Automatic conversation device and automatic conversation program

Abstract

To provide an automated conversational device that learns the user's preferences for conversation topics and selects appropriate topics without the user being aware of it or requiring complicated operations. [Solution] The prompt creation unit 122 creates prompts to conduct a conversation in line with the selected theme and provides them to the artificial intelligence chatbot. The question playback unit 123 plays back questions from the artificial intelligence chatbot and provides them to the responder, and the response receiving unit 124 receives responses from the responder and provides them to the artificial intelligence chatbot. The sentiment analysis unit 125 analyzes the responder's emotions based on the responder's responses, and the theme scoring unit 126 scores the themes based on the analysis results of the sentiment analysis unit and updates the theme score column in the theme table 105. The theme sorting unit 127 sorts the themes in the theme table 105 based on their scores and selects themes from the theme table using normally distributed random numbers to increase the selection frequency of themes with higher scores.

Description

Technical Field

[0004] , , ,

[0001] This invention relates to a device and a program that are used, for example, in elderly care facilities and function as conversation partners on behalf of people.

Background Art

[0002] As society ages, the number of elderly people with dementia is increasing. "Dementia" refers to a state in which the functions of nerve cells in the brain gradually decline due to various brain diseases, and cognitive functions (such as memory and judgment) decline, causing problems in social life. In addition, there are also elderly people with mild cognitive impairment who do not reach the stage of dementia. Also, even if they do not have dementia or mild cognitive impairment, in the case of elderly people, memory loss may become severe as they age. Thus, it is said that having conversations that activate brain activity is good for preventing dementia, mild cognitive impairment, memory loss, etc., which are common among the elderly, or for slowing the progression of these symptoms.

[0003] However, when elderly people have conversations, it is difficult to secure a conversation partner. That is, it is difficult to secure a person who can observe the feelings of the elderly and have appropriate conversations gently and patiently. Also, even if there is a person with appropriate qualities, there are often time limitations for their availability. Moreover, it is unreasonable to cover all the people who talk to the elderly with volunteers, and there may be cases where it is necessary to pay a fee to have someone become a conversation partner for the elderly. Thus, securing a conversation partner for the elderly is not an easy problem in terms of human resources, time, cost, etc. For this reason, so-called conversation robots have been conventionally utilized. A conversation robot means a robot that can communicate with people through actions such as words and gestures.

[0004] However, conventional conversational robots have had problems, such as being unable to continue natural conversations without users getting bored, due to technical issues. To solve this problem, for example, Patent Document 1, which will be described later, discloses an invention relating to a conversation control device that leads a conversation while maintaining the positive emotions of the user who is the conversation partner. The invention disclosed in Patent Document 1 is as follows: The emotion estimation unit of the conversation control device estimates the emotions of the conversation partner based on the facial expressions or the prosody of the conversation partner's voice. The conversation control unit controls the conversation with the conversation partner using topics selected based on the quality of the emotions estimated by the emotion estimation unit. Specifically, the conversation control unit controls the conversation with the conversation partner based on a topic map that includes topics preferred by the conversation partner, which is created by a topic map creation unit. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Patent No. 6601069 [Overview of the project] [Problems that the invention aims to solve]

[0006] In the case of the invention disclosed in Patent Document 1 mentioned above, a topic map is created based on the estimated emotions of the conversation partner, and only after a certain amount of information has been collected in this topic map is a topic selected that exceeds a certain threshold. Furthermore, the threshold also needs to be changed depending on the conversation partner. For this reason, in order to lead the conversation while maintaining the positive emotions of the user conversation partner, the process is complex and requires a certain amount of time and effort. For this reason, it is thought that the user may be difficult to use from the perspective of the user, especially the elderly.

[0007] In view of the above, this invention relates to a device and program for automatically conducting conversations, without requiring the user to be aware of it or to perform complicated operations, and on a topic (discussion). The goal is to learn user preferences, enable automatic optimization, and provide a user-friendly experience. [Means for solving the problem]

[0008] To solve the above problems, the automatic conversation device of the invention described in claim 1 is A theme table comprising multiple themes and a score column corresponding to each of the multiple themes, A theme selection means for selecting the theme to be used from the theme table, A prompt creation means creates a prompt for the artificial intelligence chatbot to conduct a conversation in accordance with the theme selected by the theme selection means, and provides the prompt to the artificial intelligence chatbot. A question playback means that accepts and plays back a question created by the artificial intelligence chatbot in response to the prompt created by the prompt creation means or in response to a response from the responder, A response receiving means that receives the response from the responder to the question reproduced through the question reproduction means and provides it to the artificial intelligence chatbot, A sentiment analysis means for analyzing the emotions of the responder regarding the response received through the response receiving means, A theme scoring means that scores the theme selected by the theme selection means based on the analysis results of the emotion analysis means and updates the score column of the corresponding theme in the theme table. A theme sorting means sorts the data stored in the theme table, which has been scored by the theme scoring means, in order of score. Equipped with, The theme selection means selects themes to be used by increasing the selection frequency of themes with high scores in the score column of the theme table using normally distributed random numbers. It is characterized by the following:

