Information processing device, information processing method, and program

The information processing device optimizes and analyzes user histories to associate prompts with LLMs, addressing the high cost and time of log analysis, enabling efficient and cost-effective LLM selection.

WO2026234787A1PCT designated stage Publication Date: 2026-11-12NTT DOCOMO INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2025/016987
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2025-05-09
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

The high cost and time-consuming nature of analyzing logs to select appropriate Large Language Models (LLMs) for efficient utilization due to the vast volume of logs, necessitating a method to reduce the size of logs for analysis.

Method used

An information processing device that acquires user LLM prompt and behavior histories, optimizes and analyzes these histories to associate prompts with LLMs, reducing the size of logs by excluding low-value logs and labeling high-value prompts, thereby selecting the most suitable LLM for tasks.

Benefits of technology

Reduces the cost and time required for LLM selection by analyzing a smaller, optimized log set, maintaining accuracy and efficiency in LLM utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2025016987_12112026_PF_FP_ABST
    Figure JP2025016987_12112026_PF_FP_ABST
Patent Text Reader

Abstract

An information processing device according to one embodiment of the present disclosure comprises: an acquisition means for acquiring an LLM prompt history of each user and an action history of each user; a size reduction means for reducing the size of the acquired prompt history on the basis of the action history; and an analysis means for analyzing the reduced-size prompt history and associating a prompt with an LLM.
Need to check novelty before this filing date? Find Prior Art

Description

Information Processing Apparatus, Information Processing Method, and Program

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and a program. In particular, the present disclosure relates to an information processing apparatus, an information processing method, and a program for selecting an LLM.

[0002] It is practiced to improve work efficiency by utilizing LLMs (Large Language Models) such as generative AI (Artificial Intelligence) models. An LLM is a language model constructed with a large volume of data and deep learning technology. Various types of LLMs exist, and the fields of expertise and costs of the respective LLMs differ from each other.

[0003] An analysis method for analyzing only logs of problematic communications among logs to be analyzed is known (for example, Patent Document 1).

[0004] An analysis method is known, in which the importance level of a log is determined using a learning model, and analysis is performed focusing on logs with high importance levels (for example, Patent Document 2).

[0005] It is known to analyze page transition logs and on-page operation logs in order to evaluate the usability of an application (for example, Patent Document 3).

[0006] It is known to identify a problem by comparing an operation log and a recommended operation log, and select a countermeasure method (for example, Patent Document 4).

[0007] It is known that both prompts transmitted to an LLM provider and output results can be described in logs (for example, Non-Patent Document 1).

[0008] Japanese Unexamined Patent Application Publication No. 2017-117429, International Publication No. 2018 / 235252, International Publication No. 2011 / 099170, Japanese Unexamined Patent Application Publication No. 2015-087875

[0009] Cognigy LLM Prompt (URL: https: / / cognigy.tdse.jp / docs / virtual_agents / flow-nodes / other-nodes / llm-prompt / )

[0010] Users tend to use LLMs (Limited Logistics Models) that are highly versatile and accurate. However, using such LLMs is expensive. To utilize LLMs while reducing costs, it is necessary to select the appropriate LLM. While it is possible to make decisions regarding LLM selection by analyzing logs, the logs are enormous, making log analysis time-consuming and costly. Therefore, it is necessary to reduce the size of the logs to be analyzed.

[0011] An information processing device in one aspect of the present disclosure includes: acquisition means for acquiring each user's LLM prompt history and each user's behavior history; optimization means for optimizing the acquired prompt history based on the behavior history; and analysis means for analyzing the optimized prompt history and associating prompts with LLMs.

[0012] According to this disclosure, the logs used for analysis can be made smaller.

[0013] Example of the overall system configuration Example of information stored in a memory device Example of an LLM characteristic table Example of the configuration of an information processing device Example of a functional block diagram of an information processing device Example of a flowchart executed by an information processing device Another example of a flowchart executed by an information processing device Example of a prompt history Example of a processed prompt history Another example of a prompt history Another example of a processed prompt history

[0014] The embodiments of this disclosure will be described in detail below with reference to the drawings as appropriate.

[0015] (Embodiment 1) In Embodiment 1, the information processing device 120 provides information for selecting which LLM to use according to the task.

