialogue system 1 is a system that processes information regarding information preferred by the user.

It is assumed that the content database 106 stores a plurality of contents to be narrowed down. An example of the content is information indicating a cooking menu (sometimes simply referred to as a "cooking menu"), and this case will be described later, but the content may be various other information such as information indicating a writer or real estate. There can be something like that. The content database 106 corresponds to a storage unit and can be realized by a storage device, a database management system, or the like.

The content database 106 stores a plurality of pieces of information, each of which indicates a cooking menu. Then, the dialogue system 1 provides information indicating the user's desired cooking menu by using a dialogue including a question regarding the search of the cooking menu to the user and a response by the user to the question.

The input unit 101 acquires an utterance from the user to the dialogue system 1 and converts it into an utterance text. The display unit 102 receives the utterance text input by the user from the input unit 101, the response text from the dialogue system 1 to the user, and other response information. The input unit 101 can be realized by a processor, a memory, or the like. The input unit 101 may include a microphone that picks up the user's utterance and acquires a sound signal, or may acquire a sound signal generated by an external microphone by picking up the sound without the microphone. In addition, the input unit 101 obtains the utterance text by voice recognition processing for the acquired sound signal. A known technique can be adopted for the voice recognition process.

The display unit 102 displays the utterance text, the response text, and the like. The display unit 102 can be realized by a processor, a memory, a display (display screen), and the like. The display unit 102 may be realized as an image displayed on the display. The display unit 102 is an example of a presentation unit. Further, as another example of the presentation unit, the dialogue system 1 may include a voice output unit that outputs a response text or the like by voice instead of the display unit 102.

FIG. 2 is an explanatory diagram showing an example of an input unit 101 and a display unit 102 according to the present embodiment. FIG. 2 shows an example of an image in which the input unit 101 and the display unit 102 are realized as a GUI (graphical interface) displayed on a display.

The image 201 shown in FIG. 2 corresponds to the input unit 101. Further, the image 202 shown in FIG. 2 corresponds to the display unit 102.

The user operates the image 201 as the input unit 101, inputs the utterance text in the input field 201A, and operates the transmission button 201B to complete the input of the utterance text. The "operation" refers to general operations on the GUI such as pressing, touching, tapping, and clicking.

On the image 202 corresponding to the display unit 102, the utterance text input by the user and the response text from the dialogue system 1 are displayed in a balloon type. FIG. 2 shows the response texts 202A and 202B from the dialogue system 1 and the utterance text 202C input by the user. The response texts 202A and 202B are displayed on the left side of the image 202 in the order of passage of time from top to bottom. The utterance text 202C is displayed on the right side of the image 202 in the order of passage of time from top to bottom. When the response texts 202A and 202B or the utterance text 202C are displayed up to the lower end in the image 202, the image 202 scrolls upward and a new response text or the utterance text is displayed in the image 202. ..

The language understanding unit 103 linguistically analyzes the user's utterance text input from the input unit 101, and converts the utterance text into semantic information. The language understanding unit 103 can be realized by a known language analysis process executed by a processor, a memory, or the like.

FIG. 3 is an explanatory diagram showing an example of conversion from spoken text to semantic information by the language understanding unit 103 according to the present embodiment.

The language comprehension unit 103 converts the expression of the utterance text having the same meaning into the same meaning information.

For example, the language comprehension unit 103 converts utterance text indicating an affirmative answer to a question, such as "it is good" and "yes", into semantic information of "YES". It also converts utterance texts that indicate a negative response to a question, such as "I don't like it" and "No", into semantic information that means "NO". In addition, the utterance text that is neither an affirmative response nor a negative response, such as "Random" and "Neither", is converted into semantic information of "NOT_CARE".

Various methods can be adopted to realize the function of the language understanding unit 103. For example, there is a rule-based method in which a table of utterance texts assumed in advance is created to correspond with the semantic information, or a method in which the utterance text is labeled with the semantic information by machine learning.

The dialogue management unit 104 receives the semantic information converted by the language understanding unit 103, controls the narrowing down of the content according to the type of the semantic information, controls the operation of the dialogue system 1 with reference to the dialogue history, and the dialogue system 1. Performs the operation of determining the response content. Specifically, the dialogue management unit 104 determines the mode of appearance of a third response that is neither an affirmative response nor a negative response in the history of the responses included in the dialogue history held by the dialogue history holding unit 109. Judge whether or not the condition of is satisfied. The dialogue management unit 104 determines as a response content a closed question that provides options and causes the answer to be selected. The dialogue management unit 104 can be realized by a processor, a memory, a storage device, and the like.

