Division device, method for division, and division program

Abstract

To enable appropriate division of tasks.SOLUTION: A division device 100 acquires a task which is information input to a generation model to cause the generation model to generate predetermined information, and rules concerning execution of the task. In a case where it is determined that the acquired task is complex, the division device 100 inputs the task and the rules concerning execution of the task to a division search-augmented generation model trained to divide a task into one or more subtasks, thereby dividing the task into one or more subtasks. The division device 100 outputs the divided subtasks.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a division device, a division method, and a division program. [Background technology]

[0002] In some cases, the processing instructions and other information (hereinafter simply referred to as "tasks") input to large-scale language models such as generative AI (artificial intelligence) to generate predetermined information are ambiguous or complex. Specifically, there are cases where the processing steps for a task are unclear or the task consists of many steps.

[0003] Conventionally, there are known techniques for using large-scale language models to generate solutions for ambiguous or complex tasks, but generative AI trained with general knowledge may not be able to generate accurate solutions for unknown tasks. Therefore, in order to enable generative AI to generate accurate answers, Chain of Thought (CoT) prompts are known as a method for leading to an answer by layering step-by-step inferences (thought chains) through intermediate inference steps (see, for example, Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Chain-of-Thought Prompting Elicits Reasoning in Large Language Models,Jason Wei et al,<URL:https: / / arxiv.org / pdf / 2201.11903.pdf> ,<Searched on March 13, 2021> Summary of the Invention [Problem to be solved by the invention]

[0005] However, the above-mentioned conventional technology has problems in appropriately dividing tasks. For example, the conventional technology has a generation AI generate an answer through intermediate inference, but if the task cannot be appropriately divided because the intermediate inference method for the problem is presented in advance, it may be difficult to generate a highly accurate answer. [Means for solving the problem]

[0006] Therefore, in order to solve the above-mentioned problems and achieve the objective, the segmentation device of the present invention is characterized by having an acquisition unit that acquires a task and specifications regarding the execution of the task, which are information input to a generative model to generate specified information; a segmentation unit that inputs the task and specifications regarding the execution of the task to a search extension generative model for segmentation that has been trained to split the task into one or more subtasks when the task acquired by the acquisition unit is determined to be complex, and segments the task into one or more subtasks; and an output unit that outputs the subtasks split by the segmentation unit. [Effects of the Invention]

[0007] The present invention has the effect of enabling appropriate division of tasks. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating an overall image of processing by a dividing device according to this embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the configuration of the dividing device according to this embodiment. [Figure 3] FIG. 3 is a table diagram showing an example of task information according to this embodiment. [Figure 4] FIG. 4 is a table diagram showing an example of subtask information according to this embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of processing according to this embodiment. [Figure 6]FIG. 6 is a flowchart showing an example of the procedure of the division process according to this embodiment. [Figure 7] FIG. 7 is a hardware configuration diagram showing an example of a computer that realizes the functions of the dividing device according to this embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, an embodiment (hereinafter referred to as "embodiment") will be described with reference to the drawings. In the following description, common components are given the same reference numerals, and repeated description will be omitted. Furthermore, the description of the embodiment described below does not limit the dividing device, dividing method, and dividing program according to the present invention.

[0010] <Introduction> First, an introduction to this embodiment will be provided. Fig. 1 is a diagram illustrating an overview of processing by a division device 100 according to this embodiment. The division device 100 shown in Fig. 1 is an example of a computer that provides a technology for determining whether a task, which is information input to a generative model or the like to generate predetermined information, is complex, and for dividing a task determined to be complex into one or more subtasks.

[0011] (background) When inputting tasks to large-scale language models such as generative AI, the processing steps for the task may be unclear or the task may consist of many steps. To enable large-scale language models to accurately generate solutions for such ambiguous or complex tasks, a reference technique called "chain of thought (CoT) prompt" is known, which leads to an answer by layering step-by-step inferences (chain of thought) through intermediate inference steps.

[0012] However, since the above-mentioned reference technology is realized by presenting a method of intermediate inference for a problem in advance, it may be difficult to have a generative AI or the like generate an accurate answer if the task cannot be appropriately divided.

[0013] (Overall Overview of Processing by Splitting Device 100) In order to solve the above-mentioned problems, the dividing device 100 according to this embodiment divides an input task into subtasks when the task is complex. Here, a series of processing steps of the dividing device 100 will be explained using FIG.

