Information processing system and program

Abstract

An information processing system is provided with at least one processor that acquires a combination of information identifying each of raw materials received from a user and the amount of each of the raw materials. Inputting at least one of a chemical fingerprint, an SMILES character string, chemical graph structure data, a product name or a substance name corresponding to each of the raw materials, and the amount of each of the raw materials into a first machine learning model; inputting, into a second machine learning model, at least one of a chemical fingerprint, an SMILES character string, chemical graph structure data, a product name or a substance name corresponding to each of the raw materials, and a numerical value group based on each amount of the raw materials; a predicted value for a physical property value of a physical property name to be predicted for a composition comprising each of the raw materials is obtained. The first machine learning model uses, as inputs, at least one of a chemical fingerprint, an SMILES character string, chemical graph structure data, a product name or a substance name corresponding to each of the raw materials, and the amount of each of the raw materials. The parameters of the learning dataset, which outputs the physical property value of the physical property name of the prediction object, are adjusted, so that the output model can be predicted according to the input. The second machine learning model uses, as inputs, at least one of a chemical fingerprint, an SMILES character string, chemical graph structure data, a product name or a substance name corresponding to each of the raw materials, and a numerical value group converted using the amount of each of the raw materials. The parameters of the learning data set which outputs the physical property value of the physical property name of the prediction object are adjusted, so that the output model can be predicted according to the input.

Description

Technical Field

[0001] The present invention relates to an information processing system and a program. Background Art

[0002] A method for predicting the physical property data of a composition has been developed. For example, Patent Document 1 describes the following: using physical property data related to multiple vulcanized rubber compositions, the identification names of the raw materials in the vulcanized rubber compositions, the mixing ratio of the raw materials, and information on the processing conditions, a computer prediction module performs machine learning on the physical property data. Patent Document 1 also describes the following: In the prediction module that has performed machine learning, the physical property data of the predicted vulcanized rubber composition is predicted using the names of the raw materials that constitute the unvulcanized rubber composition before vulcanization, the mixing ratio of the raw materials, and the processing conditions used to produce the predicted vulcanized rubber composition.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Laid-Open No. 2020-038495

[0006] Non-patent literature

[0007] Non-patent literature 1: https: / / arxiv.org / abs / 1712.02034

[0008] Non-Patent Document 2: https: / / docs.dgl.ai / en / 0.8.x / guide / training-graph.html Summary of the Invention

[0009] Technical solutions to technical problems

[0010] The information processing system involved in the first aspect of the present invention has at least one processor, which obtains a combination of information for identifying raw materials received from a user and the respective amounts of the raw materials, and inputs the chemical fingerprints, SMILES character strings or chemical graph structure data or at least one of the product names or substance names corresponding to the raw materials and the respective amounts of the raw materials into a first machine learning model, or, alternatively, inputs the chemical fingerprints, SMILES character strings or chemical graph structure data or at least one of the product names or substance names corresponding to the raw materials and a numerical group based on the respective amounts of the raw materials into a second machine learning model, thereby obtaining physical property names of predicted object physical properties for compositions composed of the raw materials. The first machine learning model is a model that adjusts parameters so as to predict output according to input by using a learning data set that takes chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and the respective amounts of the raw materials as input, and outputs the physical property values of the physical property names of the predicted objects; the second machine learning model is a model that adjusts parameters so as to predict output according to input by using a learning data set that takes chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and the numerical groups converted using the respective amounts of the raw materials as input, and outputs the physical property values of the physical property names of the predicted objects.

[0011] The information processing system according to the second aspect of the present invention is based on the information processing system according to the first aspect, wherein the first machine learning model or the second machine learning model is constructed for each prediction target physical property name, and the processor obtains a predicted value of the physical property value for the prediction target physical property name by using the first machine learning model or the second machine learning model corresponding to the prediction target physical property name received from the user. Alternatively, the information processing system according to the second aspect of the present invention is based on the information processing system according to the first aspect, wherein the first machine learning model or the second machine learning model can be constructed by collectively learning multiple physical properties for raw materials used in multiple fields.

[0012] The information processing system according to the third embodiment of the present invention is based on the information processing system according to the first or second embodiment, and further includes, as input to the second machine learning model, a second chemical fingerprint, a second SMILES string, or a second chemical graph structure corresponding to each additive, and a numerical value group obtained by converting the amount of the additive. The processor obtains a predicted value of the physical property value of the prediction target physical property name by inputting the numerical value group based on any one of the chemical fingerprints, SMILES strings, or chemical graph structures corresponding to each raw material, and the numerical value group based on any one of the second chemical fingerprints, second SMILES strings, or second chemical graph structures corresponding to each additive into the second machine learning model. Alternatively, the information processing system according to the third embodiment of the present invention is the information processing system according to the first or second embodiment, wherein the raw material may include an additive, the input to the first machine learning model or the second machine learning model includes the type of raw material, and the processor obtains a predicted value of the physical property value of the prediction target physical property name for a composition composed of each raw material by further inputting the type of raw material obtained from a user into the first machine learning model or the second machine learning model.

[0013] The information processing system involved in the fourth embodiment of the present invention is based on the information processing system involved in the first or second embodiment, and further includes a second SMILES string or a second chemical graph structure corresponding to each additive as an input to the first machine learning model. The processor obtains a predicted value of the physical property value of the prediction object physical property name by inputting the SMILES string or chemical graph structure corresponding to each raw material and the respective amount of the raw material, and the second SMILES string or second chemical graph structure corresponding to each additive and the amount of the additive into the first machine learning model. Alternatively, the information processing system involved in the fourth embodiment of the present invention is based on the information processing system involved in any one of the first to third embodiments, and further includes the physical properties of the raw material or a numerical value group based on the physical properties in the input of the first machine learning model or the second machine learning model. The processor obtains a predicted value of the characteristic value of the physical property name of the prediction object by further inputting the physical properties of the raw material or the numerical value group based on the physical properties obtained from the user into the first machine learning model or the second machine learning model.

[0014] The information processing system involved in the fifth embodiment of the present invention is based on the information processing system involved in any one of the first to fourth embodiments, and the input of the first machine learning model or the second machine learning model includes the characteristics of the raw materials or the numerical group based on the type, and the processor obtains the predicted value of the physical property value of the physical property name of the prediction object by further inputting the characteristics of the raw materials or the numerical group based on the type obtained from the user into the first machine learning model or the second machine learning model.

[0015] The information processing system involved in the sixth embodiment of the present invention is based on the information processing system involved in any one of the first to fifth embodiments, and the input of the first machine learning model or the second machine learning model includes information related to the process conditions and / or the device used in the process or a numerical group based on the information, and the processor obtains the predicted value of the physical property value of the predicted object physical property name for the composition composed of the raw materials by further inputting the information related to the process conditions and / or the device used in the process or a numerical group based on the information obtained from the user into the first machine learning model or the second machine learning model.

[0016] The information processing system involved in the seventh embodiment of the present invention is based on the information processing system involved in any one of the first to sixth embodiments, and the processor extracts at least one raw material candidate based on at least one of the product name of the raw material, the manufacturer name of the raw material, the category of the raw material, and the purpose of the raw material received from the user, and outputs information for selectively displaying the extracted raw material candidate.

[0017] The information processing system involved in the eighth aspect of the present invention has at least one processor, which regards a combination of raw materials and the respective quantities of the raw materials as a parent group, produces a next-generation offspring group according to a genetic algorithm, uses the combination of raw materials and the respective quantities of the raw materials shown in the produced next-generation offspring group to calculate the value of the fitness function, repeats the process of producing the next-generation offspring group until the end condition of the physical property value range specified by the user is met, and when the end condition is met, outputs information including the combination of raw materials and the respective quantities of the raw materials shown in the next-generation offspring group produced last.

[0018] The information processing system involved in the ninth embodiment of the present invention is based on the information processing system involved in the eighth embodiment, and the processor obtains at least one group of expected physical property value ranges input by the user for the expected physical property name, and inputs the chemical fingerprints, SMILES character strings or chemical graph structure data corresponding to the raw materials included in the initial combination of raw materials and the respective quantities of the raw materials into the first machine learning model, or inputs the chemical fingerprints, SMILES character strings or chemical graph structure data corresponding to the raw materials and the numerical groups based on the respective quantities of the raw materials into the second machine learning model, thereby obtaining the predicted values of the physical property values of the physical property names, and screening out the combinations of raw materials that meet the range of expected physical property values specified by the user from the obtained predicted values of the physical property values. The group composed of the respective quantities of the raw materials is regarded as the parent group, and the next generation offspring group is produced according to the genetic algorithm. The first machine learning model is a model that adjusts the parameters of a learning data set that uses the chemical fingerprints, SMILES strings or chemical graph structure data corresponding to the raw materials and the respective quantities of the raw materials as inputs, and the physical property values of the predicted property names as outputs so as to predict the output according to the input. The second machine learning model is a model that adjusts the parameters of a learning data set that uses the chemical fingerprints, SMILES strings or chemical graph structure data corresponding to the raw materials and the numerical group converted using the respective quantities of the raw materials as inputs, and the physical property values of the property names as outputs so as to predict the output according to the input.

[0019] The information processing system involved in the tenth embodiment of the present invention is based on the information processing system involved in the ninth embodiment, the first machine learning model or the second machine learning model is constructed according to each physical property name, and the processor obtains the predicted value of the physical property value of the predicted object physical property name by using the first machine learning model or the second machine learning model corresponding to the physical property name received from the user.

[0020] The information processing system according to the eleventh aspect of the present invention is the information processing system according to the ninth or tenth aspect, wherein the processor further outputs predicted values of the physical property values of the desired physical property names shown by each of the next generation groups finally produced.

[0021] The information processing system involved in the twelfth embodiment of the present invention is based on the information processing system involved in any one of the eighth to eleventh embodiments, and when the user selects one of the raw materials to be output, the processor outputs the contact information of the sales company that sells the raw material and / or detailed information of the raw material.

[0022] The information processing system involved in the thirteenth embodiment of the present invention comprises: at least one storage device which stores data related to embodiments of patent documents; and at least one processor which extracts embodiments of input raw materials from the embodiments stored in the storage device, which use embodiments of raw materials specified by a user and / or which do not use raw materials with similar physical property values achieved by the combination of the raw materials, and outputs information for displaying a list of applicants or patent holders of patent documents that record the extracted embodiments.

[0023] The information processing system involved in the fourteenth embodiment of the present invention is based on the information processing system involved in the thirteenth embodiment. After extracting the embodiment, the processor replaces the target raw material specified by the user with other raw materials and screens out raw materials with similar physical property values.

[0024] The information processing system involved in the fifteenth embodiment of the present invention is based on the information processing system involved in the thirteenth embodiment. When the object raw material specified by the user is replaced with other raw materials, when screening raw materials with similar physical property values, the processor inputs the group of other raw materials and the amount of other raw materials into the machine learning model that has been learned in the first embodiment to respectively determine the predicted value of the physical property value of the predicted object physical property name, and by comparing the predicted value of the physical property value with the physical property value of the object raw material, extracts raw materials whose predicted value of the physical property value is equal to or higher than the physical property value of the raw material specified by the user or similar raw materials.

[0025] The information processing system according to the sixteenth aspect of the present invention is the information processing system according to any one of the thirteenth to fifteenth aspects, wherein the list includes similar raw materials described in the extracted embodiment as replacement raw materials.

[0026] The information processing system involved in the seventeenth embodiment of the present invention has at least one processor, which obtains a target physical property value specified by a user, obtains a range of a mixing amount specified by the user, determines a plurality of candidate mixing amounts for each candidate raw material within the range of the mixing amount specified by the user, predicts physical property values for all combinations of a group of candidate raw materials and the candidate mixing amounts of each candidate raw material, and outputs a group of candidate raw materials and the amounts of each candidate raw material based on a comparison between the target physical property value and each predicted physical property value.

