Information processing system and program

Abstract

To provide an information processing system and a program for facilitating acquisition of physical property values of a composition obtained by combining raw materials.SOLUTION: In an information processing system S, a computer system 2 acquires a combination of identification information of raw materials received from a terminal 1 that a user uses and amounts of the respective raw materials, and acquires prediction values of physical property values of prediction object physical property names for a composition composed of each of raw materials by inputting, to a first machine learning model, one of a chemical fingerprint, a SMILES character string, chemical graph structure data, a product name, and a substance name of each of raw materials, and the amount of each of the raw materials, or by inputting, to a second machine learning model, a numerical value group based on at least one of a chemical fingerprint, a SMILES character string, chemical graph structure data, a product name, and a substance name of each of the raw materials, and the amount of each of the raw materials.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] A method for predicting the physical property data of a composition has been developed. For example, Patent Document 1 describes that the physical property data of a plurality of vulcanized rubber compositions, the identification names of raw materials in the vulcanized rubber composition, the compounding ratio of the raw materials, and information on the processing conditions are used to machine-learn the physical property data in a prediction module of a computer. Patent Document 1 also describes that the machine-learned prediction module predicts the physical property data of the prediction target vulcanized rubber composition using the names of the constituent raw materials constituting the unvulcanized rubber composition before vulcanization of the prediction target vulcanized rubber composition, the compounding ratio of the constituent raw materials, and the processing conditions in the processing for producing the prediction target vulcanized rubber composition. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-038495 A [Non-patent literature]

[0004] [Non-Patent Document 1] https: / / arxiv.org / abs / 1712.02034 [Non-Patent Document 2] https: / / docs.dgl.ai / en / 0.8.x / guide / training-graph.html Summary of the Invention [Means for solving the problem]

[0005] An information processing system according to a first aspect of the present invention includes 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, and inputs at least one of a chemical fingerprint, a SMILES character string, a 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 to a first machine learning model, or inputs a group of numerical values ​​based on at least one of a chemical fingerprint, a SMILES character string, a 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 to a second machine learning model, thereby acquiring a predicted value of a physical property value of a predicted target physical property name for a composition composed of each of the raw materials. The first machine learning model is a model in which at least one of a chemical fingerprint, a SMILES character string, a 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 are input, and parameters are adjusted so that an output can be predicted from the input using a learning dataset in which the physical property value of a prediction target physical property is output; and the second machine learning model is a model in which at least one of a chemical fingerprint, a SMILES character string, a 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 are input, and parameters are adjusted so that an output can be predicted from the input using a learning dataset in which the physical property value of the prediction target physical property is output.

[0006] The information processing system according to a second aspect of the present invention is 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 predicted target property name, and the processor acquires a predicted value of the physical property value of the predicted target property name by using the first machine learning model or the second machine learning model corresponding to the predicted target property name received from a user. Alternatively, the information processing system according to the second aspect of the present invention is the information processing system according to the first aspect, wherein the first machine learning model or the second machine learning model is constructed by collectively learning a plurality of physical properties of raw materials used in a plurality of fields.

[0007] The information processing system according to the third aspect of the present invention is the information processing system according to the first or second aspect, further comprising a set of numerical values ​​obtained by converting a second chemical fingerprint, a second SMILES character string, or a second chemical graph structure corresponding to each additive and the amount of each additive as an input of the second machine learning model, and the processor inputs a set of numerical values ​​based on any one of the chemical fingerprints, the SMILES character string, or the chemical graph structure corresponding to each raw material, and a set of numerical values ​​based on any one of the second chemical fingerprints, the second SMILES character string, or the second chemical graph structure corresponding to each additive, to the second machine learning model, thereby obtaining a predicted value of the physical property value of the predicted target physical property name. Alternatively, the information processing system according to the third aspect of the present invention is the information processing system according to the first or second aspect, wherein the raw materials include additives, and the input of the first machine learning model or the second machine learning model includes the type of raw material, and the processor may further input the type of raw material acquired from a user to the first machine learning model or the second machine learning model, thereby obtaining a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials.

[0008] An information processing system according to a fourth aspect of the present invention is the information processing system according to the first or second aspect, further comprising a second SMILES character string or a second chemical graph structure corresponding to each of the additives as an input of the first machine learning model, and the processor inputs a SMILES character string or a chemical graph structure corresponding to each of the raw materials and the amount of each of the raw materials, and a second SMILES character string or a second chemical graph structure corresponding to each of the additives and the amount of each of the additives to the first machine learning model, thereby obtaining a predicted value of a property value of the predicted target physical property name. Alternatively, the information processing system according to the fourth aspect of the present invention is the information processing system according to any of the first to third aspects, wherein the input of the first machine learning model or the second machine learning model further comprises a physical property of a raw material or a set of numerical values ​​based on the physical property, and the processor may further input the physical property of a raw material or a set of numerical values ​​based on the physical property acquired from a user to the first machine learning model or the second machine learning model, thereby obtaining a predicted value of a characteristic value of the predicted target physical property name.

[0009] An information processing system according to a fifth aspect of the present invention is an information processing system according to any one of the first to fourth aspects, wherein the input to the first machine learning model or the second machine learning model includes a group of numerical values ​​based on the characteristics of raw materials or the type of raw materials, and the processor obtains a predicted value of the physical property value of the physical property name to be predicted by further inputting the characteristics of raw materials or the group of numerical values ​​based on the type of raw materials obtained from a user into the first machine learning model or the second machine learning model.

[0010] An information processing system according to a sixth aspect of the present invention is an information processing system according to any one of the first to fifth aspects, wherein the input to the first machine learning model or the second machine learning model includes information regarding process conditions and / or equipment used in the process or a group of numerical values ​​based on said information, and the processor further inputs information regarding the process conditions and / or equipment used in the process obtained from a user or a group of numerical values ​​based on said information into the first machine learning model or the second machine learning model, thereby obtaining a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials.

[0011] An information processing system according to a seventh aspect of the present invention is an information processing system according to any one of the first to sixth aspects, wherein the processor 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 use received from a user, and outputs information for displaying the extracted raw material candidate in a selectable manner.

[0012] An information processing system according to an eighth aspect of the present invention comprises at least one processor, which regards a combination of raw materials and the amounts of each of the raw materials as a parent population, creates a next-generation child population according to a genetic algorithm, calculates a fitness function value using the combination of raw materials and the amounts of each of the raw materials indicated by the created next-generation child population, repeats the process of creating a next-generation child population until a termination condition based on a physical property value range specified by a user is satisfied, and if the termination condition is satisfied, outputs information including the combination of raw materials indicated by each of the last created next-generation child populations and the amounts of each of the raw materials.

[0013] An information processing system according to a ninth aspect of the present invention is the information processing system according to the eighth aspect, wherein the processor acquires at least one set of desired property value ranges input by a user for a desired physical property name, and acquires predicted values ​​of the physical property values ​​of the physical property name by inputting chemical fingerprints, SMILES strings or chemical graph structure data corresponding to each of the raw materials included in the initial combination of raw materials and the amounts of each of the raw materials into a first machine learning model, or inputting a group of numerical values ​​based on the chemical fingerprints, SMILES strings or chemical graph structure data corresponding to each of the raw materials and the amounts of each of the raw materials into a second machine learning model, and selects a combination of raw materials that satisfies the desired property value range specified by the user from the acquired predicted values ​​of the physical property values ​​and the amounts of each of the raw materials. the first machine learning model is a model in which input is a chemical fingerprint, SMILES string or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials, and parameters are adjusted so that the output can be predicted from the input using a learning dataset in which the chemical fingerprint, SMILES string or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials are output; and the second machine learning model is a model in which input is a group of numerical values ​​obtained by conversion using the chemical fingerprint, SMILES string or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials, and parameters are adjusted so that the output can be predicted from the input using a learning dataset in which the physical property value of the physical property name is output.

[0014] An information processing system according to a tenth aspect of the present invention is an information processing system according to the ninth aspect, wherein the first machine learning model or the second machine learning model is constructed for each physical property name, and the processor obtains a predicted value of the physical property value of the physical property name to be predicted by using the first machine learning model or the second machine learning model corresponding to the physical property name received from a user.

[0015] An information processing system according to an eleventh aspect of the present invention is an information processing system according to the ninth or tenth aspect, wherein the processor further outputs predicted values ​​of the physical property value of the desired physical property name exhibited by each of the offspring population of the next generation finally produced.

[0016] An information processing system according to a 12th aspect of the present invention is an information processing system according to any one of the 8th to 11th aspects, wherein when one of the outputted raw materials is selected by a user, the processor outputs contact information for the sales company selling the raw material and / or detailed information about the raw material.

[0017] An information processing system according to a thirteenth aspect of the present invention comprises at least one storage device in which data relating to examples of patent documents are stored, and at least one processor, wherein the processor extracts, from the examples stored in the storage device, examples that use raw materials specified by a user and / or examples that do not use the input raw materials among raw materials having similar physical property values ​​realized by the combination of the raw materials, and outputs information for displaying a list of applicants or patent holders of patent documents in which the extracted examples are described.

[0018] An information processing system according to a 14th aspect of the present invention is an information processing system according to the 13th aspect, in which the processor extracts the examples and then narrows down the target raw material specified by the user to raw materials that have similar physical property values ​​when replaced with other raw materials.

[0019] An information processing system according to a 15th aspect of the present invention is an information processing system according to the 13th aspect, wherein, when narrowing down a target raw material specified by a user to raw materials having similar physical property values ​​when replaced with other raw materials, the processor inputs pairs of the other raw materials and the amounts of the other raw materials into the trained machine learning model of the first embodiment to determine predicted values ​​of the physical property values ​​of the predicted target physical property name, and compares each of the predicted physical property values ​​with the physical property values ​​of the target raw material to extract raw materials or similar raw materials whose predicted physical property values ​​are equal to or greater than the physical property values ​​of the raw material specified by the user.

[0020] An information processing system according to a sixteenth aspect of the present invention is an information processing system according to any one of the thirteenth to fifteenth aspects, wherein the list includes similar raw materials described in the extracted examples as raw materials to be replaced.

[0021] An information processing system according to a seventeenth aspect of the present invention includes at least one processor, which acquires target physical property values ​​specified by a user, acquires a range of blend amounts specified by the user, determines a plurality of candidate blend amounts within the range of blend amounts specified by the user for each of the candidate raw materials, predicts physical property values ​​for all combinations of a set of candidate raw materials and each of the candidate raw materials' candidate amounts, and outputs the set of candidate raw materials and each of the candidate raw materials based on a comparison between the target physical property values ​​and each of the predicted physical property values.

[0022] An information processing system according to an 18th aspect of the present invention is an information processing system according to the 17th aspect, wherein the at least one processor obtains a substance category specified by a user, and extracts each of the raw materials belonging to the substance category specified by the user as the candidate raw materials.