The response generation unit 105 converts the response content determined by the dialogue management unit 104 into a response text. The response content determined by the dialogue management unit 104 is a symbol string such as "Q, genre (Chinese food)", and the response generation unit 105 indicates a question sentence "Is Chinese style good?" Convert to response text. The response generation unit 105 can be realized by a processor, a memory, or the like.

When the dialogue management unit 104 receives the semantic information of the user's utterance text and determines the response content (corresponding to the semantic information) from the dialogue system 1 to the user, each of the semantic information and the response content is set. Stored in the dialogue history holding unit 109. The dialogue history holding unit 109 holds a history of dialogue including a question to the user and a response by the user to the question. The dialogue history holding unit 109 can be realized by a storage device.

FIG. 4 is an explanatory diagram showing an example of the dialogue history held in the dialogue history holding unit 109 according to the present embodiment.

As shown in FIG. 4, the dialogue history is recorded in a table format, and each row indicates the utterance of one user or the response of the dialogue system 1.

The leftmost column of the dialogue history is information indicating an ID (or number) representing the order of individual utterances and responses included in the dialogue. The ID starts from 1, and the ID is incremented by 1 each time one history is added.

The second column from the left of the dialogue history is information indicating the speaker, which is recorded as "System" in the case of a response from the dialogue system 1 to the user, and "User" in the case of a speech from the user to the dialogue system 1. Is recorded.

The third column from the left of the dialogue history is the semantic information of the user's utterance or the information indicating the response content of the dialogue system 1.

In the dialogue management unit 104, the response content of the dialogue system 1 recorded in the dialogue history holding unit 109 is a question form such as "Q, genre (Chinese)", and the user's utterance (answer) to the question is If it is any of "YES", "NO", and "NOT_CARE", the immediately preceding response content "genre (Chinese)" is treated as the narrowing key. In this case, the dialogue management unit 104 treats the narrowing-down key as if the narrowing-down condition of "YES", "NO", or "NOT_CARE" is given. Then, the dialogue management unit 104 stores the set of the narrowing down key and the narrowing down condition in the narrowing down information holding unit 107 as the narrowing down information. The narrowing down information holding unit 107 can be realized by a storage device.

FIG. 5 is an explanatory diagram showing an example of narrowing down information held in the narrowing down information holding unit 107 according to the present embodiment.

The leftmost column of the narrowing information is the narrowing key, and the second column from the left is the narrowing condition. The dialogue management unit 104 acquires a plurality of such narrowing down information from the utterance and response between the dialogue system 1 and the user, that is, the exchange of dialogue, and stores the narrowing down information in the narrowing down information holding unit 107.

The dialogue management unit 104 instructs the narrowing down processing unit 108 to perform narrowing down processing of a plurality of contents stored in the content database 106 by using the narrowing down information stored in the narrowing down information holding unit 107.

FIG. 6 is an explanatory diagram of a content table showing an example of content stored in the content database 106 according to the present embodiment. FIG. 6 is an example of a content table when the content is a cooking menu. The content database 106 is also simply referred to as a database.

Each row of the content table shown in FIG. 6 represents one content information.

The leftmost column of the content table is information indicating a content ID for identifying the content. The content ID is a numerical value such as 101 or 109, but may include alphabets or symbols.

The second column from the left of the content table is information indicating the content name representing the name of the content. The content name is, for example, a character string such as sweet and sour pork or eight treasures.

The third and subsequent columns from the left of the content table are refinement information that is a combination of the refinement key assigned to each content and the refinement condition. The narrowing down information corresponds to the classification information indicating the classification of each content.

For example, in the content table shown in FIG. 6, the content of "content ID = 101" has the content name "sweet and sour pork", and the narrowing key and the narrowing condition are "temperature (warm) = YES" and "genre (Chinese food)". ) = YES ”,“ Genre (Japanese style) = NO ”, etc. are included. The narrowing information shown in FIG. 6 is a part, and may further include a narrowing key of a type not shown in FIG. 6 and a narrowing condition.