[0027] The information processing system involved in the eighteenth embodiment of the present invention is based on the information processing system involved in the seventeenth embodiment, and the at least one processor obtains the category of the substance specified by the user and extracts each raw material belonging to the category of the substance specified by the user as the candidate raw material.

[0028] The information processing system involved in the nineteenth embodiment of the present invention comprises at least one processor, which performs the following steps: a step of determining a plurality of quantity candidate values within the range of the amount of the other raw material specified by the user when a part of the raw materials constituting the target composition is replaced with other raw materials specified by the user; a step of determining the predicted value of the physical property value for all of the determined plurality of quantity candidate values; and a step of outputting information for prompting the predicted value of the physical property value corresponding to each of the plurality of quantity candidate values, wherein in the step of determining the predicted value of the physical property value, the at least one processor inputs at least one of the SMILES character string or chemical structure data or product name or substance name corresponding to each raw material and the respective amount of the raw material into the first machine learning model, or inputs the chemical fingerprint, SMILES character string or chemical structure data corresponding to each raw material into the first machine learning model. At least one of the chemical fingerprint, SMILES character string, chemical graph structure data, product name or substance name and a numerical group based on the respective amounts of the raw materials are input into a second machine learning model to obtain a predicted value of the physical property value. The first machine learning model is a model that adjusts parameters based on the learning data set that uses the SMILES character string, chemical graph structure data, product name or substance name corresponding to the raw materials and the respective amounts of the raw materials as input, and the physical property value of the predicted object physical property name as output so as to predict the output based on the input. The second machine learning model is a model that adjusts parameters based on the learning data set that uses the chemical fingerprint, SMILES character string, chemical graph structure data, product name or substance name corresponding to the raw materials and a numerical group converted using the respective amounts of the raw materials as input, and the physical property value of the predicted object physical property name as output.

[0029] The program involved in the twentieth embodiment of the present invention is a program for causing a computer to execute the following steps, wherein the computer is capable of referring to at least one storage device, wherein the storage device stores a first machine learning model or a second machine learning model, wherein the first machine learning model is a model that adjusts parameters so as to be able to predict output based on input by using a learning data set that uses chemical fingerprints, SMILES strings or chemical graph structure data corresponding to the raw materials and the respective amounts of the raw materials as inputs and outputs the physical property values of the predicted object physical property names, and the second machine learning model is a model that uses a combination of chemical fingerprints, SMILES strings or chemical graph structure data corresponding to the raw materials and a numerical group obtained by converting the respective amounts of the raw materials as outputs. A model in which parameter learning is adjusted based on an input and a learning data set that takes as input a physical property value of a predicted object physical property name as output, wherein the following steps are: obtaining a combination of information for identifying raw materials received from a user and the respective amounts of the raw materials, and inputting the chemical fingerprints, SMILES character strings or chemical graph structure data corresponding to the raw materials and the respective amounts of the raw materials into a first machine learning model, or inputting the chemical fingerprints, SMILES character strings or chemical graph structure data corresponding to the raw materials and a numerical group based on the respective amounts of the raw materials into a second machine learning model, thereby obtaining predicted values of the physical property values of the predicted object physical property name for a composition composed of the raw materials.

[0030] The program involved in the twenty-first embodiment of the present invention is a program for causing a computer to execute the following steps: a step of treating a group consisting of a combination of raw materials and the respective quantities of the raw materials as a parent population, and producing a next-generation offspring population according to a genetic algorithm; a step of calculating the value of a fitness function using the combination of raw materials and the respective quantities of the raw materials shown in the produced next-generation offspring population; and a step of repeatedly performing the process of producing the next-generation offspring population until an end condition based on a range of physical property values specified by a user is met, and when the end condition is met, outputting information including the combination of raw materials and the respective quantities of the raw materials shown in the next-generation offspring population finally produced.

[0031] The program involved in the twenty-second embodiment of the present invention is a program for causing a computer to execute the following steps, the computer being able to refer to at least one storage device that stores data related to embodiments of a patent document, the following steps being: a step of extracting, from the embodiments stored in the storage device, embodiments using raw materials specified by a user and / or embodiments not using raw materials having similar physical property values achieved by the combination of the raw materials, the input raw materials; and a step of outputting information for displaying a list of applicants or patent holders of patent documents that record the extracted embodiments.

[0032] The program involved in the twenty-third embodiment of the present invention is used to enable a computer to execute the following steps: a step of obtaining a target physical property value specified by a user; a step of obtaining a range of a mixing amount specified by a user; a step of determining a plurality of candidate mixing amounts for each candidate raw material within the range of a mixing amount specified by the user, and predicting physical property values for all combinations of a group of candidate raw materials and their respective candidate mixing amounts; and a step of outputting a group of candidate raw materials and their respective amounts based on a comparison between the target physical property value and each predicted physical property value.

[0033] The program involved in the twenty-fourth embodiment of the present invention causes a computer to execute the following steps: a step of determining a plurality of quantity candidate values within the range of the amount of the other raw material specified by the user when a part of the raw materials constituting the target composition is replaced with other raw materials specified by the user; a step of determining predicted values of the physical property value for all of the plurality of determined quantity candidate values; and a step of outputting information for prompting predicted values of the physical property value corresponding to each of the plurality of quantity candidate values, wherein in the step of determining the predicted values of the physical property value, at least one of the SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and the respective amounts of the raw materials are input into the first machine learning model, or the chemical fingerprint, SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials are input into the first machine learning model. At least one of the names and a numerical group based on the respective quantities of the raw materials are input into a second machine learning model to obtain a predicted value of the physical property value. The first machine learning model is a model that uses a learning data set that takes the SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and the respective quantities of the raw materials as input, and the physical property values of the predicted object physical property names as output, and adjusts the parameters so that the output can be predicted based on the input. The second machine learning model is a model that uses a learning data set that takes the chemical fingerprints, SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and the numerical group converted using the respective quantities of the raw materials as input, and the physical property values of the predicted object physical property names as output, and adjusts the parameters so that the output can be predicted based on the input. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a schematic configuration diagram of an information processing system common to each embodiment.

[0035] Figure 2 This is a diagram schematically illustrating the configuration of a terminal common to all embodiments.

[0036] Figure 3 This is a schematic diagram of the configuration of a computer system common to each embodiment.

[0037] Figure 4 This is an example of a table stored in a storage device of a computer system.

[0038] Figure 5 This is a diagram showing an example of a screen for selecting raw materials displayed on the terminal according to the first embodiment.

[0039] Figure 6 It is a continuation Figure 5 An example of a screen.

[0040] Figure 7 This is a diagram showing an example of a screen for selecting raw materials displayed on the terminal according to the second embodiment.

[0041] Figure 8 It is a continuation Figure 7 An example of a screen.

[0042] Figure 9 This is a flowchart showing an example of the flow of the search process according to the second embodiment.

[0043] Figure 10 It shows Figure 9 This is a flowchart of an example of the flow of the search process in the genetic algorithm in step S250.

[0044] Figure 11 This is a diagram showing an example of screen transition in a terminal according to the third embodiment.

[0045] Figure 12 It is a continuation Figure 11 An example of a screen.

[0046] Figure 13 This is a flowchart showing an example of a process flow when a sales destination candidate button is pressed according to the third embodiment.

[0047] Figure 14 This figure explains the grid search of raw material types and blending amounts.

[0048] Figure 15A This is a diagram showing an example of a screen of a terminal in the fourth embodiment.

[0049] Figure 15B This is a diagram showing an example of a patent document search screen in the fourth embodiment.

[0050] Figure 16 This is an example of the example selection screen in the fourth embodiment.

[0051] Figure 17 is Figure 16This is an example of a screen when the “AI product search” button B121 associated with Example 1 of the selected patent document is pressed in the screen G12 of FIG.

[0052] Figure 18 is Figure 17 This is an example of a screen when the "Change" button B131 is pressed in the screen G13.

[0053] Figure 19 This is an example of a screen showing prediction results of physical property values in the fourth embodiment. DETAILED DESCRIPTION

[0054] The following describes various embodiments with reference to the accompanying drawings. Some embodiments may be omitted from the description. For example, detailed descriptions of well-known matters or repeated descriptions of substantially identical components may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art.

[0055] <First Topic>

[0056] There is a problem in that it is difficult to obtain the physical property values of a composition obtained by combining raw materials.

[0057] One embodiment of the present invention has been made in view of the above-mentioned problems, and a first object of the present invention is to easily obtain the physical property values of a composition obtained by combining raw materials.

[0058] <Second Topic>

[0059] In addition, there is a problem that it is difficult to obtain a combination of raw materials that satisfies the material properties desired by the user. One embodiment of the present invention has been made in view of the above problem, and a second object is to facilitate obtaining a combination of raw materials that satisfies the desired material properties.

[0060] <Third Topic>

[0061] In addition, it is difficult for material manufacturers to search for companies that may use the target materials. One embodiment of the present invention has been made in view of the above problem, and a third object is to facilitate the search for companies that may use the target materials.

[0062] Each embodiment has been made in view of the above-mentioned problems, and an object of the present invention is to provide an information processing system and a program that can improve at least one of these problems.

[0063] <First embodiment>

[0064] In the first embodiment, an information processing system that solves the first problem will be described. Figure 1This is a schematic diagram of the information processing system common to each embodiment. Figure 1 As shown, information processing system S includes terminals 1-1, ..., 1-N (N is a natural number) used by users and a computer system 2. Terminals 1-1, ..., 1-N are communicatively connected to computer system 2 via a communication network CN. Examples of terminals include smartphones, tablets, laptop computers, and personal computers. Hereinafter, terminals 1-1, ..., 1-N are collectively referred to as terminal 1.

[0065] Figure 2 This is a schematic diagram of the terminal structure common to each embodiment. Figure 2 As shown, as an example, terminal 1 includes an input interface 11, a communication module 12, a storage device 13, a memory 14, an output interface 15, and a processor 16. It should be noted that, while the description herein assumes that terminal 1 includes a single processor 16, it may also include multiple processors, that is, it may include more than one processor. Furthermore, while the description herein assumes that terminal 1 includes a single storage device 13, it may also include multiple storage devices, that is, it may include more than one storage device.

[0066] The input interface 11 receives input from the user of the terminal 1 and outputs an input signal corresponding to the received input to the processor 16. The communication module 12 is connected to the communication network CN and communicates with the computer system 2. This communication can be wired or wireless.

[0067] The storage device 13 is, for example, a memory, and stores programs and various data for the processor 16 to read and execute. The memory 14 temporarily stores data and programs. The memory 14 is a volatile memory, for example, a RAM (Random Access Memory).

[0068] The output interface 15 can be connected to an external display 17 and can output signals to the display 17. The processor 16 loads a program from the storage device 13 into the memory 14 and performs various processes by executing a series of commands included in the program.

[0069] Figure 3 This is a schematic diagram of the computer system common to each embodiment. Figure 3As shown, as an example, computer system 2 includes an input interface 21, a communication module 12, a storage device 23, a memory 24, an output interface 25, and a processor 26. It should be noted that, while the computer system 2 is described herein as including a single processor 26, it may also include multiple processors, that is, it may include more than one processor. Furthermore, while the computer system 2 is described herein as including a single storage device 13, it may also include multiple storage devices, that is, it may include more than one storage device.

[0070] Input interface 21 receives input from an administrator of computer system 2 (e.g., a staff member of a management organization) and outputs an input signal corresponding to the received input to processor 26. Communication module 22 is connected to communication network CN and communicates with terminals 1-1, ..., 1-N. This communication can be wired or wireless.

