Device, method, and program

The apparatus addresses the challenges of high costs and bias in few-shot classification by prioritizing category shots based on priority, enhancing accuracy and reducing user burden.

WO2026234822A1PCT designated stage Publication Date: 2026-11-12NTT DOCOMO INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2025/017139
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

Existing few-shot classification methods face challenges with increased costs and longer execution times due to the increased number of categories, leading to potential bias in accuracy and a significant burden on users, especially when selecting categories to reduce the number of input tokens.

Method used

An apparatus that acquires priority information for categories, generates prompts with selected shots based on priority, and instructs an AI to prioritize high-priority categories, adjusting the number of shots to maintain accuracy and reduce costs.

Benefits of technology

This approach prioritizes accuracy for high-priority categories, reducing user burden while keeping costs down by optimizing the number of shots based on category priority and token limits.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2025017139_12112026_PF_FP_ABST
    Figure JP2025017139_12112026_PF_FP_ABST
Patent Text Reader

Abstract

A device according to one aspect of the present disclosure comprises: an acquisition unit that acquires priority information pertaining to the priority of a category; a generation unit that generates a prompt including a shot selected for each category on the basis of the priority information; and an instruction unit that instructs a generative AI by using the prompt.
Need to check novelty before this filing date? Find Prior Art

Description

Apparatus, method, and program

[0001] The present disclosure relates to an apparatus, a method, and a program. In particular, the present disclosure relates to multi-class classification of text using an LLM.

[0002] Text input by a user is classified according to predefined categories. Zero-Shot classification and Few-Shot classification are known as classification methods. Zero-Shot classification performs classification by inputting category information and text to be classified into a prompt. Few-Shot classification performs classification by inputting category information, several classification examples, and the text to be classified into a prompt. A classification example is called a Shot, and is composed of a text to be classified and the category of the classification result. Few-Shot classification is generally more accurate than Zero-Shot classification.

[0003] In Few-Shot learning for image learning, it has been proposed that the greater the number of images per episode, the higher the accuracy of the learning model, and to control parameters related to learning according to the content desired to be learned (see, for example, Patent Document 1).

[0004] Although accuracy varies depending on the difference in the number of learned samples, it has been proposed to select K classes from M classes as training images, and select N training images for each of the selected K classes (see, for example, Patent Document 2).

[0005] It has been proposed to perform weighted correction so that classification accuracy does not deteriorate even when there is a difference in the number of training data samples for each class (see, for example, Patent Document 3).

[0006] Japanese Patent No. 6995262, Japanese Patent Laid-Open No. 2023-538662, Japanese Patent Laid-Open No. 2023-011313

[0007] In few-shot classification, as the number of categories increases, the number of shots also increases, leading to increased costs due to the increased number of input tokens and longer execution times. Furthermore, with a large number of categories, depending on the token limit, it may not be possible to input shots for all categories, potentially leading to bias in the accuracy of outputting the correct answer for each category. Carefully selecting categories to reduce the number of input tokens places a significant burden on the user.

[0008] In few-shot classification, reducing the number of shots is effective in reducing the number of input tokens. To reduce the number of shots, it is necessary to carefully select the shots. However, reducing the number of shots may lead to undesirable bias in the output. Adjusting for undesirable bias in the output requires changing the prompt, but changing the prompt requires expertise and is time-consuming. In the first place, classification has a bias in accuracy for each category. The accuracy of a generative AI is determined, for example, by Top1 accuracy. Top1 accuracy is the probability that the category that the generative AI determined to have the highest probability was correct. The accuracy of a generative AI may also be determined by TopN accuracy or by other methods.

[0009] There is a need to reduce the burden on users while keeping costs down.

[0010] An apparatus in one aspect of this disclosure includes an acquisition unit that acquires priority information relating to the priority of categories, a generation unit that generates prompts including Shots selected for each category based on the priority information, and an instruction unit that instructs a generating AI using the prompts.