[0009] The automated conversation device according to claim 1 comprises a theme table, a theme selection means selects a theme from the theme table, and a prompt creation means creates prompts to conduct a conversation in line with the selected theme and provides them to the artificial intelligence chatbot. In response to the prompts and responses from the responder, a question playback means plays back questions provided by the artificial intelligence chatbot and provides them to the responder, and a response receiving means receives responses from the responder and provides them to the artificial intelligence chatbot. An emotion analysis means analyzes the responder's emotions based on the responder's responses, and a theme scoring means scores the current theme based on the analysis results of the emotion analysis means and updates the score column for the corresponding theme in the theme table. A theme sorting means sorts the themes in the theme table based on their scores. The theme selection means uses normally distributed random numbers to select themes in a way that increases the frequency of selection of themes with high scores. [Effects of the Invention]

[0010] This invention provides a user-friendly device that automatically engages in conversation without requiring the user to be aware of it or to perform complicated operations, and that can learn the user's preferences for topics and automatically optimize the conversation. [Brief explanation of the drawing]

[0011] [Figure 1] This diagram illustrates an example of a network system configuration using the automated conversation device of the embodiment. [Figure 2] This is a block diagram illustrating an example configuration of the automated conversation device according to the embodiment. [Figure 3] This figure illustrates an example of data stored in the theme table of an automated conversation device according to an embodiment. [Figure 4] This is a diagram illustrating the processing of the theme selection unit of the automated conversation device according to the embodiment. [Figure 5] This figure illustrates an example of an emotion score table for an automated conversation device according to an embodiment. [Figure 6] It is a flowchart for explaining the processing performed by the automatic conversation device according to the embodiment. [Figure 7] It is a diagram for explaining an example of a conversation conducted through the automatic conversation device according to the embodiment.

Embodiments for Carrying Out the Invention

[0012] Hereinafter, an embodiment of the automatic conversation device and the automatic conversation program of this invention will be described with reference to the drawings. The automatic conversation device and the automatic conversation program according to this invention are capable of realizing a natural conversation even when there is no person to be a conversation partner. Conducting a conversation activates brain activity. It is said that having a conversation is good for preventing dementia, mild cognitive impairment, forgetfulness, etc., which are common among the elderly, or delaying the progression of these symptoms. In the embodiment described below, for example, in a senior care facility or at home, the case where a senior has a conversation using the automatic conversation device of this invention or a device equipped with the automatic conversation program according to this invention will be described as an example.

[0013] The term "theme" means the central idea or theme of a certain work, topic, discussion, etc., and can be paraphrased as topic, subject, title, thesis, theme, main topic, etc. In the following description, the term "theme" is used to mean the subject matter that is the center of the conversation when making the device conduct a conversation automatically.

[0014] [Configuration Example of Network System] FIG. 1 is a diagram for explaining a configuration example of a network system configured by using the automatic conversation device 1 of the embodiment. In FIG. 1, the automatic conversation device 1 is an application of an embodiment of the automatic conversation device according to this invention, and in this embodiment, it is used by elderly people as a conversation partner. The automatic conversation device 1, which will be described in detail later, includes, as a user interface, a touch panel 113 composed of a display unit and a touch sensor, a speaker 115, and a microphone 116, and has an appearance similar to that of a so-called tablet PC. The automatic conversation device 1 is connectable to a generation AI providing server device 3 through the Internet 2.

[0015] The Internet 2 is a computer network that enables various terminal devices to be interconnected by using so-called Internet Protocol Suite technology. In this embodiment, the Internet 2 includes a public switched telephone network, a mobile phone network, and a wireless LAN (Local Area Network) conforming to, for example, the Wi-Fi (registered trademark) standard that connect from each terminal device to the Internet 2. Therefore, in FIG. 1, the base station BS(1) corresponds to a base station of the mobile phone network or an access point of the wireless LAN. In addition, various Web servers are provided on the Internet 2 other than the generation AI providing server device 3. However, in this embodiment, for simplicity of explanation, the generation AI providing server device 3, which is a particularly important access destination, is shown as the Web server connected to the Internet 2.

[0016] The generation AI providing server device 3 provides a so-called artificial intelligence chatbot service. The artificial intelligence chatbot service is such that an AI (Artificial Intelligence) self-learned based on various available information receives a question from a user's terminal device, generates an answer to the question, and can return this to the user's terminal device. The question and the answer are transmitted and received as text data or voice data.

