Idea generation support system and method

Abstract

To provide an idea assistance system and method for assisting a user to easily produce a new idea without requiring time and effort.SOLUTION: A system is configured to: receive a user idea, which is a current idea that a user has; select an idea candidate related to the user idea, as a first idea candidate, from among idea candidates accumulated in an idea candidate database that accumulates idea candidates that are existing ideas; generate a second idea candidate by converting classification of the selected first idea candidate into classification of the user idea; and present the user idea, the first idea candidate, and the second idea candidate in association with each other to the user.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present invention relates to an idea support system and method, and is suitable for application to, for example, an idea support system that presents information for supporting human ideas.

Background Art

[0002] Those engaged in research and development or planning may sometimes be required to have new ideas different from known ones. In such cases, those persons often deepen their own thinking and collect relevant information in order to obtain new ideas.

[0003] As a method for expanding ideas, there has been a method called brainstorming in which multiple people successively come up with ideas and expand ideas based on them. According to brainstorming, there is an advantage that ideas can be expanded by borrowing the ideas of others.

[0004] When trying to obtain new ideas in this way, it is often difficult to expand ideas alone, so it often takes the form of cooperation among multiple people. However, according to this method, there are problems such as it being difficult to gather the cooperation of other persons because multiple people need to gather face-to-face or online at the same time for a meeting, or there may be a difference in prerequisite knowledge among the gathered multiple persons.

[0005] Therefore, in recent years, a system has been developed that promotes accidental expansion of ideas, such as randomly presenting information, so that ideas can be expanded even by one person.

[0006] For example, Patent Document 1 discloses a method of creating a sentence that more clearly expresses an idea by schematizing the idea based on a sentence expressing the idea of a user and presenting the relationship between its constituent elements in an easy-to-understand manner.

[0007] Furthermore, Patent Document 2 discloses a method for presenting elements that would normally be included in an idea, based on the words appearing in a draft expressing the user's idea and the frequency of those words in known data. [Prior art documents] [Patent Documents]

[0008] [Patent Document 1] Japanese Patent Publication No. 2006-313567 [Patent Document 2] Japanese Patent Publication No. 2021-190090 [Overview of the project] [Problems that the invention aims to solve]

[0009] According to the technologies disclosed in Patent Documents 1 and 2, it is possible to receive a person's ideas in written form, analyze them, and present the results, or related words, thereby encouraging the person to gain insights and generate new ideas. However, if the information presented at this time is not new to the user, it will not be possible to broaden their thinking.

[0010] Therefore, by acquiring and presenting information from fields different from the user's area of ​​interest, it is possible to broaden the user's thinking by adapting ideas from other fields to their own. This method is expected to lead to a wider range of new ideas.

[0011] However, this method presents information that is outside the user's area of ​​expertise. In other words, the information presented using this method is highly specialized, making it difficult for non-experts to interpret. To utilize it as inspiration in one's own field of expertise requires understanding and consideration of the other field, which poses a problem as it demands considerable effort and time.

[0012] This invention was made in consideration of the above points, and aims to propose an idea generation support system and method that can help users easily generate new ideas without requiring effort or time. [Means for solving the problem]

[0013] To solve these problems, the present invention provides an idea support system that presents information to support new ideas, comprising: a reception unit that receives user ideas, which are the user's current ideas; an idea candidate database that stores existing idea candidates; a selection unit that selects idea candidates related to the user idea from among the idea candidates stored in the idea candidate database as first idea candidates; a classification transfer unit that generates second idea candidates by converting the classification of the first idea candidates selected by the selection unit to the classification of the user idea; and a presentation unit that presents the user idea, the first idea candidate, and the second idea candidate to the user in association with each other.

[0014] Furthermore, the present invention provides an idea support method executed by an idea support system that presents information to support new ideas, comprising: a first step of receiving a user idea, which is the user's current idea; a second step of selecting an idea candidate related to the user idea as a first idea candidate from among the idea candidates stored in an idea candidate database, which stores existing idea candidates; a third step of generating a second idea candidate by converting the classification of the selected first idea candidate to the classification of the user idea; and a fourth step of presenting the user idea, the first idea candidate, and the second idea candidate to the user in association.

[0015] According to the idea generation support system and method of the present invention, information from other fields that is relevant to the user's idea field can be presented in an easy-to-understand manner. This allows for the generation of analogies and enriches and expands the user's ideas.

Advantages of the Invention

[0016] According to the present invention, it is possible to realize an idea support system and method that can assist a user in easily generating new ideas without requiring labor and time.

Brief Description of the Drawings

[0017] [Figure 1] It is a block diagram showing the logical configuration of the idea support system according to the first to third embodiments. [Figure 2] It is a block diagram showing the hardware configuration of the idea support system. [Figure 3] It is a sequence diagram for explaining the overall flow of idea support by the idea support system. [Figure 4] It is a diagram showing an example of the screen configuration of the initial idea input screen. [Figure 5] It is a flowchart showing the processing procedure of the analogy recall information presentation process. [Figure 6] It is a chart for explaining user idea data. [Figure 7] It is a chart for explaining idea candidate data. [Figure 8] It is a diagram showing an example of the screen configuration of the analogy recall screen. [Figure 9] It is a flowchart showing the processing procedure of the analogy recall data selection process. [Figure 10] It is a conceptual diagram for explaining the analogy recall data selection process. [Figure 11] It is a diagram showing an example of the screen configuration of the classification transfer result presentation screen. [Figure 12] It is a flowchart showing the processing procedure of the classification transfer result presentation process. [Figure 13] It is a chart for explaining the data system of the idea classification conversion model. [Figure 14] It is a conceptual diagram for explaining the conversion process of idea candidate data using the idea classification conversion model. [Figure 15]This diagram illustrates the method for creating training data for an idea classification and transformation model. [Figure 16] This figure shows an example of the screen layout for the review results screen. [Figure 17] This flowchart shows the processing procedure for the classification transfer result presentation process according to the second embodiment. [Figure 18] This figure illustrates the method for creating training data for an idea classification transformation model according to the second embodiment. [Figure 19] This figure shows an example of the screen configuration of the analogy recall screen according to the third embodiment. [Figure 20] This figure shows an example of the screen configuration of the classification transfer result presentation screen according to the third embodiment. [Figure 21] This figure shows an example of the configuration of training data for the idea classification transformation model according to the third embodiment. [Modes for carrying out the invention]

