JP2023123306A - Information processor and program - Google Patents

Abstract

To provide an information processor and a program, which make a response to utterance of a user so that the user and a chatbot can smoothly continue conversation.SOLUTION: An information processor includes: a first estimation part for estimating intention of user input on the basis of the user input; a first specification part for specifying an utterance promotion word for promoting continuous utterance by a user on the basis of the intention and a history of the user input; and a generation part for generating a response sentence on the basis of the intention and the utterance promotion word.SELECTED DRAWING: Figure 1

Description

TECHNICAL FIELD Embodiments of the present invention relate to an information processing apparatus and a program.

2. Description of the Related Art In recent years, there has been known a system (hereinafter also referred to as a chatbot) that automatically responds to input of voice, characters, etc. by a user. Conventionally, in order for the conversation between the user and the chatbot to proceed continuously and smoothly, based on the history of the lines selected by the user from among multiple options prepared by the system in advance, the chatbot will generate "uh" Techniques have been proposed for inserting words (hereinafter also referred to as fillers) that express emotions and the like that are interposed between conversations, such as "ah".

JP 2020-24522 A

In the method described above, in which filler is inserted based on the selection history of lines prepared in advance by the system, assuming a conversation in which filler words are inserted from a normal conversation, the assumed line is uttered by the user. may not be available without Also, the filler inserted based on the selection history may not be emotionally relevant to the current conversation. For this reason, the method of inserting fillers may result in an unconfident response or a response consisting of complicated sentences.

Accordingly, one example of the problem to be solved by the present invention is to provide an information processing device and program capable of responding to user's utterances so that the user and the chatbot can smoothly continue the conversation.

A first aspect of the present invention includes: a first estimating unit that estimates an intention of the user input based on the user input; and a response sentence to the user input based on the intention and the history of the user input. an information processing apparatus comprising: a first identifying unit that identifies a speech-promoting word that encourages continuous utterance by a user; and a generating unit that generates the response sentence based on the intention and the speech-promoting word is.

Further, the speech prompting word may include an interest word that attracts the interest of the user who has performed the user input, and the generation unit may generate the response sentence including the interest word.

The information processing device further includes: a second estimation unit that estimates a topic of the user input based on the user input; a second specifying unit that specifies linked words that form a response sentence in collaboration with the interesting word that is used, and the generator may create the response sentence that includes the sympathy word and the linked word. .

an extraction unit for extracting an important word in the topic from a plurality of words included in the content of the user input; and the utterance-promoting word used for generating a response sentence to the user input based on the topic and the important word. and a third specifying unit that specifies a sympathy word that expresses sympathy for the user input included in the above, and the generating unit generates the response sentence that includes the sympathy word.

Further, the information processing apparatus includes: a second estimating unit that estimates a topic of the user input based on the user input; A fourth specifying unit may further comprise a fourth specifying unit for specifying, among words, repeated words included in the speech-prompting words to be used for generating a response sentence to the user input.

Also, the fourth identifying unit may identify a plurality of the repeated words.

A second aspect of the present invention includes processing for estimating an intention of the user input based on the user input, and generating a response sentence to the user input based on the intention and the history of the user input. The computer of the information processing device executes a process of identifying a speech-promoting word that promotes continuous utterance by the user, and a process of generating the response sentence based on the intention and the speech-promoting word. It is a program that allows

ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to respond to a user's utterance so that a user and a chatbot can continue conversation smoothly.

FIG. 1 is a diagram schematically showing an example of the configuration of an information processing system according to an embodiment. FIG. 2 is a diagram schematically showing an example of the structure of a word category database (DB) according to the embodiment; FIG. 3 is a diagram schematically showing an example of the structure of an intention estimation database (DB) according to the embodiment; FIG. 4 is a diagram schematically showing an example of the structure of an important word database (DB) according to the embodiment. FIG. 5 is a diagram schematically showing an example of the structure of a conversation estimation database (DB) according to the embodiment; FIG. 6 is a diagram schematically showing an example of the structure of an interesting word database (DB) according to the embodiment. FIG. 7 is a diagram schematically showing an example of the structure of a linked word database (DB) according to the embodiment. FIG. 8 is a diagram schematically showing an example of the structure of a fixed phrase database (DB) according to the embodiment; FIG. 9 is a flowchart illustrating an example of information processing executed in the information processing system according to the embodiment; FIG. 10 is a diagram schematically illustrating an example of conversation realized by information processing according to the embodiment; FIG. 11 is a diagram schematically illustrating an example of a hardware configuration of a computer that implements the information processing system according to the embodiment;

Embodiments of an information processing apparatus and a program according to embodiments will be described in detail below with reference to the accompanying drawings.

FIG. 1 is a diagram schematically showing an example of the configuration of an information processing system 1 according to an embodiment. The information processing system 1 according to the embodiment receives input from a user and analyzes the input content. Further, the information processing system 1 automatically generates a response to the input content according to the analysis result, and outputs the response to the user. In the present embodiment, voice data indicating the contents of the user's utterance is exemplified as the input by the user, but the input is not limited to this. The input by the user may be text data.

As shown in FIG. 1, the information processing system 1 has a terminal device 100 and a server device 200 . The terminal device 100 and the server device 200 are communicably connected to each other via a network (not shown) such as the Internet.

As an example, the information processing system 1 is a chatbot system that implements a conversation between a user using the terminal device 100 and a chatbot implemented by the server device 200 . Hereinafter, the function realized by the server device 200 may also be described as a chatbot.

The terminal device 100 has a communication section 110, an input section 120 and an output section 130, as shown in FIG.

The communication unit 110 performs communication between the terminal device 100 and another device (the server device 200 in the example shown in FIG. 1).

The input unit 120 receives an input of voice data according to the user's speech of the terminal device 100 . Input unit 120 transmits the received voice data to server device 200 through communication unit 110 .

When the communication unit 110 receives a response from the server device 200, the output unit 130 notifies the user of the response by voice or image.