The narrowing down processing unit 108 performs the narrowing down processing of the contents stored in the content database 106 according to the instruction of the dialogue management unit 104, and holds the narrowing down result as the candidate contents. The narrowing down processing unit 108 performs the narrowing down processing by extracting the contents matching each narrowing down information from the contents in the content database 106 by using the narrowing down information held in the narrowing down information holding unit 107. The narrowing down processing unit 108 can be realized by a processor, a memory, or the like.

When the narrowing condition of "YES" or "NO" is set for the narrowing key, the narrowing process is performed by extracting only the contents that match the combination of the narrowing key and the narrowing condition. Further, when the narrowing condition of "NOT_CARE" is set for the narrowing key, the narrowing key may or may not match, and the content is extracted by extracting the content. In this way, the narrowing-down processing unit 108 extracts content that simultaneously satisfies a plurality of narrowing-down information, and holds the extracted content as candidate content.

For example, when the content of the content database 106 of FIG. 6 is narrowed down by the narrowing down information shown in FIG. 5, the content matching the search information "genre (Chinese food) = YES" is extracted, so that the content ID of FIG. = 101, 109 and 121 "sweet and sour pork", "eight treasure vegetables" and "Tenshin rice" are included in the results narrowed down by the narrowing down process.

The user intention estimation unit 110 estimates the user's intention according to the state of the dialogue history series. The dialogue management unit 104 determines what kind of dialogue operation is to be performed next according to the user intention estimation result determined by the user intention estimation unit 110 according to the state of the series of the dialogue history. The user intention estimation unit 110 can be realized by a processor, a memory, or the like.

In the user intention estimation unit 110, the response content immediately before the dialogue system 1 is "Q, narrowing down key", and the semantic information of the user's utterance to it is either "YES", "NO", or "NOT_CARE". If so, it is presumed that the user's intention is a "narrowing request". In that case, the dialogue management unit 104 adds the narrowing down key included in the response content and the narrowing down condition obtained from the semantic information of the user's utterance to the narrowing down information held in the narrowing down information holding unit 107. Above, the narrowing down processing unit 108 is instructed to perform the narrowing down processing.

For example, when "ID = 4" and "semantic information = YES" are obtained for the response contents of "ID = 3" and "Q, genre (Chinese)" of the dialogue history holding unit 109 in FIG. 4, the user The intention estimation unit 110 estimates that the user's intention is "narrowing down", and the dialogue management unit 104 instructs the narrowing down processing unit 108 to perform the content narrowing down processing.

On the other hand, the user intention estimation unit 110 estimates that the user's intention is a "candidate list request" when the dialogue history series held in the dialogue history holding unit 109 satisfies a predetermined condition. Here, as an example, a predetermined condition is a case where the meaning information of the user's utterance is "NOT_CARE" appears twice in a row with respect to the response content of "Q, narrowing down key".

For example, when the meaning information of the user's utterance of "ID = 8" of the dialogue history holding unit 109 in FIG. 4 is obtained, "NOT_CARE" is set for "Q, narrowing key" twice in a row in the response history. Has appeared. More specifically, utterances of semantic information of "ID = 6" and "NOT_CARE" have appeared in response to the response contents of "ID = 5" and "Q, category (salad)", and "ID = 7". Since the utterance of the semantic information of "ID = 8" and "NOT_CARE" appears in response to the response contents of "Q, ingredients (pork)", "Q, narrowing key" is displayed twice in a row in the response history. The utterance of the semantic information of "NOT_CARE" has appeared. Since the predetermined condition is satisfied at this point, the user intention estimation unit 110 estimates that the user's intention is a "candidate list request".

Here, the user's intention is provided by the user intention estimation unit 110 on the condition that the utterance of the semantic information of "NOT_CARE" appears for the response content of the "Q, narrowing key" twice in a row in the response history. Was presumed to be a "candidate list request", but other conditions can be used. For example, it may be a condition that "NOT_CARE" appears for "Q, narrowing down key" three or more times in a row. Further, it may be a condition that "NOT_CARE" appears twice in total for "Q, narrowing down key", or a predetermined number of times of 3 or more in total.

The judgment method itself may be a method that requires that a specific dialogue pattern, that is, a pattern of response and utterance occurs, as described above, or the history of dialogue between people and the user's intention estimation result for the history. A method of realizing the data annotated with and as learning data by machine learning may also be used.