[0071] The storage device 23 is, for example, a memory, and stores programs and various data for the processor 26 to read and execute. The memory 24 temporarily stores data and programs. The memory 24 is a volatile memory, for example, a RAM (Random Access Memory).

[0072] The output interface 25 can be connected to an external device and can output a signal to the external device. The processor 26 loads a program from the storage device 23 into the memory 24 and executes a series of commands included in the program to perform various processes described below.

[0073] Figure 4 is an example of a table stored in a storage device of a computer system. Figure 4 As shown, the data of each chemical composition is stored in Table T1. For example, in Table T1, a record includes the name of the first raw material constituting a chemical composition, its amount, its unit, ..., the name of the mth (m is a natural number) raw material, its amount, its unit. Further, a record includes the process conditions for making the chemical composition, the first characteristic name of its chemical composition, its measurement conditions, its measured value, ..., the nth (n is a natural number) characteristic name of its chemical composition, its measurement conditions, its measured value group. By chemical composition, sometimes it is composed of one raw material, and sometimes it is composed of multiple raw materials. Even in the case of being composed of multiple raw materials, the number of raw materials composed of chemical compositions may also be different.

[0074] Next, use Figure 5 as well as Figure 6 , the screen transition in the terminal of the first embodiment is described. Figure 5 This is a diagram showing an example of a screen for selecting raw materials displayed on the terminal according to the first embodiment. Figure 6It is a continuation Figure 5 An example of a screen. Figure 5 As shown, screen G1 includes a screen area R1 for setting search conditions for products serving as raw materials, and a screen area R2 for displaying the search results. Screen area R1 includes a text box R11 for inputting the current product name, a manufacturer name R12 for inputting a manufacturer name, a drop-down selection box R13 for selecting a category (specific examples include acrylic resin, titanium oxide, zirconium oxide, etc.), a text box R14 for inputting a use (specific examples include adhesives, etc.), a drop-down selection box R15 for selecting the name of a predicted physical property, and a "Filter" button B1. To set conditions for filtering, press the "Filter" button B1. When the "Filter" button B1 is pressed, the search results are displayed in screen area R2. Screen area R2 displays a list of product candidates found by the search, for example, a selection box R21, a product name, a manufacturer name, and a CAS registration number are displayed for each product found.

[0075] A database is constructed in the storage device 23, storing a product table, a main application table, a main category table, and a physical property value table. In the product table, for example, a product ID, which is an example of product identification information used to identify a product, is associated with the product name, manufacturer ID, CAS registration number, application ID for identifying the application, and category ID for identifying the category. Furthermore, the main application table stores application IDs and application names in association. Furthermore, the physical property value table stores product IDs and the physical property values of the product in association.

[0076] On pressing Figure 5 The following describes the processing of processor 26 until the product candidate list is displayed when the "Filter" button B1 on screen G1 of the terminal 1 is pressed. Processor 26 extracts at least one raw material candidate based on at least one of the raw material product name, raw material manufacturer name, raw material category, and raw material application received from the user, outputs information for selectively displaying the extracted raw material candidate, and transmits it to terminal 1. Consequently, the product candidate list resulting from the search is displayed in screen region R2 of screen G1 of terminal 1, which receives this information.

[0077] If the add button B2 is pressed while the selection box R21 of the screen area R2 of at least one screen G1 is selected, the screen transitions to Figure 6 Screen G2. Figure 6Screen G2 displays a text box R22 for inputting the name of a raw material, a text box R23 for inputting the amount, and a text box R24 for inputting the unit. For each raw material name, the name, amount, and unit can be entered. Furthermore, for additives, a text box R25 for inputting the name and amount of the additive can be entered, and a text box R26 for inputting the amount can be entered. Furthermore, a text box R27 for inputting the name of the physical property to be predicted can be entered.

[0078] Here, as an example, the raw material name corresponding to the selection box R21 selected in the screen area R2 of the screen G1 is displayed in the text box R22. Also, as an example, the prediction target physical property name selected in the pull-down selection box R15 in the screen area R1 of the screen G1 is displayed in the text box R27.

[0079] It should be noted that Figure 6 In the screen G2, the name of the device used for processing can be displayed in a manner that allows input, and the processing conditions can also be displayed in a manner that allows input. In this case, the device name and / or processing conditions can also be added to the input of the machine learning model.

[0080] After at least one raw material and its amount and unit, the name of the predicted physical property, and additives as needed are input, if the physical property value prediction button B12 is pressed, the screen transitions to screen G3.

[0081] On screen G3, prediction results of the physical property values of the input prediction target physical property names are displayed for the composition produced by the input combination of raw materials and their respective amounts, and the combination of additives and their respective amounts. Figure 6 In the example of screen G3, the predicted value of the glass transition temperature is displayed as an example.

[0082] <Method for predicting the physical property value of the prediction target physical property name>

[0083] Next, regarding the method for predicting the physical property value of the prediction target physical property name, an outline of the processing will be first described, and then a specific processing method will be described.

[0084] The storage device 13 stores a machine learning model that has been learned. Specifically, for example, the storage device 13 stores a machine learning model constructed for each prediction object property name. Depending on the situation, the prediction object property name can also be added, and a machine learning model can be constructed for the added prediction object property name. It should be noted that the processor can learn from scratch to build a machine learning model, or it can add or update teacher data to the stored machine learning model for re-learning. Here, the machine learning model is, for example, a model that has been learned through a learning data set, which takes as input a chemical fingerprint obtained by converting a combination of the names of the raw materials constituting the chemical composition and the respective amounts of the raw materials, and outputs the physical property value of the prediction object property name. In this case, if the user inputs a combination of the names of the raw materials and the respective amounts of the raw materials, the processor 26, for example, when multiple raw materials are input, can also convert the fingerprints obtained by converting the multiple raw materials input into fingerprints into corresponding chemical fingerprints by weighted averaging the respective values of the fingerprints by their respective amounts. Here, in the process of converting to fingerprint, when the raw material is input by the user, the processor 26 can convert the structural formula of the raw material into a one-line character string (for example, CC1=CC2, etc.) (or read it as a character string) using characters and symbols through the expression of SMILES (Simplified Molecular Input Line Entry System), and further convert the converted character string into a chemical fingerprint represented by a numerical column.

[0085] On the other hand, in the case where only one raw material is input, the processor 26 may convert the raw material into a chemical fingerprint.

[0086] Alternatively, a machine learning model may be one that has been trained using a learning dataset that includes chemical fingerprints obtained by converting the raw materials that make up a chemical composition and the amounts of each raw material as input, and outputs the physical property values of the predicted physical properties. In this case, if a user inputs a combination of raw material names and the amounts of each raw material, processor 26 can convert each of the input raw materials into a corresponding chemical fingerprint.

[0087] As another method, the machine learning model can also be a model that is learned using a learning data set, which includes separate information (such as index, name, CAS registration number, etc.) for identifying the raw materials that constitute the chemical composition and the respective quantities of the raw materials as input, and outputs the physical property value of the predicted object physical property name.

[0088] Processor 26 obtains a combination of raw material names and the amounts of each raw material received from the user, converts the combination of raw material names and the amounts of each raw material into a chemical fingerprint, and inputs the obtained chemical fingerprint into a learned machine learning model stored in storage device 23 to obtain a predicted value of the physical property value of the prediction target physical property name. More specifically, for example, processor 26 inputs the chemical fingerprint into the machine learning model corresponding to the prediction target physical property name received from the user to obtain a predicted value of the physical property value of the prediction target physical property name.

[0089] It should be noted that multiple fingerprints can be input into the machine learning model, and the number of input fingerprints is not limited to two, and can be three or more. Specifically, the machine learning model can also be a model that takes multiple fingerprints as input and learns the predicted value of the physical property value of the predicted object physical property name as output. As an example, as an input to the second machine learning model, it can also include a second chemical fingerprint obtained by weighted averaging and converting the chemical fingerprints corresponding to the additives with the respective amounts of the additives. In this case, the processor 26 can also obtain the predicted value of the physical property value of the predicted object physical property name by inputting at least one of the chemical fingerprints and at least one of the second chemical fingerprints into the learned second machine learning model. For example, in the case of distinguishing additives by type, in addition to the second chemical fingerprint, there can also be a third fingerprint and subsequent fingerprints. In addition, it can also be a case where there are more than four fingerprints of two or more raw materials (such as polymers) with different functions and more than two fingerprints of additives with different functions, and it is also possible to obtain the predicted value of the physical property value of the predicted object physical property name by inputting these four or more fingerprints into the learned second machine learning model.

[0090] It should be noted that the example of conversion into a chemical fingerprint has been described so far, but it is not limited to this. The processor 26 can also convert the chemical formula of at least one raw material (and / or at least one additive) into a character string (hereinafter also referred to as a SMILES character string) through the expression of SMILES, and input the converted SMILES character string or the numerical group (such as a vector) obtained by converting the SMILES character string respectively into a machine learning model (for example, a machine learning model of a natural language processing system in the case of an input character string, for example, a neural network in the case of an input numerical group), and output the predicted value of the physical property value of the predicted object physical property name. Here, as a method of converting to a vector, for example, a technology such as Smiles2vec described in non-patent document 1 can also be used to convert the SMILES character string into a vector. The technology called Smiles2vec is based on the idea of RNN (recurrent neural network: recurrent neural network) and layers such as LSTM (Long short-term memory), GRU (Gated recurrent unit), etc. Here, a machine learning model is a model that learns by taking a SMILES string or a numerical array based on a SMILES string (e.g., the converted vector described above) as input and outputting predicted values of the physical property names of the predicted objects as learning data. This numerical array can be a vector (e.g., a one-dimensional sequence) or a matrix (a multidimensional sequence).

[0091] Alternatively, the processor 26 can convert the chemical formula of at least one raw material (and / or at least one additive) into chemical graph structure data (e.g., including an adjacency matrix) representing the chemical graph structure, input the converted information into a machine learning model (e.g., a machine learning model including a graph neural network at the front stage and a fully connected neural network at the back stage), and output a predicted value of the physical property value of the predicted object physical property name. For example, in the case of an adjacency matrix, the graph convolutional neural network is used to convert the chemical graph structure data into a vector, and the converted vector is input into a fully connected neural network. On the other hand, for example, the chemical graph structure data can be converted into a vector (e.g., converted to 0 or 1 based on whether it meets specific rules such as fingerprints, MACCS Keys, etc.), and the vector can be input into a (e.g., fully connected) neural network.

[0092] Here, a machine learning model is a model that learns by taking chemical graph structure data or a numerical array (e.g., a vector) converted from chemical graph structure data as input and outputting predicted values of physical property names as learning data. This numerical array can be a vector (one-dimensional sequence) or a matrix (multidimensional sequence).

[0093] Here, as a method for converting to a numerical sequence (e.g., a vector) representing the graph structure, for example, a fingerprint can be obtained by converting the sequence to 0 or 1 based on whether specific rules such as MACCS Keys are met. Alternatively, in the case of an adjacency matrix representing the graph structure, this adjacency matrix can be input into a graph convolutional network, and the vector output from the final layer of the graph convolutional network can be used as the converted vector. To convert the graph structure to a vector, for example, one or more convolutions followed by the Readout function described in Non-Patent Document 2 (https: / / docs.dgl.ai / en / 0.8.x / guide / training-graph.html) can be used. In this case, for example, specific graph features (e.g., the number of CC connections) can be counted and included in the vector. Furthermore, an operation can be created in a neural network (e.g., a fully connected neural network) that converts the graph structure to a vector and then creates a weighted average of the components to obtain the vector.

[0094] Here, a specific example of the process of creating a vector obtained by weighted averaging of compound a and compound b according to the components is described. Let the adjacency matrix of compound a be G_a. This adjacency matrix is, for example, a square matrix that sets atoms as nodes, sets connections as edges, and stores the presence or absence (1 or 0) of edges between each node. H_a is a vector obtained by, for example, performing one or more convolution operations and a subsequent readout function on the adjacency matrix of compound a.

