Information processing methods, information processing systems, and programs

Abstract

This invention provides an information processing method that solves the problem that response information generated by large-scale language models has a uniform level of detail regardless of the user's usage status, such as frequency of use or depth of dialogue, and therefore fails to provide the user with an optimal dialogue experience. [Solution] The acquisition unit 300 of the information processing device 3 acquires at least one of the following: usage frequency information, which shows the usage history of a user of a large-scale language model over a predetermined period, and conversation depth information, which shows the degree of repetition of dialogue in a session. The calculation unit 301 calculates interaction density information, which indicates an appropriate value for the explanation granularity, based on at least one of the acquired usage frequency information and conversation depth information. The adjustment unit 302 adjusts the explanation granularity for the response information by setting constraints on the explanation granularity for the generation process based on the interaction density information, before executing the response information generation process by the large-scale language model.

Description

Technical Field

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

Background Art

[0002] In recent years, technologies related to document generation, dialogue processing, and inference support using large language models (LLMs) have been developed. For example, Patent Document 1 discloses an information processing apparatus that generates a response sentence for an input text using a large language model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the prior art such as Patent Document 1, the processing mainly focuses on generating a transient response sentence for a single input text from the user. Therefore, regardless of the user's usage status such as the continuous usage record of the user within a predetermined period and the accumulation of exchanges in a series of dialogues, response sentences are provided with a uniform explanation depth and text structure. As a result, there is a problem that a response sentence with an optimal explanation granularity may not be obtained according to the user's usage status.

[0005] The present disclosure has been made in view of the above problems, and an object thereof is to provide an information processing method, an information processing system, and a program that can appropriately adjust the explanation granularity in the response information generated by a large language model based on the usage status such as the user's usage frequency and conversation depth, and provide an optimal dialogue experience for the user.

Means for Solving the Problems

[0006] To achieve the above objective, an information processing method according to one aspect of the present invention is: An information processing method performed in an information processing system that utilizes a large-scale language model, An acquisition step of acquiring at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user; The process involves adjusting the level of detail in the response information generated by the large-scale language model in response to user input, based on at least one of the acquired frequency of use information and conversation depth information. [Effects of the Invention]

[0007] According to one aspect of this disclosure, the descriptive granularity in response information generated by a large-scale language model can be adjusted based on at least one of the following: usage frequency information indicating the user's usage history over time, and conversation depth information indicating the degree of repetition of dialogue in a session. This makes it possible to automatically and appropriately optimize the descriptive granularity of response information output from the large-scale language model in accordance with the user's usage status.

[0008] Other issues, configurations, and effects will be clarified in the embodiments for carrying out the invention described later. [Brief explanation of the drawing]

[0009] [Figure 1] This is a schematic diagram showing an example of information processing system 1. [Figure 2] This is a block diagram showing the functional configuration of the information processing device 3. [Figure 3] This is a hardware configuration diagram showing an example of the Computer 900. [Figure 4] This figure shows the data structure and management examples for usage frequency information and conversation depth information. [Figure 5]This is a flowchart showing the overall processing steps of the information processing method. [Figure 6] This flowchart shows the processing procedure for calculating and updating interaction density information. [Figure 7] This flowchart shows the processing procedures for the setup process and the regeneration process. [Figure 8] This figure shows an example of a user interface provided to user terminal 2. [Modes for carrying out the invention]

[0010] The following describes embodiments for implementing this disclosure with reference to the drawings. The following sections schematically show the extent necessary to explain the objectives of this disclosure, and primarily describe the extent necessary to explain the relevant parts of this disclosure. Any parts omitted from the explanation will be considered to be covered by prior art.

[0011] <Embodiment> (Overview of Information Processing System 1) Figure 1 is a schematic diagram showing an example of information processing system 1. Information processing system 1 uses a generative artificial intelligence model 50, such as a large-scale language model, to perform information processing such as document generation, dialogue processing, and reasoning support, and also functions as a system that dynamically adjusts the explanatory granularity of the response information D2 generated by the generative artificial intelligence model 50 based on the user's usage status.

[0012] The information processing system 1, as shown in Figure 1, comprises a user terminal 2, an information processing device 3, external storage 4, and a generative AI device 5. Each device 2 to 5 is, for example, a general-purpose or dedicated computer (see Figure 3 below) and is connected to a wired or wireless network 6, enabling the mutual transmission and reception of various types of data. The number of each device 2 to 5 and the connection configuration of the network 6 are not limited to the example in Figure 1 and may be changed as appropriate.

[0013] User terminal 2 functions as a user interface for obtaining the results of the generation process by the generative AI device 5 via the information processing device 3, and is an information processing terminal that performs various instruction inputs and dialogue operations. User terminal 2 transmits prompt information D1, which is input information from the user, to the information processing device 3, receives response information D2 generated from the generative AI device 5 via the information processing device 3, and outputs such as display and audio playback. User terminal 2 may be any form of information processing terminal, such as a desktop or notebook personal computer, smartphone, tablet terminal, wearable terminal, or VR / AR device.

[0014] The information processing device 3 is composed of, for example, a server-type computer or a cloud-type computer. The information processing device 3 is a device that provides an environment for the generative AI device 5 to execute generation processing based on prompt information D1 transmitted from the user terminal 2. The information processing device 3 may run the generative AI device 5 within its own device, or it may be used by calling an external API. In this embodiment, the information processing device 3 is responsible for setting constraints on the level of explanation for the generation processing of response information D2 in the generative AI device 5, based on the user's usage frequency information and conversation depth information. The reception of prompt information D1 (input information) by the information processing device 3 is not limited to interaction with the user via the user terminal 2, but may also be done by a management user interface, an application programming interface (API), or automated processing within the system.

[0015] The external storage 4 functions as a database that stores data such as past dialogue logs and usage histories between the user and the generative artificial intelligence model 50. In the external storage 4, usage records over time within a predetermined period for each user and dialogue histories for each session, which is a series of dialogue units, are accumulated. The external storage 4 may be implemented, for example, as cloud storage, a storage area of another server device, network-connected storage, a storage area provided in a local device, or a dedicated data vault area. Note that in the present disclosure, "external" does not necessarily mean physically separate enclosures. The external storage 4 may be a logically separated storage area within the same enclosure as the information processing device 3.