[0018] An embodiment of the present invention will be described in detail below with reference to the drawings.

[0019] (1) First embodiment (1-1) Configuration of the idea generation support system according to this embodiment Figure 1 shows an example of the logical configuration of the idea generation support system 1 according to this embodiment. This idea generation support system 1 comprises a user idea reception unit 3 that receives input from a user 2, an idea generation support information presentation unit 4 that presents information to the user 2, an analogy recall data selection unit 5 that selects information that evokes analogies to present to the user 2, an idea candidate database 6 that stores data of idea candidates described later (hereinafter referred to as idea candidate data), which are examples of ideas being expressed, an idea candidate classification transfer unit 7 that generates an idea that has been reinterpreted as an idea in a different field, and an idea classification conversion model database 8 that stores machine learning model data for reinterpreting ideas.

[0020] Figure 2 shows an example of the physical configuration of the idea generation support system 1. As shown in Figure 2, the idea generation support system 1 consists of a general-purpose computer device equipped with a processor 10 with computing power, a DRAM (Dynamic Random Access Memory) 11 which is a volatile temporary storage medium that can be read and written at high speed, a storage device 12 which provides a persistent storage area using an HDD (Hard Disk Drive) or flash memory, an input device 13 consisting of a mouse or keyboard for the user to operate the idea generation support system 1, a monitor 14 for showing the operation to the user, and a communication interface 15 such as a serial port for communicating with the outside.

[0021] The user idea reception unit 3, idea support information presentation unit 4, analogy recall data selection unit 5, and idea candidate classification transfer unit 7 in Figure 1 are functional units that are realized when the processor 10 executes a program stored in the storage device 12. The idea candidate database 6 and idea classification conversion model database 8 are also realized when the processor 10 executes a program that stores data in the storage device 12.

[0022] (1-2) Idea generation support processing (1-2-1) Flow of idea generation support using an idea generation support system Figure 3 shows the flow of idea generation support by the idea generation support system. Figure 3 focuses on the idea generation support process that the idea generation support system 1 performs when supporting user 2's ideas, and shows the interactions between user 2 and the idea generation support system 1, as well as the operations within each element of the idea generation support system 1.

[0023] As shown in Figure 3, when User 2 activates the Idea Generation Support System 1 (S1), the Idea Generation Support System 1 first displays a predetermined initial draft input screen on the monitor 14 (Figure 2) for User 2 to input an initial draft (hereinafter, this will be referred to as "presentation" as appropriate) (S2). Then, when User 2 inputs an initial draft into this initial draft input screen (S3), the Idea Generation Support System 1 performs the analogy recall information presentation process described later, presenting User 2 with one or more idea candidates to encourage a leap in thinking, thereby guiding User 2's thinking into a new realm (S4). Details of the analogy recall information presentation process will be described later.

[0024] Then, when User 2 selects a desired idea candidate from the idea support system 1 presented (S5), the idea support system 1 executes a classification transfer result presentation process (S6) which generates and presents a text in which the idea candidate selected by User 2 has been reclassified according to the classification of the initial text. Details of the classification transfer result presentation process will be described later.

[0025] User 2 can expand on the presented draft by updating it (S7). When the presented draft is updated by User 2, the idea generation support system 1 again performs analogy recall information presentation processing to present one or more idea candidates to User 2 to encourage further leaps in thinking, guiding User 2's thinking into new areas (S8).

[0026] After this, steps S5 to S8 are repeated in the same manner, and finally, when user 2 determines that they have made sufficient changes in their thinking and performs the predetermined termination operation (S9), the idea generation support system 1 confirms the final draft as the result of user 2's consideration (S10). This completes the idea generation support provided by the idea generation support system 1. The following describes this series of processes in detail.

[0027] (1-2-2) Initial draft submission screen Figure 4 shows an example of the configuration of the initial idea input screen 20, which is presented to user 2 as the initial draft submission screen in step S2 described above. The initial idea input screen 20 is provided with an initial idea input text box 21, which allows user 2 to input their ideas as text. In the following, the term "text" includes "sentence".

[0028] Furthermore, the initial idea input text box 21 on the initial idea input screen 20 can contain not only text but also diagrams representing the user's initial ideas (hereinafter referred to as idea content representation diagrams). If an idea content representation diagram is embedded, the data of that diagram is included in the user idea data 25 as an idea content representation diagram 25C, which will be described later with respect to Figure 6.

[0029] Furthermore, on the initial idea input screen 20, it is possible to obtain data from an external source, assuming that there is another system that assists User 2 in the process of summarizing the content of their ideas into written form. In the example in Figure 4, by entering a URL (Uniform Resource Locator) on the internet into the URL input text box 23 and clicking the import button 24, information such as text and figures is automatically obtained from the web page of that URL, and the text and figures based on this obtained information can be displayed in the initial idea input text box 21.

