Information processing systems, information processing methods, programs

Abstract

To enable information output tailored to the user's situation. [Solution] An information processing system characterized by comprising: a determination means for determining a category suitable for a user using a first trained model; an output means for outputting information about the determined category using a second trained model; and a learning means for training the first trained model based on the category related to the information output by the output means and the user's response.

Description

Technical Field

[0001] The present invention relates to an information processing system, an information processing method, and a program.

Background Art

[0002] There is a mechanism for communicating with a user in the form of a dialogue using a large language model (so-called generative AI).

[0003] For example, Patent Document 1 describes a mechanism for generating a response sentence for communicating with a user using information related to a topic selected based on the user's private information.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Disclosure of the Invention

Problems to be Solved by the Invention

[0005] Since the user's interests and understanding level always change, in order to communicate smoothly with the user or for the user to enjoy the conversation, it is necessary to grasp the user's situation and output a message with appropriate content.

[0006] Therefore, an object of the present invention is to enable information output according to the user's situation.

Means for Solving the Problems

[0007] Determination means for determining a category suitable for the user using a first learned model, Output means for outputting information related to the determined category using a second learned model, A learning means for training the first trained model based on the categories related to the information output by the output means and the user's response, An information processing system characterized by comprising the following features. [Effects of the Invention]

[0008] According to the present invention, it becomes possible to output information tailored to the user's situation. [Brief explanation of the drawing]

[0009] [Figure 1] A diagram showing an example of a system configuration. [Figure 2] A diagram showing an example of the hardware configuration of the information processing device 101. [Figure 3] A flowchart illustrating an example of the processing details of the present invention. [Figure 4] A flowchart showing an example of the reinforcement learning process in step S102. [Figure 5] A flowchart illustrating an example of topic update processing. [Figure 6] Conceptual diagram of reinforcement learning [Figure 7] An illustrative diagram showing the content of the topic update process. [Figure 8] A diagram illustrating the process of generating dialogue messages. [Figure 9] An illustrative diagram showing the interaction between a large-scale language model and a user. [Figure 10] A conceptual diagram illustrating the process of generating new topics. [Modes for carrying out the invention]

[0010] An example of an embodiment of the present invention will be described below using Figures 1 to 10.

[0011] Figure 1 shows an example of the system configuration of the present invention.

[0012] As shown in FIG. 1, in the information processing system of the present invention, an information processing apparatus 101 is communicably connected to a voice input / output apparatus 102, an imaging apparatus 103, a topic selection AI 104 (first learned model), and a large language model 105 (second learned model).

[0013] The topic selection AI 104 has a function of determining which topic among the topics (categories) registered in the topic candidate list should be conversed about based on matters of interest to the user. It transmits the determined topic to the information processing apparatus 101, and the information processing apparatus 101 instructs the large language model 105 (so-called generation AI) to generate a conversation sentence based on the topic. The large language model 105 generates a conversation sentence based on the instruction from the information processing apparatus 101 and transmits the generated conversation sentence to the information processing apparatus 101. Also, the topic selection AI 104 updates the selection value related to the topic registered in the topic candidate list based on an instruction from the information processing apparatus 101.

[0014] The information processing apparatus 101 conducts a conversation with the user by displaying the conversation sentence generated by the large language model 105 as text on the display unit or outputting it as voice from the voice input / output apparatus 102. Also, the voice input / output apparatus 102 acquires the voice (response to the conversation) uttered by the user and transmits it to the information processing apparatus 101. The transmitted voice is processed (such as text conversion) by the information processing apparatus 101, and a prompt based on the response content by the user is transmitted from the information processing apparatus 101 to the large language model 105, instructing it to generate the next conversation sentence. The response from the user may be input not by voice but by the user operating the keyboard of the information processing apparatus 101.

[0015] The imaging apparatus 103 is an apparatus that photographs the user engaged in the conversation, inputs the actions and expressions of the user during and after the conversation as video / images, and transmits them to the information processing apparatus 101. The information processing apparatus 101 calculates the degree of interest of the user by analyzing the expression of the user in the transmitted image, and performs reinforcement learning on the topic selection AI 104 based on the degree of interest.