[0016] The generative AI device 5 is a computer system that provides computing resources for executing inference processing using the generative artificial intelligence model 50. The generative AI device 5 is configured by, for example, one or more server devices, a cloud computing system, or a distributed computing environment. The generative AI device 5 has a function of receiving prompt information D1 from the user and constraint conditions set by the information processing device 3 through communication with the information processing device 3, and returning the calculation result by the generative artificial intelligence model 50 as answer information D2. Note that the generative AI device 5 may be composed of a single model or may be configured such that a plurality of models operate in cooperation.

[0017] The generative artificial intelligence model 50 is a trained model constructed using a vast learning dataset and deep learning techniques, and a large-scale language model or the like, which is a base model, is used. The generative artificial intelligence model 50 performs advanced processing such as text generation, dialogue response, summarization, translation, and information extraction. Note that the generative artificial intelligence model 50 may use one provided as a service by an external provider, and in that case, the generative AI device 5 may be omitted from the components of the information processing system 1.

[0018] Network 6 is a communication network for data communication between devices. Network 6 is composed of the Internet, LAN, WAN, mobile communication network (such as 5G, 6G, etc.), or a combination thereof, regardless of wired or wireless. In addition, in the case where the information processing device 3 and the generative AI device 5 are implemented in the same device, some communications may be replaced by bus connections inside the device or transfers between memories.

[0019] In the information processing system 1, the information processing device 3 acquires at least one of usage frequency information indicating the usage performance over time within a predetermined period of the user, and conversation depth information indicating the degree of repetition of conversations in a session, which is a series of dialogue units conducted between the large language model and the user. Then, based on at least one of the acquired usage frequency information and conversation depth information, the information processing device 3 calculates interaction density information indicating an appropriate value of the explanation granularity, and sets a constraint condition regarding the explanation granularity for the generation process before executing the generation process of the response information D2 by the generative artificial intelligence model 50. Thereby, the information processing system 1 can provide the response information D2 having an optimal explanation granularity adapted to the usage state of the user.

[0020] (Functional configuration of the information processing device 3) FIG. 2 is a block diagram showing an example of the information processing device 3. The information processing device 3 includes a control unit 30 composed of a processor or the like, a storage unit 31 composed of an HDD, SSD, memory, etc., a communication unit 32 which is a communication interface with the network 6, an input unit 33 composed of a keyboard, mouse, etc., and a display unit 34 composed of a display or the like. Note that the input unit 33 and the display unit 34 may be omitted.

[0021] The storage unit 31 stores session management data 310 and an information processing program 311, and also stores an operating system, other programs, various data, etc. In addition, the storage unit 31 cooperates with the external storage 4 to temporarily or continuously hold history information such as dialogue logs and usage performance for each user.

[0022] Session management data 310 is used temporarily to maintain the currently running session and includes, for example, the session ID, various acquired history information, calculated interaction density information, set constraints, and dialogue logs of prompt information D1 and response information D2 in the session.

[0023] The control unit 30 functions as an acquisition unit 300, a calculation unit 301, an adjustment unit 302, a presentation unit 303, a setting unit 304, and a regeneration unit 305 by executing the information processing program 311 stored in the storage unit 31.

[0024] The acquisition unit 300 performs an acquisition process to acquire at least one of the following: usage frequency information, which shows the user's usage history over a predetermined period of time for the large-scale language model; and conversation depth information, which shows the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user. The usage frequency information is obtained, for example, based on the number of days the system is used within a predetermined period or the number of dialogue sessions per day. The conversation depth information is obtained, for example, based on the average number of turns of dialogue within a session or the degree of inference layering.

[0025] The calculation unit 301 performs a calculation process to calculate interaction density information that indicates an appropriate value for the explanatory granularity in the response information D2 generated by the large-scale language model in response to user input information, based on at least one of the usage frequency information and conversation depth information acquired by the acquisition unit 300. The calculation unit 301 also sets a time window with a predetermined time width and calculates the interaction density information based on the usage frequency information, etc., included in the time window. Furthermore, the calculation unit 301 continuously updates the interaction density information by changing the time window at a predetermined update cycle. This ensures that the appropriate value is determined based on a quasi-steady-state usage condition rather than temporary fluctuations in input.

[0026] The adjustment unit 302 performs an adjustment process to adjust the explanatory granularity of the response information D2 by setting constraints on the explanatory granularity for the generation process of the response information D2 by the large-scale language model, based on the interaction density information calculated by the calculation unit 301, before the generation process of the response information D2 is executed. Specifically, the adjustment unit 302 imposes constraints on at least one of the following: prompt information D1, which is input information input to the large-scale language model, and reference information used by the large-scale language model. The constraints are set on at least one of the following: the sentence structure of the response information D2, the degree of abstraction of each element constituting the response information D2, and the presentation order of multiple elements constituting the response information D2. In addition, the adjustment unit 302 dynamically switches the explanatory granularity adjustment mode in response to whether the change in the interaction density information over time satisfies predetermined transition conditions.

[0027] The presentation unit 303 performs presentation processing to present control state information related to interaction density information to the user via a display screen or the like provided by the user terminal 2. As a result, the current control state at the explanatory granularity is not implicitly optimized, but rather becomes visible to the user.

[0028] The setting unit 304 receives instructions from the user regarding whether or not to adjust the explanatory granularity based on the interaction density information, and performs setting processing to enable or disable the adjustment. Furthermore, based on the user's instructions regarding the control state information presented in the presentation unit 303, the setting unit 304 sets one of the following: continue, change, fix, or disable the adjustment of the explanatory granularity.

[0029] When the setting unit 304 is configured to change, fix, or disable the adjustment of the explanation granularity, the regeneration unit 305 reconstructs the prompt information D1 and constraints according to the setting and performs a regeneration process to regenerate the answer information D2 for the large-scale language model. This allows the user to correct the explanation granularity of the answer information D2 presented to them at any time if they feel that the granularity is inappropriate.

[0030] (Hardware configuration of each device) Figure 3 is a hardware configuration diagram showing an example of a computer 900 as an information processing device. The user terminal 2, information processing device 3, and generative AI device 5 in the information processing system 1 are composed of a general-purpose or dedicated computer 900.