[0030] (1-2-3) Analogy recall information presentation processing Figure 5 shows the flow of the analogy recall information presentation process executed in step S4 of Figure 3 in the idea generation support system 1. In this analogy recall information presentation process, the user idea reception unit 3 first creates user idea data 25 (Figure 6) based on user input to the initial idea input screen 20 (Figure 4) as described above, and sends the created user idea data 25 to the analogy recall data selection unit 5 (S20).

[0031] Figure 6 shows an example of the structure of user idea data 25. As shown in Figure 6, user idea data 25 is information that includes the idea date and time 25A, the idea content expression string 25B, and the idea content expression diagram 25C. The idea date and time 25A is information that represents the date and time when the user idea data 25 was generated, and the idea content expression string 25B is information that represents a string of text (hereinafter referred to as the idea content expression string) that expresses the initial idea content of user 2, which was entered by user 2 in the initial idea input text box 21 of the initial idea input screen 20.

[0032] The analogy recall data selection unit 5 performs an analogy recall data selection process (S21) to select idea candidate data 26 (Figure 7) of idea candidates that are thought to contribute to the analogy from among the idea candidate data stored in the idea candidate database 6 based on the user idea data 25 sent from the user idea reception unit 3. Details of the analogy recall data selection process will be described later. The processing result of the analogy recall data selection process is a set of selected idea candidate data 26.

[0033] Figure 7 shows an example of the structure of such idea candidate data 26. Idea candidate data 26 consists of candidate ID 26A, attribute information 26B, classification code 26C, idea representation data 26D, and representation vector data 26E.

[0034] Candidate ID 26A is a unique identifier assigned to the corresponding idea candidate data 26. Attribute information 26B is supplementary information about the idea candidate data 26, such as the author, and classification code 26C represents a code value that indicates the classification of the idea candidate based on the idea candidate data 26, for example, "01" for chemistry, "02" for electricity, and "03" for information.

[0035] Furthermore, the idea expression data 26D is data of a string of text (idea content expression string) that expresses the content of an idea candidate based on the corresponding idea candidate data 26, and the expression vector data 26E represents the expression vector data obtained by applying the conversion process described later to the idea expression data 26D.

[0036] The idea candidate database 6 can be prepared, for example, by obtaining and processing patent publications. Furthermore, the idea candidate database 6 may be periodically updated with additional data by crawling the database regularly.

[0037] The idea candidate data 26 obtained as a result of the analogy recall data selection process is used in combination with user ideation data 25 (Figure 6) to generate the analogy recall screen 30, which will be described later with respect to Figure 8 (S22). The generated analogy recall screen 30 is presented to user 2 (S23).

[0038] Figure 8 shows an example of the configuration of such an analogy recall screen 30. As shown in Figure 8, the analogy recall screen 30 is configured to include a current idea display area 31, an analogy recall information display area 32, a user idea data display text box 33, a next button 34, and an exit button 35.

[0039] Currently, the idea display area 31 displays an idea content expression string based on the idea content expression string 25B (Figure 6) of the corresponding user idea data 25 (Figure 6), and an idea content expression diagram based on the idea content expression diagram 25C (Figure 6). In this case, such idea content expression strings and idea content expression diagrams are displayed in a form that cannot be edited.

[0040] Furthermore, the analogy recall information display area 32 displays the content of each idea candidate data 26 (Figure 7) selected by the analogy recall data selection process described above, to serve as a reference when editing the idea content expression string or idea content expression diagram currently displayed in the idea display area 31.

[0041] Specifically, the content of each of these candidate idea data 26 is displayed, including attribute information 26B (Figure 7) and idea expression data 26D (Figure 7) contained in the candidate idea data 26. If the attribute information 26B includes a URL, the text and image information may be obtained from the web page at that URL and displayed in the analogy recall information display field 32.

[0042] Furthermore, the analogy recall information display area 32 is provided with selection buttons 36 corresponding to each of the idea candidate data 26 selected by the analogy recall data selection process. By clicking one of these selection buttons 36, the user can select the idea candidate data 26 associated with that button 36. The idea expression data 26D based on the selected idea candidate data 26, combined with the idea content expression string and idea content expression diagram based on the user idea data 25 currently displayed in the idea display area 31, can then be displayed in the user idea data display text box 33.

[0043] In this case, the user can edit the displayed text in the user idea data display text box 33, allowing user 2 to further develop their ideas and express them in written form, using the displayed text as a reference.

[0044] On the other hand, when the user clicks the "Next" button 34 on the analogy recall screen 30, new user idea data 25 is generated, with the date and time at that moment being the idea date and time, and the strings and figures stored in the user idea data display text box 33 at that time being used as the idea content expression string and idea content expression figure, respectively. The analogy recall information presentation process is then executed again based on the generated user idea data 25. This corresponds to step S8 of the process described above for Figure 3.

[0045] Furthermore, clicking the "End" button 35 on the analogy recall screen 30 confirms the results of the examination. Clicking the "End" button 35 corresponds to step S9 of the process described above in Figure 3, and the process of confirming the examination results in this way corresponds to step S10 in Figure 3.

[0046] Next, we will explain the specific processing details of the analogy recall data selection process performed by the analogy recall data selection unit 5 in step S21 of Figure 5.