[0016] Further, the information processing apparatus 101 requests the large language model 105 to generate a new topic based on the adjusted selection value, and requests the topic selection AI 104 to update the topic candidate list based on the topic generated by the large language model 105. The topic candidate list is a list in which topics (candidates) of interest to each user are registered. The topic selection AI 104 may store and manage the list, or the topic selection AI 104 may learn topics of interest and degrees of interest for each user, or the list may be stored in the information processing apparatus 101 or other devices so that the topic selection AI 104 can appropriately access the list.

[0017] FIG. 2 is a block diagram showing an example of the hardware configuration of the information processing apparatus 101 of the present invention.

[0018] As shown in FIG. 2, the information processing apparatus 101 is connected to a CPU (Central Processing Unit) 201, a ROM (Read Only Memory) 202, a RAM (Random Access Memory) 203, a storage device 204, an input controller 205, an audio controller 206, a video controller 207, a memory controller 208, and a communication I / F controller 209 via a system bus 200.

[0019] The CPU 201 comprehensively controls each device and controller connected to the system bus 200.

[0020] The ROM 202 or the external memory 213 holds a BIOS (Basic Input / Output System), an OS (Operating System), which are control programs executed by the CPU 201, a computer-readable executable program for realizing the present information processing method, and various necessary data (including data tables).

[0021] RAM203 functions as the main memory, work area, etc., of the CPU201. The CPU201 loads the necessary programs, etc., from ROM202 or external memory213 into RAM203, and then executes the loaded programs to perform various operations.

[0022] The input controller 205 controls input from input devices such as a keyboard 210 or a pointing device such as a mouse (not shown). If the input device is a touch panel, the user can give various instructions by pressing (touching with a finger, etc.) icons, cursors, or buttons displayed on the touch panel.

[0023] Furthermore, the touch panel may be a multi-touch screen or other touch panel capable of detecting the positions of multiple fingers touching it.

[0024] The video controller 207 controls the display to an external output device such as the display 212. The display may include the display of a notebook computer integrated with the main unit. The external output device is not limited to a display; for example, it may be a projector. Furthermore, for the aforementioned touch-enabled device, an input device is also provided.

[0025] The video controller 207 can control the video memory (VRAM) used for display control. It can utilize a portion of the RAM 203 as the video memory area, or it can provide a separate, dedicated video memory.

[0026] The memory controller 208 controls access to the external memory 213. The external memory can include an external storage device (hard disk), a flexible disk (FD), or a CompactFlash® memory connected to a PCMCIA card slot via an adapter, which stores boot programs, various applications, font data, user files, editing files, and other data.

[0027] The communication interface controller 209 connects to and communicates with external devices via a network and performs communication control processing over the network. For example, it can handle communication using TCP / IP, telephone lines such as ISDN, and mobile phone 4G and 5G lines.

[0028] Furthermore, the CPU 201 enables display on the display 212 by, for example, performing the process of expanding (rasterizing) outline fonts into the display information area in RAM 203. The CPU 201 also enables user input via a mouse cursor (not shown) on the display 212.

[0029] Next, an example of the processing content of the present invention will be explained using flowcharts from Figures 3 to 5.

[0030] The processes shown in the flowcharts of Figures 3 to 5 are processes in which the CPU 201 of the information processing device 101 reads a predetermined control program and executes it.

[0031] The processes in steps S102 to S104 of Figure 3 are to be repeatedly executed while the conversational application, which is an application for carrying out the present invention, is running (S101).

[0032] In step S102, reinforcement learning for topic selection is performed. The details of the process are explained using the flowchart in Figure 4.

[0033] Step S103 determines whether the topic update process in Step S104 needs to be performed. This determination is made using the elapsed time since the last topic update process; if a predetermined time has passed, it is determined that the topic needs to be updated. Alternatively, the user's level of interest in the currently selected topic (the value of the topic) is used; if the level of interest no longer meets a predetermined standard, it is determined that the topic needs to be updated. The information used to determine the need for a topic update can be set as appropriate and is not limited to the elapsed time since the last topic update process or the user's level of interest as described above.

[0034] The topic selection AI 104 learns the selection value of each topic through reinforcement learning (by selecting a topic, engaging in a dialogue, and analyzing the quality of the user's response to perform reinforcement learning). The selection value is learned so that topics with a high probability of eliciting a positive response from the user are given a higher value. Topics that are likely to elicit a positive response from the user are considered to be of high user interest, and here the selection value of the learned topic is used as the user's level of interest. Reinforcement learning will be explained in S210.