[0017] The automated conversation device 1 of this embodiment utilizes an artificial intelligence chatbot service provided by the generation AI provision server device 3, enabling elderly users of the automated conversation device 1 to converse as if they were talking to a real person. Furthermore, the automated conversation device 1 of this embodiment learns the user's preferences for conversation topics and automatically optimizes the selected topics to suit the elderly user without the user being aware of them or requiring complicated operations. In short, from a variety of topics, it learns the user's preferences for conversation topics using emotional feedback and automatically personalizes the conversation. The configuration example and operation of the automated conversation device 1 of this embodiment will be described in detail below.

[0018] [Example configuration of automatic conversation device 1] Figure 2 is a block diagram illustrating an example configuration of the automatic conversation device 1. The transmitting / receiving antenna 101A and the wireless communication unit 101 are components that enable communication with wireless base stations such as base station BS(1), and enable connection to the Internet 2 via the wireless base station. Therefore, communication between the device and the AI ​​generation server device 3 via the Internet 2 is performed through the transmitting / receiving antenna 101A and the wireless communication unit 101. The control unit 102, although not shown, is a microprocessor equipped with a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), non-volatile memory, etc., and controls each part of the automatic conversation device 1.

[0019] The storage device 103 is a device consisting of a recording medium and its driver, such as an HDD (hard disk drive) or SSD (solid state drive), and performs various operations such as recording, modifying, deleting, and reading data. In addition to storing necessary data and programs, the storage device 103 is also used as a work area to temporarily store intermediate data generated in various processes. The operation unit 104 is configured with, for example, a power on / off key, a volume adjustment key, and several other function keys, and can receive operation input from the user and supply a corresponding signal to the control unit 102.

[0020] The theme table 105 and the response voice storage unit 106 are formed on the recording medium of the device unit, which consists of a recording medium and its driver, such as an HDD or SSD, and store and hold data used for processing. As shown in Figure 2, the theme table 105 and the response voice storage unit 106 can be formed on different recording media, or they can be formed on the same recording medium with different storage areas.

[0021] The theme table 105 stores multiple themes (text data) used in conversations and points associated with each theme. Figure 3 is a diagram illustrating an example of the data stored in the theme table 105 of the automatic conversation device 1. As shown in Figure 3, the theme table 105 is composed of a storage field for themes (text data) and a storage field for points associated with each theme. The theme storage field stores text data indicating themes to be discussed when having a conversation with an elderly person (hereinafter referred to as "the elderly person") who is a user of the automatic conversation device 1, such as "recent climate and seasonal changes," "local events and festivals," and "garden and plant care."

[0022] As shown in Figure 3, the score storage column for each theme stores a score for each theme, such as 90 points for "recent climate and seasonal changes," 85 points for "local events and festivals," and 82 points for "garden and plant care." These scores are calculated based on the results of sentiment analysis performed on the voice response data of the elderly person engaging in the conversation, as will be explained in more detail later. For example, if the sentiment analysis based on the elderly person's voice response data analyzes their emotions as bright, positive, and optimistic, the score for that theme will be higher, and conversely, if the analysis analyzes their emotions as dark, negative, and pessimistic, the score for that theme will be lower.

[0023] The response voice storage unit 106 stores the collected response voice data of the elderly person. The response voice data of the elderly person stored in this response voice storage unit 106 will be used for emotional analysis of the elderly person, as will be described in more detail later. The touch panel 113, voice output processing unit 114 and speaker 115, microphone 116 and voice input processing unit 117 constitute the user interface. The touch panel 113 is an input / output interface composed of a thin display unit such as an LCD (Liquid Crystal Display) or an organic EL (Electro Luminescence) display and a touch sensor 112 provided on the display screen of the display unit. The display information displayed on the display screen of the display unit and the touch sensor on the display information make it possible to configure operation keys, etc.

[0024] The voice output processing unit 114 processes the digital voice data from the control unit 102 to form an analog voice signal and supplies it to the speaker 115. The speaker 115 emits sound corresponding to the analog voice signal from the voice output processing unit 114. This makes it possible to convert questions (digital voice data) from the generation AI provision server device 3 into analog voice signals and emit them. The microphone 116 collects sound, converts it into an analog voice signal, and supplies it to the voice input processing unit 117. The voice input processing unit 117 converts the analog voice signal from the microphone 116 into digital voice data and takes it in. This makes it possible to receive the response voice of an elderly person, convert it into digital voice data, and store it in the response voice storage unit 106, etc.

[0025] The AI ​​generation unit 120 is the main component of the automatic conversation device 1 that enables it to engage in conversation with the elderly person using themes appropriate for that elderly person, ensuring that the elderly person can maintain interest and engage in a stable conversation. The AI ​​generation unit 120 includes a theme selection unit 121, a prompt creation unit 122, a question playback unit 123, a response reception unit 124, an emotion analysis unit 125, a theme scoring unit 126, a theme sorting unit 127, and a theme selection instruction unit 128.

