Action suggestion device, action suggestion method, and program

Abstract

The present invention provides an action suggestion device capable of generating optimal actions that are appropriate to the conditions of the environment in which the user is acting. [Solution] A management computer 11 equipped with a processor 17 that performs processing, wherein the processor 17 is configured to perform: a first process that obtains prompts for when user P2 takes action in the environment; a second process that determines the state of the environment in which user P2 takes action; a third process that extracts actions that user P2 can perform in the environment based on the state of the environment; a fourth process that estimates feature quantities including comfort, efficiency, safety, and economy for user P2, assuming that user P2 will perform the actions extracted in the third process; and a fifth process that generates the optimal action that user P2 can perform in the state of the environment based on the prompts obtained in the first process and the feature quantities estimated in the fourth process.

Description

Technical Field

[0001] The present disclosure relates to an action proposal device, an action proposal method, and a program for proposing to a user an action corresponding to a state in the user's environment.

Background Art

[0002] An example of an action proposal device that proposes to a user an action corresponding to a state in the user's environment is disclosed in Patent Document 1. The action proposal device disclosed in Patent Document 1 aims to provide a technique for improving the availability of a device equipped with an artificial intelligence model by making it possible to respond to changes in indicators requested by a user.

[0003] To achieve the above object, the action proposal device disclosed in Patent Document 1 includes an acquisition unit, a generation unit, and an output unit. The acquisition unit acquires information related to the situation of an instruction target as situation information. The instruction target is, for example, an unmanned aerial vehicle, an automobile, a person, or a robot. The generation unit uses an artificial intelligence model generated based on the relationship between the surrounding situation of the instruction target and the action of the instruction target based on a preset indicator, to generate the action of the instruction target corresponding to the situation information as proposal instruction content. In addition, the generation unit generates a plurality of pieces of proposal instruction content corresponding to the situation information using a plurality of the artificial intelligence models respectively based on a plurality of mutually different indicators. Further, the output unit outputs the generated plurality of pieces of proposal instruction content.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0005] The inventors of this application recognized that the behavior suggestion device described in Patent Document 1 uses an artificial intelligence model generated based on the relationship between the suggested behavior and a predetermined index, and therefore still has room for improvement in terms of generating the optimal behavior that is suitable for the state of the environment in which the user is acting.

[0006] The purpose of this disclosure is to provide an action suggestion device, an action suggestion method, and a program that can generate optimal actions that are suited to the conditions of the environment in which the user acts. [Means for solving the problem]

[0007] This embodiment discloses an action suggestion device equipped with a processor that performs processing, wherein the processor performs: a first process of obtaining prompts for when a user acts in an environment; a second process of determining the state of the environment in which the user acts; a third process of extracting actions that the user can perform in the environment based on the state of the environment; a fourth process of estimating feature quantities including the user's comfort, efficiency, safety, and economy, assuming that the user will perform the actions extracted in the third process; and a fifth process of generating the optimal action that the user can perform in the state of the environment based on the prompts obtained in the first process and the feature quantities estimated in the fourth process. [Effects of the Invention]

[0008] According to this embodiment, the processor generates the optimal action that the user can perform given the environmental conditions, based on features including prompts, user comfort, efficiency, safety, and economy. Therefore, it can generate the optimal action that matches the environmental conditions and the prompts generated by the user. Consequently, the accuracy of the processor's optimal action generation is improved. [Brief explanation of the drawing]

[0009] [Figure 1] This is a conceptual diagram showing an example configuration of an action suggestion system, including an action suggestion device. [Figure 2] This is a conceptual diagram showing an example configuration of a management server, which is an example of an action suggestion device. [Figure 3] This flowchart shows an example of an action suggestion method implemented by the action suggestion system. [Modes for carrying out the invention]

[0010] (Example of an action suggestion system configuration) In this embodiment, several specific examples of the action suggestion system, action suggestion device, action suggestion method, and program are described with reference to the drawings. Figure 1 shows an example of the action suggestion system 10. The action suggestion system 10 consists of a management computer 11, terminal equipment 12, an external computer 13, etc. The management computer 11 is connected to the terminal equipment 12 and the external computer 13 via a network 14. The external computer 13 has a processor, storage device, communication device, etc., and the external computer 13 provides a website to the network 14.

[0011] The management computer 11 is a computer managed and operated by administrator P1. Terminal devices 12 are operated and used by user P2. Multiple terminal devices 12 operated by different users P2 may be provided. User P2 may be an individual, a company representative, a government representative, etc. External computers 13 are managed by administrator P3, who is different from administrator P1 of the management computer 11. Multiple external computers 13 managed by different administrator P3 may be implemented.

[0012] Network 14 is a communication line for transmitting various types of information, and is composed of base stations, communication satellites, communication antennas, communication equipment (access points), repeaters, communication circuits, communication cables (copper wires, optical fibers), etc. Repeaters include modems (devices that convert analog signals to digital signals), hubs (concentrators), routers (devices that connect terminals to the internet), etc. The communication line includes one or more communication methods, including wireless communication and wired communication. Wireless communication includes short-range wireless communication. Short-range wireless communication includes, for example, wireless LAN (Wi-Fi®) and Bluetooth®. Wireless communication includes electrical signals, optical signals, radio waves, infrared rays, satellite communication, etc. Signals used for communication include digital signals and analog signals.

[0013] (Example configuration for the management server) The management computer 11 is a computer implemented by various hardware and various software. The software includes an operating system and applications. The management computer 11 may have hardware such as a main unit (casing), ROM (Read Only Memory) 16, a processor 17, main memory 18, auxiliary memory 19, and communication device 20. The ROM 16 is a data read-only memory and the data written to it during manufacturing is not modified afterward. The ROM 16 stores programs that are executed first when the management computer 11 starts up, such as the IPL (Initial Program Loader). ROM is a non-volatile storage device.

[0014] The processor 17 is located inside the main body of the management computer 11 and may be composed of a central processing unit (CPU) that integrates an arithmetic unit (arithmetic circuit) and a control unit (control circuit). The processor 17 is connected to the ROM 16, main memory 18, auxiliary memory 19, communication device 20, etc. via a communication bus 21. The processor 17 comprehensively controls other devices and circuits located inside the main body and devices and circuits located outside the main body.

[0015] Furthermore, the processor 17 may have, in place of or in addition to the central processing unit, arithmetic processing circuits such as a digital signal processor, an application-specific integrated circuit (ASIC), a GPU (Graphics Processing Unit), or an MPU (Micro Processing Unit). The GPU processes image data and audio data.

[0016] The processor 17 performs various processes by running a program. The processes performed by the processor 17 include arithmetic, judgment, learning, comparison, identification, classification, control, generation, etc. The processes performed by the processor 17 include reading a program from the auxiliary storage device 19, running the program read from the auxiliary storage device 19 in the main memory 18, retrieving information and data from the auxiliary storage device 19 according to the program and processing it, processing information and data acquired from the external computer 13, processing information and data acquired from the terminal device 12, storing the processing results performed in the main memory 18 in the auxiliary storage device 19, processing information and data to be transmitted to the terminal device 12, etc.