[0035] Step S104 updates the topic candidates that can be selected during the conversation. The details of the topic update process are explained using the flowchart in Figure 5, but it is a process that replaces topics with low user interest (topic selection value) among the currently recorded topic candidates with new topics.

[0036] The above process is repeatedly executed while the application is running.

[0037] Next, we will explain the reinforcement learning process in step S102 using the flowchart in Figure 4.

[0038] In step S201, the user's image, captured by the imaging device 103 before the dialogue begins, is acquired and analyzed to perform face detection and calculate the positivity of the facial expression from the image.

[0039] One example of a method for determining positivity is to use facial recognition AI to calculate Valence (emotional valence), which indicates whether a facial expression is positive or negative. Facial recognition AI is an AI that estimates emotions from human facial images and is trained on a dataset consisting of facial images and emotion labels that represent various emotions. By inputting a user's facial image into the facial recognition AI, Valence can be calculated based on the distribution of emotions obtained.

[0040] In this embodiment, we have described a method for calculating positivity (a value based on emotions obtained from facial images), but any information obtained from the user's body (user's reaction), such as vital data or physical data like brain waves, pulse rate, and voice tone, may be used.

[0041] Step S202 determines whether a dialogue has started. Specifically, this determination can be made by checking whether the dialogue start button has been pressed (if the dialogue start button is pressed, it is determined that a dialogue has started), or by analyzing the user's facial expressions and motions from the video (for example, if it is detected that the user's mouth is moving, it is determined that a dialogue has started). Alternatively, the start of a dialogue may be determined by detecting the user's voice as audio.

[0042] In step S203, the topic selection AI 104 is asked to select a topic, and a topic is obtained from the topic selection AI 104. Specifically, the topic selection AI 104 selects one topic from a pre-registered list of topic candidates based on the selection value of each topic. The selection algorithm can be one in which topics with high value are selected with a high probability.

[0043] The topic suggestions are set for each user, and the user to be interacted with is identified by facial recognition or by having them log in to the system in advance. The topic is then selected from the suggestions corresponding to the identified user.

[0044] Step S204 generates a dialogue message for the user. Specifically, it generates a prompt instructing the system to generate a message related to the topic selected in step S203 and sends it to the large-scale language model 105. The large-scale language model 105 generates a dialogue message based on the transmitted prompt and returns the generated dialogue message to the information processing device 101. The large-scale language model 105 can also generate dialogue messages related to the topic based on information obtained from news sites and other sources using RAG.

[0045] The generated dialogue message is output to the user via the audio input / output device 102 or the display unit of the information processing device 101 (when output from the audio input / output device 102, it is output as voice; when output to the display unit of the information processing device 101, it is output as text information, etc.). For example, if "entertainment" is selected as the topic, a message is generated from the content of news related to entertainers.

[0046] Figure 8 illustrates the process of generating dialogue messages. The example in Figure 8 illustrates the generation of a dialogue message when the topic "reading" is selected in S203. The information processing device 101 creates a prompt to cause the large-scale language model to generate a dialogue message about the topic "reading". The created prompt 801 is the prompt in Figure 8 that says, "Please create a message to speak to the user on the topic of 'reading'". Note that this prompt can be created by, for example, applying the selected topic to a pre-prepared template (for example, a template such as "Please create a message to speak to the user on the topic of '***'") and replacing it with the "***" part of the template, but any method can be used to create the prompt.

[0047] The generated prompt is input into a large-scale language model, and the large-scale language model outputs a sample response. Sample response 802 output is "Hello, have you read any interesting books recently?".

[0048] In step S206, the user's response to the message generated in step S204 is obtained. For example, this could involve obtaining text information entered via the keyboard or obtaining spoken audio from the user via the audio input / output device 102. The information obtained as audio is assumed to be converted into text information.

[0049] Step S207 generates a response message for the user response obtained in step S206. To generate the response message, a prompt is created based on the user response message obtained in step S206 (the response message can also be used as the prompt), and this prompt is sent to the large-scale language model 105, causing the large-scale language model to create the response message.

[0050] The generated response message will be output in the same way as S204.