[0016] Figure 1 shows an example of the overall system 100 configuration. System 100 includes an input / output terminal 110, an information processing device 120, a storage device 130, and an LLM 140.

[0017] The input / output terminal 110 is a terminal that allows the user to input and output information, such as a display, speaker, keyboard, or touch panel. The input / output terminal 110 may also be a wireless terminal installed in the vehicle. The input / output terminal 110 may also be a sensor installed in the vehicle that notifies the battery level. The input / output terminal 110 transmits information entered by the user or information measured by the sensor to the information processing device 120. The input / output terminal 110 may transmit multiple pieces of information to the information processing device 120.

[0018] The information processing device 120 selects the next node. The selection method will be described later.

[0019] The storage device 130 is a storage device such as a hard disk drive (HDD) or a solid state drive (SSD). The storage device 130 may reside in the cloud. The storage device 130 may store an action history 132, a prompt history 131, and an LLM characteristics table 133.

[0020] LLM140 includes multiple LLMs, for example, LLM1 and LLM2. The number of LLMs included in LLM140 is arbitrary. LLM140 may include three or more LLMs. Each LLM differs in its developer, available functions, input token limit, output token limit, and the types of tasks it excels at.

[0021] Figure 2 shows an example of information stored in the storage device 130. The storage device 130 may also store a prompt history 131, an action history 132, and an LLM characteristics table 133.

[0022] The prompt history 131 is a log of each user's LLM transmission and reception. The prompt history 131 includes the prompts entered into the LLM 140 by each user, and the LLM (e.g., LLM1, etc.) that the prompt was entered into, as well as the output of the LLM 140. The prompt history 131 may also include user feedback, such as satisfaction with the output results for each LLM. Satisfaction may be determined by the user. For example, satisfaction may be rated as "good" or "bad". Users are not required to rate satisfaction.

[0023] The activity log 132 is a log of operations performed by each user on the input / output terminal 110, excluding the prompt log 131. The activity log 132 includes operations performed by each user up to the point of sending a prompt, such as the operation log of the input / output terminal 110. For example, the activity log 132 includes information such as the speed of operations (key input, etc.), the number of operational errors, the functions used, the types of buttons pressed, and the circumstances under which the system 100 was accessed. The activity log 132 may also be the activity log for each user in each LLM.

[0024] The LLM characteristics table 133 is a table that describes the relationship between prompts and LLMs. The LLM characteristics table 133 includes the characteristics of each LLM. For example, the LLM characteristics table 133 includes available functions, upper limits on input tokens, upper limits on output tokens, and features. The LLM characteristics table 133 may also include input information (e.g., text, still images, videos, audio, etc.), output information, cost, use cases, response time, and processing capacity (e.g., text generation capacity, mathematical reasoning capacity, coding capacity, number of requests that can be processed within a given time, etc.). Users can use the LLM characteristics table 133 to select an appropriate LLM.

[0025] Figure 3 shows an example of the LLM characteristic table 133.

[0026] The LLM characteristics table 133 contains the characteristics of each LLM. For example, the LLM characteristics table 133 includes available functions, upper limits on input tokens, upper limits on output tokens, and features. The LLM characteristics table 133 may also include input information (e.g., text, still images, videos, audio, etc.), output information, costs, use cases, response time, processing capabilities (e.g., text generation capabilities, mathematical reasoning capabilities, coding capabilities, the amount of requests that can be processed within a given time, etc.).

[0027] Figure 4 shows an example configuration of the information processing device 120. The information processing device 120 includes a processor 121, an input device 122, and an output device 123. In addition to the devices shown in Figure 4, it may also include a communication device, memory, and storage. Each device, such as the processor 121 and the input device 122, is connected by a bus 124 for communicating information. The bus 124 may be configured using a single bus, or different buses may be configured for each device.

[0028] The processor 121 is composed of a computer including a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 121 receives a query from the input / output terminal 110 via the input device 122 and sends a response to the query to the input / output terminal 110. The processor 121 may also retrieve the query by searching the storage device 130.

[0029] The input device 122 may be an input device that receives input information from an external source (e.g., a keyboard, mouse, microphone, switch, button, sensor, etc.), or it may be an interface or communication device that receives data. The input device 122 may be connected to the storage device 130 and other devices not shown.