[0023] An information processing system according to a nineteenth aspect of the present invention includes at least one processor, and the at least one processor executes the steps of: when a part of a raw material constituting a target composition is replaced with another raw material specified by a user, determining a plurality of candidate values ​​for the amount of the other raw material within a range specified by the user; determining predicted values ​​of physical properties for all of the determined plurality of candidate values ​​for the amount; and outputting information for presenting the predicted values ​​of the physical properties in association with each of the plurality of candidate values ​​for the amount, wherein in the step of determining the predicted values ​​of the physical properties, the at least one processor inputs at least one of a SMILES string or chemical graph structure data, or a product name or a substance name corresponding to each of the raw materials and the amount of each of the raw materials to a first machine learning model, or inputs a chemical fingerprint, a SMILES string, or a chemical graph structure data corresponding to each of the raw materials to a first machine learning model. a prediction value of the physical property value is obtained by inputting at least one of the raw material data, the product name, or the substance name, and a group of numerical values ​​based on the amount of each of the raw materials into a second machine learning model, the first machine learning model is a model in which at least one of the SMILES character string or chemical graph structure data, or the product name or the substance name corresponding to each of the raw materials and the amount of each of the raw materials are input, and parameters are adjusted so that the output can be predicted from the input using a learning dataset in which the physical property value of the prediction target physical property name is output; and the second machine learning model is a model in which at least one of the chemical fingerprint, the SMILES character string, the chemical graph structure data, or the product name or the substance name corresponding to each of the raw materials and the amount of each of the raw materials are input, and parameters are adjusted so that the output can be predicted from the input using a learning dataset in which the physical property value of the prediction target physical property name is output.

[0024] A program according to a twentieth aspect of the present invention records a first machine learning model in which parameters are adjusted so that an output can be predicted from an input using a learning data set in which a chemical fingerprint, SMILES character string or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials are input, and a physical property value of a prediction target physical property name is output; or a trained second machine learning model in which parameters are adjusted so that an output can be predicted from an input using a learning data set in which a combination of a chemical fingerprint, SMILES character string or chemical graph structure data corresponding to each of the raw materials and a numerical group obtained by converting using the amount of each of the raw materials is input, and a physical property value of a prediction target physical property name is output. and inputting a combination of information identifying each of the raw materials received from a user and the amount of each of the raw materials into a first machine learning model, or inputting a group of numerical values ​​based on the chemical fingerprint, SMILES string or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials into a second machine learning model, thereby obtaining a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials.

[0025] A program related to a 21st aspect of the present invention is a program for causing a computer to execute the following steps: treating a set of a combination of raw materials and the amounts of each of the raw materials as a parent population, and creating a next-generation child population according to a genetic algorithm; calculating a fitness function value using the combination of raw materials and the amounts of each of the raw materials indicated by the created next-generation child population; repeating the process of creating a next-generation child population until a termination condition based on a physical property value range specified by a user is satisfied, and if the termination condition is satisfied, outputting information including the combination of raw materials indicated by each of the last created next-generation child populations and the amounts of each of the raw materials.

[0026] A program according to a 22nd aspect of the present invention is a program for causing a computer capable of referencing at least one storage device in which data relating to examples of patent documents is stored to execute the steps of: extracting, from the examples stored in the storage device, examples that use raw materials specified by a user and / or examples that do not use the input raw materials among raw materials having similar physical property values ​​realized by the combination of the raw materials; and outputting information for displaying a list of applicants or patent holders of patent documents in which the extracted examples are described.

[0027] A program according to a 23rd aspect of the present invention is a program for causing a computer to execute the steps of acquiring target physical property values ​​specified by a user, acquiring a range of blend amounts specified by the user, determining a plurality of candidate blend amounts within the range of blend amounts specified by the user for each of the candidate raw materials, and predicting physical property values ​​for all combinations of the set of candidate raw materials and the candidate blend amounts of each of the candidate raw materials, and outputting the set of candidate raw materials and the amounts of each of the candidate raw materials based on a comparison between the target physical property values ​​and each of the predicted physical property values.

[0028] A program according to a 24th aspect of the present invention is a program for causing a computer to execute the steps of: when a part of a raw material constituting a target composition is replaced with another raw material specified by a user, determining a plurality of candidate values ​​for the amount of the other raw material within a range specified by a user; determining predicted values ​​of physical properties for all of the determined plurality of candidate values ​​for the amount; and outputting information for presenting the predicted value of the physical property in association with each of the plurality of candidate values ​​for the amount, wherein in the step of determining the predicted value of the physical property, at least one of a SMILES character string, chemical graph structure data, product name, or substance name corresponding to each raw material and the amount of each of the raw materials is input to a first machine learning model, or at least one of a chemical fingerprint, a SMILES character string, chemical graph structure data, product name, or substance name corresponding to each of the raw materials is input to a first machine learning model. a second machine learning model that obtains a predicted value of a physical property value by inputting at least one of a SMILES character string, a 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 second machine learning model, wherein the first machine learning model is a model in which at least one of a SMILES character string, a 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 are input, and parameters are adjusted so that an output can be predicted from an input using a learning dataset in which a physical property value of a prediction target physical property name is output; and the second machine learning model is a model in which at least one of a chemical fingerprint, a SMILES character string, a 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 are input, and parameters are adjusted so that an output can be predicted from an input using a learning dataset in which a physical property value of a prediction target physical property name is output. [Brief description of the drawings]

[0029] [Figure 1] FIG. 1 is a schematic configuration diagram of an information processing system common to each embodiment. [Diagram 2] FIG. 2 is a schematic configuration diagram of a terminal common to each embodiment. [Diagram 3]FIG. 1 is a schematic configuration diagram of a computer system common to each embodiment. [Figure 4] 1 is an example of a table stored in a storage device of a computer system. [Diagram 5] FIG. 4 is a diagram showing an example of a screen for selecting raw materials displayed on the terminal of the first embodiment. [Figure 6] 6 is an example of a continuation of the screen in FIG. 5. [Figure 7] FIG. 11 is a diagram showing an example of a screen for selecting raw materials displayed on a terminal according to the second embodiment. [Figure 8] 8 is an example of a continuation of the screen in FIG. 7. [Figure 9] 13 is a flowchart showing an example of the flow of a search process according to the second embodiment. [Figure 10] 10 is a flowchart showing an example of the flow of a search process using a genetic algorithm in step S250 of FIG. 9. [Figure 11] 13A to 13C are diagrams illustrating an example of a screen transition on a terminal according to a third embodiment. [Figure 12] 12 is an example of a continuation of the screen in FIG. 11. [Figure 13] 13 is a flowchart showing an example of a process flow when a sales destination candidate button according to the third embodiment is pressed. [Figure 14] FIG. 13 is a diagram for explaining a grid search of raw material types and blend amounts. [Figure 15A] FIG. 13 is a diagram showing an example of a screen of a terminal in the fourth embodiment. [Figure 15B] FIG. 13 is a diagram showing an example of a patent document search screen in the fourth embodiment. [Figure 16] 23 is an example of an example selection screen in the fourth embodiment. [Figure 17] This is an example of a screen displayed when the "AI product search" button B121 associated with Example 1 of the selected patent document is pressed on the screen G12 of FIG. [Figure 18] This is an example of a screen that appears when, for example, the "Change" button B131 is pressed on the screen G13 of FIG. [Figure 19] 13 is a diagram showing an example of a screen showing a prediction result of a physical property value in the fourth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0030] Hereinafter, each embodiment will be described with reference to the drawings. However, more detailed description than necessary may be omitted. For example, detailed description of already well-known matters or duplicate description of substantially the same configuration may be omitted. This is to avoid the following description becoming unnecessarily redundant and to facilitate understanding by those skilled in the art.

[0031] <First issue> There is a problem in that it is difficult to obtain the physical properties of the composition obtained by combining raw materials.

[0032] One embodiment of the present invention has been made in view of the above problems, and has as its first object to facilitate obtaining the physical property values ​​of a composition obtained by combining raw materials.

[0033] <Second issue> Another problem is that it is difficult for a user to obtain a combination of raw materials that satisfies the material properties that the user desires. One embodiment of the present invention has been made in consideration of the above problems, and a second object of the present invention is to facilitate obtaining a combination of raw materials that satisfies the material properties that the user desires.

[0034] <Third issue> Another problem is that it is difficult for raw material manufacturers to search for companies that may use the target raw material. One embodiment of the present invention has been made in consideration of the above problem, and a third object of the present invention is to facilitate the search for companies that may use the target raw material.

[0035] Each embodiment has been made in consideration of the above problems, and aims to provide an information processing system and a program that can improve at least one of these problems.

[0036] <First embodiment> In the first embodiment, an information processing system for solving the first problem will be described. FIG. 1 is a schematic configuration diagram of an information processing system common to each embodiment. As shown in FIG. 1, the information processing system S includes terminals 1-1, ..., 1-N (N is a natural number) used by users and a computer system 2. Each of the terminals 1-1, ..., 1-N is communicatively connected to the computer system 2 via a communication network CN. The terminals are, for example, computers such as smartphones, tablet terminals, notebook computers, or personal computers. Hereinafter, the terminals 1-1, ..., 1-N are also collectively referred to as terminals 1.

[0037] Fig. 2 is a schematic diagram of a terminal common to each embodiment. As shown in Fig. 2, the terminal 1 includes, as an example, an input interface 11, a communication module 12, a storage device 13, a memory 14, an output interface 15, and a processor 16. Note that, in one embodiment, the terminal 1 is described as including one processor 16, but may include multiple processors, i.e., one or more processors. Also, in one embodiment, the terminal 1 is described as including one storage device 13, but may include multiple processors, i.e., one or more storage devices.

[0038] The input interface 11 accepts input from a user of the terminal 1 and outputs an input signal corresponding to the accepted 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 may be wired or wireless.

[0039] The storage device 13 is, for example, a storage device, and stores programs and various data to be read and executed by the processor 16. The memory 14 temporarily holds the data and programs. The memory 14 is a volatile memory, for example, a RAM (Random Access Memory).

[0040] The output interface 15 can be connected to an external display 17 and can output a signal 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 instructions included in the program.

[0041] Fig. 3 is a schematic diagram of a computer system common to each embodiment. As shown in Fig. 3, the computer system 2 includes, as an example, an input interface 21, a communication module 12, a storage device 23, a memory 24, an output interface 25, and a processor 26. Note that, as one embodiment, the computer system 2 is described as including one processor 26, but may include multiple processors, i.e., one or more processors. Also, as one embodiment, the computer system 2 is described as including one storage device 13, but may include multiple processors, i.e., one or more storage devices.

[0042] The input interface 21 accepts input from an administrator of the computer system 2 (e.g., an employee of a management organization) and outputs an input signal corresponding to the accepted input to the processor 26. The communication module 22 is connected to the communication network CN and communicates with each of the terminals 1-1, ..., 1-N. This communication may be wired or wireless.

[0043] The storage device 23 is, for example, a storage device, and stores programs and various data to be read and executed by the processor 26. The memory 24 temporarily holds the data and programs. The memory 24 is a volatile memory, for example, a RAM (Random Access Memory).

[0044] 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 instructions included in the program to perform various processes described below.