[0051] Figure 9 shows an example of a user response message to a dialogue message generated in S204, and a reply message generated in S207 to the user's response message. The dialogue message 901 generated in S204 is "Hello, have you read any interesting books recently?", and the user's response 902 is "I recently read the science fiction novel XX. It was very exciting, and I finished it in one go. I recommend it." The information processing device 101 then creates a prompt based on this user response (in the example in Figure 9, the user's response is directly input to the large-scale language model 105 as a prompt, but a prompt created by adding a pre-set instruction to the user's response can also be input to the large-scale language model 105), and inputs it to the large-scale language model 105. The reply message 903 generated by the large-scale language model is "That sounds interesting. I'd love to read it."

[0052] Step S208 determines whether the conversation with the user has ended. This determination can be made using metrics such as the level of engagement in the conversation. The level of engagement is a numerical indicator of the level of engagement in the conversation. For example, it can be the user's response speed (the time from when a message is output in S204 or S207 until a response from the user is received in S206) or the amount of messages exchanged at once (data size, number of characters, etc., of the user's response message). Alternatively, metrics based on this information may be used. Based on this information, if the response is fast or the amount of messages is large (meeting predetermined criteria), it can be interpreted as a state of high engagement. If the level of engagement is high, the conversation will not end; if it is low, the conversation will end.

[0053] If the dialogue is to be terminated (S208:YES), the process proceeds to step S209. If the dialogue is not to be terminated (S208:NO), the process returns to step S206.

[0054] In step S209, the user's image captured by the imaging device 103 after the dialogue ends is acquired and analyzed to perform face detection and calculate the positivity of the facial expression from the image. The positivity is the same index as in S201.

[0055] In step S210, reinforcement learning is performed on the topic selection AI 104. Specifically, the positivity score calculated in steps S201 and S209 is analyzed, and if the value improves before and after the dialogue (positive change), a positive reward is given for the topic selected in S203; if it decreases (negative change), a negative reward is given to reinforce the topic selection AI.

[0056] In this embodiment, the positivity level is compared before and after the dialogue (before the dialogue begins and after it ends). However, it is also possible to evaluate the change in positivity level during the dialogue and use the evaluation results for reinforcement learning, rather than limiting the comparison to the difference between before and after the dialogue. In this case, the positivity level will be repeatedly calculated and the changes recorded not only before and after the dialogue begins but also during the dialogue. As mentioned above, the positivity level is not limited to what is calculated from facial images; any value that can be detected / evaluated by the user from their body or actions may be used.

[0057] Figure 6 shows an illustrative diagram of reinforcement learning. Figure 6 illustrates a scenario where a dialogue takes place about topic A, and the selection value of topic A before the dialogue was 30. By comparing the facial expressions before and after the dialogue, if the positivity of the facial expression improves, a positive reward is given for topic A. As a result, the selection value of topic A increases by 5 points to 35. The amount of the reward given may be determined based on how much the positivity improved (degree of improvement), or it may be predetermined.

[0058] If the level of positivity in facial expressions decreased before and after the conversation, the value of choosing topic A decreased by 5 points, to 25.

[0059] Next, we will explain the details of the topic update process (S104) using the flowchart in Figure 5.

[0060] In step S301, topics with high selection value are retrieved from the list of topic candidates. Topics that meet a predetermined selection value are retrieved and interpreted as topics of high user interest (topic selection value). In cases where no such topics exist, it is assumed that the topic update in step S303 will be based only on topics of low interest.

[0061] In step S302, topics with low selection value are retrieved from the list of topic candidates. Topics whose selection value does not meet a predetermined threshold are retrieved and interpreted as topics of low user interest (topic selection value), and are replaced with newly generated topic candidates in step S304. Here, if there are no topics with low selection value (topics that do not meet the predetermined threshold), there is no need to update the topics, so it is assumed that the judgment in step S103 will branch to No, and such cases are not considered in subsequent processing.

[0062] Step S303 generates new topic candidates. Based on the user's level of interest (topic selection value) for each topic obtained in steps S301 and S302, prompts are created and sent to the large-scale language model 105 to request the generation of new topics, causing the large-scale language model 105 to generate new topics. In cases where high-value topics could not be obtained in step S301, topics are generated using only low-value topics. It is also confirmed that there is no overlap between the generated topics and existing topics.

[0063] In this embodiment, we have described a method of generating prompts and new topics based on the user's level of interest in each topic. However, it is also possible to generate topics that have not been discussed before, regardless of the user's level of interest, or to generate topics that are trending or current events.