When the user intention estimation unit 110 estimates that the user's intention is a "candidate list request", the dialogue management unit 104 generates a candidate list display generation unit so as to generate a list of candidate contents held by the narrowing down processing unit 108. Instruct 111.

The candidate list display generation unit 111 generates candidate list text in the form of text information that can be displayed on the display unit 102 from the candidate contents obtained by the previous narrowing down process held by the narrowing down processing unit 108. The generated candidate list text is added as a part of the response text generated by the response generation unit 105, and is displayed by the display unit 102 as the response information of the dialogue system 1. In this way, when the dialogue management unit 104 determines that the predetermined condition is satisfied, the display unit 102 presents, that is, displays the information stored in the content database 106. An example of the above information is at least a part of the information narrowed down by the narrowing down process stored in the content database 106. The candidate list display generation unit 111 can be realized by a processor, a memory, or the like.

At the time of presentation by the display unit 102, a question may be presented to the user as to whether or not at least a part of the information stored in the content database 106 is presented in advance. Then, at least a part of the information stored in the content database 106 may be presented only when the user gives an affirmative response to the above question.

When the user intention estimation unit 110 estimates that the user's intention is a "candidate list request", and the narrowing condition is not set in the narrowing information holding unit 107, in other words, the response history has a number of 1 or more. When it is determined that only the three responses are included, the candidate list text may be generated by the candidate list display generation unit 111 after performing the narrowing process by the narrowing down processing unit 108 using predetermined information. .. This is to avoid that all the contents stored in the content database 106 are candidates when there is no narrowing condition.

Next, the processing flow of the dialogue system 1 of the present embodiment will be described. FIG. 7 is a flow chart showing a processing flow of the dialogue system 1 according to the present embodiment. FIG. 8 is a flow chart showing a process related to a determination as to whether or not to present a candidate list in the dialogue system 1 according to the present embodiment.

In step S101, the dialogue management unit 104 generates a question for narrowing down and outputs it as a response content. The question for narrowing down is to select one of the narrowing down keys assigned to the contents of the content database 106 from the narrowing down keys that do not appear in the narrowing down information held in the narrowing down information holding unit 107. To choose. This selection method may be a method of randomly selecting, a method of determining which narrowing key to be prioritized by a decision tree algorithm, or a method of following a fixed order in advance. good.

In step S102, the response generation unit 105 converts the response content (that is, the question) selected in step S101 into the response text.

In step S103, the display unit 102 displays the response text (that is, the question sentence) generated by the response generation unit 105 in step S103.

For example, when the response content is "Q, genre (Chinese food)", the text such as "Do you like Chinese food?" Is displayed as in the response text 801 of FIG.

In step S104, the input unit 101 acquires the utterance text of the user's utterance.

In step S105, the language understanding unit 103 performs language understanding processing of the utterance text acquired in step S104 and converts the user's utterance text into semantic information.

In step S106, the dialogue management unit 104 adds the semantic information converted in step S105 to the dialogue history holding unit 109. In other words, the dialogue management unit 104 holds a history of dialogues including a question to the user and a response by the user to the question.

In step S107, the user intention estimation unit 110 estimates the user's intention based on the dialogue history. The dialogue management unit 104 determines whether or not to perform the content candidate list display creation process based on the intention estimation result estimated by the user intention estimation unit 110 based on the dialogue history. In other words, the dialogue management unit 104 and the user intention estimation unit 110 determine whether or not the mode of appearance of the third response in the history of the response included in the history of the held dialogue satisfies a predetermined condition. The details of the determination in step S107 will be described later.

If it is determined in step S107 that the content candidate list display creation process is to be performed (YES in step S107), the process proceeds to step S110, and if not (NO in step S107), the process proceeds to step S108.

In step S108, the dialogue management unit 104 extracts the narrowing down key from the immediately preceding response content, extracts the narrowing down condition from the semantic information of the user's utterance, combines them, and adds them to the narrowing down information holding unit 107 as the narrowing down information. .. Further, the narrowing down processing unit 108 is instructed to perform the narrowing down processing, and the narrowing down processing unit 108 performs the narrowing down processing of the contents. Then, the process returns to step S101.

In step S110, the dialogue management unit 104 instructs the candidate list display generation unit 111 to create the candidate list display. The candidate list display generation unit 111 creates text information of a list of narrowed-down content candidates from the narrowing-down result of the content held in the narrowing-down processing unit 108 up to immediately before.