[0045] FIG. 4 is an example of a table stored in a storage device of a computer system. As shown in FIG. 4, data for each chemical composition is stored in table T1. For example, in table T1, one record includes the name of the first raw material constituting one chemical composition, its amount, its unit, ..., the name of the mth (m is a natural number) raw material, its amount, and its unit. Furthermore, one record includes a set of a process condition for making the chemical composition, the first property name of the chemical composition, its measurement condition, its measured value, ..., the nth (n is a natural number) property name of the chemical composition, its measurement condition, and its measured value. Some chemical compositions are composed of one raw material, while others are composed of multiple raw materials. Even if a chemical composition is composed of multiple raw materials, the number of raw materials constituting the chemical composition may differ.

[0046] Next, screen transitions in the terminal of the first embodiment will be described with reference to FIG. 5 and FIG. 6. FIG. 5 is a diagram showing an example of a screen for selecting raw materials displayed on the terminal of the first embodiment. FIG. 6 is an example of a screen following FIG. 5. As shown in FIG. 5, the screen G1 has a screen region R1 for setting search conditions for raw material products and a screen region R2 for displaying the search results. The screen region R1 has a text box R11 for inputting the current product name, a manufacturer name input R12, a select box R13 for selecting a category (specific examples include acrylic resin, titanium oxide, zirconia, etc.), a text box R14 for inputting an application (specific examples include adhesives, etc.), a select box R15 for selecting the predicted target physical property name, and a "narrow down" button B1. When narrowing down the search by setting conditions, the "narrow down" button B1 is pressed. When the "narrow down" button B1 is pressed, the search results are displayed in the screen region R2. The screen region R2 displays a list of product candidates hit by the search, and for example, a selection box R21, product name, manufacturer name, and CAS registration number are displayed for each hit product.

[0047] A database is constructed in the storage device 23, and a product table, a use master table, a category master table, and a physical property table are stored. In the product table, for example, a product ID, which is an example of product identification information for identifying a product, is stored in association with a product name, a manufacturer ID, a CAS registration number, a use ID for identifying a use, and a category ID for identifying a category. In the use master table, a use ID and a use name are stored in association with each other. In the physical property table, a product ID and a physical property value of the product are stored in association with each other.

[0048] The following describes the process of the processor 26 up to displaying the product candidate list when the "Refine" button B1 on the screen G1 in Fig. 5 is pressed. The 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 use received from the user, outputs information for displaying the extracted raw material candidate in a selectable manner, and transmits it to the terminal 1. As a result, the product candidate list hit by the search is displayed in the screen region R2 of the screen G1 of the terminal 1 that received this information.

[0049] When the Add button B2 is pressed with at least one selection box R21 in the screen region R2 of the screen G1 selected, the screen transitions to the screen G2 of Fig. 6. In the screen G2 of Fig. 6, a text box R22 in which the name of the raw material can be input, a text box R23 in which the amount can be input, and a text box R24 in which the unit can be input are displayed, and the name, amount, and unit can be input for each raw material name. In addition, a text box R25 in which the name of the additive can be input, and a text box R26 in which the amount can be input are displayed, and the name and amount of the additive can be input. In addition, a text box R27 in which the name of the predicted object physical property can be input is displayed, and the name of the predicted object physical property can be input.

[0050] Here, as an example, the name of the raw material corresponding to the selection box R21 selected in the screen region R2 of the screen G1 is displayed in the text box R22. Also, as an example, the name of the predicted target physical property selected in the selection box R15 in the screen region R1 of the screen G1 is displayed in the text box R27.

[0051] 6, the name of the device used for processing may be displayed so as to be input, or processing conditions may be displayed so as to be input. In this case, the device name and / or processing conditions may be added to the input of the machine learning model.

[0052] When at least one raw material, its amount and unit, the name of the physical property to be predicted, and additives are input as necessary, the physical property prediction button B12 is pressed, and the screen transitions to screen G3.

[0053] Screen G3 displays the predicted property values ​​of the input prediction target property name for the composition produced from the input combination of raw materials and the amounts of each, and the combination of additives and the amounts of each. In the example of screen G3 in Figure 6, the predicted value of glass transition temperature is displayed as an example.

[0054] <How to predict the physical property value of the predicted physical property name> Next, regarding the method for predicting the physical property value of the prediction target physical property name, first, an outline of the process will be described, and then, a specific processing method will be described. The storage device 13 stores a trained machine learning model. Specifically, for example, the storage device 13 stores a machine learning model constructed for each prediction target property name. In some cases, a prediction target property name may be added and a machine learning model may be constructed for the added prediction target property name. The processor may build a machine learning model by learning from scratch, or may add or update teacher data to the stored machine learning model and re-train it. Here, the machine learning model is, for example, a model trained by a learning data set that inputs a chemical fingerprint obtained by converting a combination of names of raw materials constituting a chemical composition and the amount of each of the raw materials, and outputs a physical property value of the prediction target property name. In this case, when a user inputs a combination of names of raw materials and the amount of each of the raw materials, the processor 26 may, for example, convert the input raw materials into fingerprints and convert the numerical values ​​of each fingerprint obtained by weighting the amount of each of the raw materials into a corresponding chemical fingerprint. In the process of converting into a fingerprint, when a raw material is input by a user, processor 26 may, for example, convert (or read as a string) the raw material into a one-line string (e.g., CC1=CC2) using letters and symbols representing the structural formula in SMILES (Simplified Molecular Input Line Entry System) notation, and then further convert the converted string into a chemical fingerprint represented by a numeric string. Alternatively, processor 26 may convert a single ingredient into a chemical fingerprint if only one ingredient is input.

[0055] Alternatively, the machine learning model may be a model trained by a training data set that includes, as input, the converted chemical fingerprints of each of the raw materials constituting the chemical composition and the amount of each of the raw materials, and outputs the physical property value of the predicted physical property name. In this case, when a user inputs a combination of the names of raw materials and the amount of each of the raw materials, processor 26 may convert each of the input raw materials into a corresponding chemical fingerprint. As yet another alternative, the machine learning model may be a model trained from a training data set that includes, as input, information identifying each of the raw materials that make up the chemical composition (e.g., index, name, CAS registry number, etc.) and the amount of each of the raw materials, and outputs the physical property value of the predicted physical property name.

[0056] Processor 26 acquires the combination of names of raw materials and the amounts of each of the raw materials received from the user, converts the combination of names of raw materials and the amounts of each of the raw materials into a chemical fingerprint, and acquires a predicted value of the physical property value of the predicted target physical property name by inputting the acquired chemical fingerprint into a trained machine learning model stored in storage device 23. More specifically, for example, processor 26 acquires a predicted value of the physical property value of the predicted target physical property name by inputting the chemical fingerprint into a machine learning model corresponding to the predicted target physical property name received from the user.

[0057] In addition, a plurality of fingerprints may be input to the machine learning model, and the number of fingerprints input is not limited to two, but may be three or more. Specifically, the machine learning model may be a model that is trained by inputting a plurality of fingerprints and outputting a predicted value of the physical property name of the predicted target physical property. As an example, the input of the second machine learning model may further include a second chemical fingerprint obtained by converting the chemical fingerprints corresponding to each additive by, for example, weighting the amount of each additive. In this case, the processor 26 may obtain a predicted value of the physical property value of the predicted target physical property name by inputting at least one of the chemical fingerprints and at least one of the second chemical fingerprints to the trained second machine learning model. For example, when the additives are distinguished by type, in addition to the second chemical fingerprint, there may be a third fingerprint or later. In addition, there may be a total of four or more fingerprints, including fingerprints of two or more types of raw materials (e.g., polymers) with different roles and fingerprints of two or more types of additives with different roles, and the predicted value of the physical property value of the predicted target physical property name may be obtained by inputting these four or more fingerprints to the trained second machine learning model.

[0058] Although an example of conversion to a chemical fingerprint has been described so far, the present invention is not limited to this. The processor 26 may convert the chemical formula of at least one raw material (and / or at least one additive) into a character string in SMILES notation (hereinafter also referred to as a SMILES character string), input the converted SMILES character string or a group of values ​​(e.g., a vector) into which each SMILES character string is converted into a machine learning model (e.g., a machine learning model of a natural language processing system when a character string is input, and for example, a neural network when a group of values ​​is input), and output a predicted value of the physical property value of the predicted physical property name. Here, as a method of conversion to a vector, for example, a technique called Smiles2vec described in Non-Patent Document 1 may be used to convert the SMILES character string into a vector. This Smiles2vec technique uses layers such as LSTM (Long short-term memory) and GRU (Gated recurrent unit) based on the idea of ​​RNN (Recurrent Network: Recurrent Neural Network). Here, the machine learning model is a model trained by learning data in which a SMILES string or a group of values ​​based on a SMILES string (e.g., the above-mentioned converted vector) is input and the predicted value of the physical property value of the predicted physical property name is output. This group of values ​​may be a vector (e.g., a one-dimensional array) or a matrix (a multidimensional array).

[0059] Alternatively, the processor 26 may 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 a chemical graph structure, input the converted information to a machine learning model (e.g., a machine learning model including a graph neural network in the front stage and a fully connected neural network in the rear stage) and output a predicted value of the physical property value of the predicted target physical property name. For example, in the case of an adjacency matrix, it is converted into a vector by a graph convolutional neural network, and the converted vector is input to the fully connected neural network. On the other hand, for example, the chemical graph structure data may be converted into a vector (e.g., converted into 0 or 1 depending on whether or not a specific rule such as a fingerprint or MACCS Keys is satisfied), and the vector may be input to a (e.g., fully connected) neural network.

[0060] Here, the machine learning model is a model trained by learning data in which chemical graph structure data or a group of values ​​(e.g., vectors) converted based on chemical graph structure data is input, and the predicted value of the physical property name of the prediction target is output. This group of values ​​may be a vector (one-dimensional array) or a matrix (multidimensional array).

[0061] Here, as a method of converting into a numerical sequence (e.g., a vector) representing a graph structure, for example, a numerical sequence (e.g., a vector) representing the graph structure may be obtained by converting into 0 or 1 depending on whether a specific rule such as MACCS Keys is satisfied, and then converting into a fingerprint. On the other hand, in the case of an adjacency matrix representing a graph structure, this adjacency matrix may be input into a graph convolution network, and the vector output from the final layer of the graph convolution network may be used as the converted vector. Here, the conversion from the graph structure to a vector may be performed, for example, by one or more convolution processes, followed by conversion using the Readout function described in Non-Patent Document 2 (https: / / docs.dgl.ai / en / 0.8.x / guide / training-graph.html). In this case, for example, a specific feature of the graph (e.g., the number of CC bonds, etc.) may be counted and converted into a vector. Then, after converting from the graph structure to a vector, an operation of creating a vector by weighting and averaging each according to the amount may be created in a neural network (e.g., a fully connected neural network).

[0062] Here, a specific example of a process for creating a vector that is a weighted average of compounds a and b according to their amounts will be described. The adjacency matrix of compound a is G_a. This adjacency matrix is, for example, a square matrix in which atoms are treated as nodes and bonds as edges, and the presence or absence (1 or 0) of an edge between each node is stored. H_a is, for example, a vector obtained by performing one or more convolution processes and the subsequent Readout function on the adjacency matrix of compound a.