The server device 200 includes a communication unit 210, a text analysis unit 220, a word category database (DB) 221, a word analysis unit 230, an intention estimation database 231, an important word database 232, a conversation estimation database 233, an interesting word database 234, and a response generation unit. 240 , linked word database 241 , fixed phrase database 242 and management unit 250 .

The communication unit 210 performs communication between the server device 200 and another device (the terminal device 100 in the example shown in FIG. 1).

The text analysis unit 220 analyzes the voice data representing the user's utterance input by the user using the terminal device 100 through voice recognition, and generates text data representing the user's utterance. Also, the text analysis unit 220 acquires morphemes contained in text data representing user utterances by a known morphological analysis.

In other words, the sentence analysis unit 220 divides the user's utterance converted into text into words by morphological analysis. The text analysis unit 220 selects extracted words from the words obtained by word segmentation using the word category database 221 . Also, the text analysis unit 220 estimates a topic corresponding to the extracted word. The text analysis unit 220 supplies the words obtained by word segmentation and the estimated topics to the word analysis unit 230 .

The word category database 221 is a database in which extracted words and topics corresponding to each extracted word are registered.

The information registered in the word category database 221 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be generated artificially. may be

FIG. 2 is a diagram schematically showing an example of the structure of the word category database 221 according to the embodiment.

In the example shown in FIG. 2, "[clothing item]", "suitable", "fashionable", "trend", and "buy" are registered as extracted words corresponding to the topic "fashion". Note that "[clothing item]" represents a specific clothing name (for example, a dress).

Also, "spring", "early summer", "summer", "autumn", "late autumn", "winter", "rainy season" and "season" are registered as extracted words corresponding to the topic "season". As extracted words corresponding to the topic "travel", "travel", "Hawaii", "Guam", "Tokyo", "Kyoto", "Hokkaido", "go" and "next time" are registered. Also, as extracted words corresponding to the category of the topic "fashion proposal", "look good", "coordination", "clothes", "dinner", "fashionable", "budget" and "boyfriend" are registered.

If the extracted word exists in a plurality of topics, the topic that contains many extracted words may be estimated. Alternatively, when an extracted word exists in a plurality of topics, it is also possible to adopt a configuration in which the top topics are estimated in No. order.

The word analysis unit 230 uses the intention estimation database 231 to estimate the intention of the user's utterance, that is, the utterance intention from the words input from the text analysis unit 220 . In other words, the word analysis unit 230 estimates the category representing the utterance intention of the user's utterance based on the intention estimation database 231 and the words obtained by word division.

The word analysis unit 230 also uses the important word database 232 to identify important words, which are highly important words, from the words input from the text analysis unit 220 . Further, word analysis section 230 identifies repeated words based on the identified important words. For example, the word analysis unit 230 identifies repetitive words according to a predetermined priority order of important words.

Note that the word analysis unit 230 may identify multiple repeated words. Further, when the word analysis unit 230 cannot identify the important word, the word analysis unit 230 may omit the process of identifying repeated words.

Then, the word analysis unit 230 adds the identified repetitive word as a repetitive word to be used for generating a response sentence.

In addition, the word analysis unit 230 uses the conversation estimation database 233 to identify sympathy words expressing sympathy for the user's utterance based on the estimated topic and the identified repetitive words. Note that the word analysis unit 230 may omit the process of identifying the sympathy word when the sympathy word cannot be identified from the repeated words. Then, the word analysis unit 230 adds the specified sympathy words as sympathy words used for generating the response sentence.

In addition, the word analysis unit 230 uses the interesting word database 234 to determine the estimated category and the behavior of the user that occurred prior to the user's utterance (for example, the chat with the user that occurred prior to the target user's utterance). An interesting word, which is a word that attracts the user's interest, is specified from the previous conversation information representing the conversation with the bot. Then, word analysis section 230 adds the specified interesting word as an interesting word to be used for generating a response sentence.

Also, the word analysis unit 230 supplies the added repetitive words, sympathetic words, and interesting words to the response generation unit 240 .

The intention estimation database 231 is a database in which extracted words corresponding to categories representing each utterance intention are registered. In other words, the intention estimation database 231 is a database in which correspondences between words (extracted words) and utterance intentions indicating categories including the words are registered.

The information registered in the intention estimation database 231 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be generated artificially. may be

FIG. 3 is a diagram schematically showing an example of the structure of the intention estimation database 231 according to the embodiment.

In the example shown in FIG. 3, "search", "search", and "tell me" are registered as extracted words corresponding to the category "search". Also, as extracted words corresponding to the category "question", "Do you want to do it?", "Yes", "What do you think?"

Also, as extracted words corresponding to the category "greetings", "How are you?", "Hello", "Good morning", "Nice", and "Welcome" are registered. Extracted words corresponding to the category "chat" are "today", "next time", "do", "go", "wear", "saw", "hot", "cold", "how", "think" , "know", "condition", "trip", "buy", and "determine" are registered.

In addition, as extracted words corresponding to the category "backhand", "aah", "yes", "that's right", "is that true", "true", "great", and "agree" are registered.

Note that if the extracted words exist in a plurality of categories, for example, the category containing many extracted words may be estimated. Alternatively, when an extracted word exists in a plurality of categories, it is also possible to adopt a configuration in which a higher category is estimated in order of No.

The important word database 232 is a database in which important words are registered for each topic. In other words, the important word database 232 is a database in which correspondence between topics and important words in the topic is registered. The information registered in the important word database 232 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be generated artificially. may be

FIG. 4 is a diagram schematically showing an example of the structure of the important word database 232 according to the embodiment.

In the example shown in FIG. 4, "body", "material", "wear", "comfortable", "[clothing item]", and "suit" are registered as important words corresponding to the topic "fashion". . Also, as important words corresponding to the topic "season", "first visit of the year", "setsubun", "Valentine's Day", "first spring", and "doll's festival" are registered.

Also, "travel", "go", "picture", "video", "shoot", "reservation", and "[place]" are registered as important words corresponding to the topic "travel". Note that "[location]" represents the name of a specific location (for example, Tokyo). Also, "boyfriend", "birthday party" and "thank you party" are registered as important words corresponding to the topic "fashion proposal". Also, "weather", "season", "hot", "cold", "rainy", "sunny" and "cloudy" are registered as important words corresponding to the topic "weather".