[0017] The main memory 18 is a volatile memory device, implemented, for example, by RAM (Random Access Memory). The main memory 18 functions as a work area and buffer area for storing programs, data, instructions, etc., retrieved from the auxiliary memory 19 when the processor 17 processes and executes those programs, data, instructions, etc. The main memory 18 is a non-temporary storage medium.

[0018] The auxiliary storage device 19 is a non-transitory storage medium. The auxiliary storage device 19 may also be recognized as storage. The auxiliary storage device 19 may be provided inside the main body, or may be provided outside the main body and connected to the communication bus 21 via wireless communication or wired communication. The auxiliary storage device 19 has a larger capacity than the main storage device 18. The auxiliary storage device 19 operates according to input instructions and output instructions from the processor 17.

[0019] The auxiliary storage device 19 is configured by one or more of, for example, a magnetic storage method, an optical storage method, a magneto-optical method, a semiconductor storage method, and the like. The magnetic storage method is implemented by a hard disk, a floppy disk, a magnetic tape, and the like. The optical storage method is a configuration in which the disk surface of a storage medium 19A is irradiated with laser light to record data with light or heat, and the reflection of the laser light is read by an optical head.

[0020] The optical storage method is implemented by an optical drive and the storage medium 19A. The optical drive includes a combo drive, a super combo drive, a multi-drive, a super multi-drive, a hyper drive, and the like. The storage medium 19A includes a compact disc, a digital video disc, a Blu-ray disc, and the like. The magneto-optical method is implemented by, for example, a magneto-optical disk.

[0021] The magnetic storage method and the optical storage method each have a head and the storage medium 19A. The storage medium 19A may be configured to be attachable to and detachable from the main body. The semiconductor storage method may be defined as a flash memory. Examples of the flash memory include an SD memory card, a USB flash drive, a solid-state drive, and the like. The flash memory can also be recognized as the storage medium 19A configured to be attachable to and detachable from the main body. The auxiliary storage device 19 stores non-transitory programs, various types of information and data.

[0022] The processor 17 executes a non-transitory program and cooperates with various hardware constituting the management computer 11, thereby realizing various functional components shown in Figure 2. The processor 17 realizes, for example, a website management unit 22, a user information processing unit 23, an acquired information processing unit 24, an artificial intelligence unit 25, and the like.

[0023] When the web application is executed by the processor 17, the website management unit 22 provides a website at a predetermined Uniform Resource Locator (URL) on the network 14. When the terminal device 12 accesses the management computer 11 via the website, the website management unit 22 realizes a function of transmitting a web page described in HTML (Hyper Text Markup Language) format to the terminal device 12, a function of acquiring information and data transmitted from the terminal device 12, and a function of transmitting information and data to the terminal device 12.

[0024] The user information processing unit 23 processes the user information acquired from the terminal device 12 to perform user registration, and stores the user information and the user registration result in the auxiliary storage device 19. When the user P2 is an individual, the user information includes name, address, location of place of work, age, gender, hobby, product purchase history, occupation, email address, account for identifying the individual, and the like. The account includes an identification number and a password.

[0025] When the user P2 is a person in charge of a company, the user information includes company name, company location, company industry type, email address, account for identifying the company, and the like. The account for identifying the company includes an identification number, a password, and the like. When the user P2 is a person in charge of a government, the user information includes the name of a government office or ministry agency, the location of the government office or ministry agency, the public projects and operations handled, an email address, and an account for identifying the name of the government office or ministry agency. The account includes an identification number and a password.

[0026] The acquired information processing unit 24 processes information acquired from the terminal device 12 to acquire auxiliary information other than user information, and also processes various information and data acquired from the external computer 13. The acquired information processing unit 24 also stores the auxiliary information and various information and data acquired from the external computer 13 in the auxiliary storage device 19. The auxiliary information acquired from the terminal device 12 may include the location information of user P2, the biometric information of user P2, the emotional information of user P2, prompts, information on user P2's actions, etc. Of the auxiliary information, the location information of user P2, the biometric information of user P2, prompts, and information on user P2's actions will be described later.

[0027] The acquired information processing unit 24 indirectly acquires user P2's emotional information by processing information and data acquired from the terminal device 12. The acquired information processing unit 24 works in cooperation with the artificial intelligence unit 25 and the auxiliary storage device 19 to estimate user P2's emotional information from image information and image data acquired from the terminal device 12 and convert it into text data. User P2's emotions include joy, satisfaction, anger, sadness, calmness, relaxation, enjoyment, surprise, comfort, discomfort, relief, relaxation, confusion, etc. These emotions can be estimated by processing the degree of eyelid opening, gaze, eyebrow movement, presence or absence of wrinkles on the nose, mouth movement, degree of mouth opening, pupil dilation, etc., contained in the image of user P2's upper body, including the face. These emotions can also be estimated from the movements of user P2's upper body, such as whether or not the head is tilted, whether or not the head is nodded, hand movements, etc. The estimated emotions can be quantified and a table or graph can be generated.

[0028] Furthermore, the technology for analyzing user P2's emotions based on image information including user P2's face is publicly known, as shown in, for example, Japanese Patent Publication No. 6868422, Japanese Patent Publication No. 6703893, Japanese Patent Publication No. 6042015, Japanese Patent Publication No. 3953024, Japanese Patent Publication No. 4458888, Japanese Unexamined Patent Publication No. 2015-229040, Japanese Unexamined Patent Publication No. 2018-32164, Japanese Unexamined Patent Publication No. 2022-139436, Japanese Unexamined Patent Publication No. 2020-184216, etc., so a detailed explanation will be omitted.

[0029] Details of the various information and data that the management computer 11 obtains from the external computer 13 will be described later. A prompt is information entered by user P2 on the terminal device 12 when user P2 attempts to obtain the optimal action from the management computer 11. The management computer 11 obtains the prompt from the terminal device 12. Details of the prompt will be described later.

[0030] When the program is run, the artificial intelligence unit 25 works in cooperation with the auxiliary storage device 19 to perform various processes. The processes performed by the artificial intelligence unit 25 include learning processes (learning phase) and inference processes (inference phase). The information and data used by the artificial intelligence unit 25 to perform the processes include image data, sound data, text data, various models, etc.

[0031] The artificial intelligence unit 25 outputs processing results (output data) by processing the input training data with a learning model during the learning process, for example, in the machine learning process. The artificial intelligence unit 25 also calculates the error between the processing result and the correct data using a loss function and trains the learning model based on that error. Specifically, it adjusts weights, biases, etc., to minimize the error. In the inference process, the artificial intelligence unit 25 outputs inference results by processing the data to be inferred with the learning model.

[0032] The processing of input data performed by machine learning includes extracting features from the input data. These features include classification and regression. Classification is the process of classifying and identifying features. Regression is the process of understanding the trends of the classified features. The learning model is stored in auxiliary memory 19, and as machine learning is repeated, the learning model is continuously trained.