[0063] Similarly, the adjacency matrix of compound b is G_b, and H_b is a vector obtained by performing one or more convolution processes and the subsequent Readout function on the adjacency matrix of compound b. The vector H_total, which is a weighted average of compounds a and b according to their amounts (hereinafter also referred to as the weighted average vector), is expressed by, for example, the following formula.

[0064] H_total = H_a × amount of compound a + H_b × amount of compound b

[0065] The weighted average vector H_total is input to the subsequent layer (for example, the fully connected layer) of this machine learning model, and the predicted value of the physical property is output from the output layer of the machine learning model. By using such a machine learning model, it is possible to improve the prediction accuracy of the physical property from the chemical structure and its amount in various expression formats.

[0066] In this way, the input to the second machine learning model may further include a set of numerical values ​​obtained by conversion using the second chemical fingerprint, the second SMILES string, or the second chemical graph structure corresponding to each of the additives and the amount of each of the additives, and processor 26 may obtain a predicted value of the physical property value of the predicted target physical property name by inputting the set of numerical values ​​based on either the chemical fingerprint, the SMILES string, or the chemical graph structure corresponding to each of the raw materials, and the set of numerical values ​​based on either the second chemical fingerprint, the second SMILES string, or the second chemical graph structure corresponding to each of the additives into the second machine learning model.

[0067] Alternatively, the input to the first machine learning model may further include a second SMILES string or a second chemical graph structure corresponding to each of the additives, and processor 26 may obtain a predicted value of the physical property value of the predicted target physical property name by inputting the SMILES string or chemical graph structure corresponding to each of the raw materials and the amount of each of the raw materials, and the second SMILES string or second chemical graph structure corresponding to each of the additives and the amount of each of the raw materials into the first machine learning model.

[0068] The specific process will be described in detail below. For example, a database is constructed in the storage device 23, and data on working examples collected from patent documents (e.g., raw material names and structures, physical properties, amounts of each raw material used, process conditions, property values, etc.) and raw material data used for similar or the same purposes as those working examples (e.g., raw material names, structure data, property information listed in catalogs, available information such as CAS numbers, etc.) are stored.

[0069] Next, an example of a method for constructing a machine learning model will be described. (Step S10) For example, the collected data of 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 amounts of raw materials contained in the mixture. Any outliers are excluded.

[0070] (Step S20) Next, a machine learning model is constructed. The machine learning model may be, for example, a random forest, a Gaussian process regression, a neural network model, or another Bayesian model.

[0071] <Machine learning model construction process> As an example of the machine learning model construction process below, a first machine learning model (or a second machine learning model) is constructed for each prediction target physical property name by executing the process for each prediction target physical property name. Here, as an example of the first machine learning model, the machine learning model is a model in which a group of numerical values ​​obtained by conversion using the chemical fingerprints corresponding to each raw material and the amounts of each raw material are input, and parameters are adjusted so that the output can be predicted from the input using a learning data set in which the physical property values ​​of the physical property names are output. Although a model that learns for one physical property will be described as an example here, a model that learns for multiple physical properties simultaneously may also be used. For example, in the case of a neural network, multiple physical properties may be learned simultaneously using multitask learning. This machine learning model also includes a regression analysis method. This regression analysis method includes Bayesian linear regression, partial least squares regression (PLS), and the like. For example, as an example of regression analysis, a regression analysis method called partial least squares regression (PLS) may be used to target multiple physical properties. Since input variables and output variables are learned simultaneously, multiple physical properties may be estimated. PCR (Principal Component Regression) may be used, or one or more physical properties may be predicted using a Bayesian model (for example, Bayesian linear regression or hierarchical Bayesian model). (Step S21) In order to improve generalization performance, examples included in the same patent document are grouped together, and all examples are divided into K sets (K is a natural number) of training and validation data using Group-K fold. (Step S22) The optimal combination of hyperparameters for the selected machine learning model is verified using validation data. Each hyperparameter combination is trained using K sets of training data, and accuracy is calculated using the validation data. The average of the K sets of accuracy is set as the accuracy of each selected hyperparameter. The hyperparameter combination with the highest accuracy in the validation data is selected. (Step S23) The machine learning model is learned from all data using the most accurate combination of hyperparameters and saved.

[0072] <Machine learning model estimation process> For example, processor 26 obtains a weighted average of chemical fingerprints by using a combination of information (e.g., index, name, CAS registration number, etc.) for identifying raw materials input by the user and the amount of each of the raw materials in the same manner as above. Then, processor 26 obtains a predicted value of the physical property value of the prediction target physical property name by inputting the obtained chemical fingerprints into the machine learning model saved in step S23.

[0073] As described above, the information processing system S according to the first embodiment includes at least one processor 26 that acquires a combination of information (e.g., index, name, CAS registry number, etc.) identifying each raw material received from a user and the amount of each raw material, and inputs the chemical fingerprint, SMILES character string, or chemical graph structure data corresponding to each raw material and the amount of each raw material to a first machine learning model, or inputs a set of numerical values ​​based on the chemical fingerprint, SMILES character string, or chemical graph structure data corresponding to each raw material and the amount of each raw material to a second machine learning model to acquire a predicted value of the physical property value of the prediction target physical property name. Here, the first machine learning model is a model in which the parameters are adjusted so that the output can be predicted from the input by a learning data set in which the chemical fingerprint, SMILES character string, or chemical graph structure data corresponding to each raw material and the amount of each raw material are input, and the physical property value of the prediction target physical property name is output. The second machine learning model is a model whose parameters are adjusted so that the output can be predicted from the input using a learning dataset that takes as input a set of numerical values ​​obtained by converting at least one of the chemical fingerprint, 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 physical property name to be predicted.

[0074] According to this configuration, a user can obtain a predicted value of a property value of a predicted target property name of a composition obtained by combining the raw materials by inputting a combination of raw materials and the amount of each of the raw materials, which makes it easy to obtain the property value of a composition obtained by combining the raw materials.

[0075] <Second embodiment> Next, a second embodiment will be described. In the second embodiment, an information processing system that solves the second problem described above will be described. In the second embodiment, when a desired range of physical property values ​​is input, candidates for raw material combinations are output. The hardware configuration of the information processing system according to the second embodiment is similar to that of the information processing system according to the first embodiment, and therefore the description thereof will be omitted.

[0076] Fig. 7 is a diagram showing an example of a screen for selecting raw materials displayed on a terminal in the second embodiment. Fig. 8 is an example of a screen subsequent to Fig. 7. Screen G4 in FIG. 7 displays a text box R41 in which the desired physical property name can be input, a text box R42 in which the desired lower limit of the physical property value can be input, a text box R43 in which the desired upper limit of the physical property value can be input, and a text box R44 in which the unit can be input. 7, a text box R45 for inputting the name of the raw material to be used at a minimum to obtain a desired range of physical property values, a text box R46 for inputting the lower limit of the amount of the raw material, a text box R47 for inputting the upper limit of the amount of the raw material, and a text box R48 for inputting the unit of the amount are displayed. When the raw material search button B41 is pressed on the screen G4 of FIG. 7, the screen transitions to the screen G5 of FIG. 8.

[0077] Screen G5 in FIG. 8 is a screen that displays the raw material search results. For example, for the physical property name specified on screen G4 in FIG. 7, a physical property map is displayed in which the physical property values ​​of each raw material are plotted two-dimensionally. Screen G5 in FIG. 8 displays a raw material list obtained as a result of the raw material search. The raw material list includes various physical property values, combinations of substance names, and the amounts of each of the substance names. For simplicity, FIG. 8 shows the case where there is one substance name.

[0078] For each substance name, a link is provided to a web page containing information about that substance. Here, the page containing information about that substance contains, for example, a description of the substance, its physical properties, and contact information for the manufacturer (or its agent). When a specific operation (e.g., left-click) is performed on a specific substance name (here, for example, "polystyrene"), the screen transitions to screen G6 containing information about the specific substance name (here, for example, "polystyrene"). Screen G6 contains a description of the specific substance name (here, for example, "polystyrene"), its physical properties, and contact information for the manufacturer (or its agent).

[0079] As in the first embodiment, the storage device 23 in the second embodiment stores a trained 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 trained by a learning data set in which a chemical fingerprint obtained by converting a combination of raw materials and the amount of each of the raw materials is input, and the physical property value of the physical property name is output.

[0080] Next, an example of a method for constructing a machine learning model will be described. (Step S110) As in step S10 of the first embodiment, for example, the collected data of examples is processed into data for machine learning. Specifically, the units of the example data are aligned and converted into chemical fingerprints. If the example is a mixture, the mixture is converted into a chemical fingerprint by taking a weighted average based on the amounts of raw materials contained in the mixture. Any outliers are excluded.

[0081] (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 is, for example, a random forest, a Gaussian process regression, a neural network model, or another Bayesian model.

[0082] <Machine learning model construction process> As an example, the following machine learning model construction process is executed for each predicted property name, and a machine learning model (specifically, for example, a first machine learning model or a second machine learning model) is constructed for each predicted property name. (Step S121) As in step S21 in the first embodiment, in order to improve generalization performance, examples included in the same patent document are grouped together, and all examples are divided into K sets of training and validation data (K is a natural number) using Group-K fold. (Step S122) As in step S22 in the first embodiment, the optimal combination of hyperparameters for the selected machine learning model is verified using validation data. Each hyperparameter combination is trained using K sets of training data, and accuracy is calculated using the validation data. The average of the K sets of accuracy is set as the accuracy of each selected hyperparameter. The hyperparameter combination with the highest accuracy in the validation data is selected. (Step S123) As in step S23 in the first embodiment, all data is trained using the combination of hyperparameters with the highest accuracy, and the machine learning model is saved.

[0083] <Exploration process> FIG. 9 is a flowchart showing an example of the flow of a search process according to the second embodiment. (Step S210) The terminal 1 accepts input from the user, for example, on screen G4 in FIG. 7. The information accepted from the user here is the name of the physical property, the value range of the physical property name, and the unit of the value. In addition, the information accepted from the user is the candidate raw materials, the amount range of the raw material, and the unit of the amount. The terminal 1 transmits the information accepted from the user to the computer system 2.

[0084] (Step S220) When processor 26 receives the information transmitted in step S210, it determines whether or not a candidate ingredient has been designated by the user (that is, whether or not the received information includes a candidate ingredient).

[0085] (Step S230) When a candidate raw material is designated by the user (i.e., when the received information includes a candidate raw material), processor 26 acquires, for example in FIG. 7 , substitutable raw materials from the catalog for the candidate raw material designated by the user, and acquires a raw material combination obtained by replacing the candidate raw material with the acquired substitutable raw material.

[0086] (Step S240) Processor 26 may convert each of the raw materials included in the raw material combination obtained in step S230 into a chemical fingerprint, and input the obtained chemical fingerprints (and the amounts of each raw material) into the trained machine learning model obtained in step S120, thereby obtaining a predicted value of the physical property value of the physical property name specified by the user in Fig. 7, for example. Here, the initial value of the amount of each raw material may be set in advance, or may be specified by the user.

[0087] (Step S245) Processor 26 narrows down the obtained predicted values ​​of the physical property values ​​to combinations of ingredients and amounts of each of the ingredients that satisfy the desired range of the physical property values ​​designated by the user.