[0011] According to this disclosure, accuracy can be prioritized for high-priority classification categories, thereby reducing the burden on users while keeping costs down.

[0012] Diagram showing the overall system configuration. Functional block diagram of the information processing device. Example of a prompt generated by the information processing device. Flowchart executed by the information processing device. Diagram showing the number of samples for each category. Diagram showing the number of shots for each category when no priority is entered. Diagram showing an example of the number of shots for each category when the priority follows the category ratio of samples stored in the memory. Diagram showing an example of the number of shots for each category when a specific category has high priority. Diagram showing another example of the number of shots for each category when a specific category has high priority. Diagram showing an example of the hardware configuration of the information processing device.

[0013] The embodiments of this disclosure will be described in detail below with reference to the drawings as appropriate.

[0014] (Embodiment 1) Figure 1 shows the configuration of the entire system 100.

[0015] System 100 includes an input / output terminal 110, an information processing device 120, a storage device 130, and an LLM (Large Language Model) 140.

[0016] The input / output terminal 110 is a device that receives input information from the user (for example, instructions to generate variation sentences) and outputs a response to the input information (for example, generated variation sentences).

[0017] The information processing device 120 is a device that generates text to be input into the learning device based on input information entered by the user.

[0018] 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 Shot data.

[0019] LLM140 is a language model built using deep learning. It outputs in response to prompts input to LLM140. A prompt is an instruction given to an AI (Artificial Intelligence) such as LLM.

[0020] Figure 2 shows a functional block diagram of the information processing device 120.

[0021] The information processing device 120 includes an acquisition unit 121, a generation unit 122, and an instruction unit 123.

[0022] The acquisition unit 121 receives data input from the input / output terminal 110. The acquisition unit 121 also receives priority information regarding the priority of categories. The category information may be pre-entered, or it may be input from the acquisition unit 121 along with the category priority information.

[0023] The generation unit 122 determines the number of shots for each category based on the priority information input by the acquisition unit 121, and generates a prompt containing the determined number of shots for that category. The generation unit 122 selects the determined number of shots for each category from the shot data stored in the storage device 130. The generation unit 122 may also select shots based on metadata. The metadata includes the upper limit of input tokens for the generated AI, the total number of classification categories, the number of tokens for a shot, the minimum number of classification categories, the minimum number of shots, the category distribution, the required accuracy, or a combination thereof. The number of shots determined by the generation unit 122 is such that the number of shots for high-priority categories is greater than the number of shots for low-priority categories. In other words, the generation unit 122 selects more shots for high-priority categories than for low-priority categories. High-priority categories may be categories that require high accuracy, while low-priority categories may be categories that do not require high accuracy.

[0024] The instruction unit 123 uses the prompt generated by the generation unit 122 to give instructions to the generating AI such as the LLM 140.

[0025] Figure 3 shows an example of a prompt generated by the information processing device 120. The information processing device 120 can use the prompt in Figure 3 to instruct the LLM 140 to output a category (classification) for the text <User Input>.

[0026] The prompt includes a classification category 310 and an example answer (Shot) 320.

[0027] Classification category 310 is a category used to classify the text entered by the user. The generation unit 122 specifies the category to which the text should be classified as classification category 310.

[0028] Answer example 320 is called a Shot. In the case of Few-Shot, several examples of classification results for the input text (text to be classified) are provided. The generation unit 122 determines the number of Shots for each category according to the category priority and selects the determined number of Shots from the Shots stored in the storage device 130. The generation unit 122 records the selected Shots as answer example 320. Figure 3 shows an example of a prompt in which a Shot classifying the input text "xxxxxx" into category CC and a Shot classifying the input text "yyyyyy" into category AA are recorded. The total number of Shots is arbitrary.

[0029] Figure 4 shows a flowchart of the operations performed by the information processing device 120.

[0030] The information processing device 120 displays a list of categories to the user (step S401). The user inputs a priority for each category in the displayed list of categories. The user may input one of the priorities from 1-5, high, medium, low, etc., for each category, or may specify a high priority or low priority category.