Also, "Shibuya", "cafe" and "amusement park" are registered as important words corresponding to the topic "hobby". Also, "music", "reading", "fishing" and "golf" are registered as important words corresponding to the topic "entertainment". Also, "food" and "present" are registered as important words corresponding to the topic "health".

The conversation estimation database 233 is a database in which sympathetic words are registered for each set of topics and important words. In other words, the conversation estimation database 233 is a database in which correspondence between pairs of topics and important words and sympathetic words in the pairs is registered.

The information registered in the conversation estimation database 233 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be generated artificially. may be

FIG. 5 is a diagram schematically showing an example of the structure of the conversation estimation database 233 according to the embodiment.

In the example shown in FIG. 5, "perfect fit" is registered as a sympathy word corresponding to the set of the topic "fashion" and the important words "[clothing item]", "buy". Also, as a sympathy word corresponding to the set of the topic "fashion" and the important words "[clothing item]", "something", and "trend", "let's look for it" is registered. Also, "nice" is registered as a sympathy word corresponding to the set of the topic "fashion" and the important words "[clothing item]" and "sense".

Also, "I wonder if the season has changed yet" is registered as a sympathy word corresponding to the set of the topic "season" and the important words "[season]", "cold", and "hot". Also, as a sympathy word corresponding to the set of the topic "season" and the important words "[season]", "sunny", and "rain", "I wonder if this kind of weather will continue" is registered.

Also, "nice" is registered as a sympathetic word corresponding to the set of the topic "travel" and the important words "[place]", "travel", and "go". Also, "That's a good shopping" is registered as a sympathy word corresponding to the set of the topic "travel" and the important words "[item]", "travel", and "buy". Also, "[Date and time] is a nice season" is registered as a sympathy word corresponding to the set of the topic "Travel" and the important words "[Date and time]", "Travel", and "Plan"".

The interesting word database 234 is a database in which interesting words corresponding to sets of categories and extracted words are registered. In other words, the conversation estimation database 233 is a database in which correspondence between pairs of categories and extracted words and interesting words in the pairs is registered.

The information registered in the interesting word database 234 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be generated artificially. may be

FIG. 6 is a diagram schematically showing an example of the structure of the interesting word database 234 according to the embodiment.

In the example shown in FIG. 6, "[date and time]", "[place]", "seat ” is registered. Note that [reservation] represents specific reservation contents (for example, concert ticket reservation, etc.).

Also, "[store name]" and "[place]" are registered as words of interest corresponding to the combination of the category "search" and the extracted words "[reservation]" and "[store name type]". ing. It should be noted that "[store name type]" represents a specific store name and store genre (for example, ○○ya, Italian, etc.). In addition, "[place]" and "[when]" are registered as words of interest corresponding to the combination of the category "search" and the extracted words "[sale]" and "[item name]". there is Note that "[when]" represents a specific date and time (for example, February 2).

In addition, "[part]", "[season]", and "[time]" are registered as words of interest corresponding to the set of the category "chat" and the extracted words "[part]" and "tone". It is Note that "[part]" represents a specific part of the body (for example, abdomen).

Also, as interesting words corresponding to the set of the category "chat" and the extracted words "[clothing item]" and "suit", "[clothing item]", "think", "sense", and "stylish". is registered. Also, "[weather]" and "[season]" are registered as words of interest corresponding to the set of the category "chat" and the extracted words "[weather]" and "outing".

The response generation unit 240 uses the linked word database 241 to identify, from the interesting words input from the word analysis unit 230, repeated words, sympathetic words, and linked words that form a response sentence in cooperation with the interesting words. Note that the response generation unit 240 may omit the process of identifying the linked word when the linked word cannot be specified from the interesting word.

In addition, the response generation unit 240 uses the standard sentence database 242 to generate a response sentence from repeated words, sympathetic words, interesting words, and linked words. The response generation unit 240 transmits the response text to the terminal device 100 through the communication unit 210 .

Note that the response generation unit 240 may generate a response sentence without repeated words, sympathetic words, and linked words. Further, the response generation unit 240 may perform processing for determining whether or not the generated response text is meaningful based on a predetermined criterion.

The linked word database 241 is a database in which linked words corresponding to sets of conversation categories, conversation topics, and interesting words are registered. In other words, the linked word database 241 is a database in which correspondences between pairs of conversation categories, conversation topics and interesting words, and linked words in the pairs are registered.

The information registered in the linked word database 241 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be generated artificially. may be

FIG. 7 is a diagram schematically showing an example of the structure of the linked word database 241 according to the embodiment.

In the example shown in FIG. 7, "wear" is registered as a linked word corresponding to the set of the category "chat", the topic "fashion", and the word of interest "[clothing item]". Also, "[color] is cutting-edge" is registered as a linked word corresponding to the set of the category "chat", the topic "fashion", and the word of interest "trend". Note that "[color]" represents a specific color (for example, red). In addition, "[location] suits you" is registered as a linked word corresponding to the combination of the category "chat", the topic "fashion", and the word of interest "[item name]".

In addition, as a linked word corresponding to a set of the category "chatter", the topic "travel", and the word of interest "[place]", "what you have done" is registered. Also, "[clothing item]", "sense"" is registered as a linking word corresponding to the combination of the category "chat", the topic "travel", and the word of interest "[clothing item]". In addition, as a linked word corresponding to the combination of the category "chat", the topic "travel", and the word of interest "[weather]", "outing", "Let's look for a good place in "[weather]"" is registered. ing.

The standard sentence database 242 is a database in which standard sentences corresponding to sets of categories, topics, and interesting words are registered. In other words, the standard sentence database 242 is a database in which pairs of categories, topics, and interesting words and standard sentences corresponding to the pairs are registered.

The fixed phrases registered in the fixed phrase database 242 may be generated by a machine learning model or the like in which parameters are learned based on responses from past users, or may be artificially generated. can be anything.

FIG. 8 is a diagram schematically showing an example of the structure of the fixed phrase database 242 according to the embodiment.