[0064] Figure 10 is an illustrative diagram of the process in which the large-scale language model 105 generates new topics in S303. Prompt 1001 is an example of a prompt created based on the topics acquired in steps S301 and S302. In the example in Figure 10, the topics of high interest (topics acquired in S301) are "listening to music" and "watching movies," while the topic of low interest (topic acquired in S302) is "sports." The prompt created based on this information is input to the large-scale language model 105, and the topic 1002 generated by the large-scale language model 105 is output. In the example in Figure 10, "watching musicals" and "reading" are generated as topics of interest to the user.

[0065] Regarding how to create a prompt, any method is acceptable, but as an example, one method is to prepare a template in advance, as explained in Figure 8, and then insert the topics obtained in S301 and S302 into the section where the topic is to be written.

[0066] In step S304, the topic candidates are updated. Specifically, topic candidates that were interpreted in step S302 as having low user interest (topic selection value) are replaced with topics generated in step S303. (Specifically, the information processing device 101 outputs an instruction to the topic selection AI 104 to replace topic candidates that were interpreted in step S302 as having low user interest with topics generated in step S303, thereby updating the topic candidate list).

[0067] In step S305, a selection value is set for the swapped topic (the topic newly added to the topic candidates). To make the swapped topic more likely to be selected immediately after the topic update, the maximum value of the topic candidates is set.

[0068] Figure 7 is an illustrative diagram showing the content of the topic update process in Figure 5.

[0069] The pre-update topic candidate list 701 is a data table in which the selection value of each topic candidate is registered before the topic update process shown in Figure 5 is performed. Topic A has the highest selection value, with a score of 30. Topic D has the lowest selection value, with a score of -10. The topic generation prompt 703 is a prompt generated in S303 and passed to the large-scale language model, and as mentioned above, it includes topics with high selection value (interest level) (topic A in the example in Figure 7) and topics with the lowest selection value (topic D in the example in Figure 7).

[0070] The generated text 704 is text generated by the large-scale language model 105 based on the topic generation prompt 703, and includes the newly generated topic (topic E in the example in Figure 7). As shown in the updated topic candidate list 702, topic D, which had the lowest selection value in the pre-update topic candidate list 701, has been swapped with the newly generated topic E (S304). The selection value of the newly added topic E is set to a score of 30, which is the maximum value among the topic candidates (the selection value of topic A).

[0071] The above embodiment describes an example in which, in a case of interacting with a user, the topic of the conversation is determined based on the results of analyzing the user's image, and dialogue text related to the determined topic is generated.

[0072] The present invention is not limited to this embodiment and can be applied to other embodiments as well; one example is described below.

[0073] First, it can be used for learning, such as taking e-learning courses or studying learning materials (self-study). In this case, the topic selection AI 104 selects (determines) the difficulty level of the explanation, and the large-scale language model 105 outputs an explanatory text (an explanation of the learning material or content) according to the selected difficulty level. The topic selection AI 104 undergoes reinforcement learning based on the level of understanding analyzed from the user's image, and the value of selecting the difficulty level is adjusted.

[0074] More specifically, the topic selection AI 104 has a function to determine, based on the user's level of understanding, what level (difficulty level such as level understandable to elementary school students or level for experts) and content (which may include the words, kanji, and expressions used) the learning content should be explained. The determined level is transmitted to the information processing device 101, and the information processing device 101 instructs the large-scale language model 105 (a so-called generative AI) to generate an explanatory text (which may also be a question (quiz) testing knowledge and understanding) based on that level. The large-scale language model 105 generates an explanatory text about the learning content based on the instructions from the information processing device 101 and transmits the generated explanatory text to the information processing device 101.

[0075] The information processing device 101 provides explanations of the learning content to the user by displaying the explanatory text generated by the large-scale language model 105 as text on the display unit or outputting it as audio from the audio input / output device 102. The audio input / output device 102 also acquires audio from the user (such as responses regarding understanding of the outputted explanation or answers to questions) and transmits it to the information processing device 101. The transmitted audio is processed by the information processing device 101 (e.g., converted to text), and the information processing device 101 sends a prompt to the large-scale language model 105 based on the user's response, instructing it to generate the rest of the explanatory text. The user's response may also be entered by the user operating the keyboard of the information processing device 101 instead of by audio.

[0076] The imaging device 103 is a device that photographs the learning user, inputting the user's gestures and facial expressions as video and images, and transmitting them to the information processing device 101. The information processing device 101 calculates the user's level of understanding by analyzing the user's facial expressions in the transmitted images, and performs reinforcement learning on the topic selection AI 104 based on the level of understanding. Through reinforcement learning, the value of selecting difficulty levels is adjusted.