[0031] The information processing device 120 determines whether or not priority information has been acquired (step S402). Step S402 is performed by the acquisition unit 121. If priority information has not been acquired (step S402, No), the process returns to step S402 and waits for priority information to be acquired. If priority information has not been acquired even after a predetermined time has elapsed, the information processing device 120 may assume that all categories have the same priority (i.e., that the same priority has been entered for all categories). The information processing device 120 may also determine that the higher the frequency of a category, the higher the priority of that category, in accordance with the distribution of categories (frequency of being classified into each category).

[0032] When the information processing device 120 obtains priority information (step S402, Yes), it determines whether or not the text to be classified has been entered (step S403). Step S403 is performed by the acquisition unit 121. If the text to be classified has not been entered (step S403, No), it returns to step S403 and waits for the text to be classified to be entered. The information processing device 120 may also execute step S402 after step S403.

[0033] When the information processing device 120 receives text to be classified (step S403, Yes), it determines the number of shots for each category (step S404). Step S404 is performed by the generation unit 122. The information processing device 120 may determine the number of shots for each category based on the total number of shots and the priority of each category. Multiple categories may be assigned priorities, or multiple priorities may exist. The information processing device 120 may increase the number of shots for high-priority categories from the base (default) number of shots. The information processing device 120 may decrease the number of shots for low-priority categories. The number of shots increased or decreased may correspond to the priority level. For example, the information processing device 120 may increase the number of shots by 2 for a priority 1 category, increase the number of shots by 1 for a priority 2 category, decrease the number of shots by 1 for a priority 4 category, and decrease the number of shots by 2 for a priority 5 category. The information processing device 120 may determine the minimum number of shots for each category based on the required accuracy for each category.

[0034] The information processing device 120 may display the determined number of shots to the user for confirmation. The user may reset the priority or change the number of shots. Details of the method for determining the number of shots will be described later.

[0035] The information processing device 120 acquires the number of Shots for each category determined in step S404 (step S405). Step S405 is performed by the generation unit 122. The information processing device 120 may also determine a sample of Shots from the Shot data stored in the storage device 130. The information processing device 120 generates a prompt using the determined sample as a Shot. For example, if the generation unit 122 determines that the number of Shots for category CC is 3, the generation unit 122 selects 3 Shots whose classification result is category CC. The information processing device 120 may display the acquired Shots to the user for confirmation. The user may reacquire some or all of the Shots. The information processing device 120 may perform step S403 after step S404 or step S405.

[0036] The information processing device 120 generates a prompt using the Shot acquired in step S405 (step S406). For example, the information processing device 120 generates a prompt that includes a category list, the text to be classified entered in step S403, and the Shot acquired in step S405. Step S406 is performed by the generation unit 122. The prompt includes a classification category 310 and an example answer (Shot) 320.

[0037] The information processing device 120 instructs the LLM 140 using the generated prompt (step S407). The information processing device 120 may also instruct a generating AI other than the LLM 140 using the generated prompt.

[0038] The information processing device 120 determines whether or not it has obtained a classification result from the LLM 140 (step S408). If it has not obtained a classification result (step S408, No), it returns to step S408 and waits for the text to be classified to be entered.

[0039] When the information processing device 120 acquires a classification result (step S408, Yes), it outputs the acquired classification result to the user (step S409).

[0040] <Determination of Shot Number> The information processing apparatus 120 determines the number of Shots for each category such that the tokens required for the prompt including the text to be classified do not exceed the upper limit of input tokens of the LLM 140. The information processing apparatus 120 determines Shots from the Shot data stored in the storage device 130 by the number of Shots for each category. The information processing apparatus 120 selects Shots from each Shot of the Shot data stored in the storage device 130 by the number of Shots. The information processing apparatus 120 generates a prompt including the determined Shots.

[0041] A method for determining the number of Shots determined by the information processing apparatus 120 will be described using a case example.