In the example shown in FIG. 8, as a fixed sentence corresponding to a set of the category "chat", the topic "fashion", and the word of interest "[clothing item]", "[repeated word] is [repeated word], [empathy word], [interesting word] and [linked word]?" are registered. In addition, as a set of fixed sentences corresponding to the set of the category "chat", the topic "fashion", and the interesting word "fashion", "Is the [repeated word] of [repeated word], [sympathy word], [interested word] are [Collaboration word] desu" is registered.

In addition, as a fixed phrase corresponding to the set of the category "chat", the topic "fashion", and the word of interest "[item name]", "From [repeated word] to [repeated word], isn't it? [Empathy word], [ [interesting word] is registered as [linked word] kana".

Also, as a set of fixed sentences corresponding to the set of the category "chat", the topic "travel", and the interesting word "[place]", "I'm planning [repeated word] of [repeated word], [sympathy word ], and [interesting word] has [linked word]" are registered.

Also, as a set of fixed sentences corresponding to the set of the category "chat", the topic "travel", and the word of interest "[sightseeing item]", "[repeated word] is [repeated word]. If you go to [repeated word] How about [interesting words]" is registered. Note that "[sightseeing item]" represents a specific name of a souvenir or the like (for example, XX ware, etc.).

In addition, as a fixed sentence corresponding to a set of the category "chat", the topic "travel", and the interesting word "[meal]", "[repeated word] eat [repeated word]. Now [interested word] is [linked word], looks delicious" is registered. Note that "[meal]" represents a specific name of a food material (for example, mackerel, etc.).

The management unit 250 manages user information that identifies the terminal device 100 and the user. In addition, the management unit 250 transmits request information requesting the user's voice input to the terminal device 100 through the communication unit 210 .

The word category database 221, the intention estimation database 231, the important word database 232, the conversation estimation database 233, the interesting word database 234, the linked word database 241, and the phrase database 242 shown in FIGS. This is an example and can be set arbitrarily.

FIG. 9 is a flowchart showing an example of information processing executed in the information processing system 1 according to the embodiment. Here, a case where the flow of FIG. 9 is executed by the server device 200 is illustrated. Also, as a premise, it is assumed that the terminal device 100 has input a user's speech.

First, the management unit 250 cooperates with the communication unit 210 to acquire voice data representing user utterances from the terminal device 100 (S101).

The text analysis unit 220 performs voice recognition processing and sentence word segmentation processing on the acquired voice data representing the user's utterance (S102).

After that, the text analysis unit 220 extracts words for estimating the topic from the word category database 221 (S103). Specifically, the sentence analysis unit 220 extracts words registered as extracted words in the word category database 221 from among a plurality of words obtained by dividing the user's utterance into sentence words.

After that, the text analysis unit 220 estimates a topic from the extracted words (S104). Specifically, the text analysis unit 220 refers to the word category database 221 and estimates the topic corresponding to the extracted word extracted in S103 as the topic of the conversation.

After that, the word analysis unit 230 confirms whether category discrimination is possible (S105). Specifically, the word analysis unit 230 first extracts words for estimating the category from the intention estimation database 231 .

Next, the word analysis unit 230 checks whether or not there is an extracted word corresponding to the category in the intention estimation database 231 . If the extracted word is not in the intention estimation database 231, the word analysis unit 230 determines that the category cannot be determined. On the other hand, when the extracted word exists in the intention estimation database 231, the word analysis unit 230 determines that the category can be determined. If the category cannot be determined (S105: No), the word analysis unit 230 stops processing (S118). The flow of FIG. 9 then ends.

On the other hand, if the category can be determined (S105: Yes), the word analysis unit 230 estimates the category of the conversation (S106). Specifically, the word analysis unit 230 refers to the intention estimation database 231, and corresponds to a set of a plurality of extracted words included in a plurality of words divided by dividing the user's utterance into sentence words. Infer the category as the category of the conversation.

After that, the word analysis unit 230 determines whether or not there is an important word in the estimated topic among the plurality of words divided by the sentence word division processing (S107). Specifically, the word analysis unit 230 refers to the important word database 232, and among the plurality of words obtained by dividing the user's utterance into sentence words, there is an important word corresponding to the topic estimated in S104. Determine whether or not there is If there is no important word in the estimated topic (S107: No), the flow of FIG. 9 advances to the process of S118.

On the other hand, if there is an important word in the estimated topic (S107: Yes), the word analysis unit 230 identifies the important word included in the user's utterance (S108). Specifically, the word analysis unit 230 refers to the important word database 232 and identifies important words corresponding to the estimated topic and included in the plurality of words divided by the sentence word division processing. Identifies words as key words in user utterances.

After that, the word analysis unit 230 identifies repeated words from the identified important words (S109). Specifically, when one important word is identified, word analysis section 230 identifies the important word as a repeated word. On the other hand, when a plurality of important words are identified, the word analysis section 230 identifies repeated words according to the priority of the important words predetermined for each topic.

The order of priority of important words is determined, for example, based on the number of times the word has appeared in conversations that have occurred so far. Also, the word analysis unit 230 may identify a plurality of repeated words.

After that, the word analysis unit 230 identifies sympathetic words from the conversation estimation database 233 based on the estimated topic and repeated words (S110). Specifically, the word analysis unit 230 refers to the conversation estimation database 233, and identifies sympathy words corresponding to the estimated topic and set of repeated words as sympathy words in the conversation.

After that, the management unit 250 acquires the previous conversation information of the user who has input the user's utterance (S111). Specifically, the management unit 250 acquires the user's previous conversation information stored in the storage 730 or the like, which will be described later. Here, the previous conversation information includes not only information related to the previous conversation between the chatbot and the target user, but also information such as the browsing history of the target user's Web pages.

After that, the text analysis unit 220 performs text extraction processing, text recognition processing, and text word segmentation processing on the acquired previous conversation information (S112).

After that, the text analysis unit 220 extracts words for estimating the topic from the word category database 221 in the same process as S103 (S113).

After that, the text analysis unit 220 estimates a topic from the extracted words in the same process as S104 (S114).

