Prompt creation method, generative AI utilization device, generative AI utilization program, and tourist information system

Abstract

We provide technology that can increase the likelihood of obtaining responses that satisfy users. [Solution] An exemplary prompt creation method is a method for creating a prompt to be given to a generative AI performed by a computer, and the method creates a new prompt that includes information identifying the previous prompt given to the generative AI and the previous answer generated by the generative AI, information based on the user's response to the previous answer, and a command to respond again.

Description

Technical Field

[0001] The present invention relates to a technique for creating a prompt to be given to a generative AI (Artificial Intelligence) utilization device.

Background Art

[0002] Conventionally, there is a system that provides guidance on tourist spots according to a user's preferences. For example, in the information processing system disclosed in Patent Document 1, a user inputs the user's preference information into the information processing system, and the information processing system provides (proposes) a tourist experience (tourist information) that matches the preference information.

[0003] By the way, in recent years, in various systems that provide information, the use of generative AI composed of large language models (LLMs: Large Language Models) and the like has been spreading.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, generative AI does not necessarily generate an answer with high user satisfaction.

[0006] In view of the above points, an object of the present invention is to provide a technique that can increase the possibility of obtaining an answer with high user satisfaction.

Means for Solving the Problems

[0007] An exemplary prompt creation method of the present invention is a method for creating a prompt to be given to a generative AI performed by a computer, the method for creating a re-prompt that includes information identifying the previous prompt previously given to the generative AI and the previous answer generated by the generative AI, information based on the user's response to the previous answer, and a re-response command. [Effects of the Invention]

[0008] According to an exemplary version of the present invention, a revised response can be obtained from the generative AI that takes into account the user's reaction to the previous response generated by the generative AI, thereby increasing the likelihood of obtaining a revised response that satisfies the user. [Brief explanation of the drawing]

[0009] [Figure 1] Diagram showing the configuration of the tourist information system according to the first embodiment. [Figure 2] Diagram showing the configuration of the in-vehicle unit. [Figure 3] Diagram showing the tables that make up the first text table. [Figure 4] Diagram showing user profile table and response data table. [Figure 5] This diagram shows an example of a multi-dimensional (2D) model (psychological plane) for emotion estimation. [Figure 6] A diagram showing an example of a psychological plane that includes a neutral zone. [Figure 7] Diagram showing the prompt template table [Figure 8] Diagram showing the command template table. [Figure 9] Flowchart of prompt creation and route guidance processes executed by the in-vehicle controller. [Figure 10] Flowchart of prompt creation and route guidance processes executed by the in-vehicle controller. [Figure 11] Diagram showing the display screen of the display unit. [Figure 12] Diagram showing the display screen of the display unit. [Figure 13]Figure showing the display screen of the display unit [Figure 14] Figure showing the configuration of the server [Figure 15] Flowchart of prompt creation processing and route guidance processing executed by the in-vehicle device controller according to the second embodiment [Figure 16] Flowchart of processing for changing preference control parameters

Mode for Carrying Out the Invention

[0010] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings. In the description of the embodiments, the same parts are denoted by the same reference numerals, and redundant descriptions are omitted when not particularly necessary.

[0011] <1. First Embodiment> (Configuration of the sightseeing guidance system) FIG. 1 is a diagram showing the configuration of a sightseeing guidance system SYS1 according to the first embodiment of the present invention. As shown in FIG. 1, the sightseeing guidance system SYS1 includes an in-vehicle device (generative AI utilization device) 1 and a server 2.

[0012] The in-vehicle device 1 is mounted on the vehicle V1. The server 2 is arranged outside the vehicle V1 (for example, in a building such as a generative AI service center). The server 2 may be a physical server or a virtual server. The server 2 may be composed of one server or a plurality of servers. ,

[0013] The in-vehicle device 1 and the server 2 communicate via the network NT1. The in-vehicle device 1 transmits a prompt (question) given to the generative AI to the server 2. The server 2 transmits an answer to the prompt generated by the generative AI 222 (see FIG. 14) configured by an LLM (Large Language Models) or the like to the in-vehicle device 1.

[0014] (Configuration of the in-vehicle device) Figure 2 shows the configuration of the in-vehicle unit 1. The in-vehicle unit 1 comprises a communication unit 11, a storage unit 12, a controller 13, an input unit 14, a sensor 15, a display unit 16, and an audio output unit 17.

[0015] The communication unit 11 transmits and receives arbitrary signals (information) with the server 2. Although the controller 13 can transmit and receive arbitrary information with the server 2 using the communication unit 11, the description of the communication unit 11 may be omitted below.

[0016] The storage unit 12 is configured to have non-volatile memory such as ROM (Read-only memory) or flash memory, and volatile memory such as RAM (Random Access Memory). The storage unit 12 stores a first text table 121, a second text table 122, a user profile table 123A, a response data table 123B, a prompt template table 124, a preference genre table 125, preference control parameters 126, a command template table 127, a program 128, and map data 129.

[0017] The first text table 121 includes the first text tables 121A1 and 121A2.

[0018] The first text table 121A1 is a data table that stores associated operation types and text data A1. Specifically, as shown in Figure 3, the first text table 121A1 contains text data A1 for the operation types "Tourist Information Request" and "Tourist Information Update Request (Automatic)". For example, for "Tourist Information Request", the text "Please create a tourist information document that meets the following conditions." is stored, and for "Tourist Information Update Request (Automatic)", the text "Please create a tourist information document that meets the following conditions. Please consider the user's feelings regarding the previous information document." is stored in association.