[0030] The output device 123 may be an output device that outputs user responses to the outside (e.g., a display, speaker, LED (Light Emitting Diode) lamp, or it may be an interface or communication device that transmits data. The output device 123 may be connected to the storage device 130 and other devices not shown.

[0031] The input device 122 and the output device 123 may be configured as an integrated unit (for example, a touch panel), or the communication device may be the input device 122 and / or the output device 123.

[0032] Memory is a computer-readable recording medium and may consist of at least one of the following: ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically EPROM), RAM (Random Access Memory), etc. Memory may also be called registers, cache, main memory, etc. Memory can store executable programs (program code), software modules, etc., for carrying out a wireless communication method according to one embodiment of this disclosure.

[0033] Storage is a computer-readable recording medium and may consist of at least one of the following: an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk (e.g., a floppy disk), a magneto-optical disk (e.g., a compact disk, a digital multipurpose disk, a Blu-ray® disc), a smart card, flash memory (e.g., a card, stick, key drive), or a magnetic strip. Storage may also be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, server, or other suitable medium that includes at least one of memory and storage.

[0034] Communication equipment is hardware (input / output devices) for communicating between computers via at least one of a wired network and a wireless network, and is also called a network device, network controller, network card, communication module, etc.

[0035] Furthermore, the information processing device 120 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array), and some or all of each functional block may be realized by such hardware. For example, the processor 121 may be implemented using at least one of these hardware components.

[0036] Figure 5 shows an example of a functional block diagram of the information processing device 120. The information processing device 120 includes an acquisition means 125, a weight reduction means 126, and an analysis means 127.

[0037] The acquisition means 125 acquires the transmission / reception log (prompt history 131) and the action history 132. The acquisition means 125 may also acquire the prompt history 131 and the action history 132 stored in the storage device 130.

[0038] The lightweighting means 126 performs a lightweighting process on the acquired prompt history 131. The information processing device 120 performs a lightweighting process on the prompt history 131 based on the prompt history 131, and a lightweighting process on the prompt history 131 based on the action history 132.

[0039] The lightweighting means 126 determines each user's LLM utilization skill (ability) based on the acquired behavior history 132. The information processing device 120 determines that, for example, users who use advanced functions (e.g., image search), operate quickly, and make few key input errors have high LLM utilization skills, while users who use help, operate slowly, and make many key input errors have low LLM utilization skills.

[0040] The optimization means 126 performs optimization processing of the prompt history 131 based on the prompt history 131. The optimization means 126 uses a rule-based or AI-based method to extract only a portion of the prompt history from the prompt history 131, for example, only the send / receive logs that are of high value for analysis.

[0041] For example, the optimization means 126 excludes send / receive logs of low value for analysis from the acquired prompt history 131. For example, if a prompt contains only one sentence, the optimization means 126 determines that the send / receive log is of low value for analysis because it is likely that the amount of information contained in the prompt is small. For example, the optimization means 126 determines that the send / receive logs of users with a low average number of input tokens for prompts are of low value for analysis because it is likely that the amount of information contained in the prompts is small. For example, the optimization means 126 determines that the send / receive logs of users with low LLM utilization skills are of low value for analysis because it is unlikely that they will input a variety of prompts.

[0042] The optimization means 126 performs optimization processing of the prompt history 131 based on the behavior history 132. For example, the optimization means 126 performs optimization processing of the prompt history 131 based on the user's LLM utilization skill. For example, the information processing device 120 excludes the transmission and reception logs of users with low LLM utilization skill from the prompt history 131. For example, the information processing device 120 labels the prompts.

[0043] The weight reduction means 126 performs processing on the prompt history 131. The weight reduction means 126 labels the prompts in the prompt history 131.

[0044] For example, it can be inferred that a duplicate prompt from a single user is highly likely to be the result of the user inputting the same prompt into multiple LLMs and comparing the output results. Alternatively, since an LLM may produce different output results even for the same prompt, it can be inferred that a duplicate prompt from a single user is highly likely to be the result of the user inputting the same prompt multiple times to the same LLM until a satisfactory result is obtained. It is assumed that the user will copy and paste the satisfactory result for use in other processing. The weight reduction unit 126 can determine that the result has been copied and pasted based on the action history 132. In addition, the user may evaluate the satisfactory result as "good". Therefore, the weight reduction unit 126 can determine that duplicate prompts from a single user have a high degree of satisfaction corresponding to the copied and pasted output results or the output results evaluated as "good", and are highly valuable for analysis, and thus labels them as "1".