[0014] As shown in (1) of Figure 1, the dividing device 100 acquires tasks and task execution rules, which are information input to a generative model (hereinafter sometimes simply referred to as "generative AI") that produces a predetermined output according to the input information, to generate predetermined information. In this embodiment, "task execution rules" refer to rules that define individual processing procedures, etc., for accurately executing input tasks, and hereinafter may be simply referred to as "guidelines."

[0015] As shown in (2) of Figure 1, when the acquired task is determined to be complex, the division device 100 inputs the acquired task and guidelines into a search expansion generative model for division that has been trained to divide the task into one or more subtasks, and divides the task into one or more subtasks. In this embodiment, a "subtask" refers to a processing step (procedure) of a predetermined granularity that is included in the task to execute the task, and may hereinafter be simply referred to as a "subtask."

[0016] As shown in (3) of FIG. 1, the dividing device 100 outputs the divided subtasks to a terminal device operated by a user, another information processing device, or the like.

[0017] As described above, the dividing device 100 according to this embodiment determines whether an input task is complex, and divides the task into subtasks if the task is determined to be complex. Therefore, the dividing device 100 has the effect of enabling appropriate task division according to the input task.

[0018] <Description of the dividing device 100> Next, a detailed description will be given of the functions of the dividing device 100 according to this embodiment. Fig. 2 is a diagram showing an example of the configuration of the dividing device 100 according to this embodiment.

[0019] (splitting device 100) 2, the splitting device 100 has a communication unit 110, a storage unit 120, and a control unit 130. The splitting device 100 also has an input unit (not shown) such as a keyboard or a touch panel for receiving input from a user or the like, and a display unit (not shown) such as a display or a printer for displaying the results of information processing by the splitting device 100 to a user or the like.

[0020] (Communication unit 110) The communication unit 110 outputs information such as divided subtasks and prompting questions for the user to divide the target task into subtasks, and performs communication related to input of information such as tasks, guidelines, and subtasks divided by the user. The communication unit 110 is realized by a NIC (Network Interface Card) or the like. The communication unit 110 is connected to a network via wire or wirelessly as necessary, and can transmit and receive information bidirectionally.

[0021] (Storage unit 120) The storage unit 120 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 120 stores data and programs used for various processes by the control unit 130. As shown in FIG. 2 , the storage unit 120 has a task information DB 121, a subtask information DB 122, and a learning model DB 123.

[0022] (Task information DB121) The task information DB 121 is a database that stores input tasks and guidelines as task information. An example of task information stored in the task information DB 121 will now be described with reference to Fig. 3. Fig. 3 is a table diagram showing an example of task information according to this embodiment.

[0023] As shown in FIG. 3, the task information DB 121 stores "No.", which is information for identifying an individual task, "Task," which is information related to the input task, and "Guideline," which is information related to the guideline related to the task.

[0024] For example, the task information DB 121 stores the task "cause investigation" and the guideline "cause identification → countermeasure confirmation → impact assessment → action plan" in association with No. "1." The task information associated with No. "1" above means that when the task "cause investigation" is executed, the guideline "cause identification → countermeasure confirmation → impact assessment → action plan" is applied.

[0025] Furthermore, for example, the task information DB 121 stores the task "formulation of a countermeasure" and the guideline "identification of the cause → confirmation of the countermeasure → impact assessment → countermeasure plan" in association with No. "2." The task information associated with No. "2" above means that when the task "formulation of a countermeasure" is executed, the guideline "identification of the cause → confirmation of the countermeasure → impact assessment → countermeasure plan" is applied.

[0026] Furthermore, for example, the task information DB 121 stores the task "cause investigation" and the guideline "loss of reaction control" in association with No. "3." The task information associated with No. "3" above means that when the task "cause investigation" is executed, the guideline "loss of reaction control" is applied.

[0027] Furthermore, for example, the task information DB 121 stores the task "cause investigation" and the guideline "loss of control of reaction, machine failure, external heat source" in association with No. "4." The task information associated with No. "4" above means that when the task "cause investigation" is executed, the guideline "loss of control of reaction, machine failure, external heat source" is applied.

[0028] (Subtask information DB122) The subtask information DB 122 is a database that stores information (subtask information) related to subtasks divided by the division unit 134, which will be described later. An example of the subtask information stored in the subtask information DB 122 will now be described with reference to Fig. 4. Fig. 4 is a table diagram showing an example of subtask information according to this embodiment.