[0033] The machine learning performed by the Artificial Intelligence Department 25 may include three types: supervised learning, unsupervised learning, and reinforcement learning. In supervised learning, the learning model is trained by comparing the output data with the correct answer data. Unsupervised learning is a method of training using training data that does not contain the correct answer, and it involves classifying the training data into groups of data that have similar features (such as regularities).

[0034] Reinforcement learning repeatedly processes input data with a learning model to obtain output data, generating "actions to achieve this objective." Specifically, it may generate the optimal action to achieve the objective. The reinforcement learning performed by the artificial intelligence unit 25 includes the following four stages.

[0035] In the first stage, the artificial intelligence unit 25 processes the input data using a learning model and determines the state of the environment for user (acting agent) P2. The real world in which user P2 acts corresponds to the environment, and the state of the environment in the real world includes the weather in the real world, user P2's biometric information, user P2's current location, real world economic information, real world traffic information, and information on products purchased by user P2.

[0036] In the second stage, the artificial intelligence unit 25 uses the learning model to determine what actions user P2 can perform in the current environment. In the example of stock trading, the actions would be which stocks user P2 will buy, at what price, and in what quantity. The learning model may also determine what actions can be performed in the current environment under predetermined conditions in response to prompts obtained from the terminal device 12.

[0037] In the third stage, the artificial intelligence unit 25 uses a learning model to predict the degree of reward that can be obtained by adopting the "feasible actions determined in the second stage." In the example of stock trading, the profit resulting from buying and selling stocks corresponds to the degree of reward. The technical meaning of the reward varies depending on the current environment, the feasible actions, and the content of the prompts obtained from the terminal device 12.

[0038] Then, in the fourth stage, the artificial intelligence unit 25 repeatedly trains its learning model so that, based on the environmental assessment results, feasible actions, and the predicted degree of reward obtained by adopting the feasible actions, it can select the action that will result in a relatively large (maximized) reward in the future (practical application stage).

[0039] For example, the learning model may be trained so that the value (Q-value / state-action value) of an action taken by user P2 in the environment is relatively large, and preferably so that the value is maximized. The Q-value is a function that includes the intermediate reward (immediate reward) in the short term and the long term reward (expected value) in the long term. In practice, it is difficult to calculate the expected value. Therefore, the management computer 11 proposes the optimal action to the terminal device 12, and user P2 performs the optimal action. Next, the result of the action is sent from the terminal device 12 to the management computer 11. The artificial intelligence unit 25 updates the Q-value by evaluating the result of the action obtained from the terminal device 12.

[0040] Furthermore, when the processor 17 runs the program, the artificial intelligence unit 25 in this embodiment has the function of automatically generating the optimal action for user P2 by having the learning model execute processing using various information, data and prompts acquired from the terminal device 12, various information and data acquired from the external computer 13, and various information and data stored in the auxiliary storage device 19 during the inference stage.

[0041] The artificial intelligence unit 25 can use a neural network as a learning model to realize machine learning. A neural network has an input layer, an intermediate layer (hidden layer), and an output layer. The intermediate layer is configured to make the input information useful for the output layer. Furthermore, the artificial intelligence unit 25 can perform deep learning in realizing the intermediate layer of the neural network. In deep learning, the intermediate layer is multi-layered with multiple types of stages (nodes), and weights and biases are assigned to the functions (features) contained in the information. Deep learning can reduce noise in the information and data input to the learning model because it automatically instructs the intermediate layer to focus on the functions (features).

[0042] Furthermore, the artificial intelligence unit 25 may use the functions of the large language model to process prompts, generate sentences that suggest optimal actions, and so on. Large language models (LLMs) are a type of generative artificial intelligence specialized in natural language processing (NLP). Large language models are language models that realize advanced natural language generation (LNG) by being constructed through deep learning of a vast amount of text data. Natural language processing processes natural language by combining processes such as morphological analysis, syntactic analysis, semantic analysis, contextual analysis, and intent analysis, enabling machine translation, text summarization, speech recognition, and more.

[0043] A language model is a system that predicts the word that follows a given word by understanding the patterns of natural language used by humans and analyzing the probability of word occurrence. Examples of large-scale language models include ChatGPT (Chat Generative Pre-trained Transformer) (registered trademark), PaLM2 (registered trademark), and LaMDA2 (registered trademark). The words and sentences used as instructions to input into these large-scale language models are called prompts.

[0044] When the artificial intelligence unit 25 functions as a generative artificial intelligence unit, it can perform text generation, image generation, video generation, and speech generation. Text generation is a function that, when a prompt (in text format), such as an instruction or question, is input, analyzes the content and intent of the prompt and uses the information and data stored in the auxiliary storage device 19 to automatically generate a response (in text format) to the prompt. Because a large-scale language model is used, natural and highly accurate responses can be obtained.

[0045] Image generation is a function that automatically generates an original image similar to the prompt's image by using information and data stored in the auxiliary storage device 19 in response to an input prompt (in text format). Examples of artificial intelligence that generate images include stable diffusion, mid-journey, and Dali-2.

[0046] Video generation is a function that automatically generates an original video that closely resembles the image of a prompt (in text format) by using information and data stored in the auxiliary storage device 19 in response to an input prompt. Gentoo is a well-known example of artificial intelligence that generates videos.

[0047] Voice generation is a function that automatically generates original voice data (optimal actions, responses, music, etc.) in response to input prompts (in voice or text format) using information and data stored in the auxiliary storage device 19.

[0048] The auxiliary storage device 19 stores information and data resulting from processing performed by the processor 17, and the information stored in the auxiliary storage device 19 includes hypertext data that constitutes web pages on a website.

[0049] Furthermore, as shown in Figure 2, the auxiliary storage device 19 is implemented including a program storage unit 26, a user information storage unit 27, a model storage unit 28, various information storage units 29, a processing result storage unit 30, and the like. The program storage unit 26 stores applications and various programs run by the processor 17. The applications stored in the program storage unit 26 include non-temporary programs, manuals, configuration files, data storage files, data, various libraries, and the like.

[0050] The user information storage unit 27 stores user information and registration information acquired from the terminal device 12. The model storage unit 28 stores learning models and large-scale language models, etc. The various information storage unit 29 stores information and data acquired from the external computer 13, prompts and auxiliary information acquired from the terminal device 12, map data, etc. The processing result storage unit 30 stores various processing results performed by the processor 17, various decision results, and the generated optimal actions, etc.

[0051] The communication device 20 includes devices, equipment, and standards for connecting the management computer 11 to the network 14 via at least one of either a wireless communication system or a wired communication system. The hardware constituting the communication device 20 may consist of, for example, a LAN card, a network adapter, a network interface card, a communication cable, a communication antenna, a communication port, a modem, a hub (concentrator), a router, etc.