[0031] As shown in Figure 3, the computer 900 comprises, as its main components, a bus 910, a processor 912, memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication interface unit 922, an external device interface unit 924, an I / O device interface unit 926, and a media input / output unit 928. Note that the above components may be omitted as appropriate depending on the intended use of the computer 900.

[0032] The processor 912 consists of one or more arithmetic processing units (CPU (Central Processing Unit), MPU (Micro-processing unit), DSP (digital signal processor), GPU (Graphics Processing Unit), etc.) and operates as a control unit that oversees the entire computer 900. The memory 914 stores various data and programs 930 and consists of volatile memory (DRAM, SRAM, etc.) that functions as main memory, and non-volatile memory (ROM), flash memory, etc.

[0033] The input device 916 consists of, for example, a keyboard, mouse, numeric keypad, electronic pen, microphone, etc., and functions as an input unit. The output device 917 consists of, for example, a sound (voice) output device, a vibration device, etc., and functions as an output unit. The display device 918 consists of, for example, a liquid crystal display, organic EL display, electronic paper, projector, etc., and functions as an output unit. The input device 916 and the display device 918 may be configured as an integrated unit, such as a touch panel display. The storage device 920 consists of, for example, an HDD, SSD, etc., and functions as a storage unit. The storage device 920 stores various data necessary for the execution of the operating system and program 930.

[0034] The communication I / F unit 922 is connected to a network 940 such as the Internet or an intranet by wire or wireless, and functions as a communication unit that sends and receives data with other computers according to a predetermined communication standard. The external device I / F unit 924 is connected to external devices 950 such as cameras, printers, scanners, and reader / writers by wire or wireless, and functions as a communication unit that sends and receives data with external devices 950 according to a predetermined communication standard. The I / O device I / F unit 926 is connected to I / O devices 960 such as various sensors and actuators, and functions as a communication unit that sends and receives various signals and data with the I / O devices 960, such as detection signals from sensors and control signals to actuators. The media input / output unit 928 is composed of a drive device such as a DVD drive or CD drive, and reads and writes data to media (non-temporary storage medium) 970 such as DVDs and CDs.

[0035] In the computer 900 having the above configuration, the processor 912 calls and executes the program 930 stored in the storage device 920 in the memory 914, and controls various parts of the computer 900 via the bus 910. The program 930 may also be stored in the memory 914 instead of the storage device 920. The program 930 may be recorded on the media 970 in an installable or executable file format and provided to the computer 900 via the media input / output unit 928. The program 930 may also be provided to the computer 900 by downloading it via the network 940 through the communication interface unit 922. Furthermore, the computer 900 may implement the various functions realized by the processor 912 executing the program 930 using hardware such as an FPGA or ASIC.

[0036] Computer 900 is an electronic device of any form, consisting of, for example, a stationary computer or a portable computer. Computer 900 may be a client computer, a server computer, a cloud computer, or an embedded computer such as a control panel or controller (including microcontrollers, programmable logic controllers, and sequencers).

[0037] (Data structure and management examples) Figure 4 shows an example of the data structure and management of various types of information stored in the external storage 4 or memory unit 31. In the information processing system 1, user-specific dialogue logs and usage records are continuously recorded in a database table format that can be managed individually.

[0038] As shown in Figure 4, the data structure is configured in association with identification information that uniquely identifies the user. This identification information records usage frequency information, which shows the user's usage history of the large-scale language model over time, and conversation depth information, which shows the degree of repetition of dialogue in a session.

[0039] Frequency of use information is a quantitative indicator that shows how frequently and continuously a user uses the information processing system 1. Specific examples of frequency of use information include the number of days the system is used within a given period (e.g., the most recent month), the average number of dialogue sessions per day or week, the cumulative number of prompt information D1 transmissions within a given period, or the login interval to the system. These indicators allow for an evaluation of the density of temporal interaction, determining whether the user has established the system as a daily tool or is only using it temporarily or sporadically.

[0040] Conversation depth information is an indicator that shows how deeply a dialogue is explored between a user and a large-scale language model within a single session initiated on a single theme or purpose. Specific examples of conversation depth information include the number of dialogue turns, which represents the average number of rounds of prompt information D1 and response information D2 exchanged within a single session; the percentage of follow-up questions that build upon the content of the previous response information D2 generated by the large-scale language model; the duration of dialogue on the same topic; or the average number of characters and logical complexity of prompt information D1 entered by the user. These indicators allow for an evaluation of the depth of interaction on a session basis, indicating whether the user tends to be satisfied with a single answer and end the dialogue, or whether they prefer exploratory dialogue that deepens their thinking based on the answer.

[0041] Furthermore, the data structure records interaction density information calculated by the calculation unit 301 based on the above-mentioned frequency of use information and conversation depth information. This interaction density information is a value calculated in units of a time window having a predetermined time width, and is maintained as a quasi-stationary index that smooths out short-term input fluctuations. As the time window progresses at a predetermined update cycle, this item is continuously updated with a newly calculated appropriate value.

[0042] Furthermore, the data structure records control state information determined based on the calculated interaction density information. The control state information defines constraints on the level of explanatory detail for the generation process of response information D2. For example, if the interaction density information shows a relatively low value, restrictive constraints focusing on key points are recorded; if it shows a moderate value, standard constraints are recorded; and if it shows a high value, exploratory constraints that provide a multi-layered and detailed explanation are recorded. The adjustment unit 302 refers to this control state information before executing the generation process of response information D2 and determines the constraints to be imposed on the large-scale language model.

[0043] In addition, the data structure records the user's settings. These settings include information indicating whether the function to adjust the granularity of explanation based on interaction density information is enabled or disabled, and whether or not a specific control state is fixed by the user's explicit instruction. This prevents the explanation granularity and sentence structure of response information D2 from being implicitly optimized in a way unintended by the user, ensuring transparency in dialogue control.

[0044] (Overall flow of information processing method) Figure 5 is a flowchart showing the overall processing procedure of the information processing method. This process starts, for example, when a request to start a new session is received from user terminal 2, or when prompt information D1 is sent from the user via user terminal 2.