[0026] The theme selection unit 121 uses the data stored in the theme table 105 to select a theme to be used in the conversation. In this embodiment, the theme selection unit 121 does not randomly select a theme from the data stored in the theme table 105. The theme selection unit 121 selects a theme by increasing the frequency of selecting themes that the elderly person prefers. Figure 4 is a diagram illustrating the processing of the theme selection unit 121 of the automatic conversation device 1.

[0027] As explained using Figure 3, the theme table has a score column for each theme. For each theme, a score is assigned based on the results of an emotional analysis of the elderly person, using the voice response data from the elderly person when a conversation is conducted with them. The score is updated based on the results of this scoring. In this embodiment, the score for a theme is higher if the elderly person's emotions are analyzed as bright, positive, and optimistic, and conversely, the score is lower if the emotions are analyzed as dark, negative, and pessimistic.

[0028] Therefore, as shown in Figure 4, the themes stored in the theme table 105 are sorted in descending order of their scores, and the distribution of these scores is considered to be a normal distribution dr. A normally distributed random number (a random number following a normal distribution) is then used to increase the frequency of selecting themes with high scores. As a result, the elderly user will have more of their preferred themes selected, allowing them to converse through the automatic conversation device 1 without getting bored. Furthermore, by repeatedly conversing with the elderly person about each theme stored in the theme table 105, each theme will be appropriately scored, as shown in Figure 4. Until then, for example, in Figure 4, a uniform variable ur is used to randomly select each theme, and by conversing about the selected theme, the scores of each theme in the theme table 105 are appropriately updated and adjusted to the elderly person's preferences. In other words, it becomes personalized for the elderly person.

[0029] The prompt creation unit 122 creates a prompt for the AI ​​generation server device 3 to conduct a conversation according to the theme selected by the theme selection unit 121, and sends this prompt to the AI ​​generation server device 3. Here, a prompt refers to an instruction or question that the user inputs to a system such as an AI. In this example, the prompt creation unit creates a prompt such as, "You are a conversational robot that interacts with elderly people. Please ask questions based on the following theme. Please speak in a friendly manner. Theme: Cooking or favorite foods," and sends this prompt to the AI ​​generation server device 3.

[0030] The question playback unit 123 forms question audio data that can be appropriately played back by the AI ​​generation server device 3, based on the prompt created by the prompt creation unit 122 and the elderly person's response audio data described later. Furthermore, the question playback unit 123 supplies the prepared question audio data to the audio output processing unit 114, causing it to be emitted from the speaker 115. In this way, the elderly person is provided with the question from the AI ​​generation server device 3 as audio information. The elderly person, who is the user of the automatic conversation device 1, responds to the question played back from the AI ​​generation server device 3.

[0031] The response receiving unit 124 activates the microphone and voice input processing unit 117 to receive digital voice data emitted by the elderly person, transmits it to the generation AI provision server device 3, and records it in the response voice storage unit 106. This allows the response voice data received from the elderly person through the microphone and voice input processing unit 117 to be provided to the generation AI provision server device 3, enabling the conversation to continue, and also allows the response voice data to be stored in the response voice storage unit 106. The response voice storage unit 106 stores the response voice data in association with the currently selected theme, and each unit of response voice data, which serves as the unit for recording and playback, is assigned identification information, such as a sequence number, and stored.

[0032] As mentioned above, the emotion analysis unit 125 analyzes the elderly person's response voice data, which is stored in the response voice storage unit 106 in association with the selected theme, to understand the elderly person's emotional state when they are conversing according to the selected theme. In this embodiment, the unit determines whether the elderly person is in a "cheerful," "neutral," or "gloomy" emotional state, and for the "cheerful" and "gloomy" emotional states, it also determines the degree (level) of these states in three categories: low, medium, and high.

[0033] Specifically, if the voice response data of elderly people contains many positive words such as "interesting," "fun," "happy," "I'd like to try it again," etc., it is determined that they are in a "cheerful" emotional state, and the degree of cheerfulness is further assessed on a scale of low, medium, and high, depending on the frequency of their occurrence. Conversely, if there are many words indicating a lack of interest and no emotional response, such as "not really," "I don't remember," "either way is fine," etc., it is determined that they are in a "neutral (no change in emotion)" emotional state. Conversely, if there are many negative words such as "not interesting," "I don't know," "I've never done it before," etc., it is determined that they are in a "gloomy" emotional state, and the degree of cheerfulness is further assessed on a scale of low, medium, and high, depending on the frequency of their occurrence.

[0034] However, simply judging based on whether the statement is positive or negative, such as "I had a sad experience, but now I see it as a happy memory" or "Everyone else seemed to be having fun, but I didn't," can sometimes lead to the opposite result. Therefore, the context must also be considered to understand the emotional state. Furthermore, the characteristics of the response voice data, such as the frequency state (pitch) and volume, should also be considered to understand the emotional state of the elderly person. This is just one example of sentiment analysis; of course, other methods can also be used to perform sentiment analysis.