[0045] On the other hand, the weight reduction unit 126 determines that similar prompts are highly valuable for analysis, as they are highly likely to be input by the user through trial and error in order to obtain appropriate answers.

[0046] For example, the weight reduction unit 126 may label prompts with low analysis value as "0", label prompts that the user is satisfied with and have high analysis value (that is, to be analyzed as positive results) as "1", and label prompts that the user is not satisfied with but have high analysis value (that is, to be analyzed as negative results) as "-1". The number of labels assigned by the weight reduction unit 126 is not limited to three. The weight reduction unit 126 may perform labeling into two categories, regarding prompts other than those to be analyzed as positive results as having low analysis value. For example, the weight reduction unit 126 may perform labeling into four categories by labeling prompts to be analyzed in extremely high detail as "2".

[0047] The analysis unit 127 transmits a part of the prompt history, that is, the prompts that were not excluded even after the weight reduction by the weight reduction unit 126, to the LLM 140, and causes each LLM included in the LLM 140 to analyze (classify) its characteristics. The analysis unit 127 causes a predetermined LLM, for example, the LLM 1, to analyze the characteristics of each LLM. The analysis unit 127 may cause an LLM not included in the LLM 140 to analyze the characteristics of each LLM included in the LLM 140.

[0048] The information processing apparatus 120 creates the LLM characteristic table 133 based on the analyzed characteristics of each LLM.

[0049] The weight reduction unit 126 may extract transmission / reception logs with high analysis value and transmission / reception logs with low analysis value. For example, users with high LLM utilization skills and users with low LLM utilization skills may be extracted. The analysis unit 127 can perform analysis based on differences in LLM utilization skills. For example, the analysis unit 127 can analyze advice for users with low LLM utilization skills.

[0050] FIG. 6 shows an example of a flowchart executed by the information processing apparatus 120.

[0051] The information processing apparatus 120 acquires the action history 132 and the prompt history 131 (step S601). The action history 132 and the prompt history 131 may be stored in the storage device 130.

[0052] The information processing apparatus 120 reduces the weight of the prompt history 131 based on the acquired action history 132 and prompt history 131 (step S602). The information processing apparatus 120 reduces the weight of the prompt history 131 in order to analyze only logs with high analysis value.

[0053] FIG. 7 shows another example of a flowchart executed by the information processing apparatus 120.

[0054] The information processing device 120 acquires the activity history 132 (step S701). For example, step S701 is performed by the acquisition means 125. The activity history 132 includes the operations performed by each user up to the point of sending a prompt, for example, the operating state of the input / output terminal 110. Step S701 is performed by the acquisition means 125. The activity history 132 may be stored in the storage device 130.

[0055] The information processing device 120 determines each user's LLM utilization skill based on the acquired behavioral history (step S702). For example, step S702 is performed by the lightweight means 126. The information processing device 120 determines, for example, that users who use advanced functions, operate quickly, and make few key input errors have high LLM utilization skills, while users who use help, operate slowly, and make many key input errors have low LLM utilization skills.

[0056] The information processing device 120 acquires the prompt history 131 (step S703). For example, step S703 is performed by the acquisition means 125. The prompt history 131 is a transmission and reception log with each LLM contained in each user's LLM 140. The transmission and reception log includes prompts and the responses generated to those prompts. The prompt history 131 may be stored in the storage device 130.

[0057] The information processing device 120 excludes specific transmission and reception logs from the prompt history 131 based on the prompt history acquired in step S703 (step S704). For example, step S704 is performed by the lightweight means 126. The information processing device 120 extracts some prompt history from the acquired prompt history 131. For example, the information processing device 120 excludes transmission and reception logs with a small number of sentences constituting the prompt from the prompt history 131. For example, the information processing device 120 excludes transmission and reception logs with a number of sentences constituting the prompt below a threshold (for example, only one sentence) from the prompt history 131. For example, the information processing device 120 excludes transmission and reception logs of users with a low average number of tokens in their prompts from the prompt history 131.