[0052] (Configuration of terminal equipment) Terminal device 12 is a computer used and operated by user P2. Terminal device 12 includes, for example, portable computers and fixed computers. Portable computers include smartphones and tablet computers. Fixed computers include desktop computers, notebook computers, etc.

[0053] The terminal device 12 is connected to the management computer 11 and the external computer 13 via the network 14. As shown in Figure 1, the terminal device 12 includes a main unit (casing), a processor 40, main memory 41, auxiliary storage 42, an input device 43, an output device 44, a current location detection device 45, a biometric information detection device 46, a communication device 47, etc. The processor 40 is located inside the main unit and may consist of a central processing unit (CPU) in which an arithmetic unit (arithmetic circuit) and a control device (control circuit) are integrated. The processor 40 is communicably connected to the main memory 41, auxiliary storage 42, input device 43, output device 44, current location detection device 45, biometric information detection device 46, and communication device 47 via the communication bus 48. The processor 40 has a configuration that comprehensively controls other devices and circuits located inside the main unit and devices and circuits located outside the main unit.

[0054] The processor 40 performs information and data processing by running non-temporary programs. Specifically, the processor 40 processes text data, sound data, image data, etc. Text data includes characters, numbers, sentences, symbols, figures, etc. Image data includes video, moving images, still images, photographs, etc. The processing performed by the processor 40 includes calculations, decisions, comparisons, control, etc.

[0055] The processes performed by the processor 40 include reading information and data from the auxiliary storage device 42, processing information and data obtained from the management computer 11, updating information and data stored in the auxiliary storage device 42, and sending the information and data processed by the processor 40 to the management computer 11.

[0056] The main memory 41 is a volatile memory device, and functions as a work area and buffer area for storing programs, data, instructions, etc., retrieved from the auxiliary memory 42 when the processor 40 executes processing of programs, data, instructions, etc. The main memory 41 is a non-temporary storage medium.

[0057] The auxiliary storage device 42 is a non-temporary storage medium. Non-temporary applications are stored in the auxiliary storage device 42. In addition, information and data processed by the processor 40 are stored in the auxiliary storage device 42. The information and data stored in the auxiliary storage device 42 include the processing results of the processor 40, various data, the operation details of the input device 43, control signals to the output device 44, etc.

[0058] The processor 40 is configured to perform various processes by reading and executing non-temporary programs stored in the auxiliary storage device 42. For example, the processor 40 is configured to process information and data input by the input device 43 and to process the current location of user P2 detected by the current location detection device 45. The processor 40 is configured to process biometric information detected by the biometric information detection device 46 and to process information and data output from the communication device 47. The processor 40 is configured to process information and data input via the communication device 47. Furthermore, the processor 40 is configured to store the processing results in the auxiliary storage device 42.

[0059] The auxiliary storage device 42 has a larger capacity than the main storage device 41, and operates according to input and output commands from the processor 40. The auxiliary storage device 42 is implemented, for example, by a head and a non-temporary storage medium 42A. The head writes information and data to the storage medium 42A and reads information and data from the storage medium 42A. The storage medium 42A is implemented by a magnetic disk, optical disk, flash memory, etc. An example of a magnetic disk is a hard disk drive. An example of an optical disk is a compact disc, digital video disc, Blu-ray disc, etc. Flash memory is a type of semiconductor memory, and an example of a flash memory is an SD memory card, USB flash drive, solid state drive, etc. The storage medium 42A may be configured to be attachable to and detachable from the main unit.

[0060] The input device 43 is operated when inputting information and data to the terminal device 12, when the terminal device 12 performs various processes, when the terminal device 12 sends various information and data to the network 14, when the terminal device 12 acquires various information and data via the network 14, and so on.

[0061] The input device 43 is a device operated by user P2, and may consist of devices such as a keyboard 43A, mouse 43B, microphone 43C, camera 43D, and display 43E. The display 43E may be implemented as, for example, a liquid crystal display or an organic electroluminescent display. Images, videos, graphics, characters, operation buttons, operation tabs, etc., are displayed on the screen of the display 43E. The display 43E may be understood as a monitor. User P2 can input prompts as text information by operating one or more of the keyboard 43A, mouse 43B, operation buttons, and operation tabs displayed on the screen of the display 43E. User P2 can also input prompts as voice information via the microphone 43C.

[0062] Microphone 43C is an electronic component that converts acquired sound data into electrical signals and outputs them. Microphone 43C is connected to the communication bus 48 by either a wireless or wired communication system. Camera 43D is connected to the communication bus 48 by either a wireless or wired communication system. Camera 43D captures images of objects and generates images. Images include videos, still images, photographs, etc. Images generated by camera 43D are processed by processor 40 and stored in auxiliary storage device 42.

[0063] By using the input device 43, user P2 can input a destination in the map data, select a travel route in the map data, access the network 14, access the management computer 11, send information and data to the management computer 11, select the optimal action obtained from the management computer 11, and so on. Note that some of the information and data processed by the terminal device 12 may be sent in real time from the terminal device 12 to the management server 123 even when the input device 43 is not being operated.

[0064] The output device 44 includes devices such as a display 43E, a printer 60, and an audio output device 61. The display 43E is the same as the display 43E of the input device 43. The audio output device 61 is a device that converts electrical signals contained in information and data into sound information and outputs it. The audio output device 61 is implemented by a speaker, headphones, earphones, etc. The audio output device 61 is connected to the communication bus 48 by wireless or wired communication. The printer 60 prints information and data onto paper using toner or ink and outputs it to the screen of the display 43E.

[0065] The current location detection device 45 is composed of, for example, a detection circuit and a decision circuit. The current location detection device 45 processes signals received from the satellite via the communication device 47, and signals from the gyro sensor 45A and magnetic sensor 45B installed in the terminal device 12. Based on these processing results and the map data stored in the auxiliary storage device 42, the current location detection device 45 can determine the current location and travel path of the terminal device 12 in the real world by replacing them with map data. User P2 can display the map on the display 43E and visually confirm their current location and travel path.

[0066] Furthermore, the biometric information detection device 46 may also detect the user P2's heart rate (pulse rate) per predetermined time and the user P2's respiratory rate per predetermined time. Methods for detecting the user P2's heart rate include obtaining it from a wristwatch-type sensor 46A or the like worn by the user P2, and indirectly estimating it from video information from the camera 43D. The technique for estimating heart rate from video information is a known technique, as shown in Japanese Patent Publication No. 6717424, Japanese Patent Application Publication No. 2016-190022, Japanese Patent Application Publication No. 2022-141984, etc., so a detailed explanation is omitted.

[0067] The respiratory rate of user P2 per predetermined time can be indirectly estimated from the video information of camera 43D. The technique for estimating the respiratory rate of a subject from video information is a known technique, as described in Japanese Patent Publication No. 4284538 and Japanese Patent Application Publication No. 2016-190022, so a detailed explanation will be omitted.

[0068] The communication device 47 includes devices, equipment, and standards for connecting the terminal device 12 to the network 14 by at least one of the following communication systems: a wireless communication system or a wired communication system. The devices and equipment for connecting the terminal device 12 to the network 14 may include cables, antennas, routers, communication circuits, communication ports, communication connectors, communication hubs, etc.