[0045] First, in step S100 (acquisition step), the acquisition unit 300 acquires at least one of the following from the external storage 4 or storage unit 31: usage frequency information indicating the user's usage history over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session. Simultaneously, the information processing device 3 receives prompt information D1 transmitted from the user terminal 2. At this time, the acquisition of usage frequency information or conversation depth information by the information processing device 3, or the acceptance of prompt information D1 (input information), may be performed by any of the following means: interaction with the user, a management user interface, an application programming interface (API), or automated processing.

[0046] Next, in step S200 (calculation step), the calculation unit 301 calculates interaction density information indicating an appropriate value for explanation granularity based on the usage frequency information and conversation depth information acquired in step S100. The detailed procedure for calculating and updating this interaction density information will be explained later using Figure 6.

[0047] Next, in step S300 (adjustment step), the adjustment unit 302 sets constraints regarding the granularity of explanation for the generation process of response information D2 by the large-scale language model, based on the calculated interaction density information. Specifically, the adjustment unit 302 refers to the control state information corresponding to the interaction density information and determines constraints regarding the sentence structure, degree of abstraction, and presentation order of the response information D2. The determined constraints are either added to the prompt information D1 in the form of a system prompt or otherwise, or set as inference parameters of the large-scale language model, before the generation process is executed.

[0048] Specifically, the adjustment unit 302 refers to control state information corresponding to interaction density information and determines constraints on the sentence structure, level of abstraction, and presentation order of the response information D2. The determined constraints are added as invisible commands to the beginning or end of prompt information D1 in the form of a system prompt before the generation process is executed. For example, in the concise mode, a command that restricts the sentence structure and number of characters is added, such as "The output should be no more than one paragraph, without using bullet points, and only the conclusion should be output within XX characters." On the other hand, in the detailed mode, a command that says "Use headings and bullet points, and output from multiple perspectives with a structure of three or more levels: background, specific examples, and technical issues" is added. Alternatively, the granularity of the explanation may be controlled by directly increasing or decreasing the value of a parameter that defines the maximum number of tokens to be generated, as an inference parameter of the large-scale language model, according to the mode.

[0049] Next, in step S400 (generation process), the generative AI device 5 uses a large-scale language model to perform inference processing on prompt information D1 under the constraints set in step S300, and generates response information D2. This generates response information D2 with an explanation granularity that is most appropriate for the user's current usage state. In this case, since the constraints are limited to structural attributes such as explanation granularity, the direction and accuracy of the information provided are maintained.

[0050] Next, in step S500 (output and recording process), the information processing device 3 transmits the generated response information D2 to the user terminal 2 and displays it on the display unit, etc. Subsequently, the information processing device 3 records the dialogue history of the prompt information D1 and response information D2 performed in the session to the external storage 4, etc., and updates the usage frequency information and conversation depth information. After that, the information processing device 3 terminates this process and moves to a waiting state for the next prompt information D1.

[0051] (Detailed flow of the interaction density information calculation and update process) Figure 6 is a flowchart showing the detailed processing procedure for step S200 (calculation process) in Figure 5. This process is performed by the calculation unit 301 and the adjustment unit 302, and is a process of calculating indicators and state transitions to achieve quasi-steady-state control that is not affected by temporary input fluctuations.

[0052] First, in step S201, the calculation unit 301 sets a time window having a predetermined time width. The time window is defined as a fixed period, such as the past week or the past month, counting back from the current time.

[0053] Next, in step S202, the calculation unit 301 extracts usage frequency information and conversation depth information included in the set time window. Subsequently, the calculation unit 301 calculates interaction density information based on the extracted usage frequency information and conversation depth information. The interaction density information may be calculated, for example, as the product of the usage frequency information and conversation depth information within the time window, or as a product with optimized weighting. In this way, the interaction density information is calculated as a statistic equivalent to the average value or moving average value within the time window.

[0054] Furthermore, the calculation of interaction density information is not limited to using both frequency of use information and conversation depth information; it may be calculated based on only one of them. For example, a configuration may be used that determines that the interaction density information is high when the frequency of use information exceeds a predetermined threshold, without using conversation depth information, or conversely, a configuration may be used that determines the interaction density information based only on conversation depth information, without using frequency of use information.

[0055] Furthermore, the interaction density information is not limited to a configuration in which it is calculated as a simple product of usage frequency information and conversation depth information within a time window. For example, the calculation unit 301 may classify the usage frequency information and conversation depth information into multiple levels (e.g., high, medium, low) and determine the interaction density information using a two-dimensional matrix with both as axes. Specifically, if the usage frequency information is "high" but the conversation depth information is "low" (e.g., a user who only asks single questions on a daily basis), the interaction density information is classified as "low" or "medium" and controlled to maintain a concise level of explanation. This prevents a discrepancy between the user's actual intent and the system's response level.

[0056] Next, in step S203, the calculation unit 301 determines whether a predetermined update cycle has arrived. The control of the explanatory granularity based on the interaction density information is not changed immediately each time prompt information D1 is input, but is re-evaluated at predetermined update cycles. If the update cycle has not arrived, the calculation unit 301 does not update the information, and the current control conditions are maintained. This makes it possible to suppress instability of the explanatory granularity caused by short-term fluctuations in prompt information D1 or temporary changes in usage patterns.

[0057] When the update cycle arrives, in step S204, the calculation unit 301 updates the interaction density information by changing the time window. Subsequently, in step S205, the adjustment unit 302 determines whether the change in the updated interaction density information over time satisfies predetermined transition conditions.

[0058] For example, a hysteresis condition is applied to the predetermined transition conditions. That is, even if the updated interaction density information temporarily exceeds or falls below a threshold corresponding to the current control state, the adjustment unit 302 does not immediately switch the control state. The adjustment unit 302 determines that the transition condition is met only when it confirms that the state of exceeding or falling below the threshold has continued under certain conditions. This avoids excessively frequent switching of the control state near the threshold and achieves stable control of explanatory granularity.

[0059] However, regardless of the hysteresis conditions or update cycle described above, the adjustment unit 302 may execute an exception handling process to immediately switch the control state if certain conditions are met. For example, this may occur if the number of characters or the logical complexity of the prompt information D1 currently entered suddenly exceeds a predetermined threshold from the past average value, or if it is determined that the topic has changed to a specialized topic different from the previous conversation (for example, a transition from everyday conversation to debugging programming). In such cases, the adjustment unit 302 immediately treats the interaction density information as a high value without waiting for the transition conditions to be met, and switches to a mode with detailed explanation granularity. This makes it possible to achieve both stability that smooths out temporary fluctuations and agility that can follow the user's sudden task changes.