[0042] As a case example, description will be given of a scenario where there are four types of categories A, B, C, and D, and the storage device 130 stores 5 Shots for category A, 10 Shots for category B, 20 Shots for category C, and 15 Shots for category D respectively.

[0043] Figure 5 shows the Shot data stored in the storage device 130. The Shot data includes a plurality of samples, and each sample is classified into one of the categories. For example, Figure 5 shows Shot data with 5 samples classified into category A, 10 samples classified into category B, 20 samples classified into category C, and 15 samples classified into category D.

[0044] (Case 1: When priority information is not input) Figure 6 is a diagram showing Shots for each category when priority information is not input.

[0045] The information processing apparatus 120 generates a prompt that includes the same number of Shots for each category. For example, the prompt includes 2 Shots of category A, 2 Shots of category B, 2 Shots of category C, and 2 Shots of category D. Accordingly, the information processing apparatus 120 determines 2 Shots from 5 Shots of category A, determines 2 Shots from 20 Shots of category B, determines 2 Shots from 15 Shots of category C, determines 2 Shots from 10 Shots of category D, and generates a prompt including the determined 8 Shots.

[0046] (Example 2: Instruction Following Distribution) FIG. 7 is a diagram illustrating an example of the number of Shots for each category when the priority follows the category ratio of Shot data accumulated in the storage apparatus 130. In this example, the information processing apparatus 120 determines that the ratio of the number of Shots of category A : category B : category C : category D is 1:2:4:3. For example, the prompt includes 1 Shot of category A, 2 Shots of category B, 4 Shots of category C, and 3 Shots of category D.

[0047] (Example 3: Selecting a Specific Category) FIG. 8 is a diagram illustrating an example of the number of Shots for each category when a specific category, for example, category B, has high priority. For example, the information processing apparatus 120 increases the number of Shots of category B by 2 in a state where the same number of Shots is included for each category (Example 1). The prompt includes 2 Shots of category A, 4 Shots of category B, 2 Shots of category C, and 2 Shots of category D.

[0048] The information processing apparatus 120 makes a decision to increase the number of Shots of the selected category within a range that does not exceed the upper limit number of tokens even when the Shots of the selected category are increased.

[0049] For example, considering the upper limit on the number of tokens, let's consider the case where the upper limit on the number of Shots to include in a prompt is 10. Increasing the number of Shots by 3 will exceed the upper limit on the number of tokens, but increasing the number of Shots by 2 will not exceed the upper limit on the number of tokens, so the Shots of category B, which have a high priority, will be increased by 2. The upper limit on the number of tokens may be a predetermined number of tokens, or it may be the upper limit on the number of input tokens.

[0050] (Case 4: Selecting a specific category) Figure 9 shows another example of the number of shots for each category when a specific category, for example, Category B, is given high priority. Figure 9 shows an example of shots for each category when the maximum number of tokens must be considered.

[0051] If the number of Shots for a selected category cannot be increased due to the maximum number of tokens, the number of Shots for other categories will be decreased. If the number of Shots for a selected category cannot be increased due to the maximum number of tokens, or if Shots for all categories cannot be included in the prompt, some categories will not have Shots selected (i.e., the number of Shots will be set to 0). For example, categories with a small number of accumulated samples will have their number of Shots set to 0.

[0052] Figure 9 shows the case where, according to the category ratio of the number of stored samples in the memory device 130, that is, the ratio of the number of shots for category A: category B: category C: category D is 1:2:4:3 (Case 2), an input is made to further increase the priority (increase the precision) of category B. Figure 9 shows the case where the number of shots cannot be increased. That is, the number of shots for high-priority category B is increased by 3, while the number of shots for categories A, C, and D is decreased by 1 each, and the number of shots for category A, which has the fewest stored shots, is set to 0. In other words, the prompt includes 5 shots for category B, 3 shots for category C, and 2 shots for category D.

[0053] (Variations) Example 3 or Example 4 can be combined with either Example 1 or Example 2. In Example 2, the number of Shots in the selected category may be increased, provided that the number of tokens does not exceed the upper limit, and in Example 1, the number of Shots in other categories may be decreased.

