Idea generation system and idea generation method

Abstract

This idea generation system comprises: an idea generation unit that generates an idea by inputting domain knowledge to a first language model; an idea evaluation unit that evaluates the idea by inputting the idea generated by the idea generation unit to a second language model; a persona setting unit that sets a persona characterizing each of the first language model used by the idea generation unit to generate the idea and the second language model used by the idea evaluation unit to evaluate the idea; and an adjustment unit that performs, on the basis of the evaluated result of the idea by the idea evaluation unit, adjustment processing for adjusting subsequent evaluation results of the idea by the idea evaluation unit.

Description

Idea generation system, idea generation method

[0001] The present invention relates to a system and method for generating business - useful ideas.

[0002] In recent years, as a method for business operators in a certain field to obtain business - useful ideas, a method called co - creation has attracted attention. Co - creation means holding workshops in which not only employees of the business operator, who is the business entity, but also various stakeholders such as experts with rich knowledge in the field and general consumers participate, and by expressing opinions and deepening discussions with each other on the spot, aiming to generate new valuable ideas.

[0003] Generally, in order to obtain useful ideas through co - creation, it is necessary to hold workshops many times and have each stakeholder repeat discussions over a long period of time. However, due to human or time - related constraints such as a shortage of experts in the field or incompatible schedules of each stakeholder, the number of times or the time available to hold workshops may be limited. Also, depending on the implementation period of a project for promoting business, sufficient time for co - creation may not be ensured. Thus, there are various restrictive conditions for implementing co - creation with the participation of multiple people with different positions.

[0004] On the other hand, with the progress of AI (Artificial Intelligence) technology in recent years, it has also been proposed to use AI such as machine learning for idea generation. For example, in Patent Document 1, there is described an idea support device having an idea database that stores ideas, evaluations of ideas, and parent - child relationships of ideas, an arithmetic unit that selects an idea as a draft based on the evaluation of the idea for display, and an interface for displaying the draft and inputting evaluations of participants for the draft and proposals that are new ideas associated with the draft as a parent. When the arithmetic unit selects an idea, it uses machine learning or the like to identify combinations of the attributes of evaluators and viewers and evaluates the idea.

[0005] Japanese Patent Application Publication No. 2021-157509

[0006] The idea generation support device described in Patent Document 1 effectively utilizes the collective intelligence of an unspecified number of online participants to select good ideas and actively build upon them, thereby improving the quality and quantity of ideas from each participant. However, it does not resolve the aforementioned constraints on implementing collaborative creation.

[0007] This invention is based on this background and aims to realize idea generation through collaborative creation that does not require a large number of participants.

[0008] The idea generation system according to the present invention comprises: an idea generation unit that generates ideas by inputting domain knowledge into a first language model; an idea evaluation unit that evaluates the ideas generated by the idea generation unit by inputting the ideas into a second language model; a persona setting unit that sets personas to characterize the first language model used by the idea generation unit to generate the ideas and the second language model used by the idea evaluation unit to evaluate the ideas; and an adjustment unit that performs adjustment processing to adjust subsequent evaluation results of the ideas by the idea evaluation unit based on the evaluation results of the ideas by the idea evaluation unit. The idea generation method according to the present invention uses a computer to set personas to characterize the first language model and the second language model, input domain knowledge into the first language model to generate ideas, input the generated ideas into the second language model to evaluate the ideas, and perform adjustment processing to adjust subsequent evaluation results of the ideas based on the evaluation results of the ideas.

[0009] According to the present invention, it is possible to generate ideas through collaborative creation without the need for a large number of participants.

[0010] A functional block diagram of the idea generation system according to the first embodiment of the present invention. A diagram showing an example of the hardware configuration of the idea generation platform that realizes each functional block of the idea generation system. A flowchart showing the processing flow of the idea generation system according to the first embodiment of the present invention. A diagram explaining the adjustment process of evaluation criteria. A functional block diagram of the idea generation system according to the second embodiment of the present invention. A flowchart showing the processing flow of the idea generation system according to the second embodiment of the present invention. A diagram explaining the reinforcement process. A functional block diagram of the idea generation system according to the third embodiment of the present invention. A functional block diagram of the idea generation system according to the fourth embodiment of the present invention.

[0011] Embodiments of the present invention will be described below with reference to the drawings.