[0069] (An example of a method for proposing action) An example of the action suggestion method performed by the action suggestion system 10 is shown in Figure 3. Figure 3 shows the learning process and inference process performed by the management computer 11 in an integrated manner. The learning process performed by the management computer 11 includes steps S10 to S17 in Figure 3. The learning process performed by the management computer 11 does not include steps S18 to S23 in Figure 3.

[0070] In step S30, terminal device 12 may send various information to management computer 11. The various information sent from terminal device 12 to management computer 11 includes user information and auxiliary information. In step S40, external computer 13 may send various information to management computer 11. The various information sent from external computer 13 to management computer 11 will be described later.

[0071] The inference processing performed by the management computer 11 includes steps S10 to S13 in Figure 3, and also includes steps S18 to S23. The inference processing performed by the management computer 11 does not include steps S14 to S17 in Figure 3.

[0072] In step S10, the management computer 11 acquires various information from the terminal device 12 and also acquires various information from the external computer 13. This information includes instructions, information, and data. The information acquired by the management computer 11 from the terminal device 12 includes user information, auxiliary information, etc. The auxiliary information is associated with user P2's account.

[0073] A prompt is text-formatted data created by user P2 when user P2 attempts to obtain the optimal action from the management computer 11.

[0074] The various types of information that the management computer 11 obtains from the external computer 13 in step S10 may include, for example, trend information included in SNS (social networking service), weather information for the area where user P2 is active, economic data, and traffic information for the area where user P2 is active. The trend information included in SNS may include information on events that are hot topics in society, information on popular products, information on popular fashion, etc.

[0075] Economic data may include price information for the area where user P2 operates, market trends in the area where user P2 operates, stock price information, real estate price information, production volume of goods by companies, etc. Market trends may include, for example, the Consumer Price Index. Stock price information may include the stock price of each stock traded on the stock market, the amount of change in the stock price of each stock, the number of trades for each stock, etc. Traffic information for the area where user P2 operates may include road congestion, information on troubles and delays in public transportation, and delivery times in logistics. User P2's purchase history may include user P2's past purchase history of goods. Purchase history of goods may be from either e-commerce (EC) sites or physical stores. Furthermore, goods include tangible goods, intangible goods, services, etc.

[0076] Furthermore, the various data that the management computer 11 acquires from the external computer 13 in step S10 may include breaking news (fashion event in Shibuya), information analyzing user P2's emotions (sales are great on social media), energy consumption (power consumption) at user P2's home or company, schedule information for user P2 or the company, and health risk information for user P2.

[0077] Furthermore, the various data that the management computer 11 acquires from the external computer 13 in step S10 may include company inventory data information (clothing store inventory rate 50%), public transport operation information (trains running normally), restaurant congestion information (cafes have available seats), weather warnings (no disasters), exchange rates (yen weak: 1 dollar = 150 yen), ticket inventory information (50 tickets remaining for the event), medical appointment information for user P2 (no hospital appointment), and work task information for user P2 or the company (number of unprocessed tasks).

[0078] Furthermore, the various data that the management computer 11 acquires from the external computer 13 in step S10 may include crime information in the area where user P2 is active (Shibuya is safe), information detected by environmental sensors (air cleanliness), and information regarding the delivery status of goods at a company (delivery is scheduled for 12:00). The various data that the management computer 11 acquires from the external computer 13 in step S10 may also include information posted by influencers (a celebrity posts about going to a sale), and information regarding the progress of user P2's learning (50% progress in English learning). The various data that the management computer 11 acquires from the external computer 13 in step S10 may also include information on congestion at tourist spots (for example, Harajuku is somewhat crowded), electricity market prices (for example, 1 kWh = 30 yen), and information regarding the condition of pets owned by user P2 (the dog is healthy and can be walked).

[0079] Furthermore, the various data that the management computer 11 obtains from the external computer 13 in step S10 may include climate change index, space weather, regional event density, road construction information, school event schedule, public facility utilization rate, internet connection speed, game play status, music trends, movie box office revenue, book sales rankings, cooking recipe popularity, sports match results, fashion trends, detailed regional weather forecasts, medical facility congestion, traffic accident rate, public Wi-Fi (registered trademark) utilization rate, regional population density, electricity demand forecast, water usage, gas consumption, garbage collection schedule, regional SNS activity, disaster recovery progress, election-related information, public transport ridership, etc.

[0080] Furthermore, the various data that the management computer 11 acquires from the external computer 13 in step S10 may include bicycle sharing usage rates, regional temperature change rates, pollen dispersal forecasts, regional event participation rates, tourist influx, regional economic growth rates, job posting update rates, regional infrastructure utilization rates, traffic signal optimization data, regional energy self-sufficiency rates, public service satisfaction, regional cultural event frequency, regional health statistics, educational facility utilization rates, regional crime prevention rates, regional volunteer activities, regional traffic safety index, regional disaster preparedness, regional environmental protection index, regional educational outcomes index, regional employment rates, regional tourism satisfaction, regional medical access, regional traffic accident prevention rates, regional energy efficiency, and the like.

[0081] In step S11, the management computer 11 determines whether an "error" has occurred that prevents the processing from being executed normally. For example, the management computer 11 may determine that an error has occurred if the processing time of the acquired data exceeds a predetermined value (default 1 second), and that there is no error (normal) if the processing time of the acquired data is less than or equal to the predetermined value. For example, the management computer 11 may determine that an error has occurred if the cache of the main memory 18 exceeds 2 GB in 5 minutes, and that there is no error (normal) if the cache of the main memory 18 is 2 GB or less in 5 minutes. For example, the management computer 11 may determine in step S11 that an error has occurred if one or more of the data log entries, that is, the data input values ​​or score calculation results processed by the artificial intelligence unit 25, are missing or abnormal values. Conversely, if there are no missing data input values ​​or score calculation results processed by the artificial intelligence unit 25, and they are all normal values, the management computer 11 may determine in step S11 that there is no error (normal).

[0082] If the management computer 11 determines Yes in step S11, in step S12 it processes the data using alternative values ​​as scores and proceeds to step S10. For example, for user P's location information, the last location information from the previous data acquisition may be used as an alternative value. For weather information, for example, the average value for the same month and day over the past 10 years may be used as an alternative value. Also, for user P2's biometric information, standard values ​​for the same age, same sex, approximately the same weight, approximately the same height, and the same occupation may be used as an alternative value. For economic information, the average value for user P2's region may be used as an alternative value. For traffic information, predicted values ​​may be used as an alternative value. For purchase history, the average purchase history of another user who is the same age, same sex, same hobbies, etc. as user P2 may be used as an alternative value. In step S12, alternative values ​​other than user P's location information are predefined fixed values ​​or estimated values ​​and are stored in the auxiliary storage device 19 beforehand.