[0019] The first text table 121A2 is a data table that stores data types and text data A2 associated with operation types (for example, "Tourist Information Request" and "Tourist Information Update Request (Automatic)") as shown in Figure 3. For example, for "Tourist Information Request," the data type "Gender" and the text data A2 "Gender is" are stored in association.

[0020] Furthermore, the data in these first text tables 121A1 and 121A2 will store data that has been appropriately set by the designer or others during product manufacturing, etc.

[0021] The second text table 122 is a data table that stores text data B1 and parameter values ​​(sensor detection data values: characteristic values ​​calculated based on sensor output) B1V associated with the sensor detection data type, as shown in Figure 3, for example. Specifically, a data record is generated for each sensor detection data type, and the text data B1 and its parameter value B1V corresponding to the sensor detection data type are stored in the data record. In the example shown in Figure 3, for example, for the sensor detection data type "Vehicle position", the text data B1 "The vehicle position is (parameter value)." and its parameter value B1V "North latitude Y, East longitude X" are stored. The parameter value B1V data is updated as needed based on the output of each sensor. The second text table 122 shown in Figure 3 includes a data record for the sensor detection data type "Emotion (change)", and is capable of handling "Tourist information update request (automatic)". In other words, the second text table 122 is generated with the intention of appropriately updating the tourist information by including user emotion (change) data from the provision of tourist information in the prompt (re-prompt). Furthermore, the sensor detection data type and text data B1 will store data that was appropriately set by the designer or other personnel during product manufacturing.

[0022] In this embodiment, the user profile table 123A and the response data table 123B shown in Figure 4 are used for prompt generation. For example, the user profile table 123A stores user profile data for each profile type. In the example of the user profile table 123A shown in Figure 4, "male" is stored for gender and "20s" for age group. This profile data is stored in the user profile table 123A by methods such as user input or by transferring user profile data stored on other devices such as smartphones. As shown in Figure 4, the response data table 123B stores the question content (prompt) to be updated (previous) and the response content profile of the generation AI to that question, linked together. The actual data is text, etc., but in Figure 4 it is represented by encoding (Qs1, An1).

[0023] Here, we will explain how to calculate the parameter value B1V for "emotion," which is a representative example of the sensor detection data type shown in Figure 3. Note that for other sensor detection data types, it is also possible to calculate the corresponding parameter value B1V using various known methods based on the corresponding sensor output. For example, the vehicle's position can be calculated using a GPS system to determine its position (latitude and longitude).

[0024] The emotion estimation model is a model that estimates emotions based on emotion index values, which are indicators of mental and physical states related to emotions. One of the emotion index values ​​used in this embodiment is the central nervous system arousal level (hereinafter referred to as arousal level), and its index value can be calculated using the "beta wave / alpha wave of the electroencephalogram (EEG)". Another emotion index value is the autonomic nervous system activity level (hereinafter referred to as activity level), and its index value can be calculated using the "standard deviation of the heart rate LF (Low Frequency) component (low-frequency component of the heart rate waveform signal)". Therefore, an EEG sensor that detects EEG and a heart rate sensor that detects heart rate are used as biosensors.

[0025] The emotion estimation model used in this embodiment consists of an emotion estimation model based on arousal level and activity level. The emotion estimation model used in this embodiment consists of a multidimensional model (here, a two-dimensional model with arousal level and activity level as two axes) that estimates emotion using arousal level and activity level as parameters. The two-dimensional model is created based on medical evidence (papers, etc.) that shows the relationship between multiple indicators and emotion (the relationship between arousal level / activity level and emotion). Alternatively, the two-dimensional model is created based on the results of questionnaires from many subjects (data consisting of emotion declarations by subjects and their arousal level / activity level at that time (based on electroencephalogram / heart rate measurements)). It is also possible to use a multidimensional model of three or more dimensions, not just a two-dimensional model, for the emotion estimation model used in this embodiment.

[0026] Figure 5 shows an example of a multi-dimensional (2D) model (psychological plane) for emotion estimation. According to various medical evidence related to psychology, psychology can be estimated based on two types of indicators (emotional index values) that indicate physical state. In the psychological plane shown in Figure 6, the vertical axis is "arousal level (aroused-unaroused)" and the horizontal axis is "autonomic nervous system activity level (sympathetic nervous system activity (strong emotion)-parasympathetic nervous system activity (weak emotion))."

[0027] In this psychological plane, each of the four quadrants separated by the vertical and horizontal axes is assigned a corresponding emotion (type). The distance from each axis indicates the intensity of the corresponding emotion. The first quadrant is assigned to the emotions of "happiness, joy, anger, and sadness." The second quadrant is assigned to the emotion of "melancholy." The third quadrant is assigned to the emotions of "relaxation and calmness." The fourth quadrant is assigned to the emotions of "anxiety, fear, and unpleasantness."

[0028] The axis positions are set appropriately based on experiments such as measuring the alertness and activity levels of subjects and performing statistical analysis.

[0029] Furthermore, emotions can be estimated from coordinates obtained by plotting two types of emotion indicators (arousal and activity levels) obtained based on biosignals onto a psychological plane. Specifically, emotions and their intensity can be estimated based on which quadrant the plotted coordinates lie in on the psychological plane, their position within that quadrant, and their distance from the origin. Note that the emotion estimation model shown in Figure 5 is a two-dimensional plane, but it can become a multi-dimensional space of three or more dimensions depending on the number of indicators used.