[0060] If it is determined that the transition conditions are met, in step S206, the adjustment unit 302 determines new control state information corresponding to the updated interaction density information and switches the explanation granularity adjustment mode. On the other hand, if it is determined that the transition conditions are not met, the adjustment unit 302 does not switch states and maintains the current control state. After the above processing, this subroutine terminates and returns to the overall flow shown in Figure 5.

[0061] (Detailed flow of the setup and regeneration process) Figure 7 is a flowchart showing the detailed processing procedures for the setting and regeneration steps in this embodiment. This process is performed by the presentation unit 303, the setting unit 304, and the regeneration unit 305, and is a process to ensure transparency regarding the control of the level of detail of the explanation and to avoid a configuration in which the control state is automatically optimized or changed based solely on the system's judgment.

[0062] First, in step S701, the presentation unit 303 presents the user with control state information related to interaction density information. This prevents the control state from being implicitly changed by the system alone, and makes visible to the user what level of explanatory constraints the current dialogue is being conducted under.

[0063] Next, in step S702, the setting unit 304 receives instructions from the user regarding the presented control state information. Based on the user's instructions, the setting unit 304 sets whether or not to adjust the explanatory granularity based on the interaction density information. Specifically, the setting unit 304 receives and sets one of the following instructions: continue, change, fix, or disable the adjustment of explanatory granularity. This system is designed to activate control only when explicit consent is obtained from the user.

[0064] Next, in step S703, the setting unit 304 determines whether the received instruction is to change, fix, or invalidate. If the instruction is to continue the adjustment (No), the setting unit 304 maintains the current control state and terminates this subroutine.

[0065] If the received instruction is to change, fix, or disable the adjustment of the explanation granularity (Yes), in step S704, the regeneration unit 305 causes the large-scale language model to regenerate the response information D2 according to the setting. For example, if the user feels that the currently applied exploratory explanation granularity is redundant and instructs to disable it, the regeneration unit 305 removes the constraints and regenerates the standard response information D2 and presents it to the user terminal 2.

[0066] The above processing avoids a configuration in which control conditions related to interaction density information or explanatory granularity are implicitly optimized by the system, thereby achieving appropriate dialogue control that reflects the user's intentions. Furthermore, because the control target is limited to explanatory granularity such as document structure and level of abstraction, the design ensures that even if the control is disabled or changed, there will be no disadvantage in terms of reduced accuracy or completeness of the information provided.

[0067] Furthermore, while the information processing system 1 ensures stability by performing quasi-steady-state control that is not affected by short-term input fluctuations using hysteresis conditions, etc., it is designed to substantially avoid implicit automatic optimization and exceptional state transitions on the system side. Instead, by linking the explicit user operations (changing, fixing, or disabling adjustments) with the immediate regeneration process by the regeneration unit 305, agility is ensured when the user suddenly requests explanations of different granularities. This prevents black-box state transitions by the system while achieving highly transparent interactive control that balances stability and agility.

[0068] (Specific examples of user interface and response information) Figure 8 shows an example of a user interface provided to the display unit of the user terminal 2. As shown in Figure 8, the display unit is provided with a screen that includes a chat area for displaying the dialogue history of prompt information D1 and response information D2, an input area for inputting prompt information D1, a control state display area for showing the current control state, and an adjustment operation unit for executing the setting process.

[0069] In the control state presentation area, the currently applied explanatory granularity mode, based on the calculated interaction density information, is visualized and constantly displayed using text, indicators, etc. The adjustment operation section is provided as a graphical user interface in the form of buttons and sliders. By operating this adjustment operation section, the user is presented with details of the current control state via the presentation section 303, and can input instructions via the setting section 304, such as changing the explanatory granularity, fixing it to a specific granularity, or disabling the system's automatic adjustment function.

[0070] Figure 8 illustrates a scenario where a user inputs prompt information D1 regarding the mechanism of smart contracts in the blockchain.

[0071] In state (B), where interaction density information is low and the explanatory granularity is controlled to a concise mode, the large-scale language model generates response information D2 consisting only of key points, minimizing background explanations and technical details. In this case, response information D2 is displayed in the chat area as, for example, a short paragraph that concisely explains the outline of the smart contract. This reduces the cognitive load for infrequent users.

[0072] On the other hand, in state (A), where interaction density information is high and the explanatory granularity is controlled to a detailed mode, the large-scale language model generates multi-layered response information D2 that includes multiple layers and supplementary information of different levels of abstraction. In this case, the response information D2 is structured and displayed in the chat area with multiple items such as an overview of the smart contract, as well as the technical foundation, features, and specific application examples. This provides exploratory information to users seeking deeper dialogue. At all levels of granularity, the accuracy and completeness of the information are maintained, and the design ensures that no disadvantage is caused to the user.

[0073] Furthermore, if a user receives concise response information D2 and then, wanting to know more, operates the adjustment control unit and inputs an instruction to change the explanation granularity to a detailed mode, the regeneration unit 305 will function. The regeneration unit 305 instantly regenerates the previous response information D2 into a detailed structure, changing the constraints, without requiring the user to re-enter prompt information D1, and updates the display in the chat area. In this way, by linking the visualized user interface with the regeneration function, highly transparent dialogue control is achieved, eliminating the black box aspect.

[0074] <Other Embodiments> This disclosure is not limited to the embodiments described above, and can be implemented with various modifications without departing from the spirit of this disclosure. All such modifications are included in the technical concept of this disclosure.

[0075] In the above embodiment, we described a case where the information processing system 1 is configured as a cloud-type system in which individual devices such as a user terminal 2, an information processing device 3, external storage 4, and a generative AI device 5 are connected via a network 6. In contrast, the information processing system 1 may be configured as a standalone system in which all functions are implemented within a single computer (for example, a high-performance smartphone, tablet terminal, or personal computer). In this case, a secure environment is established that eliminates network latency and physically prevents the leakage of confidential information to the outside.