[0047] As shown in Figure 9, the analogy recall data selection unit 5 first extracts the idea content expression string 25B (Figure 6) from the user idea reception unit 3 (S30), and then converts the extracted idea content expression string 25B into vector data (S31).

[0048] This vector data is a sequence of numbers with a predetermined length, for example, (0.1, 2.1, 1.2, 0.1, 0.6). If the idea content expression strings 25B are similar, the vector data will also be similar. That is, the idea content expression strings 25B are converted into vector data where the distance between vectors is small compared to similar idea content expression strings 25B. This conversion process will be explained with reference to Figure 10.

[0049] A known embedding process can be applied to convert strings into vector data. Generally, embedding refers to the process of converting data into vector data while preserving as much of the mathematical structure of that data as possible. In this embodiment, a method is applied in which the similarity between sentences is reflected in the distance between the vector data.

[0050] Methods for embedding strings include using neural networks such as Seq2Seq and Word2Vec. In this embodiment, the target string (here, an idea content expression string) 40 is divided into word units and then converted into a vector sequence using a method called one-hot encoding.

[0051] In one-hot encoding, each component of a numerical sequence corresponds to a word obtained through the above-mentioned splitting process. For each word, the corresponding component is converted to "1" while all others are "0". As a result, a vector sequence 41 is generated in which only one component is "1" and the rest are "0".

[0052] Since the length of column 41 varies depending on the input string, it is input into a neural network 42 that incorporates a recurrent neural network or similar to handle variable-length data. This neural network 42 has been pre-trained using a large number of sentences. This training can be performed using any known method for training a neural network with embedding capabilities. Specifically, methods such as generative adversarial networks, variational autoencoders, and Transformers, which use attention mechanisms, can be applied.

[0053] In these embedding methods, the similarity of the input sentences correlates with the distance between the output vectors 43, but the perspective of sentence similarity used differs depending on the method. The type of embedding to use can be selected according to the purpose of idea generation support provided by this idea generation support system 1.

[0054] Using the vector data 43 generated in this manner, idea candidate data 26 (Figure 6) is obtained from the idea candidate database 6 (S32). This idea candidate data 26 includes the representation vector data 26E shown in Figure 7.

[0055] Therefore, the distance between the vector data 43 generated as described above and the representation vector data 26E of each idea candidate data 26 stored in the idea candidate database 6 is calculated, and a predetermined number of idea candidate data 26, such as 10, are selected in ascending order of distance (S33). For example, the Euclidean distance between the two vectors (the square root of the sum of the squared differences of each component) is calculated, and idea candidate data 26 whose calculated Euclidean distance is less than or equal to a predetermined threshold are selected.

[0056] Note that the distance between the two vectors may be any method other than the Euclidean distance, as long as it can evaluate similarity. Also, instead of specifying the number of idea candidate data 26 to be selected, the idea candidate data 26 may be selected by selecting all idea candidate data 26 whose distance is smaller than a predetermined threshold.

[0057] Based on the above, it is possible to present information to user 2 that will stimulate their thinking.

[0058] (1-2-4) Classification Transfer Result Presentation Process Next, the classification transfer result presentation process, which is performed by the idea candidate classification transfer unit 7 (Figure 1) in step S6 of Figure 3, will be explained in detail. The classification transfer result presentation process is performed when a selection button 36 (Figure 8) associated with any of the idea candidate data 26 whose idea expression data 26D (Figure 7) is displayed in the analogy recall information display field 32 on the analogy recall screen 30 (Figure 8) is clicked.

[0059] This classification transfer result presentation process creates a text that combines the content of the user idea data 25 entered in the user idea data display text box 33 of the analogy recall screen 30 described above in Figure 8 with the idea candidate data 26 selected by User 2 by clicking the selection button 36, and presents it to User 2.

[0060] Figure 11 shows an example of the configuration of the classification transfer result presentation screen 50 presented by the classification transfer result presentation process. This classification transfer result presentation screen 50 is configured to include a current idea display field 51, a selected idea candidate data display field 52, a user idea data display text box 53, a converted image list display field 54, a regenerate button 55, and a confirm button 56.

[0061] Currently, the idea display area 51 displays an idea content expression string based on the idea content expression string 25B (Figure 6) of the corresponding user idea data 25, and an idea content expression diagram based on the idea content expression diagram 25C (Figure 6). In this case, such idea content expression strings and idea content expression diagrams are displayed in the idea display area 51 in a form that cannot be edited.

[0062] Furthermore, the selected idea candidate data display area 52 displays strings and images representing the content of the idea candidate data (hereinafter referred to as selected idea candidate data) 26 selected by clicking the selection button 36 (Figure 8) in the analogy recall information display area 32 (Figure 8) of the analogy recall screen 30 described above.

[0063] The content of the candidate idea data 26 displayed in the candidate idea data display area 52 may be in a different field from the user idea data 25. However, a draft text that converts the candidate idea data 26 to be in the same field as the user idea data 25 is displayed in the user idea data display text box 53. This draft text is displayed in an editable format, and user 2 can freely edit this conversion result (draft text).

[0064] Furthermore, if the selected idea candidate data 26 includes an image, the content of that image is also transformed so that it falls under the same classification as the user idea data. The results of the transformation in several patterns (hereinafter referred to as the transformed images 54A) are displayed as a list in the transformed image list display area 54. These transformed images 54A can be copied and used by clicking the corresponding copy button 54B displayed in the transformed image list display area 54.

[0065] Furthermore, on the classification transfer result presentation screen 50, if the draft text displayed in the user idea data display text box 53 is not favorable to user 2, the user can click the regenerate button 55 to regenerate the draft text and display it in the user idea data display text box 53.