[0095] Similarly, let the adjacency matrix of compound b be G_b, and H_b be a vector obtained by performing one or more convolution processes and subsequent Readout functions on the adjacency matrix of compound b.

[0096] A vector H_total obtained by weighted averaging compound a and compound b according to their components (hereinafter also referred to as a weighted average vector) is represented by the following equation, for example.

[0097] H_total = H_a × amount of compound a + H_b × amount of compound b

[0098] The weighted average vector H_total is input to the subsequent layers of the machine learning model (e.g., a fully connected layer), and the predicted values of the physical properties are output from the output layer of the machine learning model. By using such a machine learning model, the accuracy of physical property prediction can be improved based on the chemical structure and its quantity in various forms.

[0099] In this way, the input of the second machine learning model may also include the second chemical fingerprint, the second SMILES string or the second chemical graph structure corresponding to the additive, and the numerical group obtained by converting the amount of the additive. The processor 26 obtains the predicted value of the physical property value of the physical property name of the prediction object by inputting the numerical group based on any one of the chemical fingerprints, SMILES strings or chemical graph structures corresponding to the raw materials, and the numerical group based on any one of the second chemical fingerprints, second SMILES strings or second chemical graph structures corresponding to the additives into the second machine learning model.

[0100] Alternatively, the input of the first machine learning model may also include a second SMILES string or a second chemical graph structure corresponding to the additives. The processor 26 obtains the predicted value of the physical property value of the physical property name of the prediction object by inputting the SMILES strings or chemical graph structures corresponding to the raw materials and the respective quantities of the raw materials, and the second SMILES strings or second chemical graph structures corresponding to the additives and the quantities of the raw materials into the first machine learning model respectively.

[0101] The following describes the details of the specific processing. For example, a database is constructed in the storage device 23, which stores data related to the examples collected from patent documents (such as the names and structures of raw materials, physical properties, the amounts of each raw material used, process conditions, characteristic values, etc.), and data on raw materials used for applications similar to or identical to the examples (such as the names of raw materials, structural data, catalog property information, CAS numbers, and other available information).

[0102] Next, an example of a method for constructing a machine learning model is described.

[0103] (Step S10) For example, the collected data from the examples is processed into data for machine learning. Specifically, the units of the example data are aligned and converted into a chemical fingerprint. If the example is a mixture, the mixture is converted into a chemical fingerprint by taking a weighted average based on the amount of raw materials contained in the mixture. If outliers exist, they are excluded.

[0104] (Step S20) Next, a machine learning model is constructed. The machine learning model may be, for example, a random forest, Gaussian process regression, a neural network model, or other Bayesian models.

[0105] <Machine Learning Model Construction Process>

[0106] The following machine learning model construction process is used as an example. By executing the process for each prediction target physical property name, a first machine learning model (or a second machine learning model) is constructed for each prediction target physical property name. Here, as an example of the first machine learning model, the machine learning model is a model that uses a learning dataset that has chemical fingerprints corresponding to raw materials and a set of numerical values converted using the respective amounts of the raw materials as inputs and outputs physical property values of physical property names, and adjusts parameters so that the model can predict output based on the inputs.

[0107] It should be noted that the model for learning one physical property is used as an example for explanation here, but it can also be a model for learning multiple physical properties at the same time. For example, in the case of a neural network, multi-task learning can also be used to learn multiple physical properties at the same time. The machine learning model also includes a regression analysis method. The regression analysis method includes Bayesian linear regression, partial least squares regression (PLS), etc. For example, as an example of regression analysis, a regression analysis method such as partial least squares regression (PLS) can also be used to take multiple physical properties as objects. Since the input variables and output variables are learned at the same time, multiple physical properties can also be inferred. It can be PCR (Principal Component Regression), or more than one physical property can be predicted by a Bayesian model (such as Bayesian linear regression, hierarchical Bayesian model).

[0108] (Step S21) In order to improve the generalization performance, each embodiment included in the same patent document is grouped, and all embodiments are divided into K groups (K is a natural number) of training and verification data using Group-K fold.

[0109] (Step S22) Validate the optimal combination of hyperparameters for the selected machine learning model using the validation data. Learn each hyperparameter combination using K sets of training data, and calculate the accuracy using the validation data. The average of the K sets of accuracies is used as the accuracy of each selected hyperparameter. Select the hyperparameter combination that achieves the best accuracy in the validation data.

[0110] (Step S23) Use the combination of hyperparameters with the highest accuracy to learn all the data and save the machine learning model.

[0111] <Machine Learning Model Estimation Process>

[0112] Processor 26 uses the same method as described above to calculate a weighted average of the chemical fingerprints for the combination of the raw material identification information (e.g., index, name, CAS registration number, etc.) input by the user and the amounts of each raw material. Processor 26 then inputs the obtained chemical fingerprints into the machine learning model stored in step S23 to obtain a predicted value for the physical property name of the prediction target.

[0113] As described above, the information processing system S involved in the first embodiment includes at least one processor 26. The at least one processor 26 obtains a combination of information (e.g., index, name, CAS registration number, etc.) for identifying each raw material received from a user and the amount of each raw material, and then inputs the chemical fingerprint, SMILES string, or chemical graph structure data corresponding to each raw material and the amount of each raw material into a first machine learning model, or inputs the chemical fingerprint, SMILES string, or chemical graph structure data corresponding to each raw material and a numerical value group based on the amount of each raw material into a second machine learning model, thereby obtaining a predicted value of the physical property value of the prediction target physical property name. Here, the first machine learning model is a model whose parameters are adjusted using a learning data set that takes the chemical fingerprint, SMILES string, or chemical graph structure data corresponding to each raw material and the amount of each raw material as input and outputs the physical property value of the prediction target physical property name as output so as to predict the output based on the input. The second machine learning model is a model that uses a learning data set that uses chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and a numerical group obtained by converting the respective quantities of the raw materials as input, and uses the physical property values of the predicted object physical property names as output, and adjusts parameters so that the output can be predicted based on the input.

[0114] According to this configuration, the user inputs a combination of raw materials and the amounts of each raw material to obtain a predicted value of the physical property name of the composition obtained by combining the raw materials. This makes it easy to obtain the physical property value of the composition obtained by combining the raw materials.

[0115] <Second embodiment>

[0116] Next, a second embodiment will be described. In this second embodiment, an information processing system that solves the second problem described above will be described. In this second embodiment, upon inputting a range of desired physical property values, candidate combinations of raw materials are output. The hardware configuration of the information processing system involved in this second embodiment is identical to that of the information processing system in the first embodiment, and therefore its description will be omitted.

[0117] Figure 7 This is a diagram showing an example of a screen for selecting raw materials displayed on the terminal according to the second embodiment. Figure 8 It is a continuation Figure 7 An example of a screen.

[0118] exist Figure 7 In the screen G4 , a text box R41 for inputting a desired property name, a text box R42 for inputting a desired property value lower limit, a text box R43 for inputting a desired property value upper limit, and a text box R44 for inputting the unit are displayed.

[0119] In addition, Figure 7 In the screen G4, for the raw material that you want to use at a minimum to obtain the desired physical property value range, a text box R45 for entering the name of the raw material, a text box R46 for entering the lower limit of the amount of the raw material, a text box R47 for entering the upper limit of the amount of the raw material, and a text box R48 for entering the unit of the amount are displayed. Figure 7 In the screen G4, when the raw material search button B41 is pressed, the screen changes to Figure 8 Screen G5.

[0120] Figure 8 The screen G5 is a screen showing the search results of raw materials. Figure 7 The physical property name specified in screen G4 is displayed, and the physical property value of each raw material is plotted in a two-dimensional form. Figure 8 In the screen G5, the search results of raw materials and the obtained raw material list are displayed. The raw material list includes various physical property values, combinations of substance names, and the amounts of each substance name. Figure 8 In the figure, for simplicity, the case where there is only one substance name is shown.

[0121] For each substance name, a link is created to a webpage containing information about the substance. The page containing information about the substance includes, for example, a description of the substance, physical property values, and contact information for the manufacturer (or distributor). Performing a specific operation (e.g., left-clicking) on a specific substance name (e.g., "polystyrene") transitions to screen G6 containing information about the specific substance name (e.g., "polystyrene"). Screen G6 includes a description of the specific substance name (e.g., "polystyrene"), physical property values, and contact information for the manufacturer (or distributor).

[0122] Similar to the first embodiment, the storage device 23 in the second embodiment stores a learned machine learning model. Specifically, for example, the storage device 23 stores a machine learning model constructed for each physical property name. Here, the machine learning model is, for example, a model learned using a learning dataset that inputs a chemical fingerprint obtained by converting a combination of raw materials and the amounts of each raw material, and outputs physical property values for the physical property names.

[0123] Next, an example of a method for constructing a machine learning model is described.

[0124] (Step S110) Similar to step S10 of the first embodiment, the collected data from the examples is processed into data for machine learning. Specifically, the units of the example data are aligned and converted into a chemical fingerprint. If the examples are mixtures, the mixture is converted into a chemical fingerprint by taking a weighted average based on the amounts of the raw materials contained in the mixture. Any outliers are excluded.

[0125] (Step S120) Next, a machine learning model is constructed in the same manner as in step S20 of the first embodiment. The machine learning model may be, for example, a random forest, Gaussian process regression, a neural network model, or another Bayesian model.

[0126] <Machine Learning Model Construction Process>

[0127] Taking the following machine learning model construction process as an example, it is executed for each prediction object property name, thereby constructing a machine learning model (specifically, for example, the first machine learning model or the second machine learning model) for each prediction object property name.

[0128] (Step S121) Similar to step S21 of the first embodiment, in order to improve generalization performance, each embodiment included in the same patent document is grouped, and all embodiments are divided into K groups (K is a natural number) of training and verification data using Group-K fold.

[0129] (Step S122) Similar to step S22 of the first embodiment, the optimal combination of hyperparameters for the selected machine learning model is verified using validation data. Each hyperparameter combination is learned using K sets of training data, and the accuracy is calculated using the validation data. The average of the K sets of accuracies is used as the accuracy of each selected hyperparameter. The hyperparameter combination with the best accuracy in the validation data is selected.

[0130] (Step S123) Similar to step S23 of the first embodiment, all data are learned using the combination of hyperparameters with the highest accuracy and the machine learning model is saved.

[0131] Search Process

[0132] Figure 9 This is a flowchart showing an example of the flow of the search process according to the second embodiment.

[0133] (Step S210) Terminal 1 is, for example, Figure 7User input is accepted on screen G4. Information accepted from the user includes the name of a physical property, the range of values for that physical property, and the unit of that value. Additionally, information accepted from the user includes candidate raw materials, the range of quantities for those raw materials, and the units of those quantities. Terminal 1 transmits this information received from the user to computer system 2.

[0134] (Step S220) When the processor 26 receives the information transmitted in step S210, it determines whether the candidate raw materials have been designated by the user (ie, whether the received information includes the candidate raw materials).

[0135] (Step S230) When the user designates a candidate raw material (ie, when the received information includes the candidate raw material), the processor 26 performs, for example, Figure 7 The method obtains a candidate raw material specified by the user, obtains an alternative raw material from the catalog, and obtains a combination of raw materials obtained by replacing the candidate raw material with the obtained alternative raw material.

[0136] (Step S240) The processor 26 converts the raw materials included in the combination of raw materials obtained in step S230 into chemical fingerprints, and inputs the obtained chemical fingerprints (and the amounts of the raw materials) into the machine learning model obtained in step S120. For example, Figure 7 Here, the initial values of the amounts of the respective raw materials may be preset or specified by the user.