[0029] 4, the subtask information DB 122 stores "No.", which is information that identifies an individual task, "Task.", which is information about the input task, and "Subtask.", which is information about the subtasks divided from the task. In this embodiment, the subtask information DB 122 can store multiple "Subtask" items, such as "Subtask 1, 2, 3, 4, . . . n," as shown in FIG.

[0030] For example, the subtask information DB 122 stores the task "Cause Investigation" and subtask 1 "Cause Identification" in association with No. 1. The subtask information associated with No. 1 above means that the task "Cause Investigation" has been divided into one subtask, "Cause Identification," based on the guideline "Cause Identification → Confirmation of Countermeasures → Impact Assessment → Action Plan."

[0031] Also, for example, subtask information DB122 stores, in association with No. "2," task "formulation of countermeasures," subtask 1 "identification of cause," subtask 2 "confirmation of countermeasures," subtask 3 "impact assessment," and subtask 4 "countermeasure plan." The subtask information associated with No. "2" above means that the task "formulation of countermeasures" has been divided into four subtasks: "identification of cause," "confirmation of countermeasures," "impact assessment," and "countermeasure plan," based on the guideline "identification of cause → confirmation of countermeasures → impact assessment → countermeasure plan."

[0032] For example, the subtask information DB 122 stores the task "Investigate Cause" and the subtask 1 "Loss of Control of Reaction" in association with No. 3. The subtask information associated with No. 1 above indicates that the task "Investigate Cause" has been divided into one subtask for investigating the cause of "Loss of Control of Reaction" based on the guideline "Loss of Control of Reaction."

[0033] Also, for example, subtask information DB122 stores task "cause investigation," subtask 1 "loss of control of reaction," subtask 2 "machine failure," and subtask 3 "external heat source" in association with No. "4." The subtask information associated with No. "4" above means that, based on the guideline "loss of control of reaction, machine failure, external heat source," the task "cause investigation" has been divided into three subtasks: investigating the cause of "loss of control of reaction," investigating the cause of "machine failure," and investigating the cause of "external heat source."

[0034] (Learning model DB123) The learning model DB 123 can store a generation model that generates an output result according to a predetermined input such as a large-scale language model.

[0035] For example, the learning model DB 123 stores a search expansion generative model for judgment that has been trained to output the number of individual subtasks for the execution of a task input to the generative model. The learning model DB 123 also stores a search expansion generative model for judgment that has been trained to output the number of individual subtasks for the execution of a task input to the generative model.

[0036] The above-mentioned "Retrieval Augmented Generation (RAG) model" is a generative model that supplements text generation with information from private or proprietary data sources, and may be realized using known techniques. Note that "retrieval augmented generation" may also be read as "retrieval augmented generation."

[0037] (control unit 130) The control unit 130 is realized by a processor, a micro processing unit (MPU), a central processing unit (CPU), or the like executing various programs stored in the storage unit 120 using RAM as a work area. The control unit 130 is also realized by an integrated circuit (IC) such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA). As shown in FIG. 2 , the control unit 130 includes an acquisition unit 131, a determination unit 132, an identification unit 133, a division unit 134, and an output unit 135.

[0038] (Acquisition part 131) The acquisition unit 131 acquires tasks and guidelines. For example, the acquisition unit 131 acquires, as tasks, a risk assessment task based on inspection results in a factory or the like, a task for determining the cause of an abnormality when it occurs, a task for formulating a countermeasure, etc. Furthermore, as guidelines, the acquisition unit 131 acquires, for example, a guideline that specifies that in a task for dealing with an abnormality, processing should be performed in the order of "identifying the cause → checking the countermeasure → assessing the impact → planning the countermeasure."

[0039] Furthermore, the acquisition unit 131 acquires information relating to a subtask input by a user in accordance with the subtask divided by the division unit 134 described below and output by the output unit 135. For example, when a subtask is output to a user by the output unit 135 described below, the acquisition unit 131 acquires information such as additional information and additional instructions relating to the subtask input by the user (hereinafter, this may be simply referred to as "additional information").