[0088] (Step S250) Then, processor 26 uses a genetic algorithm to search for a combination of ingredients and their amounts that satisfy the physical property value range set by the user in Fig. 7. As one example, processor 26 may use a genetic algorithm to search for a combination of interchangeable ingredients and their amounts that have the highest or lowest physical property value set by the user or that is within a specific range (e.g., 100 to 200) set by the user. Here, the replaceable raw materials may be obtained from a catalog stored in the storage device 23, or the replaceable raw materials may be obtained from catalog information on the Web via the Internet.

[0089] <Specific example of search method using genetic algorithm> Here, a specific example of the search method using the genetic algorithm in step S250 will be described below. Fig. 10 is a flow chart showing an example of the flow of the search process using the genetic algorithm in step S250 in Fig. 9.

[0090] (Step S251: Initial Individuals of Genetic Algorithm) The processor 26 may set the combination of ingredients narrowed down in step S245 and the amount of each ingredient as an initial individual. Here, the individual is specified by a combination of information for identifying ingredients (e.g., ingredient names) and the amount of each ingredient. Alternatively, processor 26 may randomly select an initial individual for the genetic algorithm from among the combinations of ingredient names and the amounts of each ingredient described in the examples of the patent document. Here, the combinations of information identifying ingredients (e.g., ingredient names) and the amounts of each ingredient described in the examples of the patent document may be stored in storage device 23, and in that case, processor 26 may obtain the combinations of information identifying ingredients (e.g., ingredient names) and the amounts of each ingredient from storage device 23.

[0091] (Step S252: Mutation) Processor 26 randomly replaces information identifying an ingredient (eg, ingredient name) with information identifying another ingredient (eg, ingredient name).

[0092] (Step S253: Mutation 2) Processor 26 randomly selects the amount of an ingredient, and randomly changes the amount of the randomly selected ingredient.

[0093] (Step S254: Crossover) Processor 26 randomly selects an individual by regarding a pair of information for identifying ingredients (e.g., ingredient name) and the amount of the ingredient as one individual, and exchanges the information for identifying ingredients (e.g., ingredient name) and the amount of the ingredient between the randomly selected individuals. At this time, processor 26 adjusts the amount ratio so that the total becomes 100, for example.

[0094] (Step S255) Processor 26 calculates a fitness function for each mutated individual. The fitness function may be, for example, a difference between a predicted value of a physical property value obtained by converting a chemical fingerprint corresponding to each raw material into a chemical fingerprint by weighting the amount of each raw material using a combination of information identifying the raw material (e.g., raw material name) and the amount of each raw material, and inputting the converted chemical fingerprint into a trained machine learning model and a representative value (e.g., median, average, maximum, or minimum value) of a physical property value range set by a user, or may be something like a likelihood function if the machine learning model is a model capable of outputting a probability distribution. As an example, the converted chemical fingerprint is input into a trained machine learning model (the second machine learning model described above) to obtain predicted values ​​of physical properties. However, this is not limited to the above. The SMILES string or chemical graph structure data corresponding to each raw material and the amount of each raw material may be input into the first machine learning model described above to obtain predicted values ​​of physical properties. Also, as an example, it has been described here that the chemical fingerprints corresponding to each of the raw materials and the converted chemical fingerprints, which are a set of values ​​based on the amounts of each of the raw materials, are input into a trained machine learning model (the second machine learning model described above) to obtain predicted values ​​of physical properties, but this is not limited to this. It is also possible to input SMILES character strings or chemical graph structure data corresponding to each of the raw materials and a set of values ​​based on the amounts of each of the raw materials into the second machine learning model to obtain predicted values ​​of physical properties.

[0095] (Step S256) The processor 26 determines surviving individuals (also called surviving individuals) according to the fitness function. At that time, the processor 26 may execute a process of imposing a penalty on the fitness function if it is out of a range specified by the user (for example, including raw materials specified by the user and / or a range of physical property values ​​specified by the user from 0 to 0, etc.). This allows for efficient search.

[0096] (Step S257) Processor 26 converts the chemical fingerprints corresponding to the raw materials into a chemical fingerprint by, for example, taking a weighted average of the amounts of the raw materials using the combination of raw materials corresponding to the surviving individuals and the amounts of the raw materials, and inputs the converted chemical fingerprints into the trained machine learning model to obtain a predicted value of the physical property value. In this case, processor 26 may, for example, convert each of the raw materials corresponding to the surviving individuals into a fingerprint, and convert the obtained numerical values ​​of each fingerprint into a chemical fingerprint by taking a weighted average of the respective amounts.

[0097] (Step S258) Then, processor 26 determines whether or not a termination condition based on the physical property value range specified by the user is satisfied. If the termination condition is not satisfied, the process returns to step S251 and repeats the above-mentioned process. In this manner, processor 26 repeats the above-mentioned process until the termination condition is satisfied. The termination condition may be a condition that an optimal physical property value (also called an optimal solution) is found such that the predicted physical property value obtained in step S257 satisfies the physical property value range specified by the user. Alternatively, the termination condition may be a condition that a set number or more combinations of candidate raw materials are found such that the predicted physical property value obtained in step S257 satisfies the physical property value range specified by the user. Here, the combination of candidate raw materials may consist of one candidate raw material or may consist of multiple candidate raw materials.

[0098] (Step S258) If a predetermined end condition is met in step S257, processor 26 outputs, for example, information for displaying a list (for example, a ranking display) of combinations of candidate ingredients.

[0099] (Step S260) Returning to FIG. 9, the processor 26 transmits to the terminal 1, for example, information for displaying a list of combinations of candidate raw materials.

[0100] (Step S270) When terminal 1 receives information for displaying the list, it uses this information to display a list of combinations of candidate ingredients. As a result, as shown on screen G5 in FIG. 8, terminal 1 displays a list (e.g., a ranking display) of combinations of candidate ingredients. When outputting information for displaying the list, processor 26 may output information such that, for example, a link is set to the ingredient name. As shown on screen G5 in FIG. 8, a link is set to the ingredient name when the list is displayed on terminal 1. As a result, by clicking the link set to an ingredient, the user can transition to a web page (see, for example, screen G6) of product information (e.g., a catalog, contact methods for the manufacturer, etc.) for that ingredient, as shown in FIG. 8.

[0101] As described above, the information processing system according to the second embodiment includes at least one processor. The processor regards a combination of raw materials and the amounts of each of the raw materials as a parent population, creates a child population of the next generation according to a genetic algorithm, calculates a fitness function value using the combination of raw materials and the amounts of each of the raw materials indicated by the created child population of the next generation, repeats the process of creating a child population of the next generation until a termination condition based on a physical property value range specified by a user is satisfied, and when the termination condition is satisfied, outputs information including the combination of raw materials indicated by each of the child populations of the next generation lastly created and the amounts of each of the raw materials.

[0102] With this configuration, the user can obtain a combination of ingredients that satisfy the physical property value range specified by the user, and the amount of each of the ingredients. The processor 26 may further output the predicted values ​​of the property values ​​of the desired property names exhibited by each of the offspring populations of the next generation created last, thereby enabling the user to grasp the predicted values ​​of the property values ​​of the desired property names.

[0103] In the above step S257, the combination of raw materials corresponding to the living individual and the amount of each of the raw materials are used to convert the chemical fingerprints corresponding to the raw materials into chemical fingerprints by weighting the amount of each of the raw materials, and the converted chemical fingerprints are input to the trained machine learning model to obtain the predicted value of the physical property value, but this is not limited to the above. For example, the processor may obtain at least one set of desired physical property value ranges input by the user for the desired physical property name, and input the chemical fingerprints, SMILES character strings or chemical graph structure data corresponding to each of the raw materials included in the initial combination of raw materials and the amount of each of the raw materials into the first machine learning model, or input the chemical fingerprints corresponding to each of the raw materials (however, a specific example using this chemical fingerprint corresponds to step S257), SMILES character strings or chemical graph structure data, and a set of numerical values ​​based on the amount of each of the raw materials into the second machine learning model to obtain the predicted value of the physical property value of the physical property name, as in the first embodiment. Here, the first machine learning model is a model in which the parameters are adjusted so that the output can be predicted from the input using a learning dataset in which the chemical fingerprint, SMILES character string, or chemical graph structure data corresponding to each raw material and the amount of each raw material are input, and the physical property value of the prediction target physical property name is output. The second machine learning model is a model in which the parameters are adjusted so that the output can be predicted from the input using a learning dataset in which the chemical fingerprint, SMILES character string, or chemical graph structure data corresponding to each raw material and the amount of each raw material are input, and the physical property value of the physical property name is output. The processor, for example, narrows down the obtained predicted values ​​of the physical properties to combinations of raw materials and sets of the amounts of each of the raw materials that satisfy the desired range of physical properties specified by the user, and regards the narrowed down combinations of raw materials and sets of the amounts of each of the raw materials as a parent population, and creates the next generation of child populations according to a genetic algorithm.

[0104] This allows the narrowed-down combinations of raw materials and the amounts of each of the raw materials to be treated as a parent population and a genetic algorithm to be run, thereby making it possible to obtain, in a shorter time, combinations of raw materials and the amounts of each of the raw materials that satisfy the physical property range specified by the user.

[0105] The storage device 23 may store a machine learning model constructed for each physical property name. In this manner, 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 obtain a predicted value of the physical property value of the prediction target physical property name by inputting the converted chemical fingerprint into a machine learning model corresponding to the physical property name received from the user. With this configuration, a predicted value of the physical property value can be obtained for each physical property name, so that a combination of raw materials that satisfies the physical property value range specified by the user and the amount of each of the raw materials can be obtained for the physical property name desired by the user.

[0106] Furthermore, when one of the outputted raw materials is selected by the user, the processor 26 may output contact information of the sales company that sells the raw material. This configuration allows the user to easily obtain raw materials, and also allows the administrator of the computer system 2 to earn advertising revenue or affiliate revenue from the sales company that sells the raw material. Furthermore, when one of the outputted raw materials is selected by the user, the processor 26 may output detailed information about the raw material. With this configuration, the user can easily obtain detailed information about raw materials that satisfy the physical property value range specified by the user.

[0107] <Third embodiment> Next, a third embodiment will be described. In the third embodiment, an information processing system that solves the third problem will be described. In the third embodiment, examples that use the raw material name (or product name) input by the user and / or examples that do not use the input raw material among examples that show physical property values ​​similar to those of the extracted example are extracted from data of examples of patent documents stored in a storage device, and a list of the applicants (companies that use the raw material) is output. This makes it possible to obtain the names of companies that are likely to use products with the raw material name (or product name) input by the user.

[0108] In the third embodiment, a list of ingredients included in the examples that show similar physical property values ​​may be output. This allows the user to obtain a list of ingredient names that are likely to be substituted, which can be used for sales or promotion of the ingredient name (or product name) entered by the user.

[0109] The storage device 23 stores, for example, data on examples included in patent documents (such as raw material names and / or product names, structures, physical properties, amounts of each raw material used, process conditions, physical property values, etc.), and in addition to the above, stores the filing date, the name of the applicant, and the names of manufacturers of the raw materials described in the patent documents. This information may be stored in a database constructed in the storage device 23.