[0066] Then, on the classification transfer result presentation screen 50, if user 2 is satisfied with the draft text displayed in the user idea data display text box 53 after repeated regeneration and editing, they can click the confirm button 56 to confirm that draft text as an idea content expression string or idea content expression diagram to be used as user idea data 25 in the next analogy recall information presentation process. This completes the classification transfer result presentation process.

[0067] Figure 12 shows the specific processing flow of the classification transfer result presentation process performed by the idea candidate classification transfer unit 7 (Figure 1) in step S6 of Figure 3. This classification transfer result presentation process starts when a selection button 36, which is associated with one of the idea candidate data 26 (Figure 7) whose content is displayed in the analogy recall information display field 32 of the analogy recall screen 30 described above in Figure 8, is clicked.

[0068] The idea candidate classification transfer unit 7 then acquires the user idea data 25 created by the user idea reception unit 3 (Figure 1) based on the information of the idea content expression string and idea content expression diagram of the user idea data 25 (Figure 6) displayed in the user idea data display text box 33 of the analogy recall screen 30, and also acquires the selected idea candidate data 26 (S40).

[0069] Next, the idea candidate classification transfer unit 7 estimates the classification of the user idea data 25 acquired in step S40 (S41). This estimation can be performed, for example, by assigning a predetermined classification score to the words appearing in the idea content expression string and determining the classification according to the sum of these scores.

[0070] Alternatively, for Figure 10, the idea content expression strings of the user idea data 25 may be converted into vector data using the method described above, and then classified using a known machine learning discriminant analysis method such as a decision tree or a support vector machine. Or, for Figure 10, the neural network 42 described above may be modified to enable classification, and accurate estimation may be made based on pre-training.

[0071] Furthermore, a separate user interface could be provided for users to confirm the classification, allowing for manual classification. While this would increase the workload for humans, it would result in higher classification accuracy.

[0072] Furthermore, the above processing and operations should be carried out using a classification system identical to or consistent with that shown in Figure 7.

[0073] Next, the Idea Candidate Classification Transfer Unit 7 selects a conversion model (hereinafter referred to as the Idea Classification Conversion Model) for performing the classification conversion of ideas based on the classification described above (S42). This selection of the Idea Classification Conversion Model is performed by selecting an appropriate Idea Classification Conversion Model from among the various Idea Classification Conversion Models stored in the Idea Classification Conversion Model Database 8 (Figure 1).

[0074] The Idea Classification Conversion Model Database 8, as shown in Figure 13, is a database in which, for each Idea Classification Conversion Model, which is data from the trained machine learning model described above, the source classification code and the converted classification code are stored in correspondence. As described above, the purpose is to convert the Idea Candidate Data 26 into a classification for the User Idea Data 25, so an Idea Classification Conversion Model is selected in which the classification code of the Idea Candidate Data 26 corresponds to the source classification, and the estimated classification result for the User Idea Data 25 corresponds to the converted classification.

[0075] Finally, the idea candidate data 26 is transformed using this idea classification transformation model, and the result is returned to the idea generation support information presentation unit 4 (S43). Figure 14 shows a simulated transformation, illustrating a schematic representation of the neural network 62 that returns sequence data 64 to the idea generation support information presentation unit 4 for sequence data 60. The input sequence data 60 is first transformed into a sequence of One-Hot data 61, and then input into the neural network 62. This yields another sequence of One-Hot data 63, which can then be transformed into sequence data 64 consisting of strings. This allows the system to handle the well-known Text-to-Text problem, which takes a string as input and outputs another string.

[0076] In this embodiment, the input string (sequence data 60) corresponds to the classification code of the source classification in Figure 13, and the output string (sequence data 64) is transformed to correspond to the classification code of the post-transformation classification in Figure 13. To train such an idea classification transformation model, it is necessary to prepare a large number of pairs of data resulting from the reinterpretation of strings across domains.

[0077] However, since it is not easy to prepare such data, a method for creating such data in a simulated manner may be applied. For example, as shown in Figure 15, by extracting nouns from the sentences 70 of the existing idea candidate data 26 and randomly replacing those nouns with nouns from other fields, it is possible to create sentence data 71 that simulates conversions between classifications.

[0078] Furthermore, by combining the sentences 70 of these existing idea candidate data 26 with sentences 71 obtained by replacing the nouns in sentences 70 with nouns from other fields, training data 72 can be created to simulate conversions between classifications.

[0079] When the idea candidate data 26 is classified using an idea classification transformation model trained with such training data 72, the sentences of the idea candidate data 26 after classification transformation are transformed into sentences in which nouns are replaced with nouns from other fields as described above, while maintaining the original syntax.

[0080] With this method, it is not possible to build the idea classification transformation model database 8 unless all combinations of transformations between classifications are trained in the same way. However, the prior training process can be reduced by standardizing the classification of the inputs.

[0081] In this case, one could repeatedly perform the word substitution process shown in Figure 15 for various fields, and then construct a dataset that allows for training where text from any field is converted into text with the corresponding classification code (the classification code for the converted classification in Figure 13).

[0082] Here, any method that can convert the input data into data from another specific field is applicable. For example, the word substitution process described above can be used directly for Figure 15.

[0083] Furthermore, although this embodiment has only described the process for text strings, it is also possible to perform the same process on images in a figure. In the case of figures, the same process can be performed by implementing an image style transfer method based on a known GAN (Generative Adversarial Network) instead of Text-to-Text.