After that, the word analysis unit 230 confirms whether or not the category can be determined by the same process as in S105 (S115). If the category cannot be determined (S115: No), the flow of FIG. 9 returns to the process of S111, and the management unit 250 acquires other previous conversation information. If the category cannot be determined even after repeating the processes of S111 to S115 a predetermined number of times, the process of S118 may be performed and the flow of FIG. 9 may be ended.

On the other hand, if the category can be determined (S115: Yes), the word analysis unit 230 estimates the category of the previous conversation information by the same process as in S106 (S116).

After that, the word analysis unit 230 determines whether or not the category of the current user utterance and the category of the previous conversation information are the same category (S117). If the categories are not the same (S117: No), the flow in FIG. 9 returns to the process of S111, and the management unit 250 acquires another previous conversation information.

On the other hand, if the categories are the same (S117: Yes), the word analysis unit 230 extracts the common category between the current user utterance and the previous conversation information (hereinafter also referred to as the common category) estimated from the set of extracted words. ) is specified (S119).

Specifically, first, the word analysis unit 230 refers to the interesting word database 234, and selects the common category estimated in S117 from among the plurality of words divided by the sentence word division processing of the previous conversation information. Extract the extracted word corresponding to . Word analysis unit 230 also refers to interesting word database 234 to identify an interesting word corresponding to a set of a common category and an extracted word from previous conversation information.

After that, the word analysis unit 230 checks whether or not the previous conversation information includes the interesting word specified in S119 (S120). Specifically, word analysis unit 230 confirms whether or not the interesting word specified in S119 exists in the plurality of words divided by the sentence word division processing of the previous conversation information.

If no interesting word is included in the previous conversation information (S120: No), the word analysis unit 230 identifies an interesting word to be used for generating a response sentence based on the interesting word database 234 (S121).

Specifically, when one interesting word is specified in S119, the word analysis unit 230 specifies the interesting word as an interesting word to be used for generating a response sentence. On the other hand, if a plurality of interesting words are specified in S119, the word analysis unit 230 selects the interesting words among the interesting words specified in S119 according to the predetermined priority of the interesting words for each set of the category and the extracted word. , the interesting word used to generate the response sentence is specified.

Note that the order of priority of interesting words is determined, for example, based on the number of times the word has appeared in the conversation history of a plurality of users. Moreover, the word analysis unit 230 may specify a plurality of words as interesting words to be used for generating a response sentence.

On the other hand, if the previous conversation information includes an interesting word (S120: Yes), the word analysis unit 230 identifies one interesting word to be used for generating a response sentence based on the previous conversation information (S122).

Specifically, when one interesting word is specified in S119 and the interesting word is included in the previous conversation information, the word analysis unit 230 specifies the interesting word as an interesting word to be used for generating a response sentence. . On the other hand, when a plurality of interesting words are specified in S119 and one of the plurality of interesting words is included in the previous conversation information, word analysis section 230 responds with one interesting word included in the previous conversation information. It is specified as an interesting word used for sentence generation.

In addition, when a plurality of interesting words are specified in S119 and more than one of the plurality of interesting words is included in the previous conversation information, the word analysis unit 230 selects, for example, the target user The interesting word that appears most frequently in the conversation is specified as the interesting word to be used for generating the response sentence.

After that, the response generator 240 identifies linked words from the linked word database 241 based on the estimated topic, category, and specified interesting words (S123). Specifically, the response generation unit 240 refers to the linked word database 241 and identifies linked words corresponding to a set of the estimated topic, the estimated category, and the specified interesting word.

After that, the response generation unit 240 identifies fixed phrases from the fixed phrase database 242 based on the estimated topic, category, and specified interesting words (S124). Specifically, response generation unit 240 refers to fixed phrase database 242 to identify fixed phrases corresponding to a set of the estimated topic, the estimated category, and the specified interesting word.

After that, the response generation unit 240 generates a response sentence based on the identified repetitive words, sympathetic words, interesting words, linked words, and standard sentences (S125). Specifically, the response generating unit 240 adds the repetitive word identified in S109, the sympathetic word identified in S110, the interesting word identified in S121 or S122, and the Then, a response sentence is generated by applying the linked words.

After that, the response generation unit 240 outputs the response text to the terminal device 100 through the communication unit 210 (S126). Here, when the communication unit 110 receives the response text from the server device 200, the output unit 130 of the terminal device 100 notifies the user of the response text by voice or image.

After that, the management unit 250 confirms whether or not the next user's utterance is input within a predetermined time (S127). If the next user's utterance is not input within the predetermined time (S127: No), the response generation unit 240 performs re-speech processing of outputting a sentence prompting the user to utter to the terminal device 100 (S135). For example, the response generator 240 outputs sentences such as "I don't understand, please tell me more about it." After that, the flow of FIG. 9 returns to the processing of S127.

On the other hand, if the next user's utterance is input within the predetermined time (S127: Yes), the management unit 250 acquires voice data representing the user's utterance in the same process as S101 (S128).

After that, the text analysis unit 220 performs voice recognition processing and text word segmentation processing in the same manner as in S102 (S129).

After that, the text analysis unit 220 extracts words for estimating the topic from the word category database 221 in the same process as S103 (S130).

After that, the text analysis unit 220 estimates a topic from the extracted words in the same process as S104 (S131).

After that, the word analysis unit 230 confirms whether or not the category can be determined by the same process as in S105 (S132). If the category cannot be determined (S132: No), the flow of FIG. 9 advances to the process of S135.

On the other hand, if the category can be determined (S132: Yes), the word analysis unit 230 estimates the category of the user's utterance acquired in S128 (S133) in the same process as in S106.

After that, the word analysis unit 230 determines whether or not the category of the current user utterance and the category of the previous user utterance are the same category (S134). If the categories are not the same (S134: No), the flow of FIG. 9 proceeds to the process of S135. If the category cannot be determined even after repeating the processes of S127 to S134 a predetermined number of times, the process of S118 may be performed and the flow of FIG. 9 may be ended.

On the other hand, if the categories are the same (S134: No), the flow of FIG. 9 returns to the process of S107, and the processes of S107 to S135 are repeated.