[0012] (First Embodiment) Figure 1 is a functional block diagram of the idea generation system 1 according to the first embodiment of the present invention. The idea generation system 1 shown in Figure 1 is a system that generates ideas through collaborative creation using AI, and includes an idea generation unit 10, an idea evaluation unit 20, a persona setting unit 30, and an adjustment unit 40.

[0013] The idea generation unit 10 has a pre-trained language model 11 that can be used for AI processing. Domain knowledge is input to the idea generation unit 10. Domain knowledge is a collection of various information (e.g., text, numbers, audio, images, etc.) related to the field in which the idea generation system 1 generates ideas, and is either pre-recorded in the idea generation system 1 or acquired from an external source as appropriate. The idea generation unit 10 inputs the domain knowledge into the language model 11 and generates ideas by performing AI processing using the language model 11 based on the domain knowledge. The ideas generated by the idea generation unit 10 are input to the idea evaluation unit 20.

[0014] The idea evaluation unit 20 has a trained language model 21 that can be used for AI processing. The idea evaluation unit 20 inputs the ideas generated by the idea generation unit 10 into the language model 21 and performs AI processing using the language model 21 to evaluate the ideas. The evaluation results of the idea by the idea evaluation unit 20 are input into the adjustment unit 40. If the evaluation results meet predetermined evaluation criteria, the idea is output from the idea generation system 1 as an evaluated idea and presented to the user of the idea generation system 1.

[0015] The persona setting unit 30 sets personas to characterize the language model 11 of the idea generation unit 10 and the language model 21 of the idea evaluation unit 20. The idea generation unit 10 and the idea evaluation unit 20 adjust various conditions for AI processing using the language models 11 and 21, and adjust parameters (weights) in the language models 11 and 21, respectively, according to the personas set by the persona setting unit 30, so that AI processing corresponding to the set personas is performed. Specifically, for example, the persona setting unit 30 sets personas such as "engineer" or "software engineer" for the idea generation unit 10 and "consumer" for the idea evaluation unit 20, respectively, based on information selected by the user or pre-registered persona specification information. The idea generation unit 10 and the idea evaluation unit 20 adjust the AI ​​processing performed using the language models 11 and 21 so that processing results corresponding to these set personas are obtained. Furthermore, if processing results corresponding to the persona settings can be obtained, the idea generation unit 10 and the idea evaluation unit 20 can each perform arbitrary AI processing using the trained language models 11 and 21.

[0016] The adjustment unit 40 performs adjustment processing to adjust the subsequent evaluation results of the idea by the idea evaluation unit 20, based on the evaluation results of the idea by the idea evaluation unit 20. Details of the adjustment processing by the adjustment unit 40 will be described later.

[0017] Figure 2 shows an example of the hardware configuration of the idea generation platform 100 that realizes each functional block of the idea generation system 1. The idea generation platform 100 is configured using an information processing device (computer) such as a server or PC, and includes a memory 101, a CPU (Central Processing Unit) 102, an auxiliary processor 103, a GPU (Graphics Processing Unit) 104, an I / O unit 105, a network control unit 106, an NPU (Neural network Processing Unit) 107, and an auxiliary storage device 108. In the idea generation platform 100, these are connected to each other via a communication bus.

[0018] The memory 101 is a volatile storage medium and is used as a workspace for the CPU 102, auxiliary processor 103, GPU 104, and NPU 107 when they execute processing according to a predetermined program.

[0019] The CPU 102 executes a predetermined program to realize the functions of the idea generation unit 10, the idea evaluation unit 20, the persona setting unit 30, and the adjustment unit 40 in the idea generation system 1. The auxiliary processor 103, GPU 104, and NPU 107 can cooperate with the CPU 102 to perform processing to realize any of the functions of the idea generation unit 10, the idea evaluation unit 20, the persona setting unit 30, and the adjustment unit 40.

[0020] The I / O unit 105 performs input / output processing of various data exchanged between the idea generation system 1 and the user. For example, the I / O unit 105 incorporates user-inputted information into the idea generation system 1 as domain knowledge, and presents evaluated ideas generated by the idea generation system 1 to the user. The I / O unit 105 can be configured using, for example, a keyboard, mouse, display, or touch panel.