[0084] Similarly, any form of expression is acceptable as long as it is understandable to humans and can express human thought, such as adding a text-to-speech function to expand ideas through natural dialogue, or using videos for expression. Multiple types of expression can also be combined.

[0085] Figure 16 shows an example of the screen configuration of the review results screen 80 presented to user 2 after the review results are confirmed in step S10 of Figure 3 (after the confirmation button 56 on the classification transfer result presentation screen 50 described above for Figure 11 is clicked). This review results screen 80 is composed of an intermediate text list display area 81, a selected text display area 82, a download button 83, and a full download button 84.

[0086] Then, in the intermediate draft list display area 81, intermediate draft display fields 85, which show parts of the drafts created during the review process so far, are displayed in the order in which the drafts were created. Although Figure 16 illustrates the case where only three intermediate draft display fields 85 are displayed, if there are four or more drafts created during the review process so far, each intermediate draft display field 85 can be moved horizontally or vertically all at once by a predetermined operation. By doing so, other intermediate draft display fields 85 that virtually exist horizontally in the intermediate draft list display area 81 on the review results screen 80 can be moved into the intermediate draft list display area 81 and displayed in a visible state.

[0087] Furthermore, on the review results screen 80, by clicking on one of the drafts displayed in each draft display field 85 within the draft list display area 81, the draft display field 85 in which that draft is displayed will be highlighted (for example, by making the border of that draft display field 85 thicker). In this state, by clicking the download button 83, the entirety of the draft partially displayed in that draft display field 85 will be displayed in the selected draft display field 82.

[0088] Furthermore, on the review results screen 80, clicking the "Download All" button 84 allows users to download all drafts created during the review process so far and display their details in the selected draft display area 82. At this time, the drafts can be downloaded including the record of step S42 in Figure 12, and the record can be made more convenient by including attribute information 26B (Figure 7) so that the source idea candidate data 26 can be accessed later.

[0089] (1-3) Effects of this embodiment As described above, in the idea generation support system 1 of this embodiment, when a person has an initial idea and is gathering information for the purpose of developing it, information on similar cases across fields is presented in an easy-to-understand manner. Therefore, with this idea generation support system 1, it becomes possible to generate analogies and broaden ideas richly, and it is possible to support users in easily generating new ideas without requiring effort or time.

[0090] (2) Second embodiment In Figure 1, 90 as a whole represents the idea generation support system according to the second embodiment. This idea generation support system 90 is configured similarly to the idea generation support system 1 of the first embodiment, except that the processing content of the classification transfer result presentation process executed by the idea candidate classification transfer unit 91 in step S6 of Figure 3 differs from that of the first embodiment.

[0091] Figure 17 shows the specific processing details of the classification transfer result presentation process performed by the idea candidate classification transfer unit 91 of this embodiment. This classification transfer result presentation process differs from the classification transfer result presentation process of the first embodiment described above in Figure 12 in that it does not perform classification estimation on the user idea data in step S51. Steps S50, S52, and S53 are processed in the same way as steps S40, S42, and S43 in Figure 12, respectively.

[0092] In the classification transfer result presentation process of this embodiment, instead of performing classification estimation on the user idea data 25 (Figure 6) in step S51, the domain of the idea candidate data 26 (Figure 7) is transformed using the user idea data 25 directly. Therefore, in this classification transfer result process, in step S51, a string is generated that combines both the user idea data 25 and the idea candidate data 26, and the generated string is input into an appropriate idea classification transformation model in the idea classification transformation model database 8 (Figure 1) to transform the domain. This eliminates the need to prepare a large number of idea classification transformation models, and as a result, it becomes possible to handle the process by preparing only a single idea classification transformation model.

[0093] Figure 18 illustrates the training method of the idea classification and transformation model in this embodiment. The idea classification and transformation model used in this embodiment can be any model that corresponds to text-to-text, similar to the first embodiment, but the training data is different. In this embodiment, the training data consists of 100 to 102 combinations of three sentences (hereinafter referred to as the first to third sentences, respectively).

[0094] The first sentence 100 is a sentence based on idea expression data 26D (Figure 7) of a certain idea candidate data 26 (Figure 7). A second sentence 101 is created in the same way as in the first embodiment by replacing the nouns in this sentence with those from another field (in this case, the field of user idea data 25). Furthermore, another idea candidate data 26 belonging to the same classification as the idea candidate data 26 that is the first sentence 100 is obtained from the idea candidate database 6 (Figure 1), and a sentence based on its idea expression data 26D is used as the third sentence 102. These first to third sentences 100 to 102 are then arranged in the order of the third sentence 102, a delimiter ("&" in Figure 18), the second sentence 101, a delimiter representing the input / output boundary of Text-to-Text (">" in Figure 18), and the first sentence 100 to form the training data 103.

[0095] This allows the machine learning model to adapt to the Text-to-Text problem, which involves inputting a string combining example sentences from a predetermined classification with example sentences from other fields separated by a delimiter, and outputting a string that has been transformed by classifying the latter half of the input sentence. When the idea classification and transformation model, which has been trained in this way, is given data consisting of the user idea data 25 and the idea candidate data 26 side by side, it will output a string in which the latter half, i.e., the sentence based on the idea expression data 26D of the idea candidate data 26, has been transformed into a string in the field of the user idea data 25.

[0096] Furthermore, any method that can produce similar results can be applied. For example, by applying a classification bias to the computation process of neural network 62 in Figure 14, it is expected that similar results can be obtained without using the data in Figure 18.