[0054] Furthermore, examples 3 and 4 may be combined. For example, if the maximum number of tokens is exceeded, the increase in Shots for the category being increased may be limited, while Shots for other categories may be reduced.

[0055] Multiple shot counts may be displayed to the user, allowing them to select one. For example, the number of shots based on Example 3 and Example 4 may be displayed to the user, allowing them to select the desired number of shots.

[0056] The information processing device 120 may change the order of Shots based on priority and write them in the prompt. For example, the instruction unit 123 may instruct the generating AI using a prompt in which Shots are written in order of the highest priority category.

[0057] According to this disclosure, accuracy can be prioritized for high-priority classification categories, thereby reducing the burden on users while keeping costs down.

[0058] Furthermore, according to this disclosure, increasing the number of shots is expected to improve accuracy, thus improving the accuracy of high-priority categories.

[0059] Furthermore, according to this disclosure, since Shot is selected considering metadata, the cost or accuracy can be set within the desired range.

[0060] Furthermore, according to this disclosure, since Shots are listed in order of priority, Shots with higher priority can be reliably included in the prompt, regardless of the upper limit of the input tokens.

[0061] <Example Hardware Configuration> The block diagram used in the description of the above embodiment shows functional units. These functional blocks (components) are realized by hardware, software, or a combination thereof. Furthermore, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one device that is physically or logically coupled, or it may be realized using two or more physically or logically separated devices that are directly or indirectly connected (for example, using wired, wireless, etc.). A functional block may also be realized by combining the above one device or the above multiple devices with software.

[0062] For example, the information processing device 120 in one embodiment of the present disclosure may function as a computer that performs the processing of the present disclosure. Figure 10 is a diagram showing an example of the hardware configuration of the information processing device 120 according to one embodiment of the present disclosure. The above-described information processing device 120 may be physically configured as a computer device including a processor 1001, memory 1002, storage 1003, communication device 1004, input device 1005, output device 1006, bus 1007, etc.

[0063] In this disclosure, any two terms selected from the set of terms such as circuit, device, unit, module, chip, and means may be used interchangeably. The hardware configuration of the information processing apparatus 120 may include one or more of the devices shown in the figure, or it may be configured to omit some of the devices.

[0064] Each function in the information processing device 120 is realized by loading predetermined software (programs) onto hardware such as the processor 1001 and memory 1002, which allows the processor 1001 to perform calculations, control communication by the communication device 1004, and control the reading, writing, or both of the reading and writing of data in the memory 1002 and storage 1003.

[0065] The processor 1001 controls the entire computer, for example, by running an operating system. The processor 1001 may be composed of a central processing unit (CPU) that includes interfaces with peripheral devices, control devices, arithmetic units, registers, etc. Only one processor 1001 is shown in the diagram, but there may be multiple processors.

[0066] Furthermore, the processor 1001 reads programs (program code), software modules, data, etc., from the storage 1003, the communication device 1004, or both the storage 1003 and the communication device 1004 into the memory 1002, and executes various processes accordingly. The program used is one that causes the computer to execute at least a part of the operations described in the above embodiment. For example, the control unit of the information processing device 120 may be implemented by a control program stored in the memory 1002 and running on the processor 1001, and other functional blocks may be implemented similarly. The above-mentioned various processes have been described as being executed by one processor 1001, but they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may be transmitted from a network via a telecommunications line, or provided to the computer device via the communication device 1004, for example.

[0067] Furthermore, this disclosure also provides a computer program product, which includes a computer program. This computer program product may implement the steps of the method described in the above embodiments when the computer program is executed by a computer (for example, a processor 1001).

[0068] The memory 1002 is a computer-readable recording medium and may be composed of, for example, ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), or at least a combination of these. The memory 1002 may also be called a register, cache, main memory, etc. The memory 1002 can store executable programs (program code), software modules, etc., for carrying out a wireless communication method according to one embodiment of the present disclosure.