[0076] Furthermore, although the above embodiment was described using general dialogue processing as an example, the information processing system 1 may be configured as a system specialized in a specific field such as educational support, research support, business support, or creative support. For example, when applied to an educational support system, the information processing device 3 can provide learners with low interaction density information with concise answer information D2 that extracts basic concepts, and learners with high interaction density information with exploratory answer information D2 that includes academic background and multi-layered discussions. This makes it possible to provide an optimal learning experience according to the learner's usage status and the depth of their exploration.

[0077] Furthermore, the above embodiment described a case where the generative AI device 5 performs generation processing using a general-purpose large-scale language model (LLM). In contrast, small-scale language models (SLMs) specialized for specific tasks or domains may also be used. Alternatively, the information processing device 3 may be configured to dynamically switch the model being used according to the content of the prompt information D1 (for example, switching to a medical model if the topic is medical, or to a coding model if it is programming).

[0078] Furthermore, in the above embodiments, the present invention was described as a hardware configuration consisting of an information processing device 3 and an information processing system 1. In contrast, the present invention may be configured as a program that causes a computer (processor) to execute each of the above-described processes, namely the acquisition process, calculation process, adjustment process, etc. The program may be provided via a network such as the Internet, or it may be provided stored on a computer-readable non-temporary recording medium such as an optical disc, memory card, or USB memory.

[0079] Furthermore, the above embodiment described the case where the prompt information D1 and response information D2 are mainly text data. In contrast, the prompt information D1 and response information D2 may include multimodal data such as images, audio, video, program code, or 3D model data. In this case, a multimodal compatible model may be adopted as the generative artificial intelligence model 50, and the adjustment unit 302 may be configured to adjust not only the sentence structure of the text, but also the complexity of the constituent elements of the generated images and the level of detail in the explanations in the audio output as explanation granularity.

[0080] Furthermore, the above embodiment mainly described a configuration in which constraints regarding the granularity of explanation are added to the prompt information D1. In contrast, the information processing device 3 may be configured to impose constraints on external reference information used when the large-scale language model generates the answer information D2. For example, when using a search-extended generation technology that searches for knowledge from an external database and incorporates it into the generation process, the adjustment unit 302 controls the extraction of only basic and general documents as reference information when the interaction density information is low, and to include documents with specialized and multi-layered structures as reference information when the interaction density information is high. This makes it possible to adjust the granularity of the prerequisite knowledge provided to the large-scale language model according to the user's usage state.

[0081] Furthermore, the above embodiment described a case in which interaction density information is calculated based on usage frequency information and conversation depth information. In contrast, when calculating interaction density information, other behavior-derived indicators such as the average number of characters in prompt information D1, logical complexity, the tendency of the user to ask additional questions regarding answer information D2, or the number of correction inputs may be incorporated as parameters. This makes it possible to quantify the maturity of the relationship formed between the user and the system in a more multifaceted and accurate manner, rather than simply being limited to the number of days of use or the number of dialogue turns, and to achieve even finer control over the level of detail in explanations.

[0082] Furthermore, in the above embodiment, a configuration was described in which an index called interaction density information is calculated based on usage frequency information and conversation depth information, and the granularity of explanation is adjusted based on this index. In contrast, a configuration may be adopted in which constraints on the granularity of explanation are directly determined and adjusted using a predetermined rule base or judgment table, etc., from the raw data or simple statistical values ​​of the acquired usage frequency information and conversation depth information, without explicitly calculating an intermediate index such as interaction density information.

[0083] Although embodiments of this disclosure have been described above, the information processing system 1 is not limited to being provided as a chat application (BtoC service) with a screen that end users can directly operate. For example, the information processing device 3 may be configured as a backend infrastructure system with a function to control the granularity of explanation, and may be provided to an external client system (such as a third-party application server) via an interface such as a web API (Application Programming Interface) as a cloud service (such as SaaS or PaaS).

[0084] In this API provisioning model, the external client system adds contextual information such as user identification information, usage frequency information, and conversation depth information, along with prompt information D1, to the API request parameters or payload and sends it to the information processing device 3. The control unit 30 of the information processing device 3 parses the data structure of the received API request and performs the calculation of interaction density information and setting of constraint conditions. Then, it returns the response information D2, which has been adjusted to an appropriate level of explanatory granularity as a result of the processing, to the client system as an API response.

[0085] Thus, even a loosely coupled system architecture in which prompt information D1 (input information) and information indicating usage status are dynamically input from an external system via the network as an API request, and the result of controlling the level of explanation granularity is returned as an API response, is included within the technical scope of this disclosure.

[0086] The effects of the various aspects of this disclosure are summarized below.

[0087] One aspect of the present disclosure is an information processing method to be performed in an information processing system that utilizes a large-scale language model, comprising: an acquisition step of acquiring at least one of the following: usage frequency information indicating the usage history of the large-scale language model over a predetermined period of time by a user of the large-scale language model, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units between the large-scale language model and the user; and an adjustment step of adjusting the granularity of explanation in response information generated by the large-scale language model in response to input information from the user, based on at least one of the acquired usage frequency information and conversation depth information.

[0088] According to this configuration, the system can appropriately grasp the maturity level of the relationship formed between the user and the system based on the user's usage history over time and the depth of dialogue during sessions, and automatically adjust the level of detail in the explanation of the response information. This makes it possible to provide a dialogue experience that is optimal for the user's usage state, such as providing concise explanations to reduce cognitive load for infrequent users and providing multi-layered explanations to promote thinking for frequent users.

[0089] One aspect of the present disclosure is an information processing method comprising: a calculation step of calculating interaction density information indicating an appropriate value for the descriptive granularity based on at least one of the acquired usage frequency information and conversation depth information, and further performing an adjustment step of adjusting the descriptive granularity based on the interaction density information.

[0090] According to this embodiment, by calculating the user's usage status, which tends to be qualitative, as an objective and quantitative indicator called interaction density information, it is possible to achieve consistent and stable control of explanatory granularity throughout the entire system.

[0091] One aspect of the present disclosure is an information processing method in which, in the adjustment step, the explanation granularity is adjusted by setting constraints on the explanation granularity for the generation process before the execution of the response information generation process by the large-scale language model.

[0092] In this configuration, instead of retrospectively increasing or decreasing the output text, the explanation granularity is adjusted as a constraint before the response generation is performed. This fundamentally suppresses unnecessary inference processing and excessive explanation generation, thereby improving the efficiency of the system's overall computing resources and increasing response speed.