[0030] Furthermore, when emotional intensity is high, that is, when the emotional index value swings significantly towards the maximum or minimum value, the accuracy of emotion estimation increases. However, when emotional intensity is low, that is, when the emotional index value is near the median, the accuracy of emotion estimation decreases. For this reason, a method can be considered in which the region around the median of the emotional index value is treated as a neutral region, and judgments such as "no emotion estimated" or "emotion estimation impossible" are made based on this region.

[0031] Figure 6 shows an example of a psychological plane that includes the neutral region. In Figure 6, the shaded regions Rn1 and Rn2 are the neutral regions.

[0032] The upper and lower limits of the neutral region in the emotional index for "arousal level" are YP and YN, respectively, and the region between the upper limit YP and the lower limit YN is the arousal level neutral region Rn1 for "arousal level". Similarly, the upper and lower limits of the neutral region in the emotional index for "activity level" are XP and XN, and the region between the upper limit XP and the lower limit XN is the activity level neutral region Rn2 for "activity level".

[0033] These neutral ranges (upper limit YP, lower limit YN, upper limit XP, and lower limit XN) can be set as appropriate through experiments or other means.

[0034] The prompt template table 124 is data that shows the structure of a prompt, and as shown in Figure 7, it stores templates for operation types, such as "Tourist Information Request" and "Tourist Information Update Request (Automatic)". Specifically, in the case of the prompt template table 124 in Figure 7, for example, the template for operation type "Tourist Information Request" is "Text data A1 (Tourist Information Request), Text data A2 (Tourist Information Request) + Value A2V (Data Type) × n, Text data B1 (Vehicle Position) + Value B1V (Vehicle Position), Text data B1 (Preference) + Value B1V (Preference)". Also, the template for operation type "Tourist Information Update Request (Automatic)" is "Text data A1 (Tourist Information Update Request (Automatic)), Text data A2 (Tourist Information Update Request (Automatic)) + Value A2V (Tourist Information Update Request (Automatic)) × n, Text data B1 (Emotion) + Value B1V (Emotion), Text data B1 (Preference) + Value B1V (Preference)".

[0035] The preference genre table 125 is a table that stores potential genres of user preferences (for example, "history lover," "shopping lover," etc.), and the designers will create the data as appropriate.

[0036] The preference control parameter 126 is a parameter that indicates the user's preferences. In this embodiment, the preference control parameter 126 is one of the genres listed in the preference genre table 125 (for example, "history lover"). The selection of this parameter is based on the user's input operations or user profile data stored in other devices such as smartphones (obtained through transfer, etc.).

[0037] The command template table 127 is a data table that stores data indicating the processing that the controller 13 (text analysis unit 136 and command generation unit 137) will perform in response to the response from server 2. The designer will create the data as appropriate. As shown in Figure 8, the command template table 127 is a data table in which the analysis content and command generation content are linked and stored for each operation type.

[0038] The operation type in the command template table 127 is the same data as the operation type in the first text table 121A1 and the prompt template table 124, and will store "Request for tourist information," etc. The analysis content is data indicating the analysis method for the response from server 2 corresponding to the operation type, and the controller 13 (text analysis unit 136) performs analysis according to this data. The command generation content is a command (or program) that causes the controller 13 to perform an action according to the analysis content. In reality, the command or program itself is stored, but in Figure 8, for the sake of clarity, it is shown as the content of the process (action) to be executed. The controller 13 (command generation unit 137) searches the command template table 127 based on the analysis result for the response from server 2 corresponding to the operation type and generates a command. The controller 13 will then execute the processing according to this generated command.

[0039] According to the example data shown in the command template table 127 in Figure 8, the text analysis unit 136 performs language analysis on the response from the server 2 to extract destination candidates (facilities, etc.: e.g., XX Temple), and sets the extracted destination candidates (e.g., XX Temple) as destination candidates according to the corresponding command (set as destination candidate). Subsequently, the controller 13 performs navigation functions to the destination, such as presenting (displaying) the destination candidates to the user, processing a route search to the destination if the user confirms a destination candidate as the destination, and processing route guidance based on the searched route.

[0040] In Figure 2, the program 128 in the memory unit 12 is a program that should be executed by the controller 13. The map data 129 is data used for route searching, route display, etc., and includes road data used to represent road shapes. The map data is divided into predetermined areas called meshes (for example, 100m square), and each mesh is saved as a separate file. Map data for the area used for processing, etc., will be selected from the map data 129 and used.

[0041] The controller 13 comprehensively controls the operation of each part of the in-vehicle unit 1. The controller 13 is equipped with a processing unit including a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit) as hardware resources. The controller 13 implements the functions indicated by each functional block 131 to 138.

[0042] Controller 13 is a program execution device (computer) capable of executing any program. The controller 13 executes program 128, thereby realizing each function of the controller 13 (including the functions of function blocks 131 to 138). All operations of the controller 13 described in this embodiment may be operations realized by the controller 13 executing program 128. Program 128 may consist of multiple programs.

[0043] The input unit 14 is an input device that accepts human input operations. The input unit 14 is, for example, a touch panel, operation buttons, etc. The sensor 15 includes an electroencephalogram (EEG) sensor that detects the driver's brain waves, a heart rate sensor that detects the driver's heart rate, and a GPS (Global Positioning System) sensor that detects the position of the in-vehicle unit 1. The display unit 16 is a display device that displays information on a display screen. The display unit 16 is, for example, a liquid crystal display device included in a touch panel, an organic EL (Electro Luminescence) display device, etc. The audio output unit 17 is, for example, a speaker that converts electrical signals into sound.