[0035] The theme scoring unit 126 identifies a score for a corresponding theme based on the results of the emotion analysis performed by the emotion analysis unit 125, and updates the score column for the corresponding theme in the response voice storage unit 106 using the identified score. In this embodiment, the theme scoring unit 126 performs scoring using, for example, an "emotion score table" pre-prepared in the storage device 103. Figure 5 is a diagram illustrating an example of the emotion score table of the automatic conversation device 1. The emotion score table shown in Figure 5 comprises an emotion column, a level column, and a score column. The emotion column is broadly divided into three categories: positive emotions such as "cheerful, positive, and optimistic," normal (no change in emotion), and negative emotions such as "gloomy, passive, and pessimistic." The level column is divided into three stages: low, medium, and high for both positive and negative emotions. The score column stores the score for each level of each emotion.

[0036] Therefore, as shown in Figure 5, in this emotion scoring table, positive emotions such as "cheerful, active, and optimistic" are dominant, and a high level of these emotions is assigned a score of "4 points," a moderate level a score of "3 points," and a low level a score of "2 points." In the case of a normal state (no change in emotion), there is no level classification and it is assigned a score of "1 point." In addition, negative emotions such as "gloomy, passive, and pessimistic" are dominant, and a high level of these emotions is assigned a score of "-2 points," a moderate level a score of "-1 point," and a low level a score of "0 points."

[0037] The theme scoring unit 126 uses the emotion score table explained in Figure 5 to identify a score for the corresponding theme based on the results of the emotion analysis performed by the emotion analysis unit 125, and updates the score column for the corresponding theme in the response voice storage unit 106 using the identified score. Specifically, the identified score is added to the score in the score column for the corresponding theme. Therefore, if the identified score is "-1" or "-2", it will be subtracted, and the score in the score column will decrease. As a result, as shown in Figures 3 and 4, it is possible to score the elderly person according to their emotions when they engage in conversation on each theme in the theme table.

[0038] The theme sorting unit 127 sorts the data stored in the theme table 105, using the score in the score column for each theme as the sort key, from highest score to lowest score (descending order). The theme tables shown in Figures 3 and 4 already show the data sorted using the score as the sort key. In other words, Figures 3 and 4 show the results of a conversation conducted through the automatic conversation device 1 by an elderly person who is a user of the automatic conversation device 1 in this embodiment.

[0039] Specifically, the theme tables shown in Figures 3 and 4 reveal that themes that evoke positive emotions include "Recent climate and seasonal changes" (90 points), "Local events and festivals" (85 points), and "Garden and plant care" (82 points). Conversely, themes that evoke negative emotions include "Generational differences and viewpoints" (32 points), "Volunteer activities" (30 points), and "Computer use" (29 points). The scores for each other theme can also be obtained, allowing for an understanding of the degree of emotion (level of interest) associated with each theme. The theme tables, sorted by score in this way, are then used in the theme selection unit 121 described above to select a theme.

[0040] The theme selection instruction unit 128, when an event occurs that changes the theme, instructs the theme selection unit 121, via the control unit 102, to select a theme. In this embodiment, there are five events that cause a theme to be changed. The first is when the response voice data received from the elderly person via the response reception unit 124 instructs a change of theme. For example, if the response voice data received is "Let's talk about something else," or "Please change the theme," the theme will be changed. The second is when the result of the emotion analysis by the emotion analysis unit 125 is significantly negative. For example, if the analysis shows that negative emotions are dominant and their level remains high, the theme will be changed.

[0041] The third case is when the elapsed time since the prompt creation unit 122 provided a prompt to the artificial intelligence chatbot exceeds a predetermined time. In this case, the conversation has continued on the same topic for longer than the predetermined time. The elapsed time since the prompt creation unit 122 provided a prompt to the artificial intelligence chatbot is measured through a clock circuit provided by the automatic conversation device 1, although it is not shown in Figure 2. This elapsed time is cleared each time the prompt creation unit 122 provides a prompt to the artificial intelligence chatbot, and a new measurement is started.

[0042] The fourth event is when the question-and-answer exchange with the AI ​​generation server device 3 (artificial intelligence chatbot) occurs more than a predetermined number of times. That is, if the same topic is repeated, for example, 20 times, the topic is changed. As shown in the third event, the prompt creation unit 122 can also change the topic based on the elapsed time since providing the prompt to the artificial intelligence chatbot, so either the third or fourth event may be used. However, if the elapsed time since the prompt creation unit 122 provided the prompt to the artificial intelligence chatbot is set to be relatively long, the fourth event may also be used simultaneously.

[0043] The fifth case is when a theme change instruction is received, for example, through the touch panel 113. Specifically, the display unit 111 of the touch panel 113 displays, for example, "Change Theme," and when the user touches (or makes contact with) this display with their finger, this is received as a theme change instruction, and the theme is changed. The events that trigger a theme change, as shown here, do not have different priority levels; when any of these events occur, the theme selection instruction unit 128 instructs the theme selection unit 121 to select a theme. As a result, the prompt creation unit 122 creates a new prompt, which is sent to the generation AI provision server device 3, allowing the conversation to continue with the theme changed.