[0093] One aspect of the present disclosure is an information processing method in which, in the adjustment step, the setting of the constraints is performed by imposing constraints on at least one of the input information to the large-scale language model and the reference information used by the large-scale language model.

[0094] According to this configuration, not only are direct constraints on input information controlled, but the granularity of reference information retrieved and used from external sources can also be controlled simultaneously, making it possible to consistently adapt the system to the user's usage state, starting from the level of prerequisite knowledge provided to the large-scale language model.

[0095] One aspect of the present disclosure is an information processing method in which, in the adjustment step, the constraints are set with respect to at least one of the following: the sentence structure of the response information, the degree of abstraction of each element constituting the response information, and the presentation order of the multiple elements constituting the response information.

[0096] According to this approach, by limiting the adjustments to structural attributes such as sentence structure and degree of abstraction, and not affecting the core content or direction of the information itself, it is possible to prevent information bias due to personalization and maintain the provision of accurate and objective information. This enables dialogue control that does not disadvantage the user.

[0097] One aspect of the present disclosure is an information processing method in which, in the calculation step, a time window having a predetermined time width is set, the interaction density information is calculated based on at least one of the usage frequency information and the conversation depth information included in the time window, and the interaction density information is updated by changing the time window at a predetermined update cycle.

[0098] According to this embodiment, interaction density information can be calculated as a quasi-stationary index that smooths out short-term fluctuations in input information and temporary changes in usage patterns, thereby suppressing unnecessary fluctuations in explanatory granularity due to transient factors and providing a stable dialogue experience over a long period of time.

[0099] One aspect of the present disclosure is an information processing method in which, in the adjustment step, the adjustment of the explanatory granularity is performed in accordance with the change over time of the updated interaction density information satisfying predetermined transition conditions.

[0100] According to this embodiment, by setting transition conditions such as hysteresis, it is possible to avoid excessively frequent switching of control states when interaction density information fluctuates near a threshold, and to achieve a smooth transition in explanatory granularity that does not cause discomfort to the user.

[0101] One aspect of the present disclosure is an information processing method that further includes a setting step of determining whether or not to adjust the explanatory granularity based on the interaction density information.

[0102] This approach allows for a design that enables adjustment of explanation granularity only when explicit instructions or consent are obtained from the user, thereby eliminating implicit optimization by the system and ensuring user sovereignty.

[0103] One aspect of this disclosure is an information processing method that further includes a presentation step of presenting control state information relating to interaction density information to the user.

[0104] This embodiment prevents the current state of explanation granularity control from becoming an implicit black box, making it possible to visualize to the user the constraints under which the system generates the answer information, thereby ensuring a high degree of transparency.

[0105] One aspect of this disclosure is an information processing method in which the setting step sets one of the following: continue, change, fix, or disable the adjustment of the explanatory granularity, based on instructions from the user regarding the control state information presented in the presentation step.

[0106] According to this embodiment, users can flexibly intervene in the visualized control state information according to their own intentions, and manually adjust or fix the level of detail of the explanation, thereby improving user satisfaction and convenience with interactive control.

[0107] One aspect of this disclosure is an information processing method that further includes a regeneration step of regenerating the response information in accordance with the setting when the adjustment of the explanatory granularity is set to be changed, fixed, or disabled in the setting step.

[0108] According to this configuration, if a user is dissatisfied with the current level of detail in the explanation and changes the settings, they can immediately obtain response information based on the new constraints without having to re-enter new input information, thereby significantly improving the efficiency of the dialogue.

[0109] One aspect of this disclosure is an information processing method in which the acquisition of the frequency of use information or the conversation depth information in the acquisition step, or the acceptance of the input information, is performed by one of the following: interaction with a user, a management user interface, an application programming interface, or automated processing.

[0110] According to this embodiment, various types of information can be received not only through interactive input via a terminal directly operated by the user, but also through diverse channels such as API integration from external systems and automated processing within the system. As a result, the information processing method related to this disclosure can be flexibly applied to diverse system architectures and service forms, not only as an independent application for end users, but also as a backend cloud service (such as SaaS) that provides functionality to other client systems, or as part of an automated workflow.

[0111] While several embodiments of this disclosure have been illustrated above, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. Furthermore, the embodiments described above can be implemented in combination with each other.

[0112] The various aspects of this disclosure are summarized below as an appendix.

[0113] (Note 1) An information processing method performed in an information processing system that utilizes a large-scale language model, An acquisition step of acquiring at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user; The process involves adjusting the descriptive granularity in the response information generated by the large-scale language model in response to the user input information, based on at least one of the acquired frequency of use information and conversation depth information. Information processing methods.

[0114] (Note 2) A calculation step is performed to calculate interaction density information indicating an appropriate value for the explanatory granularity, based on at least one of the acquired usage frequency information and conversation depth information. In the adjustment process described above, The explanatory granularity is adjusted based on the interaction density information. The information processing method described in Appendix 1.

[0115] (Note 3) In the adjustment process described above, Before executing the process of generating the response information using the large-scale language model, the explanation granularity is adjusted by setting constraints on the explanation granularity for the generation process. The information processing method described in Appendix 1 or Appendix 2.

[0116] (Note 4) In the adjustment process described above, The setting of the aforementioned constraints is carried out by imposing constraints on at least one of the input information to the large-scale language model and the reference information used by the large-scale language model. The information processing method described in any one of the items in Appendix 1 to Appendix 3.

[0117] (Note 5) In the adjustment process described above, The aforementioned constraints are set with respect to at least one of the following: the sentence structure of the response information, the degree of abstraction of each element constituting the response information, and the presentation order of the multiple elements constituting the response information. The information processing method described in any one of the items in Appendix 1 to Appendix 4.

[0118] (Note 6) In the calculation process described above, A time window with a predetermined time width is set, the interaction density information is calculated based on at least one of the usage frequency information and the conversation depth information included in the time window, and the interaction density information is updated by changing the time window at a predetermined update cycle. The information processing method described in any one of the items in Appendix 1 to Appendix 5.

[0119] (Note 7) In the adjustment process described above, The adjustment of the descriptive granularity is performed in response to the change over time of the updated interaction density information satisfying predetermined transition conditions. The information processing method described in any one of the items in Appendix 1 to Appendix 6.