[0044] The following describes each functional block of the controller 13. Functional blocks 131, 132, 133, 134, 135, 136, 137, and 138 are the first text conversion unit, the second text conversion unit, the text synthesis unit, the emotion estimation unit, the preference estimation unit, the text analysis unit, the command generation unit, and the guidance unit, respectively.

[0045] The first text conversion unit 131 outputs text data corresponding to the operations performed by the driver or other user on the input unit 14, based on the data in the first text conversion table 121 (121A, 121B). In other words, the first text conversion unit 131 outputs text data that converts the information based on the output of the input unit 14 into text data.

[0046] The second text conversion unit 132 outputs text data corresponding to the information based on the output of the sensor 15, based on the data in the second text conversion table 122. In other words, the second text conversion unit 132 outputs text data that converts the information based on the output of the sensor 15 into text data.

[0047] The text synthesis unit 133 synthesizes text data based on the prompt template table 124. The text synthesis unit 133 inserts the text data output from the first text conversion unit 131 into the basic text data to create a prompt. In addition, when creating the prompt, the text synthesis unit 133 also inserts the text data output from the second text conversion unit 132 into the basic text data. Finally, the text synthesis unit 133 sends the created prompt to the server 2.

[0048] The emotion estimation unit 134 estimates the user's emotions from the output of the sensor 15 (biosensor) using the emotion estimation model described above.

[0049] The preference estimation unit 135 estimates the user's preferences from the user's emotions estimated by the emotion estimation unit 134, and appropriately rewrites the preference control parameter 126 according to the estimated user preferences. In this embodiment, a simplified preference estimation method is applied in which, if the user determines based on their emotions that the response sent from the server 2 to the prompt is not appropriate, the preference data of the preference control parameter 126 included in the prompt is deemed inappropriate, and the preference control parameter 126 is changed to different preference data. The default value of the preference control parameter 126 is stored as data appropriately set by the designer or others during product manufacturing, etc.

[0050] The text analysis unit 136 receives and analyzes the text data, which is the response sent from the server 2. The command generation unit 137 then generates a command based on the command template table 127, corresponding to the analysis results of the text analysis unit 136. The content to be analyzed (for example, words to be extracted) is determined by the operation, so if the program for performing the analysis is stored in the storage unit 12 (program 128), the corresponding analysis can be performed. The guidance unit 138 executes processing according to the command generated by the command generation unit 137, and also performs related navigation function processing, providing the user with information on potential destinations, and, once a destination is set, provides route guidance to that destination.

[0051] (Operation of the in-vehicle device) Figures 9 and 10 are flowcharts of the prompt creation process and route guidance process executed by the controller 13. The prompt creation process and route guidance process are realized when the controller 13 executes the program 128 described above. The flows shown in Figures 9 and 10 are repeatedly executed at appropriate intervals (time intervals that allow for smooth operation in response to user input) during the operation of the in-vehicle device 1.

[0052] First, in step S10, the controller 13 determines the type of operation based on the user's operation. For example, the controller 13 determines that the operation type is "Tourist Information Request" by checking whether the "Tourist Information Request" button included in the input unit 14 was pressed and "Tourist Information Request" was selected. Once the operation type is determined (if "Tourist Information Request" is selected), the process proceeds to step S20.

[0053] In step S20, the controller 13 selects a template from the prompt template table 124 corresponding to the selected operation type and proceeds to step S30. In step S30, the controller 13 recognizes the operation type in the selected template, the data type related to that operation type, and the sensor detection data type, etc., and proceeds to step S40.

[0054] In step S40, the controller 13 (first text conversion unit 131) extracts text data A1 corresponding to the operation type in the template from the first text conversion table 121A1, and also extracts text data A2 related to the data type for that operation type from the first text conversion table 121A2, and proceeds to step S50. For example, this process extracts "Please create a tourist information document that meets the following conditions." as text data A1 for the operation type "Tourist information request", and extracts "Gender" and "Gender is", and "Age group" and "Age group is" as data type and text data A2.

[0055] In step S50, the controller 13 (first text conversion unit 131) extracts parameter values ​​for the data types extracted in step S40 based on the user profile data table 200, and proceeds to step S60. For example, this process extracts the parameter value "male" for the data type "gender" and the parameter value "20s" for the data type "age group". In step S60, the controller 13 (first text conversion unit 131) synthesizes the data extracted in steps S40 and S50 to generate first text data, and proceeds to step S70.

[0056] In step S70, the controller 13 (second text generation unit 132) calculates parameter values ​​for each sensor detection data type based on the output of the sensor 15, stores the parameter value B1V in the second text generation table 122 shown in Figure 3 using the preference control parameter 126, and proceeds to step S80. In step S80, the controller 13 (second text generation unit 132) synthesizes the text data B1 and its parameter value corresponding to the sensor detection data type in the selected template to generate second text data, and proceeds to step S90. For example, this process generates text data B1 "The vehicle's position is at latitude Y, longitude X. The user's preference is history." for the "tourist information request" template.