In step S111, the dialogue management unit 104 generates a response content asking which of the candidate contents to select. Then, the response generation unit 105 converts the response content generated by the dialogue management unit 104 into the response text. Further, the dialogue management unit 104 receives the text of the candidate list generated by the candidate list display generation unit 111 in step S108.

In step S112, the dialogue management unit 104 displays the response text and the candidate list text on the display unit 102. In other words, the display unit 102 presents the information stored in the content database 106 when the user intention estimation unit 110 determines that the predetermined condition is satisfied.

Next, the details of the process included in step S107 will be described with reference to FIG.

In step S201, the user intention estimation unit 110 determines whether or not the semantic information of the user's utterance obtained as a result of language understanding by the language understanding unit 103 is "NOT_CARE". If it is determined that the semantic information of the user's utterance is "NOT_CARE" (YES in step S201), the process proceeds to step S202, and if not, the process proceeds to step S205.

In step S202, the user intention estimation unit 110 counts the number of times that "NOT_CARE" is continuous in the user's utterance history in the dialogue history, going back from the present.

In step S203, the user intention estimation unit 110 determines whether or not the number of consecutive "NOT_CARE" is 2 or more in the user's utterance history in the dialogue history. If the number of consecutive "NOT_CARE" is 2 or more (YES in step S202), the process proceeds to step S204, and if not (YES in step S202), the process proceeds to step S205.

In step S204, the user intention estimation unit 110 estimates that the user's intention is a "candidate list request". Then, the dialogue management unit 104 determines that the candidate list should be presented to the user.

In step S205, the user intention estimation unit 110 estimates that the user's intention is a "narrowing request". Then, the dialogue management unit 104 determines that the candidate list should not be presented to the user.

After completing steps S204 and S205, the series of processes shown in FIG. 8 is completed, and the process proceeds to step S108 (FIG. 7).

Next, an operation example of the dialogue system 1 is shown. 9 and 10 are explanatory views showing an operation example of the dialogue system 1 according to the present embodiment.

First, the dialogue system 1 asks the user whether the genre of the dish to be searched may be Chinese style by the response text 801. The user gives an affirmative response to this inquiry by the utterance text 802. Based on the affirmative response from the user, the narrowing-down processing unit 108 refers to the content table and narrows down the content by extracting only those whose genre is Chinese.

Next, the dialogue system 1 asks the user about the cooking method and ingredients by the response texts 803 and 805. In response to each of these inquiries, the user utters the utterance texts 804 and 806 with the semantic information "NOT_CARE".

At this time, the utterance texts 804 and 806 correspond to the utterances in which the semantic information is "NOT_CARE", which appears twice in a row in the response history. Therefore, after the intention estimation by the user intention estimation unit 110, the dialogue management unit 104 determines that the candidate list should be presented to the user (step S204). Then, the response text 807 and the candidate list text 808 are presented (steps S111 and S112).

In this way, the narrowing down of the cooking menu and the display of the candidate list using the dialogue system 1 are realized.

Based on the above processing flow, the dialogue system 1 determines whether the user wants to continue narrowing down the content or browse the narrowed down content candidate list based on the history of the dialogue between the dialogue system 1 and the user. Estimate the intent appropriately. Thereby, it is possible to appropriately determine the timing of presenting the list of contents to the user from the situation of the dialogue without grasping the information as to whether or not the user wants to see the candidate list early.

Next, an example of applying the dialogue system 1 to the search of a writer, which is another content, will be described.

In this case, the content database 106 stores a plurality of information indicating each author. Then, the dialogue system 1 provides information indicating the desired writer of the user by using a dialogue including a question regarding the search for the writer to the user and a response by the user to the question.

FIG. 11 is an explanatory diagram showing a content table in another application example of the dialogue system 1 according to the present embodiment.

In the content table shown in FIG. 11, the name of the writer is included in the content name. In addition, the narrowing key includes the genre and category of the work of the artist.

12 and 13 are explanatory views showing an operation example in another application example of the dialogue system 1 according to the present embodiment.

First, the dialogue system 1 asks the user whether the genre of the writer to be searched may be a literary work by using the response text 901. The user responds affirmatively to this inquiry by the utterance text 902. Based on the affirmative response from the user, the narrowing down processing unit 108 refers to the content table and narrows down the content by extracting only those whose genre is a literary work.