[0137] (Step S245 ) From the acquired predicted values of physical properties, the processor 26 selects a combination of raw materials and a set of quantities of the raw materials that satisfy the range of desired physical property values specified by the user.

[0138] (Step S250) And, the processor 26 searches for the Figure 7 As one example, the processor 26 may use a genetic algorithm to search for a combination of alternative raw materials and their amounts that satisfy the physical property value range set by the user, or that has the highest or lowest physical property value set by the user, or that falls within a specific range (e.g., 100 to 200).

[0139] The alternative raw materials may be obtained from a directory stored in the storage device 23 or from directory information on the WEB via the Internet.

[0140] <Specific example of search method using genetic algorithm>

[0141] Here, a specific example of the search method in the genetic algorithm in step S250 will be described. Figure 10 It shows Figure 9 This is a flowchart of an example of the flow of the search process in the genetic algorithm in step S250.

[0142] (Step S251: Initial Individuals of Genetic Algorithm) Processor 26 may also use, for example, the combination of raw materials and the amounts of the raw materials selected in step S245 as initial individuals. Individuals are identified by the combination of information identifying the raw materials (e.g., raw material names) and the amounts of the raw materials.

[0143] Alternatively, processor 26 may randomly select initial individuals for the genetic algorithm from the combinations of raw material names and respective amounts described in the embodiments of the patent document. Here, the combinations of information identifying the raw materials (e.g., raw material names) and respective amounts described in the embodiments of the patent document may be stored in storage device 23. In this case, processor 26 may obtain the combinations of information identifying the raw materials (e.g., raw material names) and respective amounts from storage device 23.

[0144] (Step S252: Mutation) The processor 26 randomly replaces the information identifying the raw material (eg, the raw material name) with information identifying another raw material (eg, the raw material name).

[0145] (Step S253: Mutation 2) The processor 26 randomly selects the amount of the raw material and randomly changes the amount of the randomly selected raw material.

[0146] (Step S254: Crossover) Processor 26 considers each pair of raw material identification information (e.g., raw material name) and raw material quantity as a single entity, randomly selects entities, and exchanges the raw material identification information (e.g., raw material name) and raw material quantity between the randomly selected entities. Processor 26 adjusts the quantity ratio so that the total is 100, for example.

[0147] (Step S255) The processor 26 calculates the fitness function of each mutated individual. The fitness function can be, for example, a combination of information identifying the raw materials (e.g., the raw material name) and the amounts of each raw material. For example, the fitness function can be converted into a chemical fingerprint by taking a weighted average of the chemical fingerprints corresponding to the raw materials by amount. The difference between the predicted value of the physical property value obtained by inputting the converted chemical fingerprint into the learned machine learning model and the representative value of the physical property value range set by the user (e.g., the median value, average value, maximum value, or minimum value) is used. If the machine learning model is a model capable of outputting a probability distribution, a function such as a likelihood function can also be used.

[0148] It should be noted that here, as an example, the converted chemical fingerprint is input into the learned machine learning model (the second machine learning model mentioned above) to obtain the predicted value of the physical property value, but it is not limited to this. The SMILES character strings or chemical graph structure data corresponding to the raw materials and the respective quantities of the raw materials can also be input into the first machine learning model mentioned above to obtain the predicted value of the physical property value.

[0149] In addition, as an example, the method described here involves inputting chemical fingerprints corresponding to the raw materials and a set of numerical values based on the amounts of the raw materials, i.e., the converted chemical fingerprints, into a trained machine learning model (the aforementioned second machine learning model) to obtain predicted physical property values. However, the method is not limited to this. Alternatively, SMILES strings or chemical graph structure data corresponding to the raw materials and a set of numerical values based on the amounts of the raw materials may be input into the second machine learning model to obtain predicted physical property values.

[0150] (Step S256) Processor 26 determines the surviving individuals (also referred to as surviving individuals) based on the fitness function. It should be noted that processor 26 may also perform a process of penalizing the fitness function if the individual deviates from a user-specified range (e.g., including user-specified raw materials and / or user-specified physical property values within a range of 0 to 0). This allows for efficient search.

[0151] (Step S257) Processor 26 uses the combination of raw materials corresponding to the living individuals and the amounts of each raw material, for example, by weighting the chemical fingerprints corresponding to the raw materials by the amounts of each raw material to convert the chemical fingerprint into a chemical fingerprint. The converted chemical fingerprint is then input into the learned machine learning model to obtain a predicted value for the physical property. In this case, processor 26 may also convert the raw materials corresponding to the living individuals into fingerprints, and weight the values of each fingerprint by the amounts of each raw material to convert the chemical fingerprint into a chemical fingerprint.

[0152] (Step S258) Furthermore, the processor 26 determines whether the end condition based on the property value range specified by the user is satisfied. If the end condition is not satisfied, the process returns to step S251 and repeats the above process. In this way, the processor 26 repeats the above process until the end condition is satisfied. The end condition may also be the condition that the predicted value of the property value obtained in step S257 satisfies the optimal property value (also called the optimal solution) within the property value range specified by the user. Alternatively, it may be the condition that the number of combinations of candidate raw materials for which the predicted value of the property value obtained in step S257 satisfies the property value range specified by the user is greater than a set number. Here, the combination of candidate raw materials includes a case where one candidate raw material is included and a case where a plurality of candidate raw materials are included.

[0153] (Step S258) When the predetermined end condition is satisfied in step S257, the processor 26 outputs information for displaying a list of combinations of candidate raw materials (for example, a ranked display), for example.

[0154] (Step S260) Return to Figure 9 The processor 26 transmits information for displaying a list of combinations of candidate raw materials to the terminal 1, for example.

[0155] (Step S270) When receiving the information for displaying a list, the terminal 1 uses the information to display a list of combinations of candidate raw materials. Figure 8 As shown in the screen G5 of FIG, the combinations of candidate raw materials are displayed in a list (e.g., sorted display) in the terminal 1. When the processor 26 outputs information for the list display, it can output the information in a manner such as setting a link to the raw material name. Figure 8 As shown in the screen G5, when the terminal 1 is displayed at a glance, a link is set for the raw material name. Thus, when the user clicks the link set as the raw material, Figure 8 As shown, it can be converted into a web page (for example, refer to screen G6) with product information (for example, catalog, contact information with the manufacturer, etc.) of the raw material.

[0156] The information processing system according to the second embodiment described above includes at least one processor. The processor considers a combination of raw material combinations and their respective amounts as a parent population, creates a next-generation offspring population using a genetic algorithm, calculates the value of a fitness function using the raw material combinations and their respective amounts represented by the created next-generation offspring population, repeats the process of creating the next-generation offspring population until a termination condition within a physical property value range specified by a user is satisfied, and outputs information including the raw material combinations and their respective amounts represented by each of the last created next-generation offspring populations when the termination condition is satisfied.

[0157] According to this configuration, the user can obtain a combination of raw materials that satisfies a physical property value range specified by the user and the amount of each raw material.

[0158] Furthermore, the processor 26 may output predicted values of the physical property values of the desired physical property names shown by each of the next generation groups finally produced, thereby enabling the user to grasp the predicted values of the physical property values of the desired physical property names.

[0159] In addition, in step S257, the combination of raw materials corresponding to the living individuals and the amounts of each raw material are used, for example, by weighted averaging the chemical fingerprints corresponding to the raw materials by amount, thereby converting the chemical fingerprints into chemical fingerprints, and inputting the converted chemical fingerprints into the learned machine learning model to obtain predicted values of the physical property values, but the present invention is not limited to this. For example, the processor can obtain at least one set of a desired physical property value range input by the user for a desired physical property name, and input the chemical fingerprints, SMILES strings or chemical graph structure data corresponding to the raw materials included in the initial combination of raw materials, and the amounts of each raw material into the first machine learning model, or input the chemical fingerprints corresponding to the raw materials (wherein the specific example of using the chemical fingerprint corresponds to step S257), SMILES strings or chemical graph structure data, and a numerical set based on the amounts of each raw material into the second machine learning model, thereby obtaining predicted values of the physical property values of the physical property names in the same manner as in the first embodiment. Here, the first machine learning model is a model whose parameters are adjusted so as to predict the output based on the input, using a learning dataset that takes as input the chemical fingerprints, SMILES strings, or chemical graph structure data corresponding to each raw material, and the amounts of each raw material, and outputs the physical property values of the predicted physical property names. Separately, the second machine learning model is a model whose parameters are adjusted so as to predict the output based on the input, using a learning dataset that takes as input the chemical fingerprints, SMILES strings, or chemical graph structure data corresponding to each raw material, and the numerical values converted using the amounts of each raw material, and outputs the physical property values of the physical property names.

[0160] For example, the processor selects a group consisting of a combination of raw materials and the respective quantities of the raw materials that meets the range of the desired physical property values specified by the user from the predicted values of the physical property values obtained, regards the selected group consisting of the combination of raw materials and the respective quantities of the raw materials as the parent group, and produces the next generation offspring group according to the genetic algorithm.

[0161] Thus, the selected group consisting of the combination of raw materials and the amounts of each raw material is regarded as the parent population, and the genetic algorithm is executed, so that the combination of raw materials and the amounts of each raw material that meet the physical property value range specified by the user can be obtained in a shorter time.

[0162] The storage device 23 may also store a machine learning model constructed for each physical property name. In this way, the first machine learning model or the second machine learning model may be constructed for each physical property name. In this case, the processor 26 may also obtain a predicted value of the physical property value of the prediction target physical property name by inputting the converted chemical fingerprint into the machine learning model corresponding to the physical property name received from the user. According to this structure, the predicted value of the physical property value can be obtained for each physical property name, so that for the physical property name desired by the user, a combination of raw materials and the amount of each raw material that meets the physical property value range specified by the user can be obtained.

[0163] Furthermore, when a user selects one of the output raw materials, processor 26 may output contact information of a sales company that sells the raw material. This configuration facilitates the user's acquisition of the raw materials, and the administrator of computer system 2 can earn advertising revenue or commission income from the sales company that sells the raw materials.

[0164] In addition, when the user selects one of the output raw materials, the processor 26 can output detailed information of the raw material. According to this structure, the user can easily obtain detailed information about the raw materials that meet the physical property value range specified by the user.

[0165] <Third embodiment>

[0166] Next, the third embodiment will be described. In the third embodiment, an information processing system for solving the third problem will be described. In the third embodiment, from the data of examples in patent documents stored in a storage device, examples using the raw material name (or product name) input by the user and / or examples not using physical property values similar to those of the extracted examples are extracted, and a list of their applicants (companies using the materials) is output. This allows the names of companies with a high probability of using products using the raw material name (or product name) input by the user to be obtained.

[0167] In the third embodiment, a list of raw materials included in the examples showing similar physical property values can also be output. This allows the user to obtain a list of raw material names that are likely to be replaced, which can be used for marketing or promotion of the raw material name (or product name) entered by the user.

[0168] The storage device 23 stores, for example, data related to the embodiments included in the patent documents (e.g., raw material names and / or product names, structures, physical properties, amounts of each raw material used, process conditions, physical property values, etc.). In addition to the above, the storage device 23 also stores the application date, applicant name, and manufacturer name of the raw materials listed in the patent documents. This information can also be stored in a database constructed in the storage device 23.

[0169] Figure 11 This is a diagram showing an example of screen transition in a terminal according to the third embodiment. Figure 12 It is a continuation Figure 11 An example of a screen.

[0170] exist Figure 11 Screen G7 shows a text box R71 for the user to enter a product name. For example, if the user enters a portion of the product name in text box R71, a list of candidates can be displayed along with checkboxes, allowing the user to select a product using the checkboxes. To address situations where the user cannot think of a product name, screen G7 includes a text box R72 for entering the chemical formula of the raw materials included in the product. Similarly, to address situations where the user cannot think of a product name, a structure drawing button B71 can be provided for depicting the structure of the raw materials included in the product. When structure drawing button B71 is pressed, another screen is displayed in a pop-up format, allowing the user to depict the structure.