[0083] If the management computer 11 determines No in step S11, it performs normalization and imputation processing in step S13. Normalization is a preprocessing step that eliminates data duplication and enables consistent data handling. For example, various types of data may be converted into numerical values ​​(scores) within the range of 0.1 to 1.0. SNS information is converted to relatively higher numerical values ​​as its popularity or topicality increases. Weather information is converted to relatively higher numerical values ​​if conditions such as user P2's actions are relatively more likely to be encouraged, safety is relatively high, or visibility is relatively good are met. For example, sunny weather is converted to 1.0, cloudy to 0.8, rainy to 0.6, foggy to 0.3, snowy to 0.2, etc.

[0084] User P2's biometric data is converted to a relatively larger value the smaller the difference between User P2's heart rate and the average heart rate for its age group. User P2's biometric data is converted to a relatively larger value the smaller the difference between User P2's blood pressure and the average blood pressure for its age group. User P2's location information is converted to a relatively larger value the smaller the distance between its current location and its destination location. Economic data is converted to a relatively larger value the lower the price index is.

[0085] Traffic information, for example, is converted to a relatively larger value the lower the road congestion rate. Purchase history, for example, is converted to a relatively larger value the more times user P2 has purchased goods, or the more items purchased, etc. The imputation process performed in step S13 is to imputate missing values ​​in various data with substitute values. The substitute values ​​used in step S13 have a different technical meaning from the substitute values ​​used in step S12. The substitute values ​​used in step S13 are values ​​used to imputate missing values ​​before the normalization process. In step S13, one of the following is selected as the substitute value depending on the situation: a fixed value, an estimated value (predicted value), a previous value, an average value, etc.

[0086] In step S14, the management computer 11 may make one or more decisions regarding whether there is missing data or whether the data is abnormal. Even if data exists, the management computer 11 may determine in step S14 that the data is missing if it does not meet the criteria necessary to suggest the optimal action, that is, the quality criteria necessary for score calculation. As a prerequisite for the decision in step S14, the number and types of data necessary for score calculation are predetermined. The necessary data may include, for example, trend information included in SNS, weather information for the area where user P2 is active, biometric data of user P2, location information of user P2, economic data, traffic information for the area where user P2 is active, and purchase history of user P2.

[0087] Then, in step S13, the weather data was interpolated with the most recent value, but if a sudden weather change occurs afterward and the most recent value and the current situation do not match, it is determined in step S14 that the data is missing.

[0088] Furthermore, if some data has been completed but other data is missing, and all the data necessary for score calculation is not available, the management computer 11 may determine in step S14 that the data is missing. In addition, if the elapsed time from the completion request in step S13 until the completion of the completion process exceeds a predetermined time, for example, 1 second, the management computer 11 may consider the substitute value used in the completion process to have expired and determine in step S14 that the data is missing.

[0089] Furthermore, the management computer 11, in a histogram of a normal distribution showing the variability of each data point, the acquired data is The mean of the data ± 3σ If the data falls outside this range, it is considered abnormal. Here, "3σ" represents the range of 99.7% of the total area of ​​the histogram, and "σ" represents the standard deviation.

[0090] If the management computer 11 determines "Yes" in step S14, that is, if one or more of the following conditions are met (data is missing or data is abnormal), it proceeds to step S15 and then to step S13. In step S15, the management computer 11 performs a process using a substitute value for each data point.

[0091] The alternative value used in step S15 has a different technical meaning from the alternative values ​​used in step S12 and step S13. The alternative value used in step S15 is the real-time data value detected immediately before the "Yes" decision is made in step S14. In other words, the alternative value used in step S15 is a dynamic data value, not a fixed value.

[0092] The management computer 11 may use the last location information from the previous day's data acquisition as a substitute value for user P2's location information. The management computer 11 may use the average value for user P2's region as a substitute value for economic information. The management computer 11 may use predicted traffic information as a substitute value for traffic information. The management computer 11 may use the average purchase history of other users who are the same age, gender, and have the same hobbies as user P2 as a substitute value for purchase history.

[0093] If the management computer 11 determines "No" in step S14, in step S16 it uses a Long Short-Term Memory (LSTM) model to analyze each data point in time series and make future predictions for each data point. In the Long Short-Term Memory model, data with relatively high importance is retained for a long period, and data with relatively low importance is deleted.

[0094] Furthermore, in step S17 following step S16, the management computer 11 trains the learning model by performing reinforcement learning. Specifically, Q-learning, which is one of the off-policy time-gap methods, may be performed. The management computer 11 may also use past datasets as learning target data (training data) to train the learning model. First, the management computer 11 generates a past dataset based on information of user P2's past actions in response to the "optimal action" generated based on the future prediction performed in step S16. The information of user P2's actions includes the value of the reward that the artificial intelligence unit 25 previously assigned to user P2 for the results of actions taken in response to the proposed optimal action.

[0095] For example, the management computer 11 inputs weather information, traffic information, and the results of actual actions as initial data, and proposes an action to user P2 based on future predictions, such as indoor movement for user P2, as the optimal action. The management computer 11 then evaluates the "time saved" as a result of user P2 actually performing indoor movement after acquiring the optimal action. Based on the evaluation results, it updates the Q value (reward) and adjusts the weights for score calculation.

[0096] More specifically, the management computer 11 may suggest indoor movement based on future predictions that the weather information indicates rain and the traffic information indicates congestion. If, as a result of the user P2 actually performing indoor movement, the management computer 11 reduces the user P2's movement time to 80% of the estimated movement time of user P2 before the optimal action was suggested (100%), the management computer 11 adjusts the weight of the safety score, which is automatically adjusted. Note that the movement time is the time required for user P2 to reach the destination from a predetermined location. The data described in the processing and judgments from steps S10 to S17 above corresponds to the learning data in the learning process and the inference data in the inference process.

[0097] In the inference process performed by the management computer 11, the process is skipped from step S13 to step S18. In step S18, the management computer 11 determines whether both of the following conditions are met: that the scores of various data that are prerequisites for generating the optimal action are available, and that the elapsed time since the time the request for generating the optimal action was generated is within a predetermined time. The request for generating the optimal action is generated, for example, when the management computer 11 receives a prompt from the terminal device 12. The predetermined time is, for example, 1 second.

[0098] If the management computer 11 determines No in step S18, it proceeds to step S19 to execute the default wait process and then proceeds to step S10. The default wait is an intermediate reward. Q value = 0.5 This involves setting it up.

[0099] If the management computer 11 determines "Yes" in step S18, it inputs the prompts and various real-time data obtained from the terminal device 12 into the learning model in step S20 and starts the process of generating the optimal action "score S" using natural language processing. The algorithm used by the management computer 11 to generate the optimal action is, for example, Score S = Σ(element value × weight) It is represented as follows.