[0120] (Note 8) A setting step is performed to determine whether or not to adjust the explanatory granularity based on the interaction density information. The information processing method described in any one of the items in Appendix 1 to Appendix 7.

[0121] (Note 9) The process further includes presenting control state information relating to the interaction density information to the user, The information processing method described in any one of the items in Appendix 1 to Appendix 8.

[0122] (Note 10) The aforementioned setting process is: Based on the user's instructions regarding the control state information presented in the presentation step, one of the following is set: continue, change, fix, or disable the adjustment of the explanatory granularity. The information processing method described in any one of the items from Appendix 1 to Appendix 9.

[0123] (Note 11) If the adjustment of the explanatory granularity is changed, fixed, or disabled in the setting step, a regeneration step is further performed to regenerate the response information according to the setting. The information processing method described in any one of the items in Appendix 1 to Appendix 10.

[0124] (Note 12) An information processing system that utilizes a large-scale language model, An acquisition unit that acquires at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user. A calculation unit calculates interaction density information that indicates an appropriate value for the explanatory granularity in the response information generated by the large-scale language model in response to the user input information, based on at least one of the acquired usage frequency information and conversation depth information. The system includes an adjustment unit that adjusts the explanatory granularity of the response information by setting constraints on the explanatory granularity for the generation process based on the interaction density information, before the generation process of the response information by the large-scale language model is executed. Information processing system.

[0125] (Note 13) A program for causing an information processing device, which comprises a processor and a memory unit, to execute processing that utilizes a large-scale language model, The aforementioned processor, An acquisition step to acquire at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user; A calculation step of calculating interaction density information that indicates an appropriate value for the explanatory granularity in the response information generated by the large-scale language model in response to the user input information, based on at least one of the acquired frequency of use information and conversation depth information, Based on the interaction density information, before the execution of the response information generation process by the large-scale language model, an adjustment step is performed to adjust the explanatory granularity of the response information by setting constraints on the explanatory granularity for the generation process. program. [Explanation of symbols]

[0126] 1... Information processing system, 2... User terminal, 3... Information processing device 30...Control unit, 300...Acquisition unit, 301...Calculation unit, 302...Adjustment unit, 303...Presentation section, 304...Setting section, 305...Regeneration section, 31...Memory unit, 310...Session management data, 311...Information processing program, 32...Communication unit, 33...Input unit, 34...Display unit, 4…External storage, 5…Generative AI device, 50…Generative artificial intelligence model 6…Network

Claims

1. An information processing method performed in an information processing system that utilizes a large-scale language model, An acquisition step of acquiring at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user; The process involves adjusting the descriptive granularity in the response information generated by the large-scale language model in response to the user input information, based on at least one of the acquired frequency of use information and conversation depth information. Information processing methods.

2. A calculation step is performed to calculate interaction density information indicating an appropriate value for the explanatory granularity, based on at least one of the acquired usage frequency information and conversation depth information. In the adjustment process described above, The explanatory granularity is adjusted based on the interaction density information. The information processing method according to claim 1.

3. In the adjustment process described above, Before executing the process of generating the response information using the large-scale language model, the explanation granularity is adjusted by setting constraints on the explanation granularity for the generation process. The information processing method according to claim 1 or claim 2.

4. In the adjustment process described above, The setting of the aforementioned constraints is carried out by imposing constraints on at least one of the input information to the large-scale language model and the reference information used by the large-scale language model. The information processing method according to claim 3.

5. In the adjustment process described above, The aforementioned constraints are set with respect to at least one of the following: the sentence structure of the response information, the degree of abstraction of each element constituting the response information, and the presentation order of the multiple elements constituting the response information. The information processing method according to claim 3.

6. In the calculation process described above, A time window with a predetermined time width is set, the interaction density information is calculated based on at least one of the usage frequency information and the conversation depth information included in the time window, and the interaction density information is updated by changing the time window at a predetermined update cycle. The information processing method according to claim 2.

7. In the adjustment process described above, The adjustment of the descriptive granularity is performed in response to the change over time of the updated interaction density information satisfying predetermined transition conditions. The information processing method according to claim 6.

8. A setting step is performed to determine whether or not to adjust the explanatory granularity based on the interaction density information. The information processing method according to claim 2.

9. The process further includes presenting control state information relating to the interaction density information to the user, The information processing method according to claim 8.

10. The aforementioned setting process is: Based on the user's instructions regarding the control state information presented in the presentation step, one of the following is set: continue, change, fix, or disable the adjustment of the explanatory granularity. The information processing method according to claim 9.

11. If the adjustment of the explanatory granularity is changed, fixed, or disabled in the setting step, a regeneration step is further performed to regenerate the response information according to the setting. The information processing method according to claim 10.

12. The acquisition of the usage frequency information or the conversation depth information in the acquisition process, or the acceptance of the input information, This is done through one of the following: user interaction, administrative user interface, application programming interface, or automated processing. The information processing method according to claim 1.

13. An information processing system that utilizes a large-scale language model, An acquisition unit that acquires at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user. A calculation unit calculates interaction density information that indicates an appropriate value for the explanatory granularity in the response information generated by the large-scale language model in response to the user input information, based on at least one of the acquired usage frequency information and conversation depth information. The system includes an adjustment unit that adjusts the explanatory granularity of the response information by setting constraints on the explanatory granularity for the generation process based on the interaction density information, before the generation process of the response information by the large-scale language model is executed. Information processing system.

14. A program for causing an information processing device, which comprises a processor and a memory unit, to execute processing that utilizes a large-scale language model, The aforementioned processor, An acquisition step to acquire at least one of the following: usage frequency information indicating the usage history of the user of the large-scale language model over a predetermined period, and conversation depth information indicating the degree of repetition of dialogue in a session, which is a series of dialogue units that take place between the large-scale language model and the user; A calculation step of calculating interaction density information that indicates an appropriate value for the explanatory granularity in the response information generated by the large-scale language model in response to the user input information, based on at least one of the acquired frequency of use information and conversation depth information, Based on the interaction density information, before the execution of the response information generation process by the large-scale language model, an adjustment step is performed to adjust the explanatory granularity of the response information by setting constraints on the explanatory granularity for the generation process. program.

Images