[0040] Furthermore, when the user's level of understanding of a task input by the later-described identification unit 133 is equal to or greater than a reference value and the user has divided the task into subtasks, the acquisition unit 131 acquires the subtasks divided by the user in response to a predetermined request for dividing the task output by the later-described output unit 135. For example, when a task "cause investigation" is identified as being complex and as being understood by the user, and the task is divided by the user into four subtasks, "cause identification," "countermeasure confirmation," "impact assessment," and "action plan," the acquisition unit 131 acquires the four subtasks.

[0041] (Judgment unit 132) The determination unit 132 determines whether the input task is complex based on the number of subtasks included in the input task and the number of pieces of information used to execute the task.

[0042] Specifically, the determination unit 132 inputs the task and guidelines into a search expansion generation model for determination that has been trained to output the number of individual subtasks for executing the task input to the generation AI, and calculates the number of individual subtasks for executing the task.The determination unit 132 then determines that the task is complex if the calculated number of individual subtasks for executing the task is two or more.

[0043] For example, if the task stored in the task information DB 121 is "cause investigation," the determination unit 132 determines that a subtask related to the task corresponds to "cause identification" based on the guideline "cause identification → countermeasure confirmation → impact assessment → response plan." Next, the determination unit 132 calculates that the task "cause investigation" includes one subtask. Therefore, the determination unit 132 determines that the task "cause investigation" is "not complicated."

[0044] On the other hand, if the task stored in the task information DB 121 is "formulation of a countermeasure," the determination unit 132 determines that the subtasks related to the task correspond to "identification of the cause," "confirmation of the countermeasure," "impact assessment," and "countermeasure plan" based on the guideline "identification of the cause → confirmation of the countermeasure → impact assessment → countermeasure plan." Next, the determination unit 132 calculates that the task "formulation of a countermeasure" includes four subtasks. Therefore, the determination unit 132 determines that the task "formulation of a countermeasure" is "complex."

[0045] Furthermore, the determination unit 132 inputs the task and the rules regarding task execution into the search expansion generation model for determination, which has been trained to output the number of information groups of a predetermined granularity related to the task input to the generation AI, and calculates the number of information groups of a predetermined granularity.The determination unit 132 then determines that the task is complex if the calculated number of information groups of a predetermined granularity is two or more.

[0046] For example, the determination unit 132 inputs the task and the guideline into the search expansion generation model for determination, and then calculates the number of information groups of a predetermined granularity to be used to generate information such as countermeasures based on the task in addition to the general knowledge possessed by the generation AI.

[0047] Here, the determination unit 132 determines that the input task is "not complex" if the calculated number of information groups of the predetermined granularity is less than 2. On the other hand, the determination unit 132 determines that the input task is "complex" if the calculated number of information groups of the predetermined granularity is 2 or more.

[0048] (Specific Section 133) The specifying unit 133 specifies the user's level of understanding of the task determined by the determining unit 132 to be complex.

[0049] For example, the identification unit 133 outputs information such as the name and content of a task input to the user via the communication unit 110 or the display unit. Next, the identification unit 133 receives from the user a selection such as "understands" or "does not understand" the task as a result of the user making a judgment based on the output information such as the name and content of the task.

[0050] If the user selects "understands," the specification unit 133 specifies the user's level of understanding for the task as "understood." On the other hand, if the user selects "does not understand," the specification unit 133 specifies the user's level of understanding for the task as "does not understand."

[0051] The above-described user's level of understanding is merely an example, and the identification unit 133 can identify the user's level of understanding based on n-value options in addition to the above-described binary values ​​of "understood" or "not understood." For example, the identification unit 133 can identify the user's level of understanding as five values ​​such as "5 (not understood)," "4 (somewhat not understood)," "3 (cannot say either way)," "2 (somewhat understood)," and "1 (understood)," instead of the binary values ​​of "understood" or "not understood."

[0052] (Divided part 134) When the task acquired by the acquisition unit 131 is determined to be complex by the determination unit 132, the division unit 134 inputs the task and guidelines into a search expansion generation model for division that has been trained to divide the task into one or more subtasks, and divides the task into one or more subtasks.

[0053] For example, if the input task is a task related to "cause investigation" for identifying the cause of the event that has occurred, the dividing unit 134 determines that the input task corresponds to "cause identification" in the guidelines associated with task information No. "1" stored in the task information DB 121. Then, the dividing unit 134 divides the input task related to "cause investigation" into subtasks of "cause identification."