[0021] The network control unit 106 is responsible for performing interface processing for data input and output between the idea generation system 1 and a network (not shown, e.g., LAN, internet, etc.). For example, the network control unit 106 can be used instead of the I / O unit 105 to acquire domain knowledge from an external source via the network, or to output evaluated ideas generated by the idea generation system 1 to an external source via the network.

[0022] The auxiliary storage device 108 is a recording medium capable of storing various programs and data, and is configured using, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), etc. For example, domain knowledge taken into the idea generation system 1 by the I / O unit 105 or the network control unit 106 is recorded in the auxiliary storage device 108 and used in the processing of the idea generation unit 10. In addition, data for constructing language models 11 and 21 can also be recorded in the auxiliary storage device 108 and used in the AI ​​processing performed by the idea generation unit 10 and the idea evaluation unit 20.

[0023] The idea generation platform 100 does not necessarily have to include all of the components shown in Figure 2. For example, the idea generation platform 100 may be configured without some or all of the auxiliary processor 103, GPU 104, and NPU 107.

[0024] Alternatively, an idea generation platform 100 may be configured by combining multiple computers, thereby realizing each functional block of the idea generation system 1 shown in Figure 1. Furthermore, some or all of the functional blocks of the idea generation system 1 may be realized by computers or virtual machines on the cloud provided as a hosted service.

[0025] Figure 3 is a flowchart showing the processing flow of the idea generation system 1 according to the first embodiment of the present invention. The processing shown in this flowchart is executed by the CPU 102, etc.

[0026] In step S10, it is determined whether personas have been set for language models 11 and 21. If personas have not been set for language models 11 and 21 by the persona setting unit 30, the process proceeds to step S20; otherwise, the process proceeds to step S50.

[0027] In step S20, the persona setting unit 30 sets the personas for language models 11 and 21. As mentioned above, for example, a persona such as "engineer" or "software engineer" is set for language model 11 of the idea generation unit 10, and a persona such as "consultant" or "consumer" is set for language model 21 of the idea evaluation unit 20. At this time, it is preferable that the persona setting unit 30 sets different settings for the persona set for language model 11 and the persona set for language model 21 in order to realize idea generation through co-creation in the idea generation system 1. In order to set the personas for language models 11 and 21, the persona setting unit 30 may, for example, change the parameters of language models 11 and 21, or specify a part of the prompts for language models 11 and 21 (prompts that the idea generation unit 10 inputs to language model 11 and prompts that the idea evaluation unit 20 inputs to language model 21) so that the prompts include words or sentences that indicate the persona. The persona setting method for language models 11 and 21 by the persona setting unit 30 is not limited to these examples.

[0028] In step S30, the persona setting unit 30 checks the status of the personas set for the language models 11 and 21 in step S20. Here, for example, the output from the idea generation unit 10 and the idea evaluation unit 20 is obtained, and the status of the personas can be checked by confirming whether or not the content of these outputs matches the set personas.

[0029] In step S40, it is determined whether the persona setting status confirmed in step S30 is appropriate. If the persona setting status for language models 11 and 21 is appropriate, that is, if the output content from both the idea generation unit 10 and the idea evaluation unit 20 is consistent with the set persona, the process proceeds to step S50. On the other hand, if the persona setting status for language models 11 and 21 is not appropriate, the process returns to step S20, and the persona setting unit 30 redoes the persona setting. In this case, if it is determined that the persona setting status is inappropriate for only one of the language models 11 and 21, it is not necessary to reset the persona for the other.

[0030] In step S50, domain knowledge of the field in which ideas are to be generated is input to the idea generation unit 10, for which the persona of the language model 11 was set in step S20.

[0031] In step S60, the idea generation unit 10 generates ideas in the relevant field by performing AI processing using the language model 11 based on the domain knowledge input in step S50. At this time, the idea generation unit 10 can generate ideas in the relevant field according to the persona settings for the language model 11. That is, the idea generation unit 10 can generate ideas in the relevant field by inputting prompts related to idea generation to the language model 11 whose parameters have been adjusted by the persona setting unit 30. Alternatively, the idea generation unit 10 can generate ideas in the relevant field by inputting both the prompts set by the persona setting unit 30 and prompts related to idea generation to the language model 11.