FIG. 10 is a diagram schematically illustrating an example of conversation realized by information processing according to the embodiment; It is assumed that the conversation in FIG. 10 is presented to the user by the output unit 130 of the terminal device 100, for example.

Hereinafter, as shown in FIG. 10, the information processing according to the embodiment will be described in more detail by taking as an example a case where the management unit 250 acquires a user utterance "I'm going on a trip to Hawaii next time" in the process of S101. I will explain in detail.

In the process of S102, the text analysis unit 220 acquires text data indicating the content of the user's utterance, "I'm going on a trip to Hawaii next time," by speech recognition processing for voice data indicating the user's utterance. In addition, the sentence analysis unit 220 performs sentence word segmentation processing to divide sentences indicated by the text data into word units, that is, “next time”, “Hawaii”, “to”, “trip”, “to”, “go” and “yo”. ”.

In the process of S103, the text analysis unit 220 extracts, among the plurality of divided words, for example, "next time", "Hawaii", "travel" and "go" registered in the word category database 221 shown in FIG. is extracted as an extraction word for estimating the topic.

In the processing of S104, the sentence analysis unit 220 refers to the word category database 221 shown in FIG. 0003 "travel" is estimated as the topic of the user's utterance.

When the text analysis unit 220 estimates a topic, it is not necessary that all of the extracted words registered in the word category database 221 as extracted words corresponding to the topic are included in the user's utterance. In the above example, the text analysis unit 220 selects No. 1, which contains the most extracted words. 0003 "travel" is presumed to be the topic of the user's utterance.

In the process of S105, the word analysis unit 230 classifies, for example, "this time", "travel", and "go" registered in the intention estimation database 231 shown in FIG. Extract as extracted words for inference. The word analysis unit 230 determines that the category can be determined because the extracted word for estimating the category has been successfully extracted.

In the process of S106, the word analysis unit 230 refers to the intention estimation database 231 shown in FIG. 0004 "chat" is estimated as the category of the user's utterance.

Note that when the word analysis unit 230 estimates a topic, it is not necessary that all of the extracted words registered in the word category database 221 as extracted words corresponding to the category are included in the user's utterance. In the above example, the word analysis unit 230 selects No. 1, which contains the most extracted words. 0004 "chat" is presumed to be the category of the user's utterance.

In the processing of S107, the word analysis unit 230 extracts, for example, No. 1 registered in the important word database 232 shown in FIG. "Tour" and "Go" corresponding to "Tour" of 0003 are extracted as important words. The word analysis unit 230 uses No. Since the important word corresponding to 0003 "travel" has been extracted, it is determined that there is an important word in the estimated topic.

In the process of S108, the word analysis unit 230 identifies "travel" and "go" as important words in the user's utterance. "Travel" and "go" specified as important words are candidates for repetitive words used to generate response sentences.

In the process of S109, the word analysis unit 230 identifies "trip" and "go" as repeated words used to generate a response sentence. Here, it is assumed that the word analysis unit 230 identifies two words as repeated words, but is not limited to this. For example, the word analysis unit 230 may identify one word as the repeated word, or may identify three words as the repeated word.

Note that when the number of important words that are candidates for repeated words exceeds a predetermined number of repeated words, word analysis section 230 identifies the repeated words according to a predetermined priority order. For example, when "travel" and "go" are specified as important words in the user's utterance and one word is specified as a repeated word, the word analysis unit 230 gives higher priority to "travel" and "go". Identifies a word as a repeated word.

The order of priority may be determined by the display order of important words in the important word database 232, or the like. For example, in the important word database 232 shown in FIG. The priority order of the important words corresponding to "trip" in 0003 may be "trip", "go", "picture", "video", "shoot", "reservation" and "[location]". Therefore, in the above example, when identifying one word as a repeated word, the word analysis unit 230 identifies "travel" as a repeated word used to generate the response sentence.

In the processing of S110, the word analysis unit 230 refers to the conversation estimation database 233 shown in FIG. 0006 "nice" is specified as a sympathy word.

Here, in the processing of S111, the management unit 250 collects, as previous conversation information, information including past user utterances such as "Does this dress look good on you? I haven't decided whether to buy it or not," spoken to the chatbot by the user. assumed to have been obtained.

In this case, in the process of S112, the text analysis unit 220 obtains text data indicating the content of the past user utterance, "Does this dress look good on you? I haven't decided whether to buy it or not." . In addition, the text analysis unit 220 performs sentence word segmentation processing to divide words such as “kono”, “one piece”, “ha”, “suit”, “kana”, “buy”, “ka”, “do”, “ka”, It is divided into "wa", "determined", "te", "not" and "but".

In the processing of S113, the sentence analysis unit 220 selects, among the plurality of divided words, for example, "[Clothing item] (dress)", "suitable" and " "Buy" is extracted as an extraction word for estimating the topic.

In the processing of S114, the text analysis unit 220 refers to the word category database 221 shown in FIG. 0001 "fashion" is estimated as the topic of the user's utterance.

In the process of S115, the word analysis unit 230 estimates categories of, for example, "buy" and "determined (determined)" registered in the intention estimation database 231 shown in FIG. Extract as an extraction word for The word analysis unit 230 determines that the category can be determined because the extracted word for estimating the category has been successfully extracted.

In the processing of S116, the word analysis unit 230 refers to the intention estimation database 231 shown in FIG. 0004 "chat" is estimated as the category of the user's utterance.

In the processing of S117, word analysis section 230 determines that both the category of user utterance and the category of previous conversation information are No. 0004 "chat", it is determined that the category of the user's utterance and the category of the previous conversation information are the same.

In the process of S119, the word analysis unit 230 refers to the interesting word database 234 to determine whether the common category of the user utterance and the previous conversation information is "chat" and the extracted word of the previous conversation information is "[Clothing item]". and "looks good on you" are specified as interesting words corresponding to the common category.

In the process of S120, the word analysis unit 230 determines that the previous conversation information includes the interesting word specified in S119 because "one piece ([clothing item])" is present in the previous conversation information.