[0044] Thus, the automated conversation device 1 of this embodiment can engage in natural conversations with elderly people by utilizing an artificial intelligence chatbot service provided by the AI ​​generation server device 3 on the Internet 2. In this case, the automated conversation device 1 learns the elderly person's preferences for conversation topics and can engage in conversations optimized for that elderly person without making them aware of it or requiring them to perform complicated operations.

[0045] [Summary of processes performed by the automated conversation device 1] Next, the processes performed by the automatic conversation device 1, as explained using Figures 2 to 5, will be summarized with reference to the flowchart. Figure 6 is a flowchart illustrating the automatic conversation process performed by the automatic conversation device 1. Figure 7 is a diagram showing an example of a conversation that takes place between the generated AI providing server device 3 and the elderly person via the automatic conversation device 1 when the process in Figure 6 is executed. The process in the flowchart of Figure 6 is executed by the control unit 102, and is performed by the control unit 102 controlling each part of the generated AI utilization unit 120. When the automatic conversation device 1 is powered on and receives a conversation start instruction, for example, when the conversation start button displayed on the touch panel 113 is selected, the control unit 102 starts the automatic conversation process shown in Figure 6.

[0046] First, the control unit 102 performs a predetermined start process to perform automatic conversation processing (step S101). In step S101, it performs a predetermined start process, such as logging into the AI ​​generation server device 3, using necessary information such as the URL (Uniform Resource Locator) and password of the AI ​​generation server device 3 which is stored in its own non-volatile memory. After this, the control unit 102 controls the theme selection unit 121 and performs a process to select a theme to be used for conversation from the theme table 105 (step S102).

[0047] In step S102, the theme selection unit 121, as explained using Figure 4, assumes that the distribution of scores (high and low scores) for each theme follows a normal distribution, and uses normally distributed random numbers to increase the frequency of selecting themes with high scores, thereby selecting themes. This ensures that themes of high interest to the elderly user of the automatic conversation device 1 are selected preferentially. Note that if the scores for each theme in the theme table 105 have not been updated at all, each theme will be selected almost randomly.

[0048] Next, the control unit 102 controls the prompt creation unit 122 to create a prompt for the AI ​​generation server device 3 to start a conversation on the theme selected in step S102, and sends this prompt to the AI ​​generation server device 3 (step S103). In step S102, for example, suppose "Cooking and Favorite Foods" is selected as the theme. In this case, as shown in Figure 7(1), the first prompt is created, consisting of a request to have a conversation with the elderly person and a sentence notifying the theme, and is sent to the AI ​​generation server device 3.

[0049] In response to the prompt created and sent in step S103, the AI ​​generating server device 3 sends a question for conversation, so the control unit 102 controls the question playback unit 123 to receive the question and play it back (step S104). In step S104, the question playback unit 123 supplies the question data from the AI ​​generating server device 3 to the voice output processing unit 114, and the speaker 115 emits the voice corresponding to the question data. In this case, as shown in Figure 7(2), as question 1 in response to the first prompt, for example, a question from the AI ​​generating server device 3 asking what you ate yesterday is played back.

[0050] Next, the control unit 102 controls the response reception unit 124 to receive a response voice from the elderly user of the automatic conversation device 1, record it in the response voice storage unit 106, and transmit it to the generation AI provision server device 3 (step S105). In step S105, the response reception unit 124 collects the voice spoken by the elderly person through the microphone 116 and the voice input processing unit 117, captures it as response voice data, associates it with the current theme, and records it in the response voice storage unit 106. Furthermore, the response reception unit 124 transmits the received response voice data to the generation AI provision server device 3. In this case, as shown in Figure 7(3), for example, a response from the elderly person such as "I ate Yudofu (boiled tofu) even though it's summer" is recorded in the response voice storage unit 107 and transmitted to the generation AI provision server device 3.

[0051] Next, the control unit 102 controls the emotion analysis unit 125 to perform emotion analysis on the response voice data recorded in the response voice storage unit 106 in step S105 (step S106). In step S106, the emotion analysis unit 125 analyzes the response voice data from the elderly person to determine whether the elderly person's emotions are cheerful, neutral, or gloomy, and in the case of cheerful or gloomy, to the degree level. In the flowchart of Figure 6, emotion analysis is performed each time response voice data from an elderly person is received. In this case, the results of the emotion analysis should be stored and the emotion analysis results of response voice data on the same theme should be compiled.