[0057] In step S90, the controller 13 (text synthesis unit 133) synthesizes the first text data generated in step S60 and the second text data generated in step S80 (generating text by inserting the data into a template) to create a prompt and proceed to step S100. For example, for operation type "Tourist information request", the following is generated: "Please create a tourist information text that is suitable for the following conditions. Gender: Male. Age: 20s. Vehicle position: North latitude Y, East longitude X. User preference: History lover."

[0058] In step S100, the controller 13 sends a prompt to the server 2 via the communication unit 11 and proceeds to step S110. In step S110, the controller 13 determines whether or not it has received a response from the server 2 via the communication unit 11. If a response (guidance information) is received from the server 2, the process proceeds to step S120 in Figure 10.

[0059] In step S120, the text analysis unit 136 analyzes the response and proceeds to step S130. Specifically, the text analysis unit 136 performs linguistic analysis on the response from server 2, extracts destination information (words indicating the destination), and generates data that the controller 13 (command generation unit 137) can handle, such as location data (data for specifying destinations in route search, such as latitude and longitude data). In the case of location data, for example, the map data stores the names and locations (latitude and longitude data) of each facility, so the location data is generated by searching for location data (latitude and longitude data) from the facility names obtained by linguistic analysis based on the map data.

[0060] In step S130, the command generation unit 137 generates a command (an instruction command to the controller 13) according to the analysis results and proceeds to step S140. Specifically, this command causes the guidance unit 138 to display multiple destination candidates 161A (each destination candidate extracted by the analysis is displayed (as a selection button)), an OK button 161B to confirm the selection of a destination, and a cancel button 161C to cancel the selection of a destination on the display screen 161 of the display unit 16, as shown in Figure 11.

[0061] In step S140, with multiple destination candidates 161A displayed on the display screen 161 of the display unit 16 as shown in Figure 11, the emotion estimation unit 134 estimates the user's emotion from the output of the sensor 15 (biosensor) using the emotion estimation model described above, and proceeds to step S150.

[0062] In step S150, the controller 13 determines whether the user's emotion estimated by the emotion estimation unit 134 is similar to the target emotion stored in the memory unit 12 (for example, happiness (an emotion the user feels when they are satisfied)). For example, the controller 13 determines that the user's emotion estimated by the emotion estimation unit 134 is similar to the target emotion stored in the memory unit 12 if the coordinates plotted on the psychological plane shown in Figure 5 based on the output of the sensor 15 are in the first quadrant of the psychological plane. On the other hand, the controller 13 determines that the user's emotion estimated by the emotion estimation unit 134 is not similar to the target emotion stored in the memory unit 12 if the coordinates plotted on the psychological plane shown in Figure 5 based on the output of the sensor 15 are not in the first quadrant of the psychological plane.

[0063] If the user's emotion estimated by the emotion estimation unit 134 is similar to the target emotion stored in the memory unit 12, it is highly likely that the user has received a response that will increase their satisfaction, and therefore, a request for a re-response is not made. Accordingly, as shown in Figure 11, when multiple destination candidates 161A are displayed on the display screen 161 of the display unit 16, and the selection of a destination is confirmed (the user selects a destination candidate 161A and operates the OK button 161B), the process proceeds to step S290. If the destination selection is canceled (the user operates the cancel button 161C), the process ends with an interrupt.

[0064] In the example above, emotions were used as the user's response, but regardless of the estimated user's emotions, if the user selects destination candidate 161A and presses the OK button 161B within a predetermined time frame (a time frame in which it can be estimated that the user felt satisfied with the answer and made an immediate selection), it is possible to determine that the user has received a highly satisfactory answer. Alternatively, emotions and these judgments can be combined as appropriate to estimate user satisfaction (and also be used as the user's response).

[0065] On the other hand, if the user's emotion estimated by the emotion estimation unit 134 is not similar to the target emotion stored in the memory unit 12, it is highly likely that the user has not received a satisfactory response, and the operation type is automatically set to request a re-response. Specifically, in step S160, the preference estimation unit 135 changes the preference control parameter 126 from the current genre to another genre stored in the preference genre table 124. The other genre may be randomly selected from the preference genre table 124, or it may be selected sequentially from the preference genre table 124. After the processing in step S160 is completed, the process proceeds to step S170.

[0066] In step S170, the controller 13 automatically sets the operation type to "Tourist Information Update Request (Automatic)" (assuming the "Tourist Information Update Request (Automatic)" operation has been performed). At this time, the controller 13 switches the control content of the display unit 16 from a state in which multiple destination candidates 161A are displayed on the display screen 161 of the display unit 16 as shown in Figure 11, to a state in which a message M1 indicating that the tourist information is being updated is displayed on the display screen 161 of the display unit 16 as shown in Figure 12, and proceeds to step S180.

[0067] In step S180, the controller 13 selects a template from the prompt template table 124 corresponding to the selected operation type ("Tourist Information Update Request (Automatic)" operation) and proceeds to step S3190. In step S190, the controller 13 recognizes the operation type in the selected template, the data type related to that operation type, and the sensor detection data type, and proceeds to step S200.

[0068] In step S200, the controller 13 (first text conversion unit 131) extracts text data A1 corresponding to the operation type in the template from the first text conversion table 121A1, and also extracts text data A2 related to the data type for that operation type from the first text conversion table 121A2, and proceeds to step S210. Through this process, for the operation type "Tourist information update request (automatic)", the text data A1 extracted is "Please create a tourist information text that is suitable for the following conditions. Please consider the user's feelings regarding the previous information text.", and the data type and text data A2 extracted are "The previous information text" and "The content of the information text for the previous information request (*question content data) is (*answer text data). *Command: See answer data table 123B.".