[0032] In step S70, the idea evaluation unit 20 evaluates the ideas generated by the idea generation unit 10 in step S60. At this time, the idea evaluation unit 20 can evaluate the generated ideas according to the persona settings for the language model 21 by performing AI processing using the language model 21. Specifically, for example, the idea evaluation unit 20 inputs the generated ideas into the language model 21 and calculates a loss function between the generated ideas and the ground truth data in the language model 21 for which a specific persona has been set. Here, different ground truth data can be used in the language model 21 depending on the set persona. Then, an evaluation value is determined according to the calculated loss function. At this time, it is preferable to determine the evaluation value such that the evaluation value increases as the value of the loss function decreases. In this way, the idea evaluation unit 20 can obtain an evaluation value that represents the evaluation result for the generated ideas. Note that the idea evaluation unit 20 may obtain the evaluation value using other methods or a combination of methods as long as the evaluation result for the generated ideas can be appropriately represented. The idea evaluation unit 20 can output an evaluation value for a generated idea by comparing the frequency and uniqueness of words included in the generated idea with words included in the correct answer data, for example, using TF-IDF (Term Frequency-Inverse Document Frequency). However, the method of evaluating the generated idea by the idea evaluation unit 20 is not limited to this.

[0033] In step S80, the idea evaluation unit 20 determines whether the evaluation value obtained in step S70 is equal to or greater than a predetermined threshold. If the evaluation value is equal to or greater than the threshold, it is determined that the evaluation result for the generated idea meets the predetermined evaluation criteria, and the process proceeds to step S90. On the other hand, if the evaluation value is less than the threshold, it is determined that the evaluation result for the generated idea does not meet the evaluation criteria, and the process proceeds to step S100.

[0034] In step S90, the ideas generated by the idea generation unit 10 in step S60 are output as evaluated ideas. Here, the evaluated ideas can be output by, for example, presenting them to the user via the I / O unit 105 or transmitting them externally via the network control unit 106. After completing the process in step S90, the idea generation system 1 terminates the process shown in the flowchart of Figure 3.

[0035] In step S100, the adjustment unit 40 calculates the semantic distance between the idea generated by the idea generation unit 10 in step S60 and a pre-registered reference idea. Specifically, for example, the adjustment unit 40 can calculate the semantic distance by vectorizing the generated idea and the reference idea, and then determining the distance between each of the resulting vectors. The reference idea represents the ideal idea that the idea generation unit 10 should generate, and is pre-registered for each persona that can be set by the persona setting unit 30. Alternatively, the correct answer data used when the idea evaluation unit 20 evaluates the generated idea in step S70 may be used as the reference idea. In addition to the above, any information can be used as the reference idea.

[0036] In step S110, the adjustment unit 40 performs an adjustment process to adjust the subsequent idea evaluation results by the idea evaluation unit 20 using the semantic distance calculated in step S100. Here, for example, the adjustment unit 40 refers to the domain knowledge input to the idea generation unit 10 in step S50 and adjusts the persona setting state for the language model 11 of the idea generation unit 10 so that the semantic distance calculated for subsequent ideas is minimized. Specifically, for example, the adjustment unit 40 instructs the persona setting unit 30 to adjust "Temperature," one of the parameters used by the persona setting unit 30 when setting a persona for the language model 11, so that the ideas generated by the idea generation unit 10 in the future become semantically closer to the aforementioned reference idea. Alternatively, the adjustment unit 40 may instruct the persona setting unit 30 to change the prompt that the idea generation unit 10 inputs to the language model 11. For example, if the persona specified in the prompt input to the language model 11 is "You are an engineer," and the evaluation value of the idea generated by the idea generation unit 10 (language model 11) falls below a threshold, the adjustment unit 40 may change the persona specified to be included in the prompts input to the language model 11 in the future to "You should act as a technical consultant regarding the Internet of Vehicles." This allows the adjustment unit 40 to perform adjustment processing so that the idea evaluation unit 20 is more likely to obtain an evaluation result that meets the evaluation criteria for ideas generated in the future.