[0110] Fig. 11 is a diagram showing an example of a screen transition on a terminal of the third embodiment, and Fig. 12 is an example of a subsequent screen of Fig. 11. In the screen G7 of FIG. 11, a text box R71 is shown for the user to input a product name. For example, when the user inputs a part of the product name in the text box R71, a list of candidates with partial matches may be displayed together with check boxes, and the user may select with the check boxes. In case the user cannot think of a product name, the screen G7 may be provided with, for example, a text box R72 for inputting the chemical formula of the raw materials contained in the product. Similarly, in case the user cannot think of a product name, a structure drawing button B71 for drawing the structure of the raw materials contained in the product may be provided. When the structure drawing button B71 is pressed, another screen is displayed as a pop-up, and the user can draw the structure.

[0111] The characteristic values ​​of each raw material are stored in advance in the database of the storage device 23. In the screen G7 of Fig. 11, a text box R73 for inputting the characteristic name, a text box R74 for inputting the lower limit of the characteristic value, a text box R75 for inputting the upper limit of the characteristic value, and a text box R76 for inputting the unit are provided. Also, a search button B72 is provided. This allows the user to search for the raw material name by specifying the product name, the chemical formula of the raw material, or the structure of the raw material and pressing the search button B72. Also, the user can search for the raw material name by specifying the characteristic name and the range of the characteristic value and pressing the search button B72.

[0112] When the search button B72 is pressed, the screen transitions to screen G8s. Screen G8 displays the names of ingredients found as a result of the search. The ingredient with this ingredient name is the ingredient for which the user is searching for potential buyers. In order to search for potential buyers of this ingredient, screen G8 is provided with a potential buyer button B73. When the potential buyer button B73 is pressed, the screen transitions to screen G9 in FIG. 12. Screen G9 displays a list of potential buyers for the ingredient (hereinafter also referred to as a promising buyer list).

[0113] The list of promising customers may include, for example, candidate companies for the sales, a patent publication number (or patent number) which is an example of information identifying a patent document in which an example of the raw material to be replaced is described, the raw material to be replaced by the target raw material, characteristic values ​​of the product to be replaced (also called characteristic values ​​before replacement), and characteristic values ​​after replacement with the target raw material (also called characteristic values ​​after replacement).

[0114] The process flow for outputting a list of candidate sales destinations when the candidate sales destination button is pressed will be described below with reference to Fig. 13. Fig. 13 is a flowchart showing an example of the process flow when the candidate sales destination button according to the third embodiment is pressed. Here, as an example of the process, an example of extracting only examples that do not use raw materials specified by the user will be described.

[0115] (Step S310) The processor 26 searches the database in the storage device 23 for examples that exhibit similar physical property values ​​to the physical property values ​​of the examples in which the raw material specified by the user is used.

[0116] (Step S320) As an example, the processor 26 narrows down the examples found in the search to examples that do not use a specific raw material specified by the user. Alternatively, the processor 26 may extract examples that do not use a specific raw material specified by the user during the search.

[0117] (Step S330) For each of the narrowed down examples, processor 26 narrows down the examples to those that are similar to the raw material specified by the user (for example, those that belong to the same category in the product database or those that have a certain degree of similarity in chemical fingerprints) to those that have the same or higher physical property values ​​(or similar physical property values) when replaced with the specified raw material. Here, when narrowing down the raw materials whose physical property values ​​are equal to or greater than those of the raw material specified by the user, the machine learning model constructed in the first embodiment may be used. Specifically, for example, when narrowing down the raw materials whose physical property values ​​are similar when the target raw material of the raw material name input by the user is replaced with another raw material, the processor 26 may input a pair of the other raw material and the amount of the other raw material into the trained machine learning model of the first embodiment to determine the predicted values ​​of the physical property values ​​of the prediction target physical property name, and compare each of the predicted values ​​of the physical property values ​​with the physical property values ​​of the target raw material to extract raw materials (or similar raw materials) whose predicted values ​​of the physical property values ​​are equal to or greater than those of the raw material specified by the user.

[0118] (Step S340) The processor 26 reads out the applicants (or patent holders) of the patent documents in which examples using the narrowed-down raw materials are described from the database of the storage device 23, and outputs the list of the read out applicants (or patent holders) as a list of potential buyers (promising buyer list) of the target raw materials. This promising buyer list may include the raw materials described in the extracted examples as products to be replaced.

[0119] As described above, the information processing system according to the third embodiment includes at least one storage device in which examples of patent documents are stored, and at least one processor. The processor extracts examples that use raw materials specified by a user from the examples stored in the storage device and / or examples that do not use the input raw materials from raw materials that have similar physical property values ​​realized by the combination of the raw materials, and outputs information for displaying a list of applicants or patent holders of patent documents in which the extracted examples are described. The list may include similar raw materials described in the extracted examples as raw materials to be replaced.

[0120] With this configuration, the applicants listed in the patent documents in which the raw material specified by the user or a raw material similar to the raw material is described in the examples are output, making it easier to search for companies that may be able to use the target raw material.

[0121] Furthermore, after extracting the examples, the processor may narrow down the list to raw materials that have similar physical property values ​​when the target raw material designated by the user is replaced with another raw material.

[0122] In addition, when narrowing down the target raw material specified by the user to raw materials whose physical property values ​​are similar when replaced with other raw materials, the processor may input pairs of the other raw materials and the amounts of the other raw materials into the trained machine learning model of the first embodiment, thereby determining predicted values ​​of the physical property values ​​of the predicted target physical property name, and comparing each of the predicted physical property values ​​with the physical property values ​​of the target raw material, thereby extracting raw materials whose predicted physical property values ​​are equal to or greater than the physical property values ​​of the raw material specified by the user, or similar raw materials.

[0123] <Modification> In the above-described embodiment, the input to the first machine learning model is the SMILES character string or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials, but is not limited thereto, and may be the product name or substance name and the amount of each of the raw materials. That is, at least one of the SMILES character string or chemical graph structure data or the product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials may be input to the first machine learning model. In this case, the first machine learning model is a model in which the parameters are adjusted so that the output can be predicted from the input using a learning data set in which the SMILES character string or chemical graph structure data or the product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials are input, and the physical property value of the prediction target physical property name is output.

[0124] Similarly, the input to the second machine learning model is a set of numerical values ​​based on the chemical fingerprint, SMILES character string, or chemical graph structure data corresponding to each of the raw materials and the amount of each of the raw materials, but is not limited to this, and may be a set of numerical values ​​based on the product name or substance name and the amount of each of the raw materials. That is, at least one of the chemical fingerprint, SMILES character string, chemical graph structure data, product name, or substance name corresponding to each of the raw materials and a set of numerical values ​​based on the amount of each of the raw materials may be input to the second machine learning model. In this case, the second machine learning model is a trained model in which the parameters are adjusted so that the output can be predicted from the input using a training data set in which at least one of the chemical fingerprint, SMILES character string, chemical graph structure data, product name, or substance name corresponding to each of the raw materials and the amount of each of the raw materials are converted as input, and the physical property value of the prediction target physical property name is output.

[0125] In the above-described embodiment, the first machine learning model or the second machine learning model is constructed for each prediction target property name, but the present invention is not limited to this. The first machine learning model or the second machine learning model may be constructed by collectively learning a plurality of properties (e.g., cleaning power of a detergent, surface roughness caused by an abrasive liquid, etc.) of raw materials used in a plurality of fields. In the above-mentioned embodiment, as an example, the raw materials and additives are separated and described as the first chemical fingerprint and the second chemical fingerprint, respectively, but the present invention is not limited to this, and the raw materials may contain additives. In this case, each raw material has a raw material type (kind) such as a polymer, a polymerization initiator, a flame retardant, etc., and the input of the first machine learning model or the second machine learning model may include the type of raw material or a numerical group based on the type (e.g., a numerical string, specifically, for example, a vector). In this case, at least one processor may further input the type of raw material or the numerical group based on the type acquired from the user into the first machine learning model or the second machine learning model, thereby acquiring a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials.

[0126] The input of the first or second machine learning model may further include the physical property of the raw material (e.g., fiber length, etc.) or a group of values ​​based on the physical property (e.g., a numeric string, specifically, for example, a vector). In this case, the processor may obtain a predicted value of 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 a group of values ​​based on the physical property acquired from the user into the first or second machine learning model.

[0127] The input of the first or second machine learning model may include the features of the raw material (e.g., higher-order structure in the case of a polymer) or a group of values ​​based on the features (e.g., a numerical sequence, specifically, for example, a vector). In this case, the processor may obtain a predicted value of the physical property value of the predicted target physical property name by further inputting the features of the raw material (e.g., higher-order structure in the case of a polymer) acquired from the user or a group of values ​​based on the features into the first or second machine learning model. Here, the features may be described in text, and in this case, the text may be converted into a numerical sequence (e.g., a vector) and input.

[0128] The input of the first machine learning model or the second machine learning model may include information on manufacturing process conditions and / or equipment used in the process (e.g., equipment, machinery, or devices, etc.) or a group of values ​​based on the information (e.g., a numerical sequence, specifically, for example, a vector). The process conditions may include, for example, at least one of the conditions during manufacturing (e.g., conditions when mixing raw materials, specifically, temperature, pressure, and their time series changes, maintenance time, or time change rate, etc.), the name of a measuring device, the measurement conditions, and the measurement standards. The information on the equipment used in the process may include the name of the manufacturing device (e.g., a machine for mixing raw materials, etc.) or the specifications of the manufacturing device (e.g., the nozzle diameter of a 3D printer, etc.). In this case, the processor may further input information on the process conditions and / or the equipment used in the process acquired from the user into the first machine learning model or the second machine learning model, thereby obtaining a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials.

[0129] FIG. 14 is a diagram for explaining the grid search of ingredient types and blend amounts. In FIG. 14, for example, the blend amounts are divided into grids of 10% increments by ingredient type, and each grid corresponds to a pair of ingredient type and blend amount. For simplicity, in FIG. 14, the search range in the grid search when one ingredient type is selected is shown in two dimensions, but when multiple ingredients are selected, the grid search range increases accordingly. For example, when two ingredients are used, the grid search range becomes four-dimensional.

[0130] For example, the at least one processor may obtain target physical property values ​​specified by a user and obtain blend amount ranges specified by the user. In this case, at least one processor may determine a plurality of candidate blend amounts within the blend amount range specified by the user for each of the candidate raw materials, and predict the physical property values ​​for all combinations of the candidate raw material set and the candidate blend amounts of each of the candidate raw materials. Here, the candidate raw materials may be acquired by the processor as each of the candidate raw materials specified by the user. Alternatively, the processor may acquire a substance category (e.g., thermoplastic resin) specified by the user, and extract each of the raw materials belonging to the substance category specified by the user as the candidate raw materials. Here, the candidate blend amounts may be selected at a predetermined interval as shown in FIG. 14, and in this case, all combinations of the candidate raw material set and the candidate blend amounts of each of the candidate raw materials may be determined by a so-called grid search. In this case, the at least one processor may output the set of candidate raw materials and the respective amounts of the candidate raw materials based on a comparison between the target physical property value and each of the predicted physical property values. Specifically, for example, the at least one processor may output the set of candidate raw materials and the respective amounts of the candidate raw materials in order of proximity of the predicted physical property value to the target physical property value.