[0069] In step S200, the controller 13 (first text conversion unit 131) extracts the (question content data) "Qs1" and the (answer text data) "An1" from the answer data table 123B based on the command contained in the text data A2 (see answer data table 123B). In step S210, the controller 13 (first text conversion unit 131) combines the data extracted in steps S190 and S200 to generate the first text data "The content of the guidance text for the guidance request Qs1 is An1." and proceeds to step S220.

[0070] In step S220, the controller 13 (second text generation unit 132) calculates parameter values ​​for each sensor detection data type based on the output of the sensor 15, and using the user's emotion estimated by the emotion estimation unit 134 and the preference control parameter 126, stores the parameter value B1V in the second text generation table 122 shown in Figure 3, and proceeds to step S230. In step S230, the controller 13 (second text generation unit 132) synthesizes the text data B1 corresponding to the sensor detection data type in the selected template with its parameter value to generate second text data, and proceeds to step S90. For example, this process generates text data B1 for the "Tourist Information Update Request (Automatic)" template: "The user's emotion is depressed due to the previous information text. The user's preference is to enjoy shopping."

[0071] In step S240, the controller 13 (text synthesis unit 133) synthesizes the first text data generated in step S210 and the second text data generated in step S230 (generating text with the data embedded in a template) to create a new prompt and proceed to step S250. For example, for operation type "Tourist information update request (automatic)", the following is generated: "Please create a tourist information text that is suitable for the following conditions. Please consider the user's feelings regarding the previous information text. The content of the information text for the previous information request Qs1 is An1. The user's feelings are depressed because of the previous information text. The user's preferences are that they like shopping."

[0072] In other words, the re-prompt generated in step S240 includes information that identifies the previous prompt given to the generation system AI222 (see Figure 14) last time (the prompt sent in step S100) and the previous response generated by the generation system AI222 (the response received in step S110), information based on the output of a sensor that detects the user's reaction to the previous response (the user's emotion estimated in step S140), and a re-response command. By sending the above re-prompt to the generation system AI222, a re-response that takes into account the user's reaction to the previous response generated by the generation system AI222 can be obtained from the generation system AI222. Therefore, by sending the above re-prompt to the generation system AI222, the possibility of obtaining a re-response that satisfies the user can be increased.

[0073] Furthermore, the re-prompt generated in step S240 includes the user's preferences estimated by the preference estimation unit 135 based on the user's emotions estimated by the emotion estimation unit 134 based on the output of the sensor 15. By including the user's preferences estimated from the user's reaction to the previous response (since the previous response was unfavorable, the user's preferences will be different from those of the previous prompt) in the re-prompt, the likelihood of the user's reaction to the previous response being reflected in the content of the re-response increases, further increasing the likelihood of obtaining a re-response that satisfies the user.

[0074] In this embodiment, if the user's emotion estimated by the emotion estimation unit 134 based on the output of the sensor 15 is not similar to the target emotion, the preference estimation unit 135 changes the user's preferred genre estimated from the user's emotion estimated by the emotion estimation unit 134 based on the output of the sensor 15 (for example, changing from "history lover" to "shopping lover"). By estimating the user's preferences in this way, the process of estimating the user's preferences can be simplified. By repeatedly executing the prompt creation process and route guidance process shown in Figures 9 and 10, it is expected that the preference control parameter 126 will converge to an appropriate parameter. By repeatedly executing the prompt creation process and route guidance process shown in Figures 9 and 10, it is possible to make the preference control parameter 126 follow the changes in the user's preferences even when the user's preferences change over time.

[0075] In step S250, the controller 13 sends a re-prompt to the server 2 via the communication unit 11 and proceeds to step S260. In step S260, the controller 13 determines whether or not it has received a re-response from the server 2 via the communication unit 11. If a re-response (updated guidance information) from the server 2 is received, the process proceeds to step S270.

[0076] In step S270, the text analysis unit 136 analyzes the revised response and proceeds to step S280. Specifically, the text analysis unit 136 performs linguistic analysis on the revised response from server 2, extracts destination information (words indicating the destination), and generates data that the controller 13 (command generation unit 137) can handle, such as location data (data for specifying destinations in route search, such as latitude and longitude data). In the case of location data, for example, the map data stores the names and locations (latitude and longitude data) of each facility, so the location data is generated by searching for location data (latitude and longitude data) from the facility names obtained by linguistic analysis based on the map data.

[0077] In step S280, the command generation unit 137 generates a command (an instruction command to the controller 13) according to the analysis results. Specifically, based on this command, the guidance unit 138 displays multiple destination candidates 161A (each destination candidate extracted by the analysis is displayed (as a selection button)), an OK button 161B to confirm the destination selection, and a Cancel button 161C to cancel the destination selection on the display screen 161 of the display unit 16, as shown in Figure 13. Once the destination selection is confirmed (by the user selecting a destination candidate 161A and pressing the OK button 161B), the process proceeds to step S290. If the destination selection is canceled (by the user pressing the Cancel button 161C), the process ends via an interrupt.