[0069] The storage 1003 is a computer-readable recording medium and may consist of, for example, an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., compact disk, digital multipurpose disk, Blu-ray® disk), a smart card, flash memory (e.g., card, stick, key drive), a floppy® disk, a magnetic strip, or at least a combination of these. The storage 1003 may also be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, server, or other suitable medium including memory 1002, storage 1003, or both memory 1002 and storage 1003.

[0070] The communication device 1004 is hardware (transceiver / receiver device) for communicating between computers via a wired network, a wireless network, or both wired and wireless networks, and is also referred to as a network device, network controller, network card, or communication module, for example. The communication device 1004 may be configured to include a high-frequency switch, duplexer, filter, frequency synthesizer, or at least a combination of these.

[0071] The input device 1005 is an input device that accepts input from an external source (e.g., a keyboard, mouse, microphone, switch, button, sensor, or a combination of at least two of these). The output device 1006 is an output device that outputs to an external source (e.g., a display, speaker, LED lamp, or a combination of at least two of these). The input device 1005 and the output device 1006 may be configured as an integrated unit (e.g., a touch panel).

[0072] Furthermore, each device, such as the processor 1001 and memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or different buses may be configured for each device.

[0073] 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), an FPGA (Field Programmable Gate Array), a GPU (Graphics Processing Unit), an NPU (Neural Processing Unit), or a combination of at least two of these, and some or all of each functional block may be realized by such hardware. For example, the processor 1001 may be implemented using at least one of these hardware components.

[0074] The processing procedures, sequences, flowcharts, etc., of each aspect / embodiment described in this disclosure may be reordered, provided they are consistent. For example, the methods described in this disclosure present various step elements using exemplary order and are not limited to the specific order presented.

[0075] Each aspect / embodiment described in this disclosure may be used individually, in combination, or switched between as needed during execution.

[0076] The above description of the disclosure is illustrative, and the present invention is not limited to the embodiments described herein. The disclosure can be implemented as modified and altered forms without departing from the spirit of the invention. The disclosure and its modifications and altered forms are included within the scope of the present invention and its equivalents.

[0077] Any reference to elements using the designations “first,” “second,” etc., as used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Accordingly, references to the first and second elements do not imply that only two elements may be employed, or that the first element must precede the second element in any way.

[0078] In this disclosure, if articles are added through translation, such as a, an, and the in English, this disclosure may include the fact that the noun following these articles is plural.

[0079] In the configuration of each of the above devices, "means" may be replaced with "part," "circuit," "device," etc.

[0080] In the embodiments described above, an HDD or SSD was exemplified as the storage device 130, but it may also be composed of ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), etc., or at least a combination of these. The storage device 130 may be a flexible disk, magneto-optical disk (e.g., compact disk, digital multipurpose disk, Blu-ray® disk), smart card, flash memory device (e.g., card, stick, key drive), CD-ROM (Compact Disc-ROM), register, removable disk, hard disk, floppy® disk, magnetic strip, database, server, or other suitable storage medium.

[0081] In the embodiments described above, the information, signals, etc., may be represented using any of the various different techniques. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc., which may be referred to throughout the above description, may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0082] In the embodiments described above, the input and output information may be stored in a specific location (e.g., memory) or managed using a management table. The input and output information may be overwritten, updated, or appended to. The output information may be deleted. The input information may be transmitted to other devices.

[0083] In the embodiments described above, the determination may be made by a value represented by one bit (0 or 1), by a boolean value (true or false), or by a numerical comparison (for example, a comparison with a predetermined value).

[0084] The programs illustrated in the embodiments described above should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, etc., whether they are called software, firmware, middleware, microcode, hardware description languages, or by any other name.

[0085] Furthermore, software, instructions, information, etc., may be transmitted and received via a transmission medium. For example, if software is transmitted from a website, server, or other remote source using at least one of wired technology (such as coaxial cable, fiber optic cable, twisted pair, or digital subscriber line (DSL)) and wireless technology (such as infrared or microwave), then at least one of these wired and wireless technologies is included in the definition of a transmission medium.