[0097] According to the idea generation support system 90 of the second embodiment described above, since it is only necessary to prepare an idea classification transformation model for classifying a single idea, classification transformation can be performed even when the amount of training data is smaller compared to the idea generation support system 1 of the first embodiment.

[0098] (3) Third Embodiment This embodiment describes idea generation support for chemical formulas. The objective of this embodiment is to find a better structural formula for an organic compound by incorporating the characteristics of organic compounds used in other applications when searching for an organic compound for a specific application. The configuration of the idea generation support system in this embodiment is the same as that of the idea generation support system 1 in the first embodiment and the idea generation support system 90 in the second embodiment, so a detailed explanation is omitted here.

[0099] The structural formulas of organic compounds can be represented as strings in known formats such as SMILES (Simplified Molecular Input Line Entry Syntax). Therefore, the configuration for handling strings described in the first and second embodiments is almost identical, except that the strings being handled represent structural formulas, and the strings should be displayed in the form of structural formulas on the screen.

[0100] Figure 19 shows an example of the analogy recall screen 110 presented by the idea support information presentation unit 4 in step S23 of the analogy recall information presentation process described above with respect to Figure 5. This analogy recall screen 110 is composed of a current idea display field 111, an analogy recall information display field 112, a current idea editing field 113, and a confirm button 114.

[0101] Currently, the idea display area 111 shows a chemical formula based on the current user idea data 25 (hereinafter referred to as the first chemical formula) as a diagram. In addition, the analogy recall information display area 112 displays several chemical formulas (hereinafter referred to as the second chemical formula) that are similar to the first chemical formula, based on the idea expression data 26D (Figure 7) of any of the idea candidate data 26 (Figure 7), as analogy recall information. This allows the user to expand their ideas by referring to the second chemical formulas displayed in the analogy recall information display area 112.

[0102] In this case, for each chemical formula displayed in the analogy recall information display area 112, information included in the attribute information 26B (Figure 7) of the idea candidate data 26 described above, such as the Japanese name and usage, can be displayed together with the formula in Figure 7, making the information easier for user 2 to understand.

[0103] Furthermore, the chemical formula representing the current User 2 idea, which is currently displayed in the Idea Display area 111, is also displayed in the Idea Editing area 113, and it is possible to modify this chemical formula in the Idea Editing area 113.

[0104] In the analogy recall screen 110, a selection candidate button 115 is provided in the analogy recall information display area 112, corresponding to each chemical formula displayed in the analogy recall information display area 112. By clicking the selection button 105 corresponding to the desired chemical formula, the user can select that chemical formula, thereby causing the aforementioned field transfer result presentation process to be executed by the idea candidate classification transfer unit 7 (Figure 1). The processing results of this field transfer result presentation process are then displayed on the classification transfer result presentation screen 120 shown in Figure 20 and presented to the user 2.

[0105] As shown in Figure 20, the classification transfer result presentation screen 120 is configured to include a current idea display field 121, a selected idea candidate data display field 122, a user idea data display text box 123, a regenerate button 124, and a confirm button 125.

[0106] Currently, the idea display area 121 displays the chemical formula created from the string of characters in the user idea data 25, and the selected idea candidate data display area 122 displays the chemical formula selected on the analogy recall screen 110 as described above.

[0107] Furthermore, the user idea data display text box 123 displays, in an editable format, the result of converting the chemical formula displayed in the selected idea candidate data display field 122 to the same classification as the user idea data 25, which displays the chemical formula corresponding to the current idea display field 121. Here, classification refers to conversions between applications (e.g., topical pharmaceuticals, food additives, heat insulating materials) or conversions between compound species (e.g., Si-based compounds, monomers, polymers).

[0108] Furthermore, on the classification transfer result presentation screen 120, if the chemical formula displayed in the user-generated data display text box 123 is not favorable to user 2, the user can click the regenerate button 124 to regenerate the chemical formula and display it in the user-generated data display text box 123.

[0109] Then, on the classification transfer result presentation screen 120, if user 2 is satisfied with the chemical formula displayed in the user idea data display text box 123 after repeated regeneration and editing, they can click the confirm button 125 to confirm that chemical formula as an idea content expression string or idea content expression diagram to be used as user idea data 25 in the next analogy recall information presentation process.

[0110] Figure 21 shows an example of training data 130 for training the idea classification transformation model stored in the idea classification transformation model database 8 (Figure 1) in the third embodiment. This training data 130 is basically equivalent to the training data 103 in the second embodiment described above with respect to Figure 18, the only difference being that the strings used are not natural language but strings representing chemical formulas.

[0111] Furthermore, mathematical formulas can be treated as strings using well-known expression formats such as TeX and MathML (Mathematical Markup Language), and can therefore be handled in the same manner as in the first and second embodiments. In addition, other information that can be represented as strings, such as program code, markup data such as HTML (Hyper Text Markup Language), sequences of drawing commands for figures, network structures such as neural networks, and diagrams such as UML (Unified Modeling Language) class diagrams, can be handled in the same way.

[0112] According to this embodiment of the idea generation support system, it is expected that the system will have the effect of expanding the range of ideas to include a wider variety of types.

[0113] (4) Other embodiments In the first to third embodiments described above, various functions of the idea generation support system 1,90, such as the user idea reception unit 3, the idea generation support information presentation unit 4, the analogy recall data selection unit 5, the idea candidate database 6, the idea candidate classification transfer unit 7,91, and the idea classification conversion model database 8, were described in a case where they were mounted on a single computer device. However, the present invention is not limited to this, and these functions may be distributed and arranged on multiple computer devices that constitute a distributed computing system.