[0078] In step S290, the guidance unit 138 sets the selected destination as the destination for route guidance and searches for a route to the set destination using known route search technology. Then, it provides guidance along the searched route to the destination (displaying the route on a map, providing voice guidance at guidance points, etc.). When the vehicle V1 arrives at the destination for route guidance, the guidance unit 138 terminates the route guidance. Once the route guidance is finished, the controller 13 terminates processing. Alternatively, after the route guidance is finished, the process may return to step S140 to reconfirm user satisfaction.

[0079] (Server configuration) Figure 14 shows the configuration of Server 2. Server 2 comprises a communication unit 21, a storage unit 22, and a controller 23.

[0080] The communication unit 21 transmits and receives arbitrary signals with the in-vehicle unit 1. The controller 23 can also transmit and receive arbitrary information with the in-vehicle unit 1 using the communication unit 21; however, the description of the communication unit 21 may be omitted below.

[0081] The memory unit 22 is configured to have non-volatile memory such as ROM (Read-only memory) or flash memory, and volatile memory such as RAM (Random Access Memory). The memory unit 12 stores the program 221 and the generation system AI 222.

[0082] Program 221 is a program to be executed by controller 23. Generative AI (AI model) 222 is a trained machine learning model for generating text data that is a response to prompts sent from the in-vehicle device 1. Generative AI 222 is composed of, for example, LLM.

[0083] The controller 23 comprehensively controls the operation of each component in the server 2. The controller 23 is equipped with a processing unit including a CPU as a hardware resource. The controller 23 has a generative AI execution unit 231.

[0084] When the generative AI execution unit 231 receives a prompt from the in-vehicle device 1, it uses the prompt as input data to execute the generative AI 222 and causes the generative AI 222 to generate a response. The controller 23 transmits the response generated by the generative AI 222 to the in-vehicle device 1. Furthermore, when the generative AI execution unit 231 receives a second prompt from the in-vehicle device 1, it uses the second prompt as input data to execute the generative AI 222 and causes the generative AI 222 to generate a third response. The controller 23 transmits the third response generated by the generative AI 222 to the in-vehicle device 1.

[0085] <2. Second Embodiment> The tourist information system according to the second embodiment of the present invention, like the tourist information system SYS1 according to the first embodiment of the present invention, comprises an in-vehicle device (generative AI utilization device) 1 and a server 2. The following describes mainly the parts that differ from the first embodiment.

[0086] A key feature of this embodiment is that user preferences are estimated while also considering the intensity of the user's emotions. In other words, it is characterized by generating a re-prompt that also takes into account the intensity of the user's response to the generative AI's answer.

[0087] In this embodiment, the preference control parameter 126 includes not only the genre but also the range (intensity) of preference within the same genre.

[0088] For example, the preference control parameter 126 may be "I like history and prefer advanced history information," "I like history and prefer intermediate history information," "I like history and prefer beginner history information," "I like shopping and prefer advanced shopping information," "I like shopping and prefer intermediate shopping information," "I like shopping and prefer beginner shopping information," etc. Hereafter, for convenience, "I like history and prefer advanced history information," "I like history and prefer intermediate history information," "I like history and prefer beginner history information," "I like shopping and prefer advanced shopping information," "I like shopping and prefer intermediate shopping information," and "I like shopping and prefer beginner shopping information" will be referred to as "high level history lover," "medium level history lover," "low level history lover," "high level shopping lover," "medium level shopping lover," and "low level shopping lover," respectively.

[0089] Figure 15 is a flowchart of the prompt creation process and route guidance process executed by the controller 13 of the in-vehicle device 1 according to this embodiment. In the flowchart of this embodiment shown in Figure 15, the processing content of step S150 and the details of step S160 differ from the flowchart of the first embodiment shown in Figure 10.

[0090] In step S150, the controller 13 determines whether the user's emotion estimated by the emotion estimation unit 134 is similar to the target emotion (e.g., happiness) stored in the memory unit 12, and whether the intensity of the user's emotion estimated by the emotion estimation unit 134 is within the target range (e.g., medium intensity). The intensity of the user's emotion can be determined using a known emotion estimation method that estimates the type of emotion and its intensity based on the user's biosignals and the magnitude of facial expression changes, and can be classified into three stages, for example, low intensity, medium intensity, and high intensity.

[0091] If the user's emotion estimated by the emotion estimation unit 134 is not similar to the target emotion (e.g., happiness) stored in the memory unit 12, and / or if the intensity of the user's emotion estimated by the emotion estimation unit 134 is not within the target range (e.g., medium intensity), the process proceeds to step S160.

[0092] Figure 16 is a flowchart of the process for changing the preference control parameter 126.

[0093] First, in step S161, the controller 13 determines whether the user's emotion estimated by the emotion estimation unit 134 is similar to the target emotion (for example, happiness) stored in the memory unit 12.

[0094] If the user's emotion estimated by the emotion estimation unit 134 is not similar to the target emotion stored in the memory unit 12, the process proceeds to step S162.

[0095] In step S162, the preference estimation unit 135 changes the preference control parameter 126 from the current genre to another genre (a different genre) stored in the preference genre table 124, and completes the process of step 160. The other genre may be randomly selected from the preference genre table 124, or it may be selected sequentially from the preference genre table 124.

[0096] On the other hand, if the user's emotion estimated by the emotion estimation unit 134 is similar to the target emotion stored in the memory unit 12, the process proceeds to step S163.

[0097] In step S163, the controller 13 determines whether the intensity of the user's emotion estimated by the emotion estimation unit 134 is higher than the target range (e.g., medium intensity) stored in the memory unit 12.