[0058] The information processing device 120 excludes specific transmission and reception logs from the prompt history based on the LLM utilization skill determined in S702 (step S705). For example, step S705 is performed by the lightweight means 126. The information processing device 120 excludes transmission and reception logs of users with low LLM utilization skills, determined in step S702, from the prompt history 131.

[0059] The information processing device 120 labels the prompts (step S706). For example, step S706 is performed by the lightweight means 126. For example, the information processing device 120 labels each prompt included in the prompt history as a prompt of high value for analysis. For example, the information processing device 120 labels prompts that have been evaluated for the obtained results as prompts of high value for analysis. For example, the information processing device 120 may label prompts that the user has given a "good" rating to the obtained results with "1 (positive)" and prompts that the user has given a "bad" rating to the obtained results with "-1 (negative)". The information processing device 120 may label prompts of low value for analysis with "0".

[0060] When the same user enters duplicate prompts for different LLMs, it is highly likely that the user will continue to enter the same prompt for each LLM until a satisfactory result is obtained, and that the satisfactory result will be used for other processing. Therefore, it is presumed that the copy-and-paste (or "good") result is the satisfactory result. Accordingly, the information processing device 120 may label the prompt for which the result was copied and pasted (or "good") with a "1" in relation to the duplicate prompts of the same user as the prompt for which a satisfactory result was obtained.

[0061] On the other hand, for similar prompts from the same user, it is highly likely that the user has been experimenting and trying different inputs to obtain the appropriate answer. Prompts where the result has been copied and pasted (or given a "good" rating) are presumed to have resulted in a satisfactory outcome for the user. Therefore, prompts where the result has been copied and pasted (or given a "good" rating) may be labeled "1" as prompts that have yielded a satisfactory result.

[0062] The information processing device 120 optimizes the prompt history based on the prompt labels.

[0063] Figure 8A shows an example of a prompt history, and Figure 8B shows an example of a prompt history after it has been optimized by the optimization means 126.

[0064] Figure 8A shows the history of user X entering the same prompt (prompt A), xxxx, into LLM1 through LLM4. Figure 8A also shows an example where user X did not enter any ratings for LLM1 and LLM4, but entered a bad rating for LLM2 and a good rating for LLM3.

[0065] The lightweighting means 126 reduces the size because the same prompt is entered into different LLMs.

[0066] Figure 8B shows an example of how the lightweighting means 126 reduces the number of LLMs: LLM3, which received a good evaluation, is labeled "1"; LLM2, which received a bad evaluation, is labeled "-1"; LLM4, which was the last LLM to be entered, is labeled "1"; and LLM1, which was not evaluated by the user and was not the last LLM to be entered, is labeled "0".

[0067] Figure 9A shows another example of the prompt history, and Figure 9B shows another example of the prompt history after it has been lightened by the lightweighting means 126.

[0068] Figure 9A shows the history of user X entering prompt A, which is xxxx, and prompt B, which is xxxxxxx, into LLM2 of LLM1 to LLM4. User X did not enter any evaluation for either prompt.

[0069] Figure 9B shows an example of how the lightweighting means 126 has made the prompts xxxx and xxxxxxx similar, and xxxxxxx is the last prompt entered, so xxxxxxx is labeled "1" and lightweighted.

[0070] The optimization means 126 may optimize by deleting prompts other than those labeled "1", or by deleting prompts labeled "0".

[0071] The information processing device 120 analyzes the lightweight prompt history, i.e., a portion of the prompt history, obtained in steps S704 to S706 (step S707). For example, step S707 is performed by the analysis means 127. The information processing device 120 may also make suggestions for when to use each LLM. For example, the information processing device 120 may suggest to the user which LLM to use and how to write prompts. For example, the information processing device 120 may suggest, "When summarizing press releases, it is best to use LLM3," "When summarizing a large amount of comment text, using LLM2 is sufficient," or "When estimating sentiment, it is recommended to use LLM1."

[0072] The information processing device 120 may process the analysis results. The information processing device 120 may create an LLM characteristic table 133 based on the analysis results. The information processing device 120 may push notification of the analysis results to the user, or register them as help for the system 100.

[0073] The information processing device 120 may have an AI such as an LLM 140 analyze the relationship between the prompt and the LLM. The AI ​​used for the analysis may be an AI trained through supervised learning. The training data may be an LLM that yielded appropriate results for the target processing content and prompt description, or an LLM that yielded inappropriate results.