In S122, word analysis unit 230 identifies "one piece" in the previous conversation information as an interesting word to be used for generating a response sentence. Note that if the previous conversation information includes a plurality of interesting words, the word analysis unit 230 identifies, for example, the interesting word that has appeared most frequently in past user utterances as the interesting word to be used for generating the response sentence. do.

Further, when there is no interesting word corresponding to the common category in the previous conversation information, the word analysis unit 230 follows the predetermined priority order and selects the following from the interesting words corresponding to the common category identified in S119: Identify interesting words used to generate response sentences.

For example, in the above example, if none of "[clothing item]", "thinking", "sense" and "fashionable" exist in the previous conversation information, and the priority is "[clothing item]", "thinking , “sense” and “fashionable”, the word analysis unit 230 identifies “[clothing item]”, which has the highest priority, as an interesting word to be used for generating a response sentence.

In S123, the response generation unit 240 refers to the linked word database 241 shown in FIG. 7, and selects the No. corresponding to the set of the category "chat", the topic "fashion", and the interesting word "[Clothing item]". "wear" in 0001 is specified as a linked word to be used for generating a response sentence.

In the process of S124, the response generation unit 240 refers to the standard sentence database 242 shown in FIG. 0001 "Are [repeated words] and [repeated words], [sympathetic words] and [interesting words] [linked words]?"

In the process of S125, the response generation unit 240 divides the specified repetitive words "trip" and "go", the sympathetic word "nice", the interesting word "one piece", and the linked word "wear" into No. 0001 is applied to generate a response sentence "I'm going on a trip, that's nice, are you going to wear a dress?"

In the process of S<b>126 , the response generation unit 240 outputs the response sentence “I am going on a trip.

In the processes of S127 and S128, the management unit 250 confirms the input from the user, and acquires the user's utterance in response to the response text "Yes. I'm wearing a dress. Isn't it cold?"

In the processing of S129, the text analysis unit 220 acquires text data indicating the content of the user's utterance, "Yes, I'm wearing a dress. It's cold." In addition, the text analysis unit 220 divides the text indicated by the text data into “that’s right”, “one piece”, “wo”, “wear”, “yo”, “cold”, and “kana” through text word segmentation processing. To divide.

In the processing of S130, the text analysis unit 220 extracts, for example, "one piece" registered in the word category database 221 shown in FIG. do.

In the processing of S131, the text analysis unit 220 refers to the word category database 221 shown in FIG. 0001 "fashion" is estimated as the topic of the user's utterance for the response sentence.

In the processing of S132, the word analysis unit 230 selects, for example, “wear” and “cold” registered in the intention estimation database 231 shown in FIG. Extract as an extraction word. The word analysis unit 230 determines that the category can be determined because the extracted word for estimating the category has been successfully extracted.

In the processing of S133, the word analysis unit 230 refers to the intention estimation database 231 shown in FIG. 0004 "chat" is estimated as the category of the user's utterance.

In the process of S134, the word analysis unit 230 determines to continue the conversation because the category of the previous user's utterance, "chat", continues in the category of the current user's utterance.

FIG. 11 is a diagram schematically showing an example of the hardware configuration of a computer 700 that implements the information processing system 1 according to the embodiment. The server device 200 according to the embodiment is configured as a computer 700 shown in FIG. 11 .

As shown in FIG. 11, computer 700 has processor 710 , memory 720 , storage 730 , input/output interface (I/F) 740 and communication interface (I/F) 750 . These hardware are connected to bus 760 .

The processor 710 is configured as, for example, a CPU (Central Processing Unit) and controls the operation of each part of the computer 700 in an integrated manner.

The memory 720 includes, for example, ROM (Read Only Memory) and RAM (Random Access Memory), volatile or non-volatile storage of various data such as programs executed by the processor 710, and the processor 710 executing programs. We will realize the provision of a work area for

The storage 730 includes, for example, a HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores various data in a non-volatile manner.

The input/output interface 740 inputs data from input devices (not shown) such as a keyboard and mouse to the computer 700 and outputs data from the computer 700 to output devices (not shown) such as a display and speakers. to control the output of

Communications interface 750 allows computer 700 to communicate with other devices.

In the embodiment, each configuration shown in FIG. 1 of the server device 200 functions as a result of the processor 710 executing an information processing program stored in the memory 720 or the storage 730 through cooperation of hardware and software. Implemented as a module. However, in the embodiment, at least part of the functional module group shown in FIG. 1 may be realized only by dedicated hardware.

Note that the information processing program according to the above-described embodiments does not necessarily have to be pre-stored in the memory 720 or the storage 730 . For example, the information processing program according to the above-described embodiments can be stored in a computer-readable recording medium such as various magnetic disks such as flexible disks (FD), or various optical disks such as DVDs (Digital Versatile Disks). It may be provided as a computer program product recorded in installable or executable form.

Also, the information processing program according to the above-described embodiments may be provided or distributed via a network such as the Internet. That is, the information processing program according to the above-described embodiments may be stored in a computer connected to a network such as the Internet, and may be provided in the form of receiving a download via the network.

Note that, in the embodiment, the terminal device 100 can also be configured as a computer 700 having the hardware configuration shown in FIG. Therefore, as a modified example of the embodiment, a configuration in which at least some of the functional modules of the server device 200 are implemented in the terminal device 100 is also conceivable.

Therefore, as a modified example of the embodiment, an example in which the terminal device 100 corresponds to the "information processing device" of the present disclosure is also conceivable. Furthermore, an example in which the information processing system 1 as a combination of the terminal device 100 and the server device 200 corresponds to the "information processing device" of the present disclosure is also conceivable.

As described above, the information processing apparatus according to the embodiment performs automatic response processing for continuous conversation with the user.

As an example, in the information processing device, the word analysis unit 230 estimates the category of the user's utterance based on the user's utterance, and generates a response sentence based on the estimated category and the previous conversation information of the target user. Identify speech facilitators that encourage the user to continue speaking. Also, the response generation unit 240 generates a response sentence based on the estimated topic, the estimated category, and the specified utterance-promoting words. According to this configuration, in a conversation between the user and the chatbot realized by the information processing device, the chatbot estimates the category of the user's utterance, identifies the utterance-promoting word from the estimated category, and specifies the utterance-promoting word. can generate a response sentence using By generating a response sentence using the speech-promoting word, the chatbot can prompt the user to continue speaking. Therefore, according to the information processing apparatus according to the embodiment, it is possible to respond to the user's utterance so that the user and the chatbot can smoothly continue the conversation.