Next, the dialogue system 1 inquires the user about the category by the response texts 903 and 905. In response to each of these inquiries, the user utters the utterance texts 904 and 906 with the semantic information "NOT_CARE".

At this time, the utterance texts 904 and 906 correspond to the utterances in which the semantic information is "NOT_CARE", which appears twice in a row in the response history. Therefore, after the intention estimation by the user intention estimation unit 110, the dialogue management unit 104 determines that the candidate list should be presented to the user (step S204). Then, the response text 907 and the candidate list text 908 are presented (steps S111 and S112).

In this way, the narrowing down of writers and the display of the candidate list using the dialogue system 1 are realized.

The dialogue processing method by the dialogue system 1 can be expressed as follows.

That is, the dialogue processing method by the dialogue system 1 is an information processing method regarding a menu preferred by the user, which uses a dialogue with the user in a chat format via the user's terminal. In this information processing method, the display provided in the terminal displays the first question asking for a request for a menu in a balloon type based on the database, and the database is a classification showing a plurality of menus and the classification of each of the plurality of menus. Corresponding to information, classification information includes menu hot / cold, menu genre, menu category, menu cooking method, ingredients used in the menu, and the first question is included in the database for the menu. Ask if you want to request the first category. Receive the first answer to the first question entered by the user through the terminal. If it is determined that the first answer contains words that affirm the first classification, request the display for the second classification corresponding to the first menu corresponding to the first classification based on the database. The second question is displayed in a balloon type, and the second classification is different from the first classification. Then, the second answer to the second question input by the user is received through the terminal. If it is determined that the second answer does not contain any words that affirm the second classification or deny the second classification, the display will display the third classification corresponding to the first menu based on the database. The third question asking if you want to request is displayed in a balloon type, and the third classification is different from the first classification and the second classification. Then, the third answer to the third question input by the user is received through the terminal. (A) If it is determined that the third answer contains words that affirm the third classification, do you request the fourth classification corresponding to the second menu corresponding to the third classification in the first menu? The fourth question asking whether or not is displayed in a balloon type, and the fourth classification is different from any of the first classification, the second classification, and the third classification. (B) If it is determined that the third answer contains words that deny the third classification, do you request the fifth classification corresponding to the third menu that does not correspond to the third classification in the first menu? The fifth question asking whether or not is displayed in a balloon type, and the fifth classification is different from any of the first classification, the second classification, the third classification, and the fourth classification. (C) If it is determined that the third answer does not include a phrase that affirms the third category or a phrase that denies the third category, the display may display information indicating the first menu. Display the fifth question asking if it is in a balloon type. Then, when the fifth answer to the fifth question entered by the user is received through the terminal, and it is determined that the fifth answer contains a phrase affirming that the information indicating the first menu is displayed. , Display the information indicating the first menu on the display.

In this way, similarly to the above, the narrowing down of writers and the display of the candidate list using the dialogue system 1 are realized.

(Modified example of the embodiment)
In this modification, a configuration in which an interactive system is realized by a plurality of devices communicably connected via a network will be described.

FIG. 14 is a block diagram showing a configuration of the dialogue system 1A according to the present modification. The same components as those in the above embodiment are designated by the same reference numerals, and detailed description thereof will be omitted.

The dialogue system 1A includes a client 2 and a server 3.

The client 2 includes an input unit 101, a display unit 102, and a client communication unit 112.

The server 3 includes a language understanding unit 103, a dialogue management unit 104, a response generation unit 105, a content database 106, a narrowing information holding unit 107, a narrowing processing unit 108, a dialogue history holding unit 109, and a user intention estimation. A unit 110, a candidate list display generation unit 111, and a server communication unit 113 are provided.

The client communication unit 112 and the server communication unit 113 are communicably connected via a network (not shown). Then, communication between the input unit 101 and the language understanding unit 103, communication between the response generation unit 105 and the display unit 102, and the like are realized by communication between the client communication unit 112 and the server communication unit 113 via the network.

With this configuration, the timing of presenting the list of candidates can be controlled more appropriately, as in the above embodiment.