[0077] Furthermore, the information processing device 101 requests the large-scale language model 105 to generate new difficulty levels (difficulty level classifications) based on the adjusted selection values, and updates the difficulty level classification list based on the difficulty level classifications generated by the large-scale language model 105.

[0078] This makes it possible to adjust the level of explanation to match the user's level of understanding.

[0079] Next, I will explain how it can be used for news distribution.

[0080] The topic selection AI 104 has the function of determining which fields of news should be delivered based on the user's interests. The determined fields are transmitted to the information processing device 101, and the information processing device 101 instructs the large-scale language model 105 (so-called generative AI) to generate news related to those fields (generating summaries of news in those fields, lists of news in those fields, etc.). The large-scale language model 105 generates news based on the instructions from the information processing device 101 and transmits the generated news to the information processing device 101.

[0081] The information processing device 101 delivers news to the user by displaying the news generated by the large-scale language model 105 as text on the display unit or outputting it as audio from the audio input / output device 102. The audio input / output device 102 also acquires audio from the user (such as reactions to the delivered news) and transmits it to the information processing device 101. The transmitted audio is processed by the information processing device 101 (e.g., converted to text), and the information processing device 101 sends a prompt to the large-scale language model 105 based on the user's response, instructing it to generate news to be delivered next. The user's response may also be entered by the user operating the keyboard of the information processing device 101, rather than by audio.

[0082] The imaging device 103 is a device that photographs users who have received news broadcasts, inputting the user's gestures and facial expressions during and after the broadcast as video and images, and transmitting them to the information processing device 101. The information processing device 101 calculates the user's level of interest by analyzing the user's facial expressions in the transmitted images, and performs reinforcement learning on the topic selection AI 104 based on the level of interest.

[0083] Furthermore, the information processing device 101 requests the large-scale language model 105 to generate new fields based on the adjusted selection values, and updates the list of candidate fields based on the fields generated by the large-scale language model 105.

[0084] This makes it possible to change the news delivered based on the target audience's level of interest.

[0085] Next, we will discuss cases where stress coping is provided.

[0086] The topic selection AI 104 has the function of determining what kind of stress coping (methods for dealing with stress / stress relief) should be suggested based on what kind of stress coping is effective for each user. The determined stress coping is sent to the information processing device 101, and the information processing device 101 instructs the large-scale language model 105 (so-called generative AI) to generate text that will allow the user to practice that stress coping. Based on the instructions from the information processing device 101, the large-scale language model 105 generates text that will allow the user to implement the stress coping and sends the generated text to the information processing device 101.

[0087] Furthermore, the topic selection AI also determines the timing for suggesting stress coping strategies. As described later, the AI ​​learns effective coping strategies based on stress levels calculated by analyzing facial expressions from the user's facial images taken by the imaging device 103 before and after coping is performed.

[0088] The information processing device 101 encourages the user to engage in stress coping by displaying text generated by the large-scale language model 105 as text on the display unit or outputting it as audio from the audio input / output device 102.

[0089] The imaging device 103 is a device that photographs users for whom stress coping is proposed / has been proposed, inputting the user's gestures and facial expressions before and after the proposal (before and after implementation) as video and images, and transmitting them to the information processing device 101. The information processing device 101 calculates the user's stress level by analyzing the user's facial expressions in the transmitted images, and performs reinforcement learning on the topic selection AI 104 based on the stress level. Specifically, it learns whether the proposed coping was effective or not based on how much the stress level improved before and after the stress coping.

[0090] Furthermore, the information processing device 101 requests the large-scale language model 105 to generate new copings based on the adjusted selection values, and updates the candidate list based on the copings generated by the large-scale language model 105.

[0091] This makes it possible to propose stress coping strategies that are effective for each individual.

[0092] Although we have described four application areas (dialogue, learning, news distribution, and stress coping) above, the application areas of the present invention are not limited to these, and it can be applied to any application that outputs information according to the user's situation, such as the user's interests and level of understanding.

[0093] Furthermore, the present invention can take the form of, for example, a system, apparatus, method, program, or recording medium. Specifically, it may be applied to a system consisting of multiple devices, or to an apparatus consisting of a single device.