[0086] In the embodiments described above, the terms "system" and "network" are used interchangeably.

[0087] The information, parameters, etc., described in this disclosure may be expressed using absolute values, relative values ​​from a predetermined value, or corresponding other information.

[0088] In the embodiments described above, the mobile terminal may be a mobile station (MS). A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or several other appropriate terms. In this disclosure, terms such as “mobile station,” “user terminal,” “user equipment (UE),” and “terminal” may be used interchangeably.

[0089] In the embodiments described above, the terms “connected,” “coupled,” or any variation thereof, mean any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” with each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, “connection” may be reinterpreted as “access.” As used in this disclosure, two elements may be considered to be “connected” or “coupled” with each other using at least one of one or more wires, cables, and printed electrical connections, and, in some non-limiting and non-exclusive examples, electromagnetic energy having wavelengths in the radio frequency domain, microwave domain, and optical (both visible and invisible) domain.

[0090] In the embodiments described above, the phrase "based on" does not mean "based solely on" unless otherwise specified. In other words, the phrase "based on" means both "based solely on" and "based at least on."

[0091] As used in this disclosure, the terms “determining” and “determining” may encompass a wide variety of actions. “Determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, or inquiring (e.g., searching in a table, database, or other data structure), or ascertaining. “Determining” may also include receiving (e.g., receiving information), transmitting (e.g., sending information), inputting, outputting, or accessing (e.g., accessing data in memory). Furthermore, "judgment" and "decision" can include considering something as having been "judged" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. In other words, "judgment" and "decision" can include considering something as having been "judged" or "decided" on some action. Also, "judgment (decision)" can be reinterpreted as "assuming," "expecting," "considering," etc.

[0092] In the embodiments described above, where “include,” “including,” and variations thereof are used, these terms are intended to be inclusive, as is the term “comprising.” Furthermore, the term “or” as used in this disclosure is not intended to be exclusive OR.

[0093] In this disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "combine" may be interpreted similarly to "different."

[0094] Each aspect / embodiment described in this disclosure may be used individually, in combination, or switched between as needed during implementation. Furthermore, notification of specific information (e.g., notification that "X is") is not limited to explicit notification, but may also be implicit (e.g., by not providing such notification).

[0095] This disclosure is useful for an information processing device for multi-class classification of text using LLM.

[0096] 100 System 110 Input / Output Terminal 120 Information Processing Device 121 Acquisition Unit 122 Generation Unit 123 Instruction Unit 130 Storage Device 140 LLM 310 Classification Category 320 Example Answer 1001 Processor 1002 Memory 1003 Storage 1004 Communication Device 1005 Input Device 1006 Output Device 1007 Bus

Claims

1. An apparatus comprising: an acquisition unit that acquires priority information regarding the priority of categories; a generation unit that generates prompts including Shots selected for each category based on the priority information; and an instruction unit that instructs a generation AI using the prompts.

2. The apparatus according to claim 1, wherein the generating unit selects the number of shots for each category based on the priority information.

3. The apparatus according to claim 2, wherein the generating unit selects more Shots of the high-priority category than Shots of the low-priority category.

4. The apparatus according to claim 1, wherein the generating unit selects the Shot based on information regarding metadata.

5. The apparatus according to claim 4, wherein the metadata includes an upper limit on the input tokens, the total number of classification categories, the number of tokens in the Shot, the minimum number of classification categories, the minimum number of Shots, the category distribution, the required precision, or a combination thereof.

6. The apparatus according to claim 1, wherein the instruction unit instructs the generating AI using prompts that list Shots in order of priority category.

7. A computer method comprising: obtaining priority information regarding the priority of categories; generating prompts containing selected Shots for each category based on the priority information; and using the prompts to instruct a generating AI.

8. A program that causes a computer to obtain priority information regarding the priority of categories, to generate prompts containing selected Shots for each category based on the priority information, and to instruct a generating AI using the prompts.