[0037] Alternatively, the adjustment process in step S110 may be performed by adjusting the evaluation criteria used by the idea evaluation unit 20 when evaluating ideas. Specifically, the adjustment unit 40 may instruct the idea evaluation unit 20 to adjust (lower) the threshold value used in the judgment process in step S80. The adjustment unit 40 may also instruct the persona setting unit 30 to change the persona set in the language model 21 in order to adjust the evaluation criteria used by the idea evaluation unit 20 when evaluating ideas. For example, the adjustment unit 40 can instruct the persona setting unit 30 to include keywords such as "relax the criteria" or "allow similarity between the generated idea and the correct answer data" in the prompts that the idea evaluation unit 20 inputs to the language model 21. In this way, the adjustment process by the adjustment unit 40 can be performed so that the idea evaluation unit 20 is more likely to obtain an evaluation result that satisfies the evaluation criteria when evaluating future generated ideas.

[0038] Figure 4 is an explanatory diagram of the adjustment process of the evaluation criteria by the adjustment unit 40. Figure 4(a) shows an example of the evaluation value for each generated idea before the adjustment of the evaluation criteria, and Figure 4(b) shows an example of the evaluation value for each generated idea after the adjustment of the evaluation criteria. In these figures, points 41 to 44 represent examples of each idea and its evaluation value generated by the idea generation unit 10. The straight line 45 in the figure represents the threshold value used as the evaluation criterion when the idea evaluation unit 20 evaluates the ideas. Although Figure 4 shows an example where there is only one type of evaluation value, there may be two or more types of evaluation values.

[0039] In the adjustment process of step S110 in Figure 3, the adjustment unit 40 lowers the threshold represented by the straight line 45, for example, as shown in Figure 4(b). As a result, the evaluation values ​​of ideas 42 and 43, which were below the threshold before the adjustment process, become above the threshold even though their evaluation values ​​themselves do not change, because the threshold is lowered. Consequently, ideas 42 and 43 are determined by the idea evaluation unit 20 to meet the evaluation criteria and are output as evaluated ideas.

[0040] Incidentally, in the above description, as an example of the adjustment process by the adjustment unit 40, either the setting state of the persona for the language model 11 or the evaluation criteria used when the idea evaluation unit 20 evaluates an idea has been described. However, both of these adjustments may be performed. Alternatively, the adjustment process by the adjustment unit 40 may be performed by other methods. As long as the evaluation results of ideas by the idea evaluation unit 20 hereafter can be appropriately adjusted, the adjustment process can be performed by any method.

[0041] Returning to the description of FIG. 3, when the adjustment process by the adjustment unit 40 is performed in step S110, the process returns to step S20 to re-perform the persona setting by the persona setting unit 30. After that, the above-described process is performed again.

[0042] According to the first embodiment of the present invention described above, the following operational effects can be obtained.

[0043] (1) The idea generation system 1 includes an idea generation unit 10 that inputs domain knowledge into the language model 11 to generate ideas, an idea evaluation unit 20 that inputs the ideas generated by the idea generation unit 10 into the language model 21 to evaluate the ideas, a language model 11 used by the idea generation unit 10 for generating ideas, a language model 21 used by the idea evaluation unit 20 for evaluating ideas, a persona setting unit 30 that sets personas characterizing each of them, and an adjustment unit 40 that performs an adjustment process for adjusting the evaluation results of ideas by the idea evaluation unit 20 hereafter based on the evaluation results of ideas by the idea evaluation unit 20. By doing so, for example, even without a large number of participants with different positions such as experts or general consumers, idea generation by co-creation can be realized.

[0044] (2) The adjustment unit 40 performs the adjustment process by adjusting at least one of the setting state of the persona for the language model 11 and the evaluation criteria used when the idea evaluation unit 20 evaluates an idea (step S110). By doing so, the evaluation results of ideas by the idea evaluation unit 20 hereafter can be appropriately adjusted.

[0045] (3) The persona setting unit 30 preferably sets different personas for the language model 11 and the language model 21. In this way, the idea generation unit 10 can appropriately generate ideas using the language model 11, and the idea evaluation unit 20 can appropriately evaluate the generated ideas using the language model 21.

[0046] (4) The idea evaluation unit 20 obtains an evaluation value for the idea generated by the idea generation unit 10 (step S70). When this evaluation value is less than a predetermined threshold (step S80: No), the adjustment unit 40 performs the adjustment process of step S110. By doing so, when the evaluation result of the generated idea does not meet the evaluation criteria, it becomes possible to perform an adjustment process so that ideas that meet the evaluation criteria can be generated in the future.