[0100] Here, the element value is a weighted average of numerical values ​​representing each element for evaluating the priority (strategy) of an action, such as comfort, efficiency, safety, economy, timing of action, etc. The management computer 11 may determine the numerical value representing comfort from numerical values ​​of real-time data such as weather information, road congestion rate, train delay information, air cleanliness, pollen dispersion status, etc. The management computer 11 may also determine the numerical value representing efficiency from numerical values ​​of real-time data such as location information, economic information, traffic information, etc.

[0101] The management computer 11 may determine a numerical value indicating safety from numerical values ​​of real-time data such as weather information, disaster information, crime information, and traffic information. The management computer 11 may determine a numerical value indicating economic efficiency from numerical values ​​of real-time data such as economic information, energy consumption, and exchange rates. The management computer 11 may determine a numerical value indicating the timing of action from numerical values ​​of real-time data such as exchange rates, road construction information, regional disaster recovery rate, and learning progress.

[0102] Furthermore, the larger the numerical value representing each element, the higher the relative comfort, efficiency, safety, economy, and timing of the action, respectively. For example, the shorter the distance between the current location and the destination location, the higher the numerical values ​​representing efficiency and timing, contributing to the optimal action of user P2. In other words, "optimal action" means "the action that maximizes the score S when multiple evaluation elements are considered comprehensively." The weight is a numerical value set according to the prompt obtained from the terminal device, and the weight is set within the range of 0.0 to 1.0.

[0103] The management computer 11 may determine a numerical value indicating comfort from real-time data such as weather information, road congestion rates, train delay information, air cleanliness, and pollen dispersal status. The management computer 11 may also determine a numerical value indicating efficiency from real-time data such as location information, economic information, and traffic information.

[0104] The management computer 11 may determine a numerical value indicating safety from real-time data such as weather information, disaster information, crime information, and traffic information. The management computer 11 may determine a numerical value indicating economic efficiency from real-time data such as economic information, energy consumption, and exchange rates. The management computer 11 may determine a numerical value indicating the timing of action from real-time data such as exchange rates, road construction information, regional disaster recovery rate, and learning progress.

[0105] An example of an optimal action generated by the management computer 11 is as follows: The management computer 11 may generate an optimal action for a user P2 when moving from their current location to their destination. Based on prompts obtained from the terminal device 12, the management computer 11 processes inference target data such as location information, weather information, and traffic information using a learning model, and may generate an optimal action that, for example, can reduce travel time by 10 minutes compared to the current method and route of travel, costs 500 yen or less, and satisfies a safety level of 80% or higher.

[0106] The management computer 11 may generate the optimal actions for delivering goods to a company. Based on prompts obtained from terminal devices 12, the management computer 11 processes inference target data such as weather information and traffic information using a learning model, and may generate, for example, an optimal delivery route and delivery routine that can reduce inventory costs by 5%, shorten delivery time by 20% compared to conventional delivery times, and maintain customer satisfaction.

[0107] The management computer 11 may generate optimal actions for the government when setting up evacuation shelters in the event of a disaster. The management computer 11 may generate optimal actions such as the location of the evacuation shelters, the method of procuring materials, the construction method for the evacuation shelters, etc., so as to reduce congestion time by 30%, keep the cost of setting up the evacuation shelters within 10%, and ensure a resident safety rate of 90% or more. Here, a 30% reduction in congestion time means that the total time during which congestion occurs within a given period, for example, within 24 hours, can be reduced by 30% compared to conventional methods.

[0108] "Within 10% of the cost of setting up evacuation shelters" means keeping the cost within 10% of the standard cost of setting up evacuation shelters. "A resident safety rate of 90% or higher" means that, with a resident safety rate of 100% representing a state where a safe evacuation environment is ensured, the environment is such that the safety of the residents is largely ensured. For example, a resident safety rate of 100% is an environment where all residents can evacuate immediately and there are sufficient supplies, while a resident safety rate of 90% is an environment where 90% of the residents can evacuate and supplies are largely sufficient.

[0109] The management computer 11 may generate for individuals the optimal actions to take when attending an event, the optimal actions to take when deciding whether or not to buy a product now, the optimal actions to take to reduce home energy consumption, etc. The management computer 11 may generate for businesses the optimal actions to take to improve profits, the optimal actions to improve customer satisfaction, the optimal actions to improve ticket sales rates, etc. The management computer 11 may generate for governments the optimal actions to take to reduce the burden of healthcare costs, the optimal actions to improve the public safety situation, the optimal actions to reduce air pollution, the optimal actions to stabilize prices, etc.

[0110] In step S20, the artificial intelligence unit 25 attempts to calculate a score S for two or more combinations of different types of data and information in order to generate the optimal action. Then, the artificial intelligence unit 25 outputs the action with the highest score among the multiple types of actions obtained as a result of the trial and error as the optimal action. Note that in step S20, Score S = Σ(element value × weight) Multiple "optimal actions" may be generated that are the same. In this case, the multiple optimal actions will differ in at least part of their content.

[0111] In step S21, following step S20, the management computer 11 determines whether it was able to generate the optimal action within a predetermined time, for example, within 1 second. In other words, the optimal action generated by the artificial intelligence unit 25 is the action with the highest score among the actions that are tried out within the predetermined time. To put it another way, there is no threshold or upper limit for the score corresponding to the optimal action. Furthermore, each time an optimal action is generated in step S21, the score of the optimal action may be the same, or it may be different. In addition, an upper limit on the number of actions to be generated through trial and error may be set in advance, and the action with the highest score among the actions generated up to or below that upper limit may be treated as the optimal action. If the management computer 11 determines Yes in step S21, it stores the generated optimal action in the auxiliary storage device 19 in step S22 and transmits the generated optimal action to the terminal device 12.

[0112] If the management computer 11 determines "No" in step S21, it outputs a standard phrase (template) for when it cannot generate an optimal action, such as "Checking status," in step S23, and proceeds to step S10. Standard phrases are pre-stored in the auxiliary storage device 19.

[0113] In step S31, terminal device 12 may obtain the optimal action from management computer 11 and output the optimal action from output device 44. In step S32, terminal device 12 selects the optimal action obtained from management computer 11. If there are multiple optimal actions, user P2 can select one by operating terminal device 12. In step S32, terminal device 12 may also input the actual action taken by user P2 corresponding to the selected optimal action. The action taken by user P2 is the actual action taken by user P2 in the real world. In step S30, terminal device 12 transmits information and data including the selected optimal action and the action taken by user P2 to management computer 11. In addition, terminal device 12 stores the information and data including the selected optimal action and the action taken by user P2 in auxiliary storage device 42.

[0114] (Effects of the embodiment) The management computer 11 generates the optimal action that user P2 can perform based on the environmental conditions, using prompts obtained from terminal devices 12 and features including user P2's comfort, efficiency, safety, and economics. Therefore, it can generate the optimal action that matches the real-world conditions and the prompts generated by user P. Consequently, the accuracy of generating the "optimal action" performed by the artificial intelligence unit 25 in its inference process is improved.