[0054] Furthermore, for example, if the input task is a task related to "formulation of a countermeasure" for the event that has occurred, the dividing unit 134 determines that the input task corresponds to "countermeasure plan" in the guidelines associated with task information No. "2" stored in the task information DB 121. Then, the dividing unit 134 divides the input task related to "formulation of a countermeasure" into four subtasks: "cause identification," "confirmation of countermeasures," "impact assessment," and "countermeasure plan."

[0055] Furthermore, for example, if the input task is a task related to "cause investigation" for identifying the specific cause of the event that has occurred, the dividing unit 134 determines that the input task corresponds to "loss of control of reaction" in the guidelines associated with task information No. "3" stored in the task information DB 121. Then, the dividing unit 134 divides the input task related to "cause investigation" into subtasks for investigating the cause of "loss of control of reaction."

[0056] Furthermore, for example, if the input task is a task for performing a more detailed "cause investigation" to identify the specific cause of the event that has occurred, the dividing unit 134 determines that the input task corresponds to "loss of control of reaction, machine failure, and external heat source" in the guidelines associated with task information No. "4" stored in the task information DB 121. Then, the dividing unit 134 divides the input task related to "cause investigation" into three subtasks for investigating the causes of "loss of control of reaction," "machine failure," and "external heat source."

[0057] Furthermore, the dividing unit 134 executes a process of dividing an input task into subtasks based on the user's level of understanding. Specifically, when the user's level of understanding for the task calculated by the identifying unit 133 is less than a reference value, the dividing unit 134 divides the task determined to be complex into one or more subtasks. For example, when the identifying unit 133 identifies the user's level of understanding for the input task as "no understanding," the dividing unit 134 divides the task into one or more subtasks.

[0058] On the other hand, the dividing unit 134 may skip the dividing process if the user's understanding level calculated by the identifying unit 133 is not less than the reference value. That is, the dividing unit 134 may not perform the process, and instead, the output unit 135 (described later) may output the task in an undivided state to the user.

[0059] (output unit 135) The output unit 135 outputs the subtasks divided by the division unit 134. Specifically, the output unit 135 outputs the subtasks divided by the division unit 134 to the user, other information processing devices, etc.

[0060] Furthermore, the output unit 135 executes a process of outputting the input task to the user in an undivided state based on the user's level of understanding. Specifically, when the user's level of understanding for the task calculated by the identification unit 133 is equal to or greater than a reference value, the output unit 135 outputs a predetermined request to the user to divide the task. For example, when the identification unit 133 identifies the user's level of understanding for the input task as "understood," the output unit 135 displays a message to the user such as "Please divide the input task according to the guidelines."

[0061] On the other hand, the output unit 135 may skip the output process if the user's understanding level calculated by the identification unit 133 is not equal to or greater than the reference value. That is, the output unit 135 may not perform the process, and instead, the division unit 134 described above may perform the division process of the task.

[0062] (Example of processing) Next, an example of processing realized by the dividing device 100 according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a diagram for explaining an example of processing according to this embodiment. Fig. 5 shows an example of executing a task dividing process and a process for generating answer content corresponding to an input task based on the result of a predetermined determination process.

[0063] First, the dividing device 100 receives a task issued by the user 10 ((1) in FIG. 5). Next, the dividing device 100 determines whether the received task is complex ((2) in FIG. 5).

[0064] If the task is not complex ((3-1) in FIG. 5), the dividing device 100 issues a query to the generation AI, etc. based on the undivided task ((3-2) in FIG. 5). Then, the generation AI, etc. that receives the query generates a response according to the task ((3-3) in FIG. 5).

[0065] On the other hand, if the task is complex and not an interactive task (No in (4) of FIG. 5), the division device 100 executes a query based on Zero-shot CoT to the generation AI, etc. ((4-1) of FIG. 5). Then, the generation AI, etc. that receives the query generates an answer according to the task ((4-2) of FIG. 5).

[0066] The above-mentioned "interactive task" refers to a task that can request input from the user. Furthermore, "Zero-shot Cot" is a method based on known technology that involves a chain of thought, in which instructions such as "think step by step" are given to a generation AI to describe intermediate steps.

[0067] Furthermore, if the task is complex and an interactive task (Yes in (4) of FIG. 5), and the user does not have a level of understanding of the task (No in (5) of FIG. 5), the splitting device 100 executes a task splitting process ((6) of FIG. 5).