[0047] (5) The adjustment unit 40 calculates the semantic distance between the idea generated by the idea generation unit 10 and a pre-registered reference idea (step S100), and performs the adjustment process of step S110 using the calculated semantic distance. By doing so, it is possible to perform the adjustment process by the adjustment unit 40 so that an evaluation result that the idea meets the evaluation criteria is likely to be obtained in the evaluation performed by the idea evaluation unit 20 for the ideas generated in the future.

[0048] (Second Embodiment) FIG. 5 is a functional block diagram of an idea generation system 1A according to the second embodiment of the present invention. The idea generation system 1A shown in FIG. 5 is different from the idea generation system 1 of FIG. 1 described in the first embodiment in that it further has a reinforcement unit 50.

[0049] The reinforcement unit 50 performs a reinforcement process for reinforcing the generation of ideas by the subsequent idea generation unit 10 based on the evaluation result of the ideas by the idea evaluation unit 20. The details of the reinforcement process by the reinforcement unit 50 will be described later.

[0050] Each functional block of the idea generation system 1A in this embodiment can be realized by the idea generation platform 100 shown in Figure 2, which was described in the first embodiment. That is, in the idea generation platform 100, the CPU 102 realizes the functions of the idea generation unit 10, idea evaluation unit 20, persona setting unit 30, adjustment unit 40, and reinforcement unit 50 shown in Figure 5 by executing a predetermined program. The auxiliary processor 103, GPU 104, and NPU 107 can cooperate with the CPU 102 to perform processing to realize any of the functions of the idea generation unit 10, idea evaluation unit 20, persona setting unit 30, adjustment unit 40, and reinforcement unit 50, respectively. In addition, each functional block of the idea generation system 1A shown in Figure 5 may be realized by an idea generation platform 100 configured by combining multiple computers. Furthermore, some or all of the functional blocks of the idea generation system 1A may be realized by computers or virtual machines on the cloud provided as a hosted service.

[0051] Figure 6 is a flowchart showing the processing flow of the idea generation system 1A according to the second embodiment of the present invention. The processing shown in this flowchart is executed by the CPU 102, etc.

[0052] In the flowchart of Figure 6, the parts where the same processing as in the flowchart of Figure 3 described in the first embodiment is performed are given the same step numbers as in Figure 3. Unless otherwise necessary, the content of the processing with these common step numbers will not be explained below.

[0053] After calculating the semantic distance in step S100, step S101 determines whether the calculated semantic distance is less than a predetermined value. If the semantic distance is less than the predetermined value, the process proceeds to step S102; otherwise, the process proceeds to step S110.

[0054] In step S102, the reinforcement unit 50 performs reinforcement processing to reinforce the idea generation by the idea generation unit 10 thereafter. Here, for example, keywords that have a significant impact on the calculation of semantic distance are extracted from the correct data in the language model 21 used when the idea evaluation unit 20 evaluated the ideas in step S70, and the persona settings for the language model 11 of the idea generation unit 10 are adjusted using these keywords. For example, if the persona set in the language model 11 of the idea generation unit 10 is "a technical consultant on the Internet of Vehicles," and "personalization" is extracted as a keyword that has a significant impact on the calculation of semantic distance, the reinforcement unit 50 can change the persona for the language model 11 of the idea generation unit 10 to "a technical consultant on the Internet of Vehicles who is interested in driver personalization." This allows the reinforcement unit 50 to perform reinforcement processing so that the idea evaluation unit 20 is more likely to obtain evaluation results that meet the evaluation criteria when evaluating the ideas that will be generated in the future.

[0055] Figure 7 is an explanatory diagram of the reinforcement process by the reinforcement unit 50. Figure 7(a) shows an example of evaluation values ​​for each generated idea before the reinforcement process is performed, and Figure 7(b) shows an example of evaluation values ​​for each generated idea after the reinforcement process is performed. In these figures, points 41 to 44 show examples of each idea and its evaluation value generated by the idea generation unit 10, similar to Figure 4 described in the first embodiment. Similarly, the straight line 45 shows a threshold value used as an evaluation criterion when the idea evaluation unit 20 evaluates an idea, similar to Figure 4 described in the first embodiment.