Further, in the configuration of this modification, it is possible to realize a form in which a plurality of clients 2 and one server 3 are connected. With this configuration, for example, a terminal owned by each user can be used as a client 2, and the server 3 can be realized as a server (so-called cloud server) device capable of communicating from the terminal via a network. This facilitates the repair and maintenance work of the components included in the server 3 such as the language understanding unit 103 and the dialogue management unit 104, and has advantages such as reduction of processing load and cost reduction.

As described above, according to the information processing method of the present embodiment, the information processing method controls the timing of presenting a list of candidates from the content of the user's utterance in the dialogue with the user. The third response, which is neither an affirmative response nor a negative response, is also considered to be an expression of the user's intention not to provide clear narrowing conditions. Therefore, the mode of appearance of the third response can be a criterion for determining whether or not the user wants to see the list of content candidates. As a result, even if you do not know whether you want to see the list of content candidates narrowed down in advance by the user, you can estimate the user's intention from the content of the user's utterance to display the narrowed down list of content candidates. The timing of presentation can be controlled.

Further, as the information processing method, it is possible to realize the information processing method more concretely by using the condition that the third response appears consecutively a predetermined number of times, and to control the timing of presenting the list of candidates more appropriately. can.

Further, as the information processing method, it is possible to realize the information processing method more concretely by using the condition that the third response is accumulated and appear a predetermined number of times, and to control the timing of presenting the list of candidates more appropriately. can.

In addition, the information processing method allows the user to browse the narrowed list of content candidates based on the intention estimation. In this way, by estimating the user's intention from the content of the user's utterance, it is possible to control the timing of presenting the narrowed list of content candidates.

In addition, the information processing method can present a question as to whether or not the user is allowed to browse the narrowed list of content candidates based on the intention estimation. In the timing control based on the intention estimation, the user may try to present the list of content candidates at a timing different from the timing actually intended. Therefore, by confirming the actual intention of the user by the above question, it is possible to suppress the presentation of the list of content candidates contrary to the intention of the user.

Further, the information processing method can present the list of content candidates and allow the user to browse after confirming that the presentation of the list of content candidates matches the actual intention of the user.

In addition, the information processing method can narrow down the information stored in the storage unit by the interface of dialogue and present the narrowed down information to the user. As a result, the user can search for information based on sequential information input by dialogue, and can receive proposals from the dialogue system, unlike searching for information by a search engine or the like.

Further, the information processing method can avoid displaying all the information stored in the storage unit, that is, the information that has not been narrowed down at all, as a list when there is no narrowing down condition.

In addition, the information processing method can perform a more specific search and suggestion based on a closed question that provides options and selects an answer and a dialogue composed of the answers.

Further, the information processing method can provide the user with a search or suggestion of the cooking menu by using the narrowing down of the information indicating the cooking menu.

In addition, the information processing method can provide a user with a search or suggestion for a writer by using the narrowing down of information indicating the writer.

In addition, a part or all of the above-described embodiment may be expressed as follows, but is not limited to the following description.

(1) This is a narrowing-down result presentation control method in a dialogue system in which information for narrowing down a plurality of contents to be searched or proposed is acquired by dialogue and the contents that meet the user's request are presented. From the history of the dialogue in which the user answers the question with a predetermined intention, the intention of the user to continue narrowing down the content or to browse the narrowed down candidate list is estimated, and the above-mentioned A narrowing-down result presentation control method that controls whether to present a list of narrowed-down content candidates at that time based on the estimation result of the user's intention.

(2) The above-mentioned feature is that when the sequence of user's answers to the dialogue system satisfies a predetermined condition in the dialogue history held in the dialogue system, it is estimated that the user's intention is to browse the list of candidates. The narrowing-down result presentation control method according to (1).

(3) Dialogue history data to which information on whether or not the user's intention requests viewing the candidate list is added to the dialogue history obtained from the dialogue between the dialogue system and the user or the dialogue between people. The narrowing-down result presentation control method according to (1) above, characterized in that it estimates whether or not the user's intention wants to browse a list of candidates by a user intention estimation method configured by machine learning as learning data.

With the above configuration, even if you do not know the information whether you want to see the list of content candidates narrowed down by the user in advance, whether the user wants to narrow down by answer based on the status of the user's answer to the system question. By estimating some intention, it is possible to control the timing of presenting the narrowed list of content candidates.

In each of the above embodiments, each component may be configured by dedicated hardware or may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU or a processor reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory. Here, the software that realizes the information processing apparatus and the like of each of the above embodiments is the following program.