[0171] The characteristic values of each raw material are stored in advance in the database of the storage device 23. Figure 11 Screen G7 includes a text box R73 for entering a property name, a text box R74 for entering a lower limit for the property value, a text box R75 for entering an upper limit for the property value, and a text box R76 for entering the unit. A search button B72 is also provided. Users can search for a raw material by specifying a product name, a raw material's chemical formula, or a raw material's structure and pressing the search button B72. Alternatively, users can search for a raw material by specifying a property name and a range of property values and pressing the search button B72.

[0172] When the search button B72 is pressed, the screen switches to screen G8. Screen G8 displays the search results and the names of the raw materials that were found. The raw materials with these names are the raw materials for which the user wants to search for candidate sales destinations. In order to search for candidate sales destinations for these raw materials, screen G8 is provided with a sales destination candidate button B73. When the sales destination candidate button B73 is pressed, the screen switches to Figure 12 Screen G9. Screen G9 displays a sales destination candidate list (hereinafter also referred to as a promising sales destination list) of the target raw material.

[0173] The list of promising sales destinations includes, for example, candidate sales destination companies, a patent publication number (or patent number) as an example of information for identifying a patent document that records an embodiment of the raw material to be replaced, raw materials that are replacement objects for the target raw materials, characteristic values of the product to be replaced (also called characteristic values before replacement), and characteristic values after replacement with the target raw materials (also called characteristic values after replacement).

[0174] The following uses Figure 13 The flow of processing for outputting a candidate sales destination list when the candidate sales destination button is pressed will be described. Figure 13 This is a flowchart showing an example of a process flow when a sales destination candidate button is pressed according to the third embodiment. Here, as an example of the process, an example of extracting only raw materials that are not used that are specified by the user will be described.

[0175] (Step S310 ) The processor 26 retrieves, from the database of the storage device 23 , examples showing physical property values similar to those of the example using the raw material specified by the user.

[0176] (Step S320) As an example, the processor 26 selects the embodiments that do not use the specific raw materials specified by the user from among the embodiments hit by the search. Alternatively, the processor 26 may extract the embodiments that do not use the specific raw materials specified by the user during the search.

[0177] (Step S330) The processor 26 selects raw materials having physical property values equal to or greater than (or similar physical property values) when replacing raw materials similar to the raw materials specified by the user (for example, raw materials belonging to the same category in the product database or raw materials whose chemical fingerprint similarity is within a certain value) with the specified raw materials for each of the screened embodiments.

[0178] Here, when screening raw materials with physical property values equal to or greater than these, the machine learning model constructed in the first embodiment can also be used. Specifically, for example, when the target raw material whose raw material name is input by the user is replaced with another raw material, when screening raw materials with similar physical property values, the processor 26 can input the other raw material and the amount of the other raw material into the learned machine learning model of the first embodiment, respectively determine the predicted value of the physical property value of the predicted target physical property name, and by comparing the predicted value of the physical property value with the physical property value of the target raw material, extract raw materials (or similar raw materials) whose predicted value of the physical property value is equal to or greater than the physical property value of the raw material specified by the user.

[0179] (Step S340) Processor 26 reads from the database of storage device 23 the applicants (or patent holders) of patent documents described in the embodiments using the selected raw materials, and outputs a list of the read applicants (or patent holders) as a candidate list of sales destinations for the target raw materials (a list of potential sales destinations). This list of potential sales destinations may also include the raw materials described in the extracted embodiments as target replacement products.

[0180] The information processing system according to the third embodiment described above includes: at least one storage device storing patent document embodiments; and at least one processor. The processor extracts, from the embodiments stored in the storage device, embodiments that use user-specified raw materials and / or do not use raw materials with similar physical property values achieved with the combination of the raw materials, the input raw materials, and outputs information indicating a list of applicants or patent holders of patent documents that describe the extracted embodiments. This list may also include similar raw materials described in the extracted embodiments as replacement materials.

[0181] According to this configuration, applicants listed in patent documents described in the embodiments for raw materials specified by the user or raw materials similar to the raw materials are output, thereby facilitating searching for companies that may adopt the target raw materials.

[0182] In addition, after extracting the examples, the processor may select raw materials having similar physical property values when replacing the target raw material specified by the user with other raw materials.

[0183] In addition, when the object raw material specified by the user is replaced with other raw materials, when raw materials with similar physical property values are screened out, the processor can input the group of other raw materials and the amount of other raw materials into the machine learning model that has been learned in the first embodiment, and determine the predicted values of the physical property values of the predicted object physical property names respectively. By comparing the predicted values of the physical property values with the physical property values of the object raw materials, raw materials whose predicted property values are equal to or higher than the physical property values of the raw materials specified by the user, or similar raw materials, are extracted.

[0184] Modifications

[0185] In the above embodiment, the input to the first machine learning model is the SMILES string or chemical graph structure data corresponding to each raw material and the amount of each raw material, but this is not limited to this and may also be the product name or substance name and the amount of each raw material. In other words, at least one of the SMILES string or chemical graph structure data or product name or substance name corresponding to each raw material and the amount of each raw material may be input to the first machine learning model. In this case, the first machine learning model is a model that uses a learning dataset that takes as input the SMILES string or chemical graph structure data or product name or substance name corresponding to each raw material and the amount of each raw material, and outputs the physical property value of the prediction target physical property name, and adjusts parameters so that the model can predict the output based on the input.

[0186] Similarly, the input to the second machine learning model is set to be a numerical group based on the chemical fingerprint, SMILES string or chemical graph structure data corresponding to each raw material and the respective amount of the raw material, but is not limited to this, and can also be a numerical group based on the product name or substance name and the respective amount of the raw material. That is, the chemical fingerprint, SMILES string or chemical graph structure data, product name or substance name corresponding to each raw material and the numerical group based on the respective amount of the raw material can also be input into the second machine learning model. In this case, the second machine learning model is a model that uses a learning data set that uses the chemical fingerprint, SMILES string or chemical graph structure data, product name or substance name corresponding to each raw material and the numerical group converted using the respective amount of the raw material as input, and outputs the physical property value of the prediction object physical property name, and adjusts the parameters for learning so that the output can be predicted based on the input.

[0187] In addition, in the above embodiment, as an example, the first machine learning model or the second machine learning model is constructed for each prediction target physical property name, but the present invention is not limited to this. The first machine learning model or the second machine learning model can also be constructed by collectively learning multiple physical properties (for example, the cleaning power of detergents, the surface roughness caused by polishing fluids, etc.) for raw materials used in multiple fields.

[0188] In addition, in the above-mentioned embodiment, as an example, raw materials and additives are separated and described separately from the first chemical fingerprint and the second chemical fingerprint, but it is not limited to this, and additives may also be included in the raw materials. In this case, there are types (kinds) of raw materials such as polymers, polymerization initiators, flame retardants, etc. in the respective raw materials, and the input of the first machine learning model or the second machine learning model may include the type of raw materials or a numerical group based on the type (such as a numerical column, specifically, a vector). In this case, at least one processor can obtain a predicted value of the physical property value of the predicted object physical property name for the composition composed of the raw materials by further inputting the type of raw materials or the numerical group based on the type obtained from the user into the first machine learning model or the second machine learning model.

[0189] In addition, the input to the first machine learning model or the second machine learning model may also include a physical property of the raw material (e.g., fiber length, etc.) or a numerical set (e.g., a numerical sequence, specifically, a vector) based on the physical property. In this case, the processor can obtain a predicted value for the characteristic value of the prediction target physical property name by further inputting the physical property of the raw material (e.g., fiber length, etc.) or the numerical set based on the physical property obtained from the user into the first machine learning model or the second machine learning model.

[0190] The input of the first machine learning model or the second machine learning model may include the characteristics of the raw material (for example, in the case of a polymer, a higher-order structure, etc.) or a numerical group based on the characteristics (for example, a numerical column, specifically, a vector). In this case, the processor can obtain the predicted value of the physical property value of the prediction object physical property name by further inputting the characteristics of the raw material obtained from the user (for example, in the case of a polymer, a higher-order structure, etc.) or the numerical group based on the characteristics into the first machine learning model or the second machine learning model. Here, the characteristics can also be described in an article, in which case the article can also be converted into a numerical column (for example, a vector) and input.

[0191] In addition, the input of the first machine learning model or the second machine learning model may also include manufacturing process conditions and / or information related to the devices used in the process (such as equipment, machinery or components, etc.) or a numerical group based on the information (such as a numerical column, specifically a vector). The process conditions may, for example, include conditions during manufacturing (such as conditions when mixing raw materials, specifically temperature, pressure and their time series changes, maintenance time or time change rate, etc.), the name of the measuring device, measurement conditions, and at least one of the measurement standards. In addition, the information related to the devices used in the process may also include the name of the manufacturing device (such as a machine for mixing raw materials, etc.) or the specifications of the manufacturing device (such as the nozzle diameter of a 3D printer, etc.). In this case, the processor can obtain the predicted value of the physical property value of the predicted object property name for the composition composed of the raw materials by further inputting the process conditions and / or information related to the devices used in the process obtained from the user into the first machine learning model or the second machine learning model.

[0192] Figure 14 This is a diagram for explaining the grid search of raw material types and blending amounts. Figure 14 In the figure, for example, the amount of raw materials mixed is divided into groups of 10% intervals, and there are groups of raw material types and amounts corresponding to each grid. Figure 14 For simplicity, the grid search range is shown in two dimensions when selecting a single raw material. However, when selecting multiple raw materials, the grid search range increases accordingly. For example, when selecting two raw materials, the grid search range becomes four-dimensional.

[0193] For example, at least one processor may obtain a target physical property value specified by a user and obtain a blending amount range specified by the user.

[0194] In this case, at least one processor can determine a plurality of candidate blending quantities for each candidate raw material within the range of the blending quantity specified by the user, and predict the physical property values for all combinations of the candidate raw material group and the candidate blending quantities of the respective candidate raw materials. Here, the candidate raw materials can also be obtained by the processor as the candidate raw materials specified by the user. Alternatively, the processor can also obtain the category of the substance specified by the user (such as thermoplastic resin, etc.), and extract the raw materials belonging to the category of the substance specified by the user as the candidate raw materials. Here, the candidate blending quantities can be as follows: Figure 14 As shown, selection is performed at a predetermined interval width. In this case, all combinations of candidate raw material groups and candidate blending quantities of the candidate raw materials can be determined by a so-called grid search.

[0195] In this case, the at least one processor may output a group of candidate raw materials and the amounts of each candidate raw material based on a comparison between the target physical property value and the predicted physical property value. Specifically, for example, the at least one processor may output the group of candidate raw materials and the amounts of each candidate raw material in the order in which the predicted physical property value is closest to the target physical property value.

[0196] <Fourth embodiment>

[0197] Next, a fourth embodiment will be described. In this fourth embodiment, a user selects an example from a patent document. The user specifies one or more target raw materials, one or more candidate raw materials, and a range of blending amounts for the candidate raw materials, among the raw materials that constitute the composition of the selected example. At least one processor outputs predicted values for physical properties when the target raw materials are replaced with respective candidate raw materials, with respective candidate blending amounts falling within the specified range. The present invention aims to facilitate the optimal combination of raw materials and blending amounts that satisfies the user's desired material properties.