[0056] In the reinforcement process of step S102 in Figure 6, the reinforcement unit 50 adjusts the persona settings for the language model 11 of the idea generation unit 10. As a result, for example, as shown in Figure 7(b), the evaluation values ​​of ideas 42 and 43, which were below the threshold before the adjustment process, change according to the persona settings, and become equal to or above the threshold represented by the straight line 45. As a result, ideas 42 and 43 are determined by the idea evaluation unit 20 to meet the evaluation criteria and are output as evaluated ideas.

[0057] Furthermore, reinforcement processing by the reinforcement section 50 may be performed by methods other than those described above. If the content of the ideas generated by the subsequent idea generation section 10 can be appropriately adjusted, reinforcement processing can be carried out by any method.

[0058] Returning to the explanation of Figure 6, after the reinforcement process by the reinforcement unit 50 is performed in step S102, the process returns to step S60 and the idea generation by the idea generation unit 10 is restarted. After that, the aforementioned process is performed again.

[0059] According to the second embodiment of the present invention described above, the idea generation system 1A further includes a reinforcement unit 50 that performs reinforcement processing to reinforce the idea generation by the idea generation unit 10 based on the idea evaluation results by the idea evaluation unit 20. In this way, if the evaluation results of the generated ideas do not meet the evaluation criteria, it becomes easier to generate ideas that meet the evaluation criteria in the future.

[0060] Furthermore, the reinforcement unit 50 performs reinforcement processing (step S102) if the semantic distance calculated in step S100 is less than a predetermined value (step S101: Yes). In this way, the reinforcement processing by the reinforcement unit 50 can be performed so that in the evaluation performed by the idea evaluation unit 20 on future generated ideas, it is easier to obtain an evaluation result that the evaluation criteria are met.

[0061] (Third Embodiment) Figure 8 is a functional block diagram of the idea generation system 1B according to the third embodiment of the present invention. The idea generation system 1B shown in Figure 8 differs from the idea generation system 1 shown in Figure 1, which was described in the first embodiment, in that the idea evaluation unit 20 notifies the idea generation unit 10 of the evaluation results of the ideas.

[0062] In this embodiment, the idea generation system 1B has an idea generation unit 10 and an idea evaluation unit 20 that communicate with each other, reflecting the results of their interactions and adjusting the language models 11 and 21 accordingly. This allows the idea generation unit 10 to feed back the evaluation results of ideas notified by the idea evaluation unit 20 to subsequent idea generation, thereby adjusting the language model 11 to obtain higher evaluation values. To achieve this operation, the idea generation system 1B may implement the idea generation unit 10 and the idea evaluation unit 20 as agent services.

[0063] Although the idea generation system 1B shown in Figure 8 does not include the reinforcing part 50 described in the second embodiment, the idea generation system 1B of this embodiment may be configured to include the reinforcing part 50.

[0064] (Fourth Embodiment) Figure 9 is a functional block diagram of the idea generation system 1C according to the fourth embodiment of the present invention. The idea generation system 1C shown in Figure 9 differs from the idea generation system 1 shown in Figure 1, which was described in the first embodiment, in that it further includes a context setting unit 60.

[0065] The context setting unit 60 sets the context used by the persona setting unit 30 to adjust parameters when setting personas. Specifically, it obtains information such as the user's location and the organization to which the user belongs, from the IP address of the access source to the idea generation system 1C and domain knowledge entered by the user, and based on this information, generates a context that shows the detailed information of the persona to be set for the language models 11 and 21 by the persona setting unit 30. By notifying the persona setting unit 30 of the generated context, the persona setting unit 30 is able to set personas corresponding to that context for the language models 11 and 21, respectively. Note that the context may be set by other methods.

[0066] Each functional block of the idea generation system 1C in this embodiment can be realized by the idea generation platform 100 shown in Figure 2, which was described in the first embodiment. That is, in the idea generation platform 100, the CPU 102 realizes the functions of the idea generation unit 10, idea evaluation unit 20, persona setting unit 30, adjustment unit 40, and context setting unit 60 shown in Figure 9 by executing a predetermined program. The auxiliary processor 103, GPU 104, and NPU 107 can cooperate with the CPU 102 to perform processing to realize any of the functions of the idea generation unit 10, idea evaluation unit 20, persona setting unit 30, adjustment unit 40, and context setting unit 60. In addition, each functional block of the idea generation system 1C shown in Figure 9 may be realized by an idea generation platform 100 configured by combining multiple computers. Furthermore, some or all of the functional blocks of the idea generation system 1C may be realized by computers or virtual machines on the cloud provided as a hosted service.