In addition, the utterance-promoting words include interesting words, and response generation section 240 generates a response sentence including the interesting words. Here, the interesting word is a word that attracts the user's interest, and if the response text includes the interesting word, the user is more likely to think of continuing the conversation with the chatbot. That is, the chatbot can prompt the user to continuously speak by generating a response sentence containing the interesting word.

Also, in the information processing apparatus, the text analysis unit 220 estimates the topic of the user's utterance based on the user's utterance. In addition, word analysis section 230 identifies linked words that form a response sentence in cooperation with the interesting words based on the estimated category, the estimated topic, and the specified interesting words. Also, the response generation unit 240 generates a response sentence including the interesting word and the linked word. According to this configuration, the chatbot can attract the user's interest with a natural sentence by generating a response sentence containing a linked word that forms a response sentence in cooperation with the interesting word. Here, for example, even if a chatbot generates a response sentence containing an interesting word, if the generated sentence is unnatural, the user may lose the desire to speak. Therefore, according to the information processing apparatus according to the embodiment, it is possible to respond to the user's utterance so that the user and the chatbot can continue the conversation more smoothly.

Also, the word analysis unit 230 extracts important words in the estimated topic from a plurality of words included in the content of the user's utterance. In addition, the word analysis unit 230 identifies sympathy words expressing sympathy for the user's input, which are used to generate response sentences to the user's utterance based on the topic and the important words. Also, the response generation unit 240 generates a response sentence including sympathy words. According to this configuration, the chatbot can empathize with the user's speech. Here, when the chatbot shows sympathy, the user is more likely to think about continuing the conversation with the chatbot. In other words, the chatbot can prompt the user to continuously speak by generating response sentences containing sympathetic words.

Further, based on the user utterance and the estimated topic, the word analysis unit 230 identifies repeated words used for generating a response sentence to the user utterance among a plurality of words included in the content of the user utterance. . Also, the response generation unit 240 generates a response sentence including repeated words. According to this configuration, the chatbot can appeal to the user that it understands the important points in the user's speech by repeating the important words in the user's speech. This makes it easier for users to feel that the chatbot is listening to them, and increases the likelihood that they will continue the conversation with the chatbot. In other words, the chatbot can prompt the user to continuously speak by generating response sentences containing repeated words.

Moreover, the word analysis unit 230 may identify a plurality of words as repeated words used for generating the response sentence. This allows the chatbot to more strongly appeal to the user that it understands the important points in the user's speech.

While embodiments and variations of the present invention have been described, the embodiments and variations are provided by way of example and are not intended to limit the scope of the invention. The novel embodiments and modifications can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. This embodiment and its modifications are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and its equivalents.

Reference Signs List 1 information processing system 100 terminal device 110 communication unit 120 input unit 130 output unit 200 server device (information processing device) 210 communication unit 220 text analysis unit 221 word category Database, 230... Word analysis unit, 231... Intention estimation database, 232... Important word database, 233... Conversation estimation database, 234... Interesting word database, 240... Response generation unit, 241... Linked word database, 242... Fixed phrase database, 250 Management Department

A first aspect of the present invention includes: a first estimating unit that estimates an intention of the user input based on the user input; and a response sentence to the user input based on the intention and the history of the user input. and a generator for generating the response sentence based on the intention and the utterance-promoting word, wherein the generation The unit is an information processing device that, when the estimated intention of the user input is different from the intention of the user input by the same user that was input before the user input, generates a sentence prompting the user to re-speak. be.

A second aspect of the present invention includes processing for estimating an intention of the user input based on the user input, and generating a response sentence to the user input based on the intention and the history of the user input. a process of identifying a speech-promoting word that encourages continuous utterance by a user; a process of generating the response sentence based on the intention and the speech-promoting word; and an estimated intention of the user input. is different from the intention of a user input by the same user that was input before the user input, a process of generating a sentence prompting the user to re-speak.

Claims (7)

a first estimation unit that estimates the intention of the user input based on the user input;
a first identification unit that identifies, based on the intention and the history of the user input, a speech-promoting word that is used to generate a response sentence to the user input and that promotes continuous utterance by the user;
a generation unit that generates the response sentence based on the intention and the speech prompt word;
Information processing device. the utterance-promoting word includes an interest word that arouses the interest of the user who has performed the user input;
The generation unit generates the response sentence including the interesting word.
The information processing device according to claim 1 . a second estimation unit that estimates a topic of the user input based on the user input;
a second identification unit that identifies a linked word that forms a response sentence in cooperation with the interesting word included in the speech prompt word based on the intention, the topic, and the interesting word,
The generation unit generates the response sentence including the interesting word and the linked word.
The information processing apparatus according to claim 2. an extraction unit that extracts important words in the topic from a plurality of words included in the content of the user input;
a third specifying unit that specifies, based on the topic and the important words, an empathy word that is included in the speech-prompting word and expresses empathy for the user input, and which is used to generate a response sentence to the user input;
The generation unit generates the response sentence including the sympathy word.
The information processing apparatus according to claim 3. identifying, based on the user input and the topic, a repeated word included in the utterance-promoting word used for generating a response sentence to the user input, among a plurality of words included in the content of the user input; 4 further comprising a specific part,
The generation unit generates the response sentence including the repeated word.
The information processing apparatus according to claim 3 or 4. The fourth identification unit identifies a plurality of the repeated words,
The information processing device according to claim 5 . In the computer of the information processing equipment,
a process of estimating the intention of the user input based on the user input;
a process of identifying, based on the intention and the history of the user input, speech-promoting words that are used to generate a response sentence to the user input and that promote continuous utterance by the user;
a process of generating the response sentence based on the intention and the speech prompt word;
program to run.

Images