[0198] Use Figure 15~ Figure 19 The terminal screen transition is explained with an example of the terminal screen. Figure 15A : is a diagram showing an example of a screen of a terminal in the fourth embodiment. Figure 15A As shown, in screen G11, in order to select an embodiment to be referenced, a "Select Patent Embodiment" button B111 is shown. When the "Select Patent Embodiment" button B111 is pressed, a patent document search screen (not shown) is displayed in a pop-up manner, for example. Figure 15Bis a diagram showing an example of a patent document search screen in the fourth embodiment. For example, information related to patent documents can be collected for each application. In this case, the information related to the patent documents can be stored in the storage device 23 for each application of the technology related to the patent document (for example, lithium-ion batteries, sealing materials, or cosmetics, etc.), or the information related to the patent document and the application of the technology related to the patent document can be associated and stored in the storage device 23. Here, the information related to the patent document includes, for example, information identifying the patent document (for example, application number and / or publication number and / or patent number), invention name, applicant, application date, abstract, text, the name of the physical property of each embodiment and a group of physical property values, etc. In this case, in this patent document search screen, patent documents can be searched by specifying conditions such as keyword search. Specifically, for example, in the patent document search screen G11P, a drop-down selection box B112 for selecting a purpose, a text box B113 for inputting keywords in the invention name to be searched, a drop-down selection box B114 for selecting an applicant, a text box B115A for inputting the start date of the range of application dates, a text box B115B for inputting the end date of the range of application dates, a text box B116 for inputting keywords in the abstract text to be searched, and a text box B117 for inputting keywords in the main text to be searched are displayed. In addition, the name of a physical property, or a group of the name of a physical property and the range of its physical property value can also be specified in the patent document search screen G11P. If the "Filter" button B118 is pressed in the patent document search screen G11P, the search is performed according to the set conditions, and the search results are displayed. Here, as an example of a hit of the search result, patent documents are shown in the screen areas R111 and R112. In this patent document search screen, for example, when a patent document (here, the patent document in the screen area R112 as an example) is selected, it is switched to Figure 16 Screen G12.

[0199] Figure 16 This is an example of the example selection screen in the fourth embodiment. Figure 16 In the screen G12, information related to the selected patent document (for example, the invention title, publication number, applicant's name, application date, abstract, etc.) is displayed. Figure 16 In screen G12, the names of the raw materials constituting the composition (e.g., polyphenylene ether resin composition) and the respective amounts of the raw materials are shown for each embodiment described in the selected patent document, and the properties at that time (e.g., glass transition temperature Tg, dielectric constant, dielectric loss tangent, tensile strength, and in the case of cosmetics, moisturizing properties, hair gloss, etc.) are shown.

[0200] exist Figure 16In the screen G12, as an example, "AI product search" buttons B121, B122, B123, and B124 are shown in association with each embodiment. When one of the buttons is pressed, the screen changes to Figure 17 Here, as an example, the following describes a case where the "AI product search" button B121 associated with Example 1 of the selected patent document is pressed.

[0201] Figure 17 is Figure 16 An example of a screen when the "AI product search" button B121 associated with the embodiment 1 of the selected patent document is pressed in the screen G12. Figure 17 In the screen G13, "Change" buttons B131 to B136 are shown in association with the raw materials constituting the composition (e.g., polyphenylene ether resin composition). By pressing the "Change" buttons B131 to B136, the user can change the corresponding raw materials to other raw materials and / or change the mixing amount. Figure 17 The following describes a case where the "Change" button B131 is pressed in the screen G13.

[0202] Figure 18 is Figure 17 This is an example of a screen when the "Change" button B131 is pressed in the screen G13. Figure 18 Screen G14 shows a plus button associated with "Product." When the plus button is pressed, a pop-up window appears for selecting candidate raw materials. When the user selects a candidate raw material on this candidate raw material selection window, the candidate raw materials (here, product names, for example) are displayed as tabs on screen G14. In the example of screen G14, as a result of the user selecting "PP-600," "SMA-EF-40," and "NORYL™ PPE 640" as candidate raw materials, tabs L141 for "PP-600," L142 for "SMA-EF-40," and L143 for "NORYL™ PPE 640" are displayed. Also displayed are text boxes T141 for entering the lower limit of the blending amount for these candidate raw materials and text boxes T142 for entering the upper limit of the blending amount for these candidate raw materials. In addition, an "OK" button B141 and a "Cancel" button B142 are displayed. If the "OK" button B141 is pressed, the selected candidate raw materials and the blending amount range are confirmed. On the other hand, if the "Cancel" button B142 is pressed, the selected candidate raw materials and the blending amount range are canceled. In the case of pressing the "Search for products with the above conditions" button B143 in the screen G14, for example, the screen changes to Figure 19 The predicted results of the physical property values are displayed.

[0203] While the example in which the user directly specifies candidate raw materials has been described, the present invention is not limited thereto. Alternatively, the user may specify a material category (e.g., thermoplastic resin, etc.). In this case, at least one processor (e.g., processor 26) may also identify candidate raw materials belonging to the material category (e.g., thermoplastic resin, etc.) specified by the user. Specifically, for example, the material category may be stored in association with identification information identifying the raw materials in storage device 23. At least one processor (e.g., processor 26) may reference storage device 23 and read from storage device 23 a raw material corresponding to the material category specified by the user as a candidate raw material, thereby identifying the candidate raw material.

[0204] Figure 19 This is an example of a screen showing the predicted results of physical property values in the fourth embodiment. Figure 19 In the screen G15, it is for Figure 18 A graph showing the predicted physical property values for all combinations of candidate raw materials and candidate blending amounts of the candidate raw materials determined by grid search within the candidate raw materials and blending amount ranges set in . Figure 19 In the screen G15, a drop-down selection box B151 for selecting the horizontal axis (x-axis) of the graph and a drop-down selection box B152 for selecting the vertical axis (y-axis) of the graph are displayed. Here, as an example, the vertical axis (y-axis) of the graph is the dielectric constant, and the horizontal axis (x-axis) is the glass transition temperature Tg. Figure 19 On the graph of screen G15, as an example, the physical property values of the embodiment are plotted, and the predicted values of the physical properties are plotted for each combination of candidate raw materials and candidate blending amounts of the candidate raw materials. This allows the user to compare the predicted values of the physical properties and thus select the optimal combination of candidate raw materials and blending amounts. For example, if the mouse is hovered over a plot point in the graph, as an example, information including the name of the candidate raw material (e.g., product name or substance name), blending amount, and predicted values of physical properties (e.g., glass transition temperature Tg, dielectric constant, dielectric loss tangent, peel strength) corresponding to the plot point where the mouse is hovered is displayed in a pop-up manner. As an example, the name of the candidate raw material and blending amount corresponding to the plot point where the mouse is hovered, the predicted values of the physical properties, and a link L151 such as "View Details" are displayed toward the right side of screen G15. If the link L151 is pressed, the details of the candidate raw material are displayed.

[0205] For output Figure 19The following process may also be performed on screen G15. When a portion of a raw material constituting a target composition is replaced with another raw material specified by the user, at least one processor (e.g., processor 26) determines a plurality of candidate quantity values within the range of the amount of the other raw material specified by the user. Here, the target composition is, for example, a composition described in an embodiment selected by the user in a patent document selected by the user.

[0206] At least one processor (eg, processor 26) determines predicted values of the physical property values for all of the determined plurality of quantity candidate values. Furthermore, at least one processor (eg, processor 26) outputs information for presenting predicted values of the physical property values corresponding to each of the plurality of quantity candidate values.

[0207] Here, in the step of determining the predicted value of the physical property value, at least one of the SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and the respective quantities of the raw materials are input into the first machine learning model through at least one processor (for example, processor 26); or, at least one of the chemical fingerprints, SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and a numerical group based on the respective quantities of the raw materials are input into the second machine learning model to obtain the predicted value of the physical property value.

[0208] Here, the first machine learning model uses a learning dataset that takes as input at least one of the SMILES character strings, chemical graph structure data, product names, or substance names corresponding to the raw materials, and the amounts of the raw materials, and outputs the physical property values of the predicted physical property names, and adjusts its parameters so that it can predict the output based on the input. Separately, the second machine learning model uses as input at least one of the chemical fingerprints, SMILES character strings, chemical graph structure data, product names, or substance names corresponding to the raw materials, and the numerical values converted using the amounts of the raw materials, and outputs the physical property values of the predicted physical property names, and adjusts its parameters so that it can predict the output based on the input. This structure allows the user to compare the predicted values of the physical properties, thereby selecting the optimal combination of candidate raw materials and blending amounts.

[0209] It should be noted that at least a portion of the computer system 2 described in the above embodiment may be configured by hardware or software. In the case of software, a program that implements at least a portion of the functions of the computer system 2 may be stored in a computer-readable recording medium so that the computer can read and execute the program. The recording medium is not limited to removable media such as magnetic disks and optical disks, but may also be a fixed type of recording medium such as a hard disk device or memory.

[0210] Alternatively, the program that implements at least a portion of the functions of the computer system 2 may be distributed via a communication line such as the Internet (including wireless communication). Furthermore, the program may be encrypted, modulated, or compressed and distributed via a wired or wireless line such as the Internet, or stored in a recording medium.

[0211] The computer system 2 may function as one or more information devices. When multiple information devices are used, one of them may be a computer, and the computer may execute a predetermined program to realize the function as at least one unit of the computer system 2.

[0212] Furthermore, in the method invention, all processes (steps) may be automatically controlled by a computer. Alternatively, each process may be performed by a computer while progress control between processes is performed manually. Alternatively, at least a portion of all processes may be performed manually.

[0213] The present invention is not limited to the original state of the above-mentioned embodiment, but can make the constituent elements deform and concretize in the implementation stage without departing from its gist. In addition, by appropriately combining the multiple constituent elements disclosed in the above-mentioned embodiment, various inventions can be formed. For example, several constituent elements can be deleted from all the constituent elements shown in the embodiment. In addition, the constituent elements in different embodiments can be appropriately combined.

[0214] Description of Reference Numerals

[0215] S:Information Processing System

[0216] 1: Terminal

[0217] 11: Input interface

[0218] 12: Communication module

[0219] 13: Storage device

[0220] 14: Memory

[0221] 15: Output interface

[0222] 16: Processor

[0223] 17: Display

[0224] 2: Computer System

[0225] 21: Input interface

[0226] 22: Communication module

[0227] 23: Storage device

[0228] 24: Memory

[0229] 25: Output interface

[0230] 26: Processor.

Claims

1. An information processing system comprising at least one processor, the processor receiving a combination of information identifying each raw material received from a user and the amount of each raw material, and inputting at least one of a chemical fingerprint, a simplified molecular linear input specification (SMILES) string, chemical graph structure data, a product name, or a substance name corresponding to each raw material, and the amount of each raw material into a first machine learning model, or inputting at least one of a chemical fingerprint, a SMILES string, chemical graph structure data, a product name, or a substance name corresponding to each raw material, and a numerical value group based on the amount of each raw material into a second machine learning model, thereby obtaining predicted values of physical property values for prediction target physical property names of compositions composed of each raw material. The first machine learning model is a model that uses a learning data set that takes as input chemical fingerprints, SMILES character strings, chemical graph structure data, product names, or substance names corresponding to the raw materials, and the amounts of the raw materials, and outputs physical property values of the predicted physical properties, and adjusts parameters so that the model can predict the output based on the input. The second machine learning model is a model that uses a learning data set that uses chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and a numerical group obtained by converting the respective quantities of the raw materials as input, and uses the physical property values of the predicted object physical property names as output, and adjusts parameters so that the model can predict the output based on the input.

2. The information processing system according to claim 1, wherein: The first machine learning model or the second machine learning model is constructed by summarizing and learning multiple physical properties of raw materials used in multiple fields.

3. The information processing system according to claim 1 or 2, wherein: Including additives in the raw materials, The input of the first machine learning model or the second machine learning model includes the type of raw material or a value group based on the type, The processor obtains predicted values of the physical property values of the predicted object physical property names for the compositions composed of the raw materials by further inputting the types of raw materials obtained from the user or a numerical value group based on the types into the first machine learning model or the second machine learning model.