[0114] Furthermore, in the first embodiment described above, as shown in Figure 15, nouns are extracted from the sentences 70 of the existing idea candidate data 26, and these nouns are randomly replaced with nouns from other fields to create sentence data 71 that simulates conversion between classifications. Training data 72 is then created using the created sentence data 71, and the classification of the idea candidate data 26 is converted using an idea classification conversion model trained with the created training data 72. However, the present invention is not limited to this, and for example, the idea candidate classification transfer unit 7 may convert the classification of the existing idea candidate data 26 to the classification of the user idea data 25 by replacing words such as nouns with words from the classification of the user idea data 25 while maintaining the syntax of the sentences of the existing idea candidate data 26. [Industrial applicability]

[0115] This invention can be widely applied to idea generation support systems of various configurations that present information to support new ideas. [Explanation of symbols]

[0116] 1, 90... Idea generation support system, 2... User, 3... User idea reception unit, 4... Idea generation support information presentation unit, 5... Analogy recall data selection unit, 6... Idea candidate database, 7, 91... Idea candidate classification transfer unit, 8... Idea classification conversion model database, 20... Initial idea input screen, 25... User idea data, 26... Idea candidate data, 30, 110... Analogy recall screen, 50, 120... Classification transfer result presentation screen, 72, 103, 130... Training data, 80... Review results screen.

Claims

1. An idea support system that presents information to support new ideas, A reception desk that accepts user ideas, which are the current ideas held by users, An idea candidate database that has accumulated existing idea candidates, A selection unit that selects from the idea candidates stored in the idea candidate database an idea candidate that is related to the user idea as a first idea candidate, A classification transfer unit generates a second idea candidate by converting the classification of the first idea candidate selected by the selection unit into the classification of user ideas, A presentation unit that presents the user's idea, the first idea candidate, and the second idea candidate to the user in association with each other. An idea generation support system characterized by having the following features.

2. The system further comprises a transformation model database containing one or more transformation models, In the aforementioned database of idea candidates, each idea candidate's data includes classification information representing the classification of that idea candidate. The aforementioned classification transfer unit is The system estimates the user-generated idea classification, and based on the estimated user-generated idea classification and the classification of the first idea candidate, it uses the corresponding conversion model from the conversion model database to generate the second idea candidate, which is obtained by converting the classification of the first idea candidate to the user-generated idea classification. The idea generation support system according to feature 1.

3. The aforementioned candidate ideas are expressed in sentences or texts, The aforementioned conversion model is, The classification of the first idea candidate is converted to the user-generated classification by replacing words in the first idea candidate sentence or text with words from the user-generated classification while maintaining the sentence structure of the first idea candidate sentence or text. The idea generation support system according to feature 2.

4. The aforementioned conversion model is, This is a transformation model trained using training data to transform idea candidates of the same classification as the first idea candidate into the same classification as user-generated ideas. The idea generation support system according to claim 2 or 3.

5. The aforementioned candidate ideas are expressed in sentences or texts, The aforementioned classification transfer unit is Using a transformation model trained with training data consisting of a first sentence or text expressing the content of the first idea candidate, a second sentence or text in which the words of the first idea candidate are replaced with the user-generated words, and a third sentence or text belonging to the same classification as the idea candidate, the second idea candidate is generated by transforming the classification of the first idea candidate into the user-generated classification. The idea generation support system according to feature 1.

6. An idea generation support method implemented by an idea generation support system that presents information to support new ideas, The first step is to accept user ideas, which are the current ideas held by the users, A second step is to select an idea candidate that is related to the user's idea from among the idea candidates stored in the idea candidate database, which is a database of existing idea candidates, as the first idea candidate, A third step is to generate a second set of idea candidates by converting the classification of the selected first set of idea candidates into the classification of user ideas, A fourth step involves presenting the user's idea, the first candidate idea, and the second candidate idea to the user in correspondence. An idea generation support method characterized by comprising the following features.

7. The aforementioned idea generation support system is It has a conversion model database in which one or more conversion models are stored, In the aforementioned database of idea candidates, each idea candidate's data includes classification information representing the classification of that idea candidate. In step 3 described above, the idea generation support system, Estimate the classification of the user ideas mentioned above, Based on the estimated user idea classification and the classification of the first idea candidate, the system generates a second idea candidate by converting the classification of the first idea candidate to the user idea classification, using the corresponding conversion model from the conversion model database. The idea generation support method described in feature 6.

8. The aforementioned candidate ideas are expressed in sentences or texts, The aforementioned conversion model is, The classification of the first idea candidate is converted to the user-generated classification by replacing words in the first idea candidate sentence or text with words from the user-generated classification while maintaining the sentence structure of the first idea candidate sentence or text. The idea generation support method according to feature 7.

9. The aforementioned conversion model is, This is a transformation model trained using training data to transform idea candidates of the same classification as the first idea candidate into the same classification as user-generated ideas. The idea generation support method according to claim 7 or 8.

10. The aforementioned candidate ideas are expressed in sentences or texts, In step 3 described above, the idea generation support system, Using a transformation model trained with training data consisting of a first sentence or text expressing the content of the first idea candidate, a second sentence or text in which the words of the first idea candidate are replaced with the user-generated words, and a third sentence or text belonging to the same classification as the idea candidate, the second idea candidate is generated by transforming the classification of the first idea candidate into the user-generated classification. The idea generation support method described in feature 6.

Images