[0068] Specifically, the dividing device 100 divides the task into subtasks based on the input task and guidelines. Next, the dividing device 100 displays the divided subtasks to the user 10 ((6-1) in FIG. 5). Next, the dividing device 100 accepts input of additional information related to the displayed subtasks input by the user 10 ((6-2) in FIG. 5). Here, the dividing device 100 repeatedly executes the above-described process according to the number of divided subtasks.

[0069] Next, the division device 100 issues a query to the generation AI or the like based on the subtask including the received additional information ((6-3) in FIG. 5). Then, the generation AI or the like that has received the query generates a response according to the task ((6-4) in FIG. 5).

[0070] On the other hand, if the task is complex and an interactive task (Yes in (4) of FIG. 5), and the user has a degree of understanding of the task (Yes in (5) of FIG. 5), the splitting device 100 executes a process to request the user to split the task ((7) of FIG. 5).

[0071] Specifically, the dividing device 100 displays the task before division to the user ((7-1) in FIG. 5). Next, the dividing device 100 accepts input of the divided subtasks by the user ((7-2) in FIG. 5).

[0072] Next, the dividing device 100 issues a query based on the received subtask to the generation AI or the like ((7-3) in FIG. 5). Then, the generation AI or the like that has received the query generates a response according to the task ((7-4) in FIG. 5).

[0073] (Processing Procedure) Hereinafter, a description will be given of the procedure of the process performed by the dividing device 100 according to this embodiment. Fig. 6 is a flowchart showing an example of the procedure of the dividing process according to this embodiment.

[0074] The acquisition unit 131 acquires a task and a guideline (S101). Next, the determination unit 132 determines the complexity of the task (S102).

[0075] Here, if the task is determined to be complex (Yes in S103), the input task is an interactive task (Yes in S104), and the user's level of understanding is below the reference value (No in S105), the dividing unit 134 divides the task into subtasks (S106).

[0076] On the other hand, if the task is determined to be complex (Yes in S103), the input task is an interactive task (Yes in S104), and the user's level of understanding is equal to or greater than the reference value (Yes in S105), the acquisition unit 131 acquires the subtasks divided by the user (S107).

[0077] The output unit 135 displays the subtasks input by the user to the user (S108). Next, the acquisition unit 131 acquires information about the subtasks (S109).

[0078] Here, if information about all subtasks has not been acquired (No in S110), the splitting device 100 returns to the previous step and continues the process. On the other hand, if information about all subtasks has been acquired (Yes in S110), the splitting device 100 ends the process.

[0079] Furthermore, if the task is either not complex (No in S103) or not an interactive task (No in S104), the splitting device 100 skips steps S105 to S110 and ends the process without performing the task splitting process.

[0080] (effect) Next, the effects of the division device 100 according to this embodiment will be described. As described above, when setting a chain of thought (CoT) prompt for a generation AI or the like, the user is required to understand the implementation method and implementation procedures of the subtasks included in the target task and divide it appropriately. However, the input task may have unclear processing procedures or may be composed of many procedures, which may make it difficult for the user to understand the content of the task and divide it appropriately.

[0081] Therefore, the acquisition unit 131 of the division device 100 according to this embodiment acquires a task and a guideline. When the determination unit 132 determines that the task acquired by the acquisition unit 131 is complex, the division unit 134 of the division device 100 inputs the task and the guideline to a search expansion generation model for division, and divides the task into one or more subtasks. The output unit 135 of the division device 100 outputs the subtasks divided by the division unit 134.

[0082] Therefore, the division device 100 of this embodiment has the effect of enabling appropriate division of a task. That is, the division device 100 can divide a complex task that is difficult for a user to divide appropriately into subtasks of appropriate granularity based on guidelines. As a result, even when generating an answer corresponding to an input task, the division device 100 has the effect of enabling a generation AI or the like to generate an appropriate answer based on a Chain of Thought (CoT) prompt using subtasks divided to appropriate granularity.

[0083] Furthermore, in order to achieve the above-mentioned effects, the dividing device 100 executes the specific processing described below.

[0084] The determination unit 132 inputs the task and guidelines into a search expansion generation model for determination, which has been trained to output the number of individual subtasks for executing the task input to the generation AI, and calculates the number of individual subtasks for executing the task. Next, the determination unit 132 determines that the task is complex if the calculated number of individual subtasks for executing the task is two or more. Then, the division unit 134 divides the task determined by the determination unit 132 to be complex into one or more subtasks.