[0094] Furthermore, the program in this invention is a program that a computer can execute using the processing methods shown in the flowcharts in Figures 3 to 5, and the storage medium of this invention stores a program that a computer can execute using the processing methods in Figures 3 to 5. Note that the program in this invention may also be a separate program for each processing method of each device in Figures 3 to 5.

[0095] As described above, it goes without saying that the object of the present invention can also be achieved by supplying a recording medium containing a program that realizes the functions of the embodiments described above to a system or device, and by having the computer (or CPU or MPU) of that system or device read and execute the program stored on the recording medium.

[0096] In this case, the program read from the recording medium itself realizes the novel function of the present invention, and the recording medium on which that program is recorded constitutes the present invention.

[0097] For recording media used to supply programs, examples include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, CD-Rs, DVD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, EEPROMs, silicon disks, and the like.

[0098] Furthermore, it goes without saying that the functions of the aforementioned embodiments are realized not only by the computer executing the program it has read, but also by the operating system (OS) running on the computer performing some or all of the actual processing based on the instructions of that program, thereby realizing the functions of the aforementioned embodiments.

[0099] Furthermore, it goes without saying that this also includes cases where, after a program read from a recording medium is written to the memory of a function expansion board inserted into a computer or a function expansion unit connected to a computer, the CPU or other components of the function expansion board or function expansion unit perform some or all of the actual processing based on the instructions of the program code, and the functions of the aforementioned embodiments are realized through that processing.

[0100] Furthermore, the present invention may be applied to a system consisting of multiple devices or to a device consisting of a single device. It goes without saying that the present invention can also be applied when the results are achieved by supplying a program to a system or device. In this case, by reading a recording medium containing a program for achieving the present invention into the system or device, the system or device can enjoy the effects of the present invention.

[0101] Furthermore, by downloading and reading the program for achieving the present invention from a server, database, etc. on a network using a communication program, the system or device can enjoy the effects of the present invention. It should be noted that configurations combining the above-described embodiments and their variations are all included in the present invention. [Explanation of symbols]

[0102] 101 Information Processing Device 102 Audio Input / Output Device 103 Imaging device 104 Topic Selection AI 105 Large-scale language models

Claims

1. A decision means for determining a category suitable for the user using a first pre-trained model, An output means that outputs information about the determined category using a second trained model, A learning means for training the first trained model based on the categories related to the information output by the output means and the user's response, An information processing system characterized by comprising the following features.

2. The information processing system according to claim 1, characterized in that the learning means learns the selection value of categories related to the information based on changes in the user before and after the output of the information by the output means.

3. The information processing system according to claim 2, characterized in that the learning means trains the first trained model by giving a positive reward if the user's change in the selection value of the category related to the information output by the output means is a positive change, and a negative reward if the change is negative.

4. The system includes an evaluation means that evaluates changes in the user based on the user's facial image before and after the output of information by the output means. The information processing system according to claim 1, characterized in that the learning means trains the first trained model based on the evaluation results by the evaluation means.

5. The information processing system according to claim 2, characterized in that the determination means determines the category with the highest selection value from among a plurality of candidate categories as the category suitable for the user.

6. A category generation means that generates new candidate categories using the second trained model, An update means for replacing a category with a low selection value among the aforementioned multiple candidate categories with a new category generated by the category generation means, The information processing system according to claim 5, characterized by comprising:

7. The system includes an instruction means that gives instructions to the second trained model to generate dialogue sentences relating to the categories determined by the determination means, The information processing system according to claim 1, characterized in that the output means outputs the dialogue text generated by the second trained model based on the instruction.

8. The system includes a receiving means for receiving responses from the user to the dialogue text output by the output means, The information processing system according to claim 7, characterized in that the instruction means instructs the second trained model to generate a continuation of the dialogue based on the response received by the receiving means.

9. The system includes termination determination means that determines whether to terminate the dialogue based on the response received from the user by the reception means, The information processing system according to claim 8, characterized in that the instruction means instructs the system to generate a continuation of the dialogue based on the response received from the user until the termination determination means determines that the system has terminated.

10. The decision means of the information processing system includes a decision step of determining a category suitable for the user using a first trained model, The output means of the information processing system includes an output step of outputting information about the determined category using a second trained model, The learning means of the information processing system includes a learning step that trains the first trained model based on the categories output by the output step and the user's response, An information processing method characterized by comprising:

11. A program for causing a computer to function as one of the means described in any one of claims 1 to 9.

Images