[0074] The user can select an LLM to prompt by referring to the LLM characteristics table 133. Alternatively, the information processing device 120 may select an LLM to prompt by referring to the LLM characteristics table 133.

[0075] As described above, this embodiment allows for the analysis of a lightweight prompt history, thereby reducing the cost and time required for analysis.

[0076] In particular, by deleting send and receive logs from the prompt history that are of low value for analysis, the system can be made lighter, thus reducing the cost and time of analysis without compromising its accuracy.

[0077] Since optimization is performed based on LLM utilization skills, it is possible to analyze differences in LLM utilization skills, or to analyze users with high LLM utilization skills.

[0078] Since optimization is performed based on the number of prompt sentences and the average number of tokens in the prompt, it is possible to analyze differences in the number of prompt sentences and the average number of prompt tokens, or to analyze users with a high number of prompt sentences and an average number of prompt tokens.

[0079] The AI ​​analyzes the streamlined prompt history, reducing the number of tokens required for analysis.

[0080] The system performs an analysis to select the appropriate LLM (Language Learning Module), and based on the analysis results, it can select the most suitable LLM according to the prompt.

[0081] In the embodiments described above, the notation for each component may be replaced with other notations such as "circuitry," "assembly," "device," "unit," or "module."

[0082] Furthermore, this disclosure can be implemented in software, hardware, or software in conjunction with hardware. Each functional block used in the description of the above embodiments may be implemented in part or in whole as an integrated circuit (LSI), and each process described in the above embodiments may be controlled in part or in whole by a single LSI or a combination of LSIs. An LSI may consist of individual chips, or it may consist of a single chip that includes some or all of the functional blocks. An LSI may have data inputs and outputs. Depending on the degree of integration, LSIs may also be referred to as ICs, system LSIs, super LSIs, or ultra LSIs.

[0083] The integrated circuit implementation method is not limited to LSIs; it may also be implemented using dedicated circuits, general-purpose processors, or dedicated processors. Furthermore, a Field Programmable Gate Array (FPGA) that can be programmed after LSI manufacturing, or a reconfigurable processor that allows for the reconfiguration of the connections and settings of circuit cells within the LSI, may also be used. This disclosure may be implemented as digital or analog processing.

[0084] Furthermore, if advancements in semiconductor technology or other derived technologies lead to the emergence of integrated circuit technologies that can replace LSIs, then naturally, it would be possible to use those technologies to integrate functional blocks. The application of biotechnology, for example, is a possibility.

[0085] This disclosure is useful for an information processing device that generates a relocation schedule for shared mobility services.

[0086] 100 System 110 Input / Output Terminal 120 Information Processing Device 121 Processor 122 Input Device 123 Output Device 124 Bus 125 Acquisition Means 126 Lightweighting Means 127 Analysis Means 130 Storage Device 131 Behavioral History 132 Prompt History 133 LLM Characteristics Table

Claims

1. An information processing device comprising: an acquisition means for acquiring each user's LLM prompt history and each user's behavior history; a optimization means for optimizing the acquired prompt history based on the behavior history; and an analysis means for analyzing the optimized prompt history and associating prompts with LLM.

2. The information processing apparatus according to claim 1, wherein the lightweighting means reduces the size of the data by deleting transmission and reception logs from the acquired prompt history that are of low value for analysis.

3. The information processing apparatus according to claim 1, wherein the weight reduction means determines LLM utilization skills based on the behavioral history and reduces weight based on the LLM utilization skills.

4. The information processing apparatus according to claim 1, wherein the optimization means further optimizes based on the number of prompt sentences included in the acquired prompt history and the average number of tokens in the prompts.

5. The information processing apparatus according to claim 1, wherein the analysis means causes the AI ​​to analyze the partial Brompt history.

6. The information processing apparatus according to claim 1, wherein the analysis means associates a prompt with an LLM in order to select an LLM to be used.

7. An information processing method for an information processing device, which acquires the prompt history of each user's LLM and the behavioral history of each user, compresses the acquired prompt history based on the behavioral history, and analyzes the compressed prompt history to associate prompts with LLM.

8. A program that causes a computer to acquire each user's LLM prompt history and each user's activity history, to optimize the acquired prompt history based on the activity history, and to analyze the optimized prompt history to associate prompts with LLM.