4. The information processing system according to any one of claims 1 to 3, wherein: The input of the first machine learning model or the second machine learning model also includes the physical properties of the raw materials or a numerical value group based on the physical properties. The processor obtains predicted values of the characteristic values of the predicted object property names for the compositions composed of the raw materials by further inputting the physical properties of the raw materials obtained from the user or a numerical group based on the physical properties into the first machine learning model or the second machine learning model.

5. The information processing system according to any one of claims 1 to 4, wherein: The input of the first machine learning model or the second machine learning model includes a feature of the raw material or a set of numerical values based on the feature. The processor obtains predicted values of the physical property values of the predicted object physical property names for the compositions respectively composed of the raw materials by further inputting the characteristics of the raw materials obtained from the user or a numerical value group based on the characteristics into the first machine learning model or the second machine learning model.

6. The information processing system according to any one of claims 1 to 5, wherein: The input of the first machine learning model or the second machine learning model includes process conditions and / or information related to equipment used in the process or a set of numerical values based on the information, The processor obtains predicted values of the physical property values of the predicted object physical property names for the compositions respectively composed of the raw materials by further inputting information related to the process conditions and / or the devices used in the process obtained from the user or a numerical value group based on the information into the first machine learning model or the second machine learning model.

7. The information processing system according to any one of claims 1 to 6, wherein: The processor extracts at least one raw material candidate based on at least one of a product name of the raw material, a manufacturer name of the raw material, a category of the raw material, and a purpose of the raw material received from the user, Then, information for selectively displaying the extracted raw material candidates is output.

8. An information processing system comprising at least one processor, The processor regards a group consisting of a combination of raw materials and the amounts of each raw material as a parent group, and creates a next-generation child group according to a genetic algorithm. The processor calculates the value of the fitness function using the combination of raw materials shown by the produced offspring population of the next generation and the respective amounts of the raw materials, The processor repeatedly performs the process of producing the next generation of offspring groups until an end condition based on a range of physical property values specified by a user is met, and when the end condition is met, outputs information including the combination of raw materials shown in each of the next generation of offspring groups finally produced and the respective quantities of the raw materials.

9. The information processing system according to claim 8, wherein: The processor obtains at least one group of ranges of expected physical property values input by the user for the expected physical property name, and inputs the chemical fingerprints, simplified molecular linear input specifications, i.e., SMILES strings, chemical graph structure data, product names, or substance names corresponding to the raw materials included in the initial combination of raw materials, and the respective amounts of the raw materials into the first machine learning model, or inputs the chemical fingerprints, SMILES strings, chemical graph structure data, product names, or substance names corresponding to the raw materials, and the numerical value groups based on the respective amounts of the raw materials into the second machine learning model, thereby obtaining predicted values of the physical property values of the physical property names, The obtained predicted values of the physical property values are filtered out to select a group consisting of a combination of raw materials and the amounts of the raw materials that meet the range of the desired physical property values specified by the user. The selected groups consisting of the combination of raw materials and the amounts of the raw materials are regarded as the parent population, and the next generation population is created according to the genetic algorithm. The first machine learning model is a model that uses a learning data set that takes as input chemical fingerprints, SMILES character strings, chemical graph structure data, product names, or substance names corresponding to the raw materials, and the amounts of the raw materials, and outputs physical property values of the predicted physical properties, and adjusts parameters so that the model can predict the output based on the input. The second machine learning model is a model that uses a learning data set that uses chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and a numerical group obtained by converting the respective quantities of the raw materials as input, and uses the physical property values of the physical property names as output, and adjusts parameters so that the output can be predicted based on the input.

10. The information processing system according to claim 9, wherein: The first machine learning model or the second machine learning model is constructed according to each physical property name, The processor acquires a predicted value of a physical property value of a prediction target physical property name by using a first machine learning model or a second machine learning model corresponding to a physical property name received from a user.

11. The information processing system according to claim 9 or 10, wherein: The processor further outputs predicted values of the physical property values of the desired physical property names indicated by each of the next generation groups finally produced.

12. The information processing system according to any one of claims 8 to 11, wherein: When the user selects one of the raw materials to be output, the processor outputs contact information of a sales company that sells the raw material and / or detailed information of the raw material.

13. An information processing system comprising: at least one storage device storing data related to embodiments of the patent document; and at least one processor, The processor extracts, from the embodiments stored in the storage device, embodiments using raw materials specified by the user and / or embodiments not using raw materials having similar physical property values achieved by the combination of the raw materials, the input raw materials. The processor outputs information for displaying a list of applicants or patentees of the patent documents that describe the extracted embodiments.

14. The information processing system according to claim 13, wherein: After extracting the examples, the processor selects raw materials having similar physical property values when replacing the target raw material specified by the user with other raw materials.

15. The information processing system according to claim 14, wherein: When the processor replaces the target raw material specified by the user with other raw materials, when screening out raw materials with similar physical property values, the processor inputs the group of the other raw materials and the amount of the other raw materials into the machine learning model that has been learned in the first embodiment to determine the predicted values of the physical property values of the predicted object physical property names, and by comparing the predicted values of the physical property values with the physical property values of the target raw materials, extracts raw materials whose predicted property values are equal to or higher than the physical property values of the raw material specified by the user, or similar raw materials.

16. The information processing system according to any one of claims 13 to 15, wherein: The list includes similar raw materials extracted from the examples as replacement raw materials.

17. An information processing system comprising at least one processor, The at least one processor obtains a target property value specified by a user, The at least one processor obtains a range of a coordination amount specified by a user, The at least one processor determines, for each candidate raw material, a plurality of candidate blending quantities within a blending quantity range specified by a user, and predicts physical property values for all combinations of candidate raw material groups and candidate blending quantities of the candidate raw materials. The at least one processor outputs a group of candidate raw materials and the amount of each candidate raw material based on a comparison between the target physical property value and each predicted physical property value.

18. The information processing system according to claim 17, wherein: The at least one processor acquires a substance category specified by a user, and extracts each raw material belonging to the substance category specified by the user as the candidate raw material.

19. An information processing system comprising at least one processor, The at least one processor performs the following steps: When a portion of raw materials constituting the target composition is replaced with another raw material specified by the user, a step of determining a plurality of candidate quantity values within a range of the quantity of the other raw material specified by the user; a step of determining predicted values of physical property values for all of the determined plurality of candidate quantity values; as well as a step of outputting information for presenting predicted values of physical property values corresponding to each of the plurality of quantity candidate values; In the step of determining the predicted value of the physical property value, the at least one processor inputs the simplified molecular linear input specifications corresponding to the raw materials, i.e., at least one of the SMILES character string or chemical graph structure data or product name or substance name, and the respective amounts of the raw materials, into a first machine learning model, or inputs the chemical fingerprint, SMILES character string or chemical graph structure data or at least one of the product name or substance name, and a numerical value group based on the respective amounts of the raw materials into a second machine learning model, thereby obtaining the predicted value of the physical property value. The first machine learning model is a model that uses a learning data set that takes as input at least one of SMILES character strings or chemical structure data or product names or substance names corresponding to the raw materials and the amounts of the raw materials, and outputs the physical property values of the predicted physical property names, and adjusts parameters so as to be able to predict output based on the input. The second machine learning model is a model that uses a learning data set that uses chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and a numerical group obtained by converting the respective quantities of the raw materials as input, and uses the physical property values of the predicted object physical property names as output, and adjusts parameters so that the model can predict the output based on the input.

20. A program for causing a computer to execute the following steps, wherein the computer is capable of referencing at least one storage device storing a first machine learning model or a second machine learning model, wherein the first machine learning model is a model that uses a learning dataset that has as input at least one of a chemical fingerprint, a simplified molecular linear input specification (SMILES) string, a chemical graph structure data, a product name, or a substance name corresponding to each raw material, and the amount of each raw material, and outputs a physical property value of a physical property name to be predicted, and adjusts parameters so as to predict an output based on the input; and the second machine learning model is a model that uses as input at least one of a chemical fingerprint, a SMILES string, a chemical graph structure data, a product name, or a substance name corresponding to each raw material, and a numerical value group converted using the amount of each raw material, and outputs a physical property value of a physical property name to be predicted, and adjusts parameters for learning so as to predict an output based on the input. The following steps are: obtaining a combination of information for identifying the raw materials received from the user and the respective quantities of the raw materials, and inputting the chemical fingerprints, SMILES strings or chemical graph structure data or at least one of the product names or substance names corresponding to the raw materials and the respective quantities of the raw materials into the first machine learning model, or, inputting the chemical fingerprints, SMILES strings or chemical graph structure data or at least one of the product names or substance names corresponding to the raw materials and a numerical group based on the respective quantities of the raw materials into the second machine learning model, thereby obtaining predicted values of the physical property values of the predicted object physical property names of the compositions respectively composed of the raw materials.

21. A program for causing a computer to execute the following steps: A step of creating a next-generation offspring population using a genetic algorithm, using a group consisting of a combination of raw materials and the amounts of each raw material as a parent population; A step of calculating a value of a fitness function using a combination of raw materials represented by the produced offspring population of the next generation and the amounts of the raw materials; and The process of producing the next generation of offspring groups is repeatedly performed until an end condition based on a range of physical property values specified by a user is satisfied, and when the end condition is satisfied, information including the combination of raw materials and the respective amounts of the raw materials respectively shown in the next generation of offspring groups finally produced is output.

22. A program for causing a computer to execute the following steps, wherein the computer is capable of referring to at least one storage device storing data related to an embodiment of a patent document, the following steps being: A step of extracting, from among the embodiments stored in the storage device, embodiments using the raw materials specified by the user and / or embodiments not using raw materials having similar physical property values achieved by the combination of the raw materials, the input raw materials; and A step of outputting information for displaying a list of applicants or patent holders of the extracted patent documents describing the embodiments.

23. A program for causing a computer to perform the following steps: A step of obtaining a target property value specified by a user; a step of obtaining a range of a matching amount specified by a user; A step of determining, for each candidate raw material, a plurality of candidate blending quantities within a range of blending quantities specified by a user, and predicting physical property values for all combinations of the candidate raw material groups and the candidate blending quantities of the candidate raw materials; and A step of outputting a group of candidate raw materials and the amount of each candidate raw material based on a comparison between the target physical property value and each predicted physical property value.

24. A program for causing a computer to execute the following steps: When a portion of raw materials constituting the target composition is replaced with another raw material specified by the user, a step of determining a plurality of candidate quantity values within a range of the quantity of the other raw material specified by the user; a step of determining predicted values of physical property values for all of the determined plurality of candidate quantity values; as well as a step of outputting information for presenting predicted values of physical property values corresponding to each of the plurality of quantity candidate values; In the step of determining the predicted value of the physical property value, at least one of the simplified molecular linear input specifications (SMILES) character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and the respective amounts of the raw materials are input into a first machine learning model, or at least one of the chemical fingerprints, SMILES character strings or chemical graph structure data or product names or substance names corresponding to the raw materials and a numerical value group based on the respective amounts of the raw materials are input into a second machine learning model, thereby obtaining the predicted value of the physical property value. The first machine learning model is a model that uses a learning data set that takes as input at least one of SMILES character strings or chemical structure data or product names or substance names corresponding to the raw materials and the amounts of the raw materials, and outputs the physical property values of the predicted physical property names, and adjusts parameters so as to be able to predict output based on the input. The second machine learning model is a model that uses a learning data set that uses chemical fingerprints, SMILES strings or chemical graph structure data or at least one of product names or substance names corresponding to the raw materials and a numerical group obtained by converting the respective quantities of the raw materials as input, and uses the physical property values of the predicted object physical property names as output, and adjusts parameters so that the model can predict the output based on the input.

Images