[0052] After this, the control unit 102 controls the theme selection instruction unit 128 to determine whether or not an event to change the theme has occurred (step S107). In step S107, as mentioned above, the theme selection instruction unit 128 processes whether or not one of the five events has occurred. If the determination process in step S107 determines that no event to change the theme has occurred, the process from step S104 is repeated. Therefore, in the next step S104, a new question provided by the generation AI provision server device 3 is played back and broadcast in accordance with the response voice data shown in Figure 7(3) that was sent to the generation AI provision server device 3 in step S105. Specifically, as shown in Figure 7(4), the elderly person's comments on eating Yudofu and a new question (question 2 asking for the reason for eating Yudofu) are broadcast through the speaker 115.

[0053] In step S105, the system receives the voice response from the elderly person to the new question 2 shown in Figure 7(4), records it in the voice response storage unit 106, and transmits it to the AI ​​generation server device 3. Specifically, as shown in Figure 7(5), the voice response data from the elderly person, "I thought that eating something hot in the summer would actually make me feel cooler," is recorded in the voice response storage unit 106 and transmitted to the AI ​​generation server device 3. After this, sentiment analysis of the new voice response data (step S106) and determination of the occurrence of an event that changes the theme (step S107) are performed.

[0054] In the determination process of step S107, if it is determined that no event has occurred that would cause a change in the theme, the process from step S104 is repeated, and a new question provided by the generation AI providing server device 3 is played back and broadcast. Specifically, as shown in Figure 7(6), the elderly person's comments on the answer and a new question (question 3 asking about other hot dishes that they would like to eat in the summer) are broadcast through the speaker 115. In this way, the process from steps S104 to S107 is repeated, and the conversation continues between the generation AI providing server device 3 and the elderly person who is using the automatic conversation device 1 via the automatic conversation device 1.

[0055] In the determination process of step S107 described above, if it is determined that an event requiring a theme change has occurred, the control unit 102 controls the theme scoring unit 126 to score the current theme in the theme table 105 (step S108). After this, the control unit 102 controls the theme sorting unit 127 to sort the stored data in the theme table 105 in descending order using the score as the sort key (step S109). After this, the control unit 102 determines whether or not the end of the conversation has been instructed (step S110).

[0056] The determination process in step S110 determines that the conversation has been instructed to end if, for example, the conversation end button displayed on the touch panel 113 is touched, or if voice data from the elderly person such as "I'm ending the conversation" is collected. If the determination process in step S110 determines that the conversation has not been instructed to end, the control unit 102 repeats the process from step S102 to start a conversation on a new topic. Also, if the determination process in step S110 determines that the conversation has been instructed to end, the control unit 102 performs a predetermined termination process, such as logging off from the AI ​​generation server device 3 (step S111), and terminates the process shown in Figure 6.

[0057] Thus, the automatic conversation device 1 of this embodiment can select a theme and continue the conversation on that theme, and if a predetermined event occurs, it can change the theme and continue the conversation. Furthermore, during the continued conversation, it can perform an emotional analysis of the elderly person, assign scores to each theme, and select themes so that themes with higher scores are selected more frequently.

[0058] [Effects of the embodiment] According to the automatic conversation device 1 of the above-described embodiment, natural conversation can be conducted with simple operations. In this case, the device can learn the user's preferences for conversation topics and automatically optimize them without the user being aware of them or being required to perform complicated operations. As a result, the user can continue the conversation without getting bored and in a good mental state.

[0059] Therefore, conversations can be effectively conducted with the aim of preventing or slowing the progression of dementia, mild cognitive impairment, and memory loss, which are common in the elderly. For this reason, there is no need to find a real person to talk to, and there are no restrictions on the place or time of conversation. For example, conversations can be held anytime, anywhere with an internet connection, such as at home or in an elderly care facility.

[0060] [Implementation into the device via program] The program that performs the processing described using the flowchart in Figure 6 is an example of an embodiment of the automatic conversation program according to this invention. Therefore, by creating a program that performs the processing described using the flowchart in Figure 6 and installing it on an information processing device such as a tablet PC, notebook PC, or desktop PC, these information processing devices can be made to function as an automatic conversation device according to this invention. In other words, the functions of each part constituting the generation AI utilization unit 120 of the automatic conversation device 1 shown in Figure 2 can also be realized as functions of the control unit 102 by a program.

[0061] [Differentiation] In the embodiment described above, a case was explained in which an elderly person, who is a user of the automatic conversation device 1, converses with the AI ​​generation server device 3 via the automatic conversation device 1. In this case, the display screen of the display unit 111 of the automatic conversation device 1 can be set to display, for example, an image of a person or an image of a character such as an animation, and the display can be set to change the mouth shape in response to the audio data of the question played back from the AI ​​generation server device 3. In addition, the user may be able to select whether the audio data of the question from the AI ​​generation server device 3 is spoken in a female voice or a male voice.

[0062] Furthermore, in the embodiment described above, the automatic conversation device 1 and the AI ​​generation server device 3 are connected via the Internet 2, but this is not the only option. For example, a LAN (Local Area Network) can be configured in a facility for the elderly, and the AI ​​generation server device 3 can be connected to this LAN. By connecting the automatic conversation device 1 and the AI ​​generation server device 3 through this LAN, conversations can be conducted through the automatic conversation device 1.