[0098] If the intensity of the user's emotion estimated by the emotion estimation unit 134 is higher than the target range stored in the memory unit 12, the process proceeds to step S164.

[0099] In step S164, the preference estimation unit 135 does not change the preference control parameter 126 from the current genre, but sets the level of the preference control parameter 126 one step lower, and then completes the process in step 160.

[0100] If the intensity of the user's emotion estimated by the emotion estimation unit 134 is lower than the target range stored in the memory unit 12, the process proceeds to step S165.

[0101] In step S165, the preference estimation unit 135 does not change the preference control parameter 126 from the current genre, but sets the level of the preference control parameter 126 one level higher, and finishes the process of step 160.

[0102] When the process shown in Figure 16 is executed, if the user's emotion estimated by the emotion estimation unit 134 based on the output of sensor 15 is similar to the target emotion, and the intensity of the user's emotion estimated by the emotion estimation unit 134 based on the output of sensor 15 is outside the target range, the range of the user's preferences can be changed without changing the genre of the user's preferences estimated from the user's emotion estimated by the emotion estimation unit 134 based on the output of sensor 15. Therefore, the possibility of obtaining a revised response that increases the user's satisfaction can be increased, and the intensity of the user's emotion when a revised response is obtained can be converged to an appropriate range (target range), which can suppress, for example, the possibility of strong emotions interfering with driving.

[0103] <3. Variant> The prompt given to the generative AI221 may be audio data, which is text data converted into speech, rather than just text data. In other words, any text data in a format that the generative AI221 can handle is acceptable. If the prompt given to the generative AI221 is audio data, the text data conversion and other processing performed in the above-described embodiment can be replaced with audio data conversion and other processing.

[0104] If the generative AI221 is capable of generating responses to prompts with attached image data, then it is acceptable for image data to be attached to the prompts. Attaching image data makes it possible to include information that is difficult to verbalize in the prompts, thus improving the quality of the prompts.

[0105] Emotions may be estimated based on the output of a camera (image sensor) that detects a person's face, rather than based on the output of a biosensor. Alternatively, emotions may be estimated based on a person's speech patterns (text, volume (intonation), voice quality (intonation), etc.).

[0106] The generation AI 221 may be installed on the in-vehicle unit 1 instead of server 2. If the generation AI 221 is installed on the in-vehicle unit 1, a tourist information system, etc., can be built on the in-vehicle unit 1 alone.

[0107] <4. Things to keep in mind> The various technical features disclosed in the embodiments for carrying out the invention as specified herein can be modified in various ways without departing from the spirit of the technical creation. Furthermore, the multiple embodiments and modifications disclosed in the embodiments for carrying out the invention as specified herein may be combined to the extent possible. [Explanation of symbols]

[0108] 1...In-vehicle device 2. Server SYS1...Tourist Information System

Claims

1. A method for creating prompts to be given to a computer-generated AI, The generation AI is given the previous prompt and information that identifies the previous response generated by the generation AI, Information based on user responses to the previous response, Create a re-prompt that includes a re-response command. How to create a prompt.

2. The prompt generation method according to claim 1, wherein the user's response is the user's emotion estimated based on the output of the sensor.

3. The aforementioned prompt and subsequent prompts include user preference information. The prompt creation method according to claim 2, wherein if the user's emotions are not similar to the target emotions, the preference information to be included in the re-prompt is changed from the preference information included in the prompt.

4. The aforementioned prompt and subsequent prompts include user preference information such as preferred genre and preferred intensity. The prompt creation method according to claim 3, wherein if the user's emotion estimated based on the output of the sensor is similar to the target emotion, and the intensity of the user's emotion estimated based on the output of the sensor is outside the target range, the user's preferred genre to be included in the re-prompt is the same as the user's preferred genre included in the prompt, and the intensity of the user's preference to be included in the re-prompt is changed from the intensity of the user's preference included in the prompt.

5. Information identifying the previous prompt given to the generation AI and the previous answer generated by the generation AI, Information based on user responses to the previous response, Create a re-prompt that includes a re-response command, The aforementioned re-prompt is output to the generation system AI, Obtain a new response from the aforementioned generation AI, Control is performed based on the acquired re-response. Generation AI utilization device.

6. The generative AI utilization device according to claim 5, wherein the user's response is the user's emotion estimated based on the sensor output.

7. Information identifying the previous prompt given to the generation AI and the previous answer generated by the generation AI, Information based on user responses to the previous response, Create a re-prompt that includes a re-response command, The aforementioned re-prompt is output to the generation system AI, Obtain a new response from the aforementioned generation AI, Perform control based on the acquired re-response, A generative AI-powered program that uses a computer to execute commands.

8. The generative AI utilization program according to claim 7, wherein the user's response is the user's emotion estimated based on the sensor output.

9. A tourist information system including a server device and a tourist information device, The server device includes a generation AI, The prompt from the tourist information device is applied to the generation AI, and the resulting response is transmitted to the tourist information device. The aforementioned tourist information device, The aforementioned generation system AI includes a sensor that detects the user's response to the previous answer generated in response to the previous prompt, A new prompt is created, which includes information identifying the previous prompt and the previous response, information based on the output of the sensor, and a new response command. The aforementioned re-prompt is output to the server device, The server device obtains a new response to the re-prompt, Control is performed based on the acquired re-response. Tourist information system.

10. The tourist information system according to claim 9, wherein the user's response is the user's emotion estimated based on the output of the sensor.

Images