[0067] Although the idea generation system 1C shown in Figure 9 does not include the reinforcement part 50 described in the second embodiment, the idea generation system 1C of this embodiment may be configured to include the reinforcement part 50. Furthermore, in this embodiment as well, the idea generation unit 10 and the idea evaluation unit 20 may interact with each other, reflect the results in each other, and adjust the language models 11 and 21 accordingly.

[0068] It should be noted that the present invention is not limited to the first to fourth embodiments described above, and various modifications are possible without departing from the spirit of the invention. For example, each embodiment has been described in detail to make the present invention easier to understand, and is not necessarily limited to having all the configurations described. Furthermore, it is possible to add, delete, or replace some of the configurations in each embodiment with other configurations.

[0069] Furthermore, each of the above-mentioned configurations, functional units, processing units, processing means, etc., may be implemented in hardware, in whole or in part, for example, by designing them as integrated circuits. Alternatively, each of the above-mentioned configurations, functions, etc., may be implemented in software by having the processor interpret and execute programs that realize each function. Information such as programs, tables, and files that realize each function can be stored in memory, hard disks, SSDs, or other recording devices, or in recording media such as IC cards, SD cards, or DVDs.

[0070] 1, 1A, 1B, 1C: Idea generation system 10: Idea generation unit 11: Language model 20: Idea evaluation unit 21: Language model 30: Persona setting unit 40: Adjustment unit 50: Reinforcement unit 60: Context setting unit 100: Idea generation platform

Claims

1. An idea generation system comprising: an idea generation unit that inputs domain knowledge into a first language model to generate ideas; an idea evaluation unit that inputs the ideas generated by the idea generation unit into a second language model to evaluate the ideas; a persona setting unit that sets personas to characterize the first language model used by the idea generation unit to generate the ideas and the second language model used by the idea evaluation unit to evaluate the ideas; and an adjustment unit that performs adjustment processing to adjust subsequent evaluation results of the ideas by the idea evaluation unit based on the evaluation results of the ideas by the idea evaluation unit.

2. An idea generation system according to claim 1, wherein the adjustment unit performs the adjustment process by adjusting at least one of the setting state of the persona for the first language model and the evaluation criteria used by the idea evaluation unit when evaluating the idea.

3. An idea generation system according to claim 1, wherein the persona setting unit sets the persona set for the first language model and the persona set for the second language model to have different settings.

4. An idea generation system according to claim 1, wherein the idea evaluation unit obtains an evaluation value for the idea generated by the idea generation unit, and the adjustment unit performs the adjustment process when the evaluation value is less than a predetermined threshold.

5. An idea generation system according to claim 1, wherein the adjustment unit calculates the semantic distance between the idea generated by the idea generation unit and a pre-registered reference idea, and performs the adjustment process using the calculated semantic distance.

6. An idea generation system according to claim 5, further comprising a reinforcement unit that performs a reinforcement process to reinforce the idea generation by the idea generation unit when the semantic distance is less than a predetermined value.

7. An idea generation system according to claim 6, wherein the reinforcement unit performs the reinforcement process based on the evaluation result of the idea by the idea evaluation unit.

8. An idea generation system according to claim 1, comprising a reinforcement unit that performs reinforcement processing to reinforce the idea generation by the idea generation unit based on the evaluation result of the idea by the idea evaluation unit.

9. An idea generation system according to claim 1, wherein the idea evaluation unit notifies the idea generation unit of the evaluation result of the idea, and the idea generation unit feeds back the evaluation result of the idea notified by the idea evaluation unit to subsequent idea generation.

10. An idea generation system according to claim 1, comprising a context setting unit which sets a context used for parameter adjustment when the persona setting unit sets the persona.

11. An idea generation method comprising: setting personas that characterize a first language model and a second language model, respectively, using a computer; inputting domain knowledge into the first language model to generate ideas; inputting the generated ideas into the second language model to evaluate the ideas; and performing adjustment processing to adjust subsequent evaluation results of the ideas based on the evaluation results of the ideas.

Images