[0085] Through the above-described processing, the dividing device 100 performs a process of dividing a complex input task, which includes a plurality of subtasks, into subtasks, thereby achieving the effect of enabling appropriate division of the task.

[0086] The determination unit 132 inputs the task and specifications regarding task execution into a search expansion generation model for determination, which has been trained to output the number of information groups of a predetermined granularity related to the task input to the generation AI, and calculates the number of information groups of a predetermined granularity.The determination unit 132 then determines that the task is complex if the calculated number of information groups of a predetermined granularity is two or more.The division unit 134 then divides the task determined by the determination unit 132 to be complex into one or more subtasks.

[0087] Through the above-described processing, the division device 100 performs a division process into subtasks for complex tasks that have multiple information sources that are used, for example, when generating an answer based on an input task, thereby achieving the effect of enabling appropriate division of tasks.

[0088] Furthermore, the dividing device 100 can select whether to divide the task itself or to have the user divide the task based on the user's level of understanding of the task. Specifically, the identification unit 133 calculates the user's level of understanding of a task determined to be complex. If the user's level of understanding of the task calculated by the identification unit 133 is less than a reference value, the division unit 134 divides the task determined to be complex into one or more subtasks. Next, the output unit 135 outputs the subtasks divided by the division unit 134 to the user. Then, the acquisition unit 131 acquires information about the subtasks input by the user in accordance with the subtasks output by the output unit 135.

[0089] As described above, the dividing device 100 divides a task based on the user's level of understanding. That is, for a complex task whose contents the user does not understand, the dividing device 100 divides the task into subtasks based on additional information from the user. Therefore, the dividing device 100 has the effect of enabling appropriate division of a task based on the user's level of understanding of the task.

[0090] On the other hand, if the user's level of understanding of the task calculated by the identification unit 133 is equal to or greater than a reference value, the output unit 135 outputs a predetermined request to the user to divide the task. Then, the acquisition unit 131 acquires the subtasks divided by the user.

[0091] As described above, for a complex task whose contents the user understands, the dividing device 100 accepts subtasks divided by the user himself / herself. Therefore, the dividing device 100 has an effect of enabling appropriate task division processing to be executed based on the user's level of understanding of the task.

[0092] <Modification> The following describes modified examples realized by the dividing device 100 according to this embodiment.

[0093] (Data, etc.) The tasks, subtasks, guidelines, interactive tasks, names of functional parts of the dividing device 100, steps, processes, names of steps or processes, etc. used in the description of the above embodiments are merely examples and can be changed as desired.

[0094] For example, it has been explained that task information DB121 stores "No," which is information that identifies an individual task, "Task," which is information about the input task, and "Guideline," which is information about the guideline related to the task, but this is not limited to this.

[0095] For example, it has been explained that the subtask information DB122 stores "No," which is information identifying an individual task, "Task," which is information about the input task, and "Subtask," which is information about the subtask divided from the task, but this is not limited to this.

[0096] (Flowcharts, etc.) The steps in the flowcharts may be interchanged as long as there is no contradiction, and some steps may not be performed. In addition, conjunctions such as "next," "continue," "further," "at this time," and "on this occasion" used in the explanation of the flowcharts do not limit the order or timing of the execution of the processes in the flowcharts.

[0097] (others) Of the processes described in the above embodiments and variations, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using known methods. In addition, the information, including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings, can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown.

[0098] Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0099] The above-described components include those that can be easily imagined by a person skilled in the art, those that are substantially the same, and those that fall within the so-called equivalent range. Furthermore, the above-described embodiments and modifications can be combined as appropriate within the scope of not causing any contradiction in the processing content.

[0100] Furthermore, the above-mentioned "section, module, unit" can be read as "means" or "circuit," etc. For example, a control section can be read as control means or a control circuit.

[0101] Although some of the embodiments have been described in detail above with reference to the drawings, these are merely examples, and it is possible to implement the present embodiments in other forms that have undergone various modifications and improvements based on the knowledge of those skilled in the art, including the aspects described in the Disclosure of the Invention section.

[0102] <Hardware configuration> The dividing device 100 according to this embodiment is realized, for example, by a computer 1000 configured as shown in Fig. 7. Fig. 7 is a hardware configuration diagram showing an example of a computer that realizes the functions of the dividing device 100 according to this embodiment. The computer 1000 has a configuration in which a CPU 1100, a RAM 1200, a ROM 1300, an auxiliary storage device 1400, a communication I / F (interface) 1500, and an input / output I / F (interface) 1600 are connected by a bus 1800.