[0131] <Fourth embodiment> Next, the fourth embodiment will be described. In the fourth embodiment, when a user selects an example in a patent document and specifies one or more target raw materials among the raw materials constituting the composition of the selected example, one or more candidate raw materials and the range of the blending amount of the candidate raw materials, at least one processor outputs predicted values ​​of physical properties for each candidate blending amount included in the specified range when the target raw materials are changed to the respective candidate raw materials. The objective of the fourth embodiment is to facilitate obtaining an optimal combination of raw materials that satisfy the material properties that the user desires to obtain and the blending amount of the raw materials.

[0132] The screen transition of the terminal will be described using an example of the terminal screen of FIG. 15 to FIG. 19. FIG. 15A is a diagram showing an example of the screen of the terminal in the fourth embodiment. As shown in FIG. 15A, the screen G11 shows a "Select Patent Example" button B111 to select an example to be referred to. When this "Select Patent Example" button B111 is pressed, a patent document search screen (not shown) is displayed, for example, in a pop-up format. FIG. 15B is a diagram showing an example of the patent document search screen in the fourth embodiment. For example, information on patent documents may be collected for each application, and in that case, information on patent documents may 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), or information on patent documents and the application of the technology related to the patent document may be associated and stored in the storage device 23. Here, the information on patent documents includes, for example, information identifying the patent document (for example, application number and / or publication number and / or patent number), the name of the invention, the applicant, the filing date, the abstract, the main text, the name of the physical property for each example, and a set of the physical property values. In that case, the patent documents can be searched for by specifying conditions such as keyword search on this patent document search screen. Specifically, for example, on the patent document search screen G11P, a select box B112 for selecting a use, a text box B113 for inputting a keyword to search within the title of the invention, a select 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 a keyword to search within the abstract text, and a text box B117 for inputting a keyword to search within the main text are displayed. Also, on the patent document search screen G11P, it may be possible to specify the name of a physical property, or a combination of the name of a physical property and the range of its physical property value. When the "Refine" button B118 is pressed on the patent document search screen G11P, a search is performed under the set conditions, and the search results are displayed. Here, patent documents are shown in screen areas R111 and R112 as examples of search results.On this patent document search screen, when, for example, one patent document (here, as an example, the patent document in the screen area R112) is selected, the screen transitions to screen G12 in FIG.

[0133] Fig. 16 is an example of an example selection screen in the fourth embodiment. In the screen G12 of Fig. 16, information on the selected patent document (e.g., the title of the invention, the publication number, the name of the applicant, the filing date, the abstract, etc.) is displayed. In addition, in the screen G12 of Fig. 16, for each example described in the selected patent document, the names of the raw materials constituting the composition (e.g., a polyphenylene ether resin composition) and the blending amounts of each of the raw materials are displayed, and the properties at that time (e.g., the glass transition temperature Tg, the dielectric constant, the dielectric loss tangent, the tensile strength, and in the case of a cosmetic example, the moisturizing property and the hair gloss, etc.) are displayed. In the screen G12 of Fig. 16, "AI product search" buttons B121, B122, B123, and B124 are shown as examples associated with the respective examples. When one of these buttons is pressed, the screen transitions to screen G13 of Fig. 17. Here, as an example, the case where the "AI product search" button B121 associated with Example 1 of the selected patent document is pressed will be described below.

[0134] FIG. 17 is an example of a screen displayed when the "AI product search" button B121 associated with Example 1 of the selected patent document is pressed on the screen G12 of FIG. 16. On the screen G13 of FIG. 17, "Change" buttons B131 to B136 are displayed in association with each of the raw materials constituting the composition (e.g., a polyphenylene ether resin composition). By pressing these "Change" buttons B131 to B136, the user can change the corresponding raw material to another raw material and / or change the blending amount. Hereinafter, a case where, for example, the "Change" button B131 is pressed on the screen G13 of FIG. 17 will be described as an example.

[0135] FIG. 18 is an example of a screen when, for example, the "Change" button B131 is pressed on the screen G13 of FIG. 17. On the screen G14 of FIG. 18, a plus button associated with "Product" is shown, and when the plus button is pressed, a selection screen for candidate raw materials is displayed, for example, as a pop-up. When a user selects a candidate raw material on this selection screen for candidate raw materials, the candidate raw material (here, as an example, the product name) is displayed as a label on the screen G14. In the example of the screen G14, "PP-600", "SMA-EF-40", and "NORYLTM PPE 640" are selected by the user as candidate raw materials, and as a result, a label L141 for "PP-600", a label L142 for "SMA-EF-40", and a label L143 for "NORYLTM PPE 640" are displayed. Also shown are a text box T141 for inputting the lower limit of the blending amount of these candidate raw materials, and a text box T142 for inputting the upper limit of the blending amount of these candidate raw materials. Also displayed are a "Confirm" button B141 and a "Cancel" button B142. Pressing the "Confirm" button B141 confirms the selected candidate raw material and the blending amount range. Pressing the "Cancel" button B142 cancels the selected candidate raw material and blending amount range. When the "Search for product with the above conditions" button B143 is pressed on screen G14, the screen transitions to a screen showing the predicted results of the physical property values, for example, as shown in FIG. 19.

[0136] Although an example in which the candidate raw materials are directly specified by the user has been described, the present invention is not limited to this, and the user may specify a substance category (e.g., thermoplastic resin, etc.). In this case, at least one processor (e.g., processor 26) may determine candidate raw materials belonging to the substance category (e.g., thermoplastic resin, etc.) specified by the user. Specifically, for example, the substance category may be stored in association with identification information for identifying the raw materials in storage device 23, and at least one processor (e.g., processor 26) may determine candidate raw materials by referring to storage device 23 and reading out raw materials corresponding to the substance category specified by the user from storage device 23 as candidate raw materials.

[0137] FIG. 19 is an example of a screen showing the predicted results of physical properties in the fourth embodiment. In the screen G15 of FIG. 19, a graph is shown showing predicted physical properties for all combinations of the candidate raw materials determined by grid search in the candidate raw materials and their respective candidate blending amounts in the blending amount range set in FIG. 18. In the screen G15 of FIG. 19, a select box B151 for selecting the horizontal axis (x-axis) of the graph and a select 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. On the graph of the screen G15 of FIG. 19, as an example, the physical properties of an example are plotted, and predicted values ​​of physical properties are plotted for each combination of the candidate raw materials and the respective candidate blending amounts of the candidate raw materials. This allows the user to compare predicted values ​​of physical properties, and therefore allows the user to select the optimal combination of the candidate raw materials and their blending amounts. For example, when the mouse is placed over one of the plots in the graph, information including the name (e.g., product name or substance name) of the candidate raw material corresponding to the plot over which the mouse is placed, the blend amount, and predicted values ​​of the physical properties (e.g., glass transition temperature Tg, dielectric constant, dielectric tangent, peel strength) is displayed as an example in a pop-up, and also, as an example, on the right side of screen G15, the name of the candidate raw material corresponding to the plot over which the mouse is placed, the blend amount, predicted values ​​of the physical properties, and a link L151 labeled "Details" are displayed. When this link L151 is pressed, details of the candidate raw material are displayed.

[0138] To output screen G15 of Fig. 19, the following steps may be executed. At least one processor (e.g., processor 26) determines a plurality of candidate values ​​for the amount of another ingredient within a range specified by the user when a part of the ingredients constituting the target composition is replaced with another ingredient specified by the user. Here, the target composition is, for example, one described in an example selected by the user in a patent document selected by the user. At least one processor (e.g., processor 26) determines predicted values ​​of the physical property for all of the determined candidate values ​​of the plurality of quantities, and outputs information for presenting the predicted values ​​of the physical property in association with each of the candidate values ​​of the plurality of quantities.

[0139] Here, in the step of determining the predicted value of the physical property, at least one processor (e.g., processor 26) obtains the predicted value of the physical property by inputting at least one of a SMILES string or chemical graph structure data or product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials into a first machine learning model, or by inputting a set of numerical values ​​based on at least one of a chemical fingerprint, a SMILES string or chemical graph structure data or product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials into a second machine learning model.

[0140] Here, the first machine learning model is a model in which at least one of the SMILES character strings, chemical graph structure data, product names, or substance names corresponding to each raw material and the amount of each raw material are input, and parameters are adjusted so that the output can be predicted from the input using a learning data set in which the physical property value of the prediction target physical property name is output. The second machine learning model is a model in which at least one of the chemical fingerprints, the SMILES character strings, chemical graph structure data, product names, or substance names corresponding to each raw material and the amount of each raw material are input, and a group of numerical values ​​obtained by conversion using at least one of the chemical fingerprints, the SMILES character strings, chemical graph structure data, product names, or substance names corresponding to each raw material and the amount of each raw material are input, and parameters are adjusted so that the output can be predicted from the input using a learning data set in which the physical property value of the prediction target physical property name is output. With this configuration, the user can compare the predicted values ​​of the physical properties, and therefore can select the optimal combination of candidate raw materials and the blending amounts.

[0141] At least a part of the computer system 2 described in the above embodiment may be configured with hardware or software. When configured with software, a program that realizes at least a part of the functions of the computer system 2 may be stored in a computer-readable recording medium and read and executed by a computer. The recording medium is not limited to removable ones such as magnetic disks and optical disks, but may be fixed recording media such as hard disk drives and memories.

[0142] Also, a program that realizes at least a part of the functions of the computer system 2 may be distributed via a communication line (including wireless communication) such as the Internet. Furthermore, the program may be encrypted, modulated, or compressed and distributed via a wired line or wireless line such as the Internet, or stored on a recording medium.

[0143] Furthermore, the computer system 2 may be operated by one or more information devices. When multiple information devices are used, one of the devices may be a computer, and the computer may execute a predetermined program to realize the functions of at least one of the means of the computer system 2.

[0144] In the method invention, all the steps may be realized by automatic control using a computer. Alternatively, each step may be performed by a computer while the progress between steps is controlled manually. Furthermore, at least some of the steps may be performed manually.

[0145] As described above, the present invention is not limited to the above-described embodiment as it is, and in the implementation stage, the components can be modified and embodied without departing from the gist of the invention. In addition, various inventions can be formed by appropriately combining the multiple components disclosed in the above-described embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, components from different embodiments may be appropriately combined. [Explanation of symbols]

[0146] S Information Processing System 1 Device 11 Input Interface 12 Communication Module 13 Storage device 14. Memory 15 Output Interface 16 processors 17 Display 2. Information Processing System 21 Input Interface 22 Communication Module 23 Storage device 24 Memory 25 Output Interface 26 processors

Claims

1. at least one processor that obtains a combination of information identifying each of the raw materials received from a user and the amount of each of the raw materials, and inputs at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials into a first machine learning model, or inputs a group of values ​​based on at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials into a second machine learning model, thereby obtaining a predicted value of a physical property value of a predicted target physical property name for a composition composed of each of the raw materials; The first machine learning model is a model in which parameters are adjusted so that an output can be predicted from an input using a learning data set in which at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or a substance name corresponding to each raw material and the amount of each raw material are input, and a physical property value of a prediction target physical property name is output; The second machine learning model is an information processing system in which parameters are adjusted so that the output can be predicted from the input using a learning dataset that receives as input a group of numerical values ​​obtained by converting at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a 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 name of the physical property to be predicted.