[0063] Furthermore, it is also possible to configure the automatic conversation device 1, which is connected to a LAN configured in the elderly care facility, to be connected to the internet via a relay device connected to the LAN, and then to the generation AI provision server device 3 via the internet. In this case, all or part of the functions of the generation AI utilization unit 120 provided in the automatic conversation device 1, as described in the above embodiment, can be provided to the relay device. In this case, the theme table 105 and the response voice storage unit 106 can be those provided in the automatic conversation device 1. Alternatively, the theme table 105 and the response voice storage unit 106 may be provided in the relay device, and the theme table 105 and the response voice storage unit 106 may be used differently depending on the user of the automatic conversation device 1.

[0064] Furthermore, although the above-described embodiment assumed that the user of the automatic conversation device 1 was an elderly person, it is not limited to this. People of various ages can be users. In addition, by making it possible to select the language to be used, conversation practice can be conducted in different languages. For example, by using English for conversation, the automatic conversation device 1 according to this invention can be used for practicing English conversation. [Explanation of symbols]

[0065] 1...Automatic conversation device, 101A...Transmitting and receiving antenna, 101...Wireless communication unit, 102...Control unit, 103...Storage device, 104...Operation unit, 105...Theme table, 106...Response voice storage unit, 111...Display unit, 112...Touch sensor, 113...Touch panel, 114...Voice output processing unit, 115...Speaker, 116...Microphone, 117...Voice input processing unit, 120...Generating AI utilization unit, 121...Theme selection unit, 122...Prompt creation unit, 123...Question playback unit, 124...Response reception unit, 125...Sentiment analysis unit, 126...Theme scoring unit, 127...Theme sorting unit, 128...Theme selection instruction unit, BS(1)...Base station, 2...Internet, 3...Generating AI provision server device

Claims

1. A theme table comprising multiple themes and a score column corresponding to each of the multiple themes, A theme selection means for selecting the theme to be used from the theme table, A prompt creation means creates a prompt for the artificial intelligence chatbot to conduct a conversation in accordance with the theme selected by the theme selection means, and provides the prompt to the artificial intelligence chatbot. A question playback means that accepts and plays back a question created by the artificial intelligence chatbot in response to the prompt created by the prompt creation means or in response to a response from the responder, A response receiving means that receives the response from the responder to the question reproduced through the question reproduction means and provides it to the artificial intelligence chatbot, A sentiment analysis means for analyzing the emotions of the responder regarding the response received through the response receiving means, A theme scoring means that scores the theme selected by the theme selection means based on the analysis results of the emotion analysis means and updates the score column of the corresponding theme in the theme table. A theme sorting means sorts the data stored in the theme table, which has been scored by the theme scoring means, in order of score. Equipped with, The theme selection means selects themes to be used by increasing the selection frequency of themes with high scores in the score column of the theme table using normally distributed random numbers. An automatic conversation device characterized by the following features.

2. An automatic conversation device according to claim 1, The theme selection instruction means is provided to instruct the theme selection means to select a new theme to be used when an event occurs that changes the theme selected by the theme selection means. The theme selection instruction means instructs the theme selection means to select a new theme to use when it detects any of the following occurrences: when the response received by the response receiving means instructs a change to the selected theme; when the analysis result by the sentiment analysis means is significantly negative; when the elapsed time since the prompt creation means provided the prompt to the artificial intelligence chatbot exceeds a predetermined time; when the exchange of questions and responses with the artificial intelligence chatbot has occurred more than a predetermined number of times; or when it receives an input instructing a change to the selected theme through a predetermined operation input receiving means. An automatic conversation device characterized by the following features.

3. A theme selection step involves selecting a theme to use from a theme table that includes multiple themes and a score column corresponding to each of the multiple themes. A prompt creation step involves creating a prompt for the artificial intelligence chatbot to conduct a conversation in accordance with the theme selected in the theme selection step, and providing the prompt to the artificial intelligence chatbot. A question playback step which accepts and plays back a question created by the artificial intelligence chatbot in response to the prompt created in the prompt creation step, or in response to a response from the responder, A response receiving step which involves receiving the response from the responder to the question reproduced through the question reproduction step and providing it to the artificial intelligence chatbot, A sentiment analysis step is performed to analyze the emotions of the responder regarding the response received through the response reception step, A theme scoring step is performed to score the themes selected in the theme selection step based on the analysis results in the emotion analysis step, and to update the score column for the corresponding theme in the theme table. The theme sorting step sorts the data stored in the theme table, which has been scored in the theme scoring step, in order of score. It has, The theme selection step involves selecting themes to be used by increasing the selection frequency of themes with high scores in the score column of the theme table using normally distributed random numbers. An automated conversation program characterized by the following features.

Images