[0103] The CPU 1100 operates and controls each unit based on a program stored in the ROM 1300 or the auxiliary storage device 1400. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 starts up, programs that depend on the hardware of the computer 1000, and the like.

[0104] The auxiliary storage device 1400 stores programs executed by the CPU 1100, data used by the programs, etc. The communication I / F 1500 receives data from other devices via a predetermined communication network NW (including closed-area wireless communication in this embodiment) and sends it to the CPU 1100, and transmits data generated by the CPU 1100 to other devices via the predetermined communication network NW. The CPU 1100 controls output devices such as displays and printers, and input / output devices 1700 such as keyboards and mice, via the input / output I / F 1600. The CPU 1100 acquires data from the input / output devices 1700 via the input / output I / F 1600. The CPU 1100 also outputs generated data to the input / output devices 1700 via the input / output I / F 1600.

[0105] For example, when the computer 1000 functions as various devices according to the present embodiment, the CPU 1100 of the computer 1000 executes a program loaded onto the RAM 1200 to realize the functions of the control unit. [Explanation of symbols]

[0106] 100 splitting device 110 Communications Department 120 Storage section 121 Task Information DB 122 Subtask Information DB 123 Learning Model DB 130 control section 131 Acquisition Department 132 Judgment section 133 Specific section 134 Division 135 Output section

Claims

1. an acquisition unit that acquires a task, which is information input to a generative model to generate predetermined information, and a rule regarding the execution of the task; a division unit that divides the task acquired by the acquisition unit into one or more subtasks when the task is determined to be complex by inputting the task and a rule regarding the execution of the task into a search expansion generative model for division that has been trained to divide the task into one or more subtasks; an output unit that outputs the subtasks divided by the division unit; A dividing device comprising:

2. a task and a specification for the execution of the task are input to a search expansion generative model for judgment, the model being trained to output the number of distinct subtasks for the execution of the task; and a determination unit that determines that the task is complex when the calculated number of individual subtasks for executing the task is two or more; The dividing unit Dividing the task determined to be complex by the determination unit into one or more subtasks.

2. The dividing device according to claim 1.

3. a search expansion generative model for judgment that has been trained to output the number of information groups of a predetermined granularity related to the task input to the generative model, inputting the task and a rule regarding the execution of the task, and calculating the number of information groups of the predetermined granularity; a determination unit that determines that the task is complex when the calculated number of information groups of the predetermined granularity is two or more; The dividing unit Dividing the task determined to be complex by the determination unit into one or more subtasks.

2. The dividing device according to claim 1.

4. The method further includes an identification unit that identifies a user's understanding of the task determined to be complex, The dividing unit If the user's level of understanding of the task identified by the identification unit is less than a reference value, the task determined to be complex is divided into one or more subtasks.

2. The dividing device according to claim 1.

5. The method further includes an identification unit that identifies a user's understanding of the task determined to be complex, The output unit If the user's understanding of the task identified by the identification unit is equal to or greater than a reference value, a predetermined request to divide the task is output to the user, The acquisition unit acquiring the subtasks divided by the user in response to a predetermined request for dividing the task output by the output unit; 5. The dividing device according to claim 1 or 4.

6. The output unit outputting the subtasks divided by the dividing unit to a user; The acquisition unit Acquire information about the subtask input by the user in response to the subtask output by the output unit.

5. The dividing device according to claim 1 or 4.

7. A division method to be executed by a division device, an acquisition step of acquiring a task, which is information input to the generative model to generate predetermined information, and a rule regarding the execution of the task; a division step of dividing the task into one or more subtasks by inputting the task and a rule regarding the execution of the task into a search expansion generative model for division that has been trained to divide the task into one or more subtasks when the task obtained by the acquisition step is determined to be complex; an output step of outputting the subtasks divided by the dividing step; A division method comprising:

8. an acquisition step of acquiring a task, which is information input to the generative model to generate predetermined information, and a rule regarding the execution of the task; a division step of dividing the task into one or more subtasks by inputting the task and a rule regarding the execution of the task to a search expansion generative model for division that has been trained to divide the task into one or more subtasks when the task obtained by the acquisition step is determined to be complex; an output step of outputting the subtasks divided by the dividing step; A division program that causes a computer to execute the above.

Images