[0115] Furthermore, the management computer 11 trains the learning model used by the artificial intelligence unit 25 for processing by evaluating the results of actions taken by user P2 when they previously performed the optimal actions that the artificial intelligence unit 25 has generated in the past. As a result, the accuracy of generating the "optimal actions" that the artificial intelligence unit 25 performs in its inference processing is further improved.

[0116] Furthermore, the management computer 11 trains the learning model based on future predictions of the environment's state. Therefore, the accuracy of generating the "optimal action" performed by the artificial intelligence unit 25 in its inference processing is further improved. In addition, the user P2's actions and lifestyle become more efficient. Moreover, it is more multi-functional and has improved processing speed compared to conventional technologies.

[0117] (Other configuration examples of the action suggestion system) The terminal device 12 may have the configuration of the management computer 11 shown in Figure 2. Specifically, the processor 40 may implement the configuration and functions of the processor 17 shown in Figure 2. Also, the auxiliary storage device 42 may implement the configuration and functions of the auxiliary storage device 19 shown in Figure 2. With this configuration, the terminal device 12 can execute steps S10 to S21 and step S23 shown in Figure 3. In other words, the action suggestion system 10 can generate the optimal action in the terminal device 12 even without the management computer 11. The effect when the action management method shown in Figure 3 is implemented in the terminal device 12 is the same as the effect when the action management method shown in Figure 3 is implemented in the management computer 11.

[0118] (supplementary explanation) An example of the technical meaning disclosed in this embodiment is as follows: For example, the management computer 11 and the terminal device 12 are examples of action suggestion devices, respectively. Processors 17 and 40 are examples of processors. Step S10 shown in Figure 3 is an example of the first, second, and sixth processes. Step S20 is an example of the third, fourth, and fifth processes. Steps S16 and S17 are examples of the seventh process. The flowchart shown in Figure 3 is an example of an action suggestion method.

[0119] Processors 17 and 40 are configured to implement, respectively, a first processing unit that obtains prompts for when the user acts in the environment by running a program, a second processing unit that determines the state of the environment in which the user acts, a third processing unit that extracts actions that the user can perform in the environment based on the state of the environment, a fourth processing unit that estimates feature quantities including user comfort, efficiency, safety, and economy, assuming that the user will perform the actions extracted in the third processing unit, and a fifth processing unit that generates the optimal action that the user can perform in the state of the environment based on the prompts obtained in the first processing unit and the feature quantities estimated in the fourth processing unit.

[0120] The fifth processing unit implemented by processors 17 and 40 is configured to generate the optimal action that the user can perform using a learning model. Processors 17 and 40 are configured to implement a sixth processing unit that records the optimal action generated in the past and retrieves the action results when the user has performed that action in the past, and a seventh processing unit that trains the learning model by evaluating the action results.

[0121] The seventh processing unit, implemented by processors 17 and 40, is configured to include a process for training a learning model based on future predictions of the state of the environment.

[0122] The computers constituting the management computer 11 may be a single computer or a distributed computer composed of multiple devices. The computers constituting the terminal equipment 12 may be a single computer or a distributed computer composed of multiple devices.

[0123] The management computer 11 is composed of one or more computers, such as a server, supercomputer, mainframe, or workstation. In this embodiment, the program for operating the management computer 11 may be stored in the auxiliary storage device 19. In other configuration examples of the action suggestion system, the program may be stored in the auxiliary storage device 42 of the terminal device 12.

[0124] This embodiment discloses the following characteristic configuration. This embodiment discloses an action suggestion system having a terminal device operated by a user and a management computer that is communicably connected to the terminal device via a network, wherein the management computer performs the following actions: a first process of obtaining prompts from the terminal device for when the user takes action in the environment; a second process of determining the state of the environment in which the user takes action; a third process of extracting actions that the user can perform in the environment based on the state of the environment; a fourth process of estimating feature quantities including the user's comfort, efficiency, safety, and economy, assuming that the user will perform the actions extracted in the third process; and a fifth process of generating the optimal action that the user can perform in the state of the environment based on the prompts obtained in the first process and the feature quantities estimated in the fourth process.

[0125] This embodiment discloses a non-temporary storage medium that stores a non-temporary program that causes a computer to execute processing, wherein the program causes the computer to execute: a first process of obtaining prompts for when a user acts in an environment; a second process of determining the state of the environment in which the user acts; a third process of extracting actions that the user can perform in the environment based on the state of the environment; a fourth process of estimating feature quantities including the user's comfort, efficiency, safety, and economy, assuming that the user will perform the actions extracted in the third process; and a fifth process of generating the optimal action that the user can perform in the state of the environment based on the prompts obtained in the first process and the feature quantities estimated in the fourth process. The program can also be understood as a program product. [Industrial applicability]

[0126] This embodiment can be used as an action suggestion device, an action suggestion method, and a program. [Explanation of symbols]

[0127] 10…Action suggestion system, 11…Management computer, 12…Terminal equipment, 17,40…Processor

Claims

1. An action suggestion device equipped with a processor that performs processing, The aforementioned processor, The first process involves obtaining prompts for when the user takes action in the environment, A second process determines the state of the environment in which the user is acting, A third process extracts actions that the user can perform in the environment based on the state of the environment, A fourth process estimates feature quantities including the user's comfort, efficiency, safety, and economics, assuming that the user will perform the actions extracted in the third process. A fifth process generates the optimal action that the user can perform in the state of the environment, based on the prompt obtained in the first process and the feature quantities estimated in the fourth process. An action suggestion device configured to perform the following actions.

2. The action suggestion device according to claim 1, The fifth process performed by the processor generates the optimal action that the user can perform using a learning model, The aforementioned processor, A sixth process to obtain the action results when the user previously performed the optimal action that was generated in the past, A seventh process involves training the learning model by evaluating the results of the aforementioned actions, Furthermore, it is a configuration for implementing an action suggestion device.

3. The action suggestion device according to claim 2, An action suggestion device wherein the seventh process performed by the processor includes a process for training the learning model based on a future prediction of the state of the environment.

4. A method for suggesting actions to be performed by a computer, The aforementioned computer, The first process involves obtaining prompts for when the user takes action in the environment, A second process determines the state of the environment in which the user is acting, A third process extracts actions that the user can perform in the environment based on the state of the environment, A fourth process estimates feature quantities including the user's comfort, efficiency, safety, and economics, assuming that the user will perform the actions extracted in the third process. A fifth process generates the optimal action that the user can perform in the state of the environment, based on the prompt obtained in the first process and the feature quantities estimated in the fourth process. A method for proposing actions to be taken.

5. A non-temporary program that causes a computer to perform a process, To the aforementioned computer, The first process involves obtaining prompts for when the user takes action in the environment, A second process determines the state of the environment in which the user is acting, A third process extracts actions that the user can perform in the environment based on the state of the environment, A fourth process estimates feature quantities including the user's comfort, efficiency, safety, and economics, assuming that the user will perform the actions extracted in the third process. A fifth process generates the optimal action that the user can perform in the state of the environment, based on the prompt obtained in the first process and the feature quantities estimated in the fourth process. A program configured to execute [something].

Images