That is, this program is an information processing method for information stored in a storage unit in a computer, and retains and retains a history of dialogues including a question to a user and a response by the user to the question. In the history of the response included in the history of the dialogue, the mode of appearance of the third response which is neither the first response indicating the affirmation of the question nor the second response indicating the denial of the question is predetermined. Is determined, and if it is determined that the predetermined condition is satisfied, an information processing method for presenting the information stored in the storage unit is executed.

Although the information processing apparatus and the like according to one or more aspects have been described above based on the embodiment, the present invention is not limited to this embodiment. As long as the gist of the present invention is not deviated, various modifications that can be conceived by those skilled in the art are applied to the present embodiment, and a form constructed by combining components in different embodiments is also within the scope of one or more embodiments. May be included within.

The information processing method according to the present disclosure can be used as an information processing method for more appropriately controlling the timing of presenting a list of candidates.

1, 1A Dialogue system 2 Client 3 Server 101 Input unit 102 Display unit 103 Language understanding unit 104 Dialogue management unit 105 Response generation unit 106 Content database 107 Narrowing information holding unit 108 Narrowing processing unit 109 Dialogue history holding unit 110 User intention estimation unit 111 Candidate list display generation unit 112 Client communication unit 113 Server communication unit 201, 202 Image 201A Input field 201B Send button 202A, 202B, 801, 803, 805, 807, 901, 903, 905, 907 Response text 202C, 802, 804, 806, 902, 904, 906 Speech text 808, 908 Candidate list text

Claims (13)

It is an information processing method related to the information stored in the storage unit.
Keep a history of dialogues, including questions to users and responses by the user to the questions.
In the history of the response included in the history of the dialogue held, the mode of appearance of the third response which is neither the first response indicating the affirmation of the question and the second response indicating the denial of the question is , Judge whether or not the prescribed conditions are met,
An information processing method for presenting the information stored in the storage unit when it is determined that the predetermined condition is satisfied. The information processing method according to claim 1, wherein the predetermined condition is that the third response appears consecutively a predetermined number of times in the history of the response. The information processing method according to claim 1, wherein the predetermined condition is that the third response is accumulated and appears a predetermined number of times in the history of the response. The information processing method according to any one of claims 1 to 3, which presents at least a part of the information stored in the storage unit at the time of the presentation. The information according to any one of claims 1 to 3, which presents a question to the user as to whether or not to present at least a part of the information stored in the storage unit at the time of the presentation. Processing method. The information processing method according to claim 5, wherein at the time of the presentation, at least a part of the information stored in the storage unit is presented when an acknowledgment by the user is obtained in response to the question. In the information processing method, further
Using the response included in the dialogue, a process of narrowing down the information stored in the storage unit is performed.
The information processing method according to any one of claims 1 to 6, wherein at the time of the presentation, the presentation regarding the information stored in the storage unit after the processing is performed. In the information processing method, further
When it is determined that only one or more of the third responses are included in the response history, a process of narrowing down the information stored in the storage unit is performed using predetermined information.
The information processing method according to any one of claims 1 to 7, wherein at the time of the presentation, the presentation regarding the information stored in the storage unit after the processing is performed. The question is described in any one of claims 1 to 8, which is a closed question for causing the user to make any one of the first response, the second response, and the third response. Information processing method. In the storage unit, as the information, a plurality of information indicating each cooking menu is stored.
The information processing method according to any one of claims 1 to 9, wherein the dialogue includes a question regarding a search for a cooking menu to the user and the response by the user to the question. In the storage unit, as the information, a plurality of information indicating each writer is stored.
The information processing method according to any one of claims 1 to 9, wherein the dialogue includes a question regarding the search for a writer to the user and the response by the user to the question. An information processing device for performing information processing on information stored in a storage unit.
A dialogue history holding unit that holds a history of dialogues including a question to a user and a response by the user to the question.
In the history of the response included in the history of the dialogue held, the mode of appearance of the third response which is neither the first response indicating the affirmation of the question and the second response indicating the denial of the question , The dialogue management unit that determines whether or not the prescribed conditions are met,
An information processing device including a presentation unit that presents information stored in the storage unit when it is determined that the predetermined condition is satisfied. A program for causing a computer to execute the information processing method according to claim 1.

Images