2. The first machine learning model or the second machine learning model is constructed by collectively learning a plurality of physical properties of raw materials used in a plurality of fields. The information processing system according to claim 1 .

3. The raw material contains an additive, The input of the first machine learning model or the second machine learning model includes a type of raw material or a set of values ​​based on the type, The processor further inputs the type of raw material or a group of values ​​based on the type acquired from the user into the first machine learning model or the second machine learning model, thereby acquiring a predicted value of the physical property value of the prediction target physical property name for a composition composed of each of the raw materials.

3. The information processing system according to claim 1 or 2.

4. The input of the first machine learning model or the second machine learning model further includes a physical property of a raw material or a set of numerical values ​​based on the physical property, The processor further inputs the physical properties of the raw materials or a group of numerical values ​​based on the physical properties acquired from the user into the first machine learning model or the second machine learning model, thereby acquiring a predicted value of the characteristic value of the prediction target physical property name for a composition composed of each of the raw materials. The information processing system according to claim 1 .

5. An input of the first machine learning model or the second machine learning model includes a characteristic of a raw material or a set of values ​​based on the characteristic; The processor further inputs the characteristics of the raw materials or a group of numerical values ​​based on the characteristics acquired from the user into the first machine learning model or the second machine learning model, thereby acquiring a predicted value of the physical property value of the prediction target physical property name for a composition composed of each of the raw materials. The information processing system according to claim 1 .

6. The input of the first machine learning model or the second machine learning model includes information about process conditions and / or equipment used in the process or a set of values ​​based on the information; The processor further inputs information on process conditions and / or equipment used in the process acquired from a user, or a group of numerical values ​​based on the information, into the first machine learning model or the second machine learning model to obtain a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials. The information processing system according to claim 1 .

7. The processor, Extracting at least one raw material candidate based on at least one of the raw material product name, the raw material manufacturer name, the raw material category, and the raw material use received from the user; outputting information for displaying the extracted raw material candidates in a selectable manner; The information processing system according to claim 1 .

8. At least one processor; The processor, A combination of raw materials and the amount of each raw material is regarded as a parent population, and a child population of the next generation is produced according to a genetic algorithm. Calculating a fitness function value using the combination of raw materials and the amount of each of the raw materials represented by the offspring population of the next generation produced; The process of creating a next-generation child population is repeated until a termination condition based on a physical property range specified by the user is met, and when the termination condition is met, information including the combination of raw materials indicated by each of the last created next-generation child populations and the amount of each of the raw materials is output. Information processing system.

9. The processor acquires at least one set of desired property value ranges input by a user for a desired property name, and inputs at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or a substance name corresponding to each of the raw materials included in the initial combination of raw materials and an amount of each of the raw materials into a first machine learning model, or inputs a group of numerical values ​​based on at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or a substance name corresponding to each of the raw materials and an amount of each of the raw materials into a second machine learning model, thereby acquiring a predicted value of a property value of the physical property name; Narrowing down the obtained predicted values ​​of the physical properties to a combination of raw materials and a combination of amounts of each of the raw materials that satisfy a desired range of the physical properties specified by the user; The narrowed down combination of raw materials and the amounts of each of the raw materials are regarded as a parent population, and a child population of the next generation is created according to a genetic algorithm. The first machine learning model is a model in which parameters are adjusted so that an output can be predicted from an input using a learning data set in which at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or a substance name corresponding to each raw material and the amount of each raw material are input, and a physical property value of a prediction target physical property name is output; The second machine learning model is a model in which a group of numerical values ​​obtained by converting at least one of a chemical fingerprint, a SMILES character 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 are input, and parameters are adjusted so that an output can be predicted from an input using a learning data set in which a physical property value of a physical property name is output. The information processing system according to claim 8.

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

11. The processor further outputs a predicted value of the property value of the desired property name exhibited by each of the offspring population of the next generation finally produced.

11. The information processing system according to claim 9 or 10.

12. When one of the outputted raw materials is selected by the user, the processor outputs contact information of a sales company that sells the raw material and / or detailed information of the raw material.

12. An information processing system according to claim 8.

13. At least one storage device storing data relating to embodiments of patent documents; At least one processor; Equipped with The processor, Extracting examples that use raw materials designated by the user from among the examples stored in the storage device and / or examples that do not use the input raw materials from among raw materials that have similar physical property values ​​realized by the combination of the raw materials; Output information for displaying a list of applicants or patent owners of patent documents in which the extracted embodiments are described Information processing system.

14. After extracting the examples, the processor narrows down the examples to raw materials that have similar physical property values ​​when the target raw material designated by the user is replaced with another raw material. The information processing system according to claim 13.

15. When narrowing down raw materials whose physical property values ​​are similar when a target raw material designated by a user is replaced with another raw material, the processor inputs a pair of the other raw material and the amount of the other raw material into the trained machine learning model of the first embodiment to determine predicted values ​​of the physical property values ​​of the predicted target physical property name, and compares each of the predicted values ​​of the physical property values ​​with the physical property values ​​of the target raw material to extract raw materials whose predicted physical property values ​​are equal to or greater than the physical property values ​​of the raw material designated by the user, or raw materials similar to the raw material.

15. The information processing system according to claim 14.

16. The list includes similar ingredients described in the extracted examples as ingredients to be substituted.

16. An information processing system according to any one of claims 13 to 15.

17. At least one processor; The at least one processor obtains a target physical property value specified by a user; Obtain the range of blend amounts specified by the user; A plurality of candidate blend amounts are determined for each of the candidate raw materials within a range of blend amounts designated by a user, and physical property values ​​are predicted for all combinations of the candidate raw materials and the respective candidate blend amounts of the candidate raw materials; Outputting a set of candidate raw materials and respective quantities of the candidate raw materials based on a comparison of the target physical property value and each predicted physical property value. Information processing system.

18. The at least one processor obtains a substance category designated by a user, and extracts each of the raw materials belonging to the substance category designated by the user as the candidate raw materials.

18. The information processing system according to claim 17.

19. At least one processor; the at least one processor determines a plurality of candidate values ​​for the amount of another ingredient designated by the user when a part of the ingredient constituting the target composition is replaced with the other ingredient designated by the user; determining predicted values ​​of physical properties for all of the determined candidate values ​​of the plurality of quantities; outputting information for presenting a predicted value of the physical property value in correspondence with each of the candidate values ​​of the quantities; In the step of determining the predicted value of the physical property, the at least one processor inputs at least one of a SMILES character string, a chemical graph structure data, a product name, or a substance name corresponding to each of the raw materials and an amount of each of the raw materials to a first machine learning model, or inputs a set of numerical values ​​based on at least one of a chemical fingerprint, a SMILES character string, a chemical graph structure data, a product name, or a substance name corresponding to each of the raw materials and an amount of each of the raw materials to a second machine learning model, thereby obtaining a predicted value of the physical property; The first machine learning model is a model in which parameters are adjusted so that an output can be predicted from an input using a learning data set in which at least one of a SMILES character string, a chemical graph structure data, a product name, or a substance name corresponding to each of raw materials and the amount of each of the raw materials is input, and a physical property value of a prediction target physical property name is output; The second machine learning model is an information processing system in which parameters are adjusted so that the output can be predicted from the input using a learning dataset that receives as input a group of numerical values ​​obtained by converting at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a 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 name of the physical property to be predicted.

20. a computer that can reference at least one storage device that stores a first machine learning model in which at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials is input, and parameters are adjusted so that an output can be predicted from the input using a learning data set in which the input is a physical property value of a prediction target physical property name, or a second machine learning model in which at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials are input, and parameters are adjusted so that an output can be predicted from the input using a learning data set in which the output is a physical property value of a prediction target physical property name, A program for executing a procedure for acquiring a combination of information identifying each of the raw materials received from a user and the amount of each of the raw materials, and inputting at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials into a first machine learning model, or inputting a group of numerical values ​​based on at least one of a chemical fingerprint, a SMILES character string, or chemical graph structure data, or a product name or substance name corresponding to each of the raw materials and the amount of each of the raw materials into a second machine learning model, thereby acquiring a predicted value of the physical property value of the predicted target physical property name for a composition composed of each of the raw materials.

21. On the computer, A step of generating a child population of the next generation according to a genetic algorithm by regarding a combination of raw materials and a set of amounts of each of the raw materials as a parent population; calculating a fitness function value using the combination of raw materials and the amount of each of the raw materials represented by the offspring population of the next generation; a step of repeating the process of producing a next-generation child population until a termination condition based on a physical property range designated by a user is satisfied, and outputting information including the combination of raw materials indicated by each of the last produced next-generation child populations and the amount of each of the raw materials when the termination condition is satisfied; A program for executing the above.

22. A computer that can refer to at least one storage device in which data related to the embodiments of the patent document is stored, A step of extracting examples that use raw materials designated by the user from among the examples stored in the storage device and / or examples that do not use the input raw materials from among raw materials that have similar physical property values ​​realized by the combination of the raw materials; outputting information for displaying a list of applicants or patent owners of patent documents in which the extracted embodiments are described; A program for executing the above.

23. On the computer, A procedure for obtaining target physical property values ​​specified by a user; obtaining a range of blend amounts specified by a user; determining a plurality of candidate blend amounts within a range of blend amounts designated by a user for each of the candidate raw materials, and predicting physical property values ​​for all combinations of sets of candidate raw materials and the respective candidate blend amounts of the candidate raw materials; outputting a set of candidate raw materials and respective quantities of the candidate raw materials based on a comparison of the target physical property value with each predicted physical property value; A program for executing the above.

24. On the computer, A step of determining a plurality of candidate values ​​for the amount of another raw material specified by a user when a part of the raw material constituting the target composition is replaced with another raw material specified by a user, within a range specified by the user; determining predicted values ​​of physical properties for all of the determined candidate values ​​of the plurality of quantities; outputting information for presenting a predicted value of the physical property value in correspondence with each of the candidate values ​​of the quantities; A program for executing In the step of determining the predicted value of the physical property value, at least one of a SMILES character string, a 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 are input to a first machine learning model, or a group of numerical values ​​based on at least one of a chemical fingerprint, a SMILES character string, a 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 are input to a second machine learning model, thereby obtaining a predicted value of the physical property value; The first machine learning model is a model in which parameters are adjusted so that an output can be predicted from an input using a learning data set in which at least one of a SMILES character string, a chemical graph structure data, a product name, or a substance name corresponding to each of raw materials and the amount of each of the raw materials is input, and a physical property value of a prediction target physical property name is output; The second machine learning model is a program in which a group of numerical values ​​obtained by converting at least one of a chemical fingerprint, a SMILES character string, or a chemical graph structure data, or a product name or a substance name corresponding to each raw material and the amount of each raw material are input, and parameters are adjusted so that the output can be predicted from the input using a learning dataset that outputs the physical property value of the physical property name to be predicted.

Images