Material recommendation device, material recommendation method, and material recommendation program

Abstract

To make it easy to recommend a similar material in a different field.SOLUTION: A material recommendation device can access a material database which stores, for each material, a field for which the material is used, feature information of the material, and a numeric vector of the material being embedded expression of the feature information, and executes: an acquisition process for acquiring feature information of a first field; a first conversion process for converting the feature information of the first field acquired by the acquisition process into a first numeric vector; a second conversion process for converting the first numeric vector converted by the first conversion process into a second numeric vector of a second field different from the first field; a determination process for determining a recommend object material of the second field from the material database on the basis of the second numeric vector of the second field converted by the second conversion process; and an output process for outputting the recommend object material of the second field determined by the determination process.SELECTED DRAWING: Figure 13

Description

[Technical Field]

[0001] The present invention relates to an ingredient recommendation device, an ingredient recommendation method, and an ingredient recommendation program for recommending ingredients. [Background technology]

[0002] Patent Document 1 discloses a transaction support system that supports the conclusion of a transaction by determining provision requirements for providing a service. This transaction support system includes: a service setting means that sets a service candidate group including multiple services that can be provided to a user in association with service information for each service used for selecting a service; an information item setting means that sets, for each service included in the service candidate group, information items required by the supplier to determine the provision requirements; a service selection means that selects a service to be provided in response to a user's request from the service candidate group using the service information and selection information obtained based on a user's instruction; a request acquisition means that acquires request information for specifying the user's request for the information items based on a user's instruction; and a determination means that determines the provision requirements based on the request information.

[0003] Patent Document 2 discloses a transaction support device that automatically finds products in mail order transactions that match a seller's sales conditions and a buyer's purchase conditions. This transaction support device includes a sales condition acquisition means for acquiring information about the sales conditions of products set by the seller, a purchase condition acquisition means for acquiring information about the purchase conditions of products set by the buyer, a determination means for collating the information about the sales conditions acquired by the sales condition acquisition means with the information about the purchase conditions acquired by the purchase condition acquisition means to determine matching sales conditions and purchase conditions, a seller notification means for notifying a seller terminal used by the seller that the sales conditions determined by the determination means to match the purchase conditions are available for sale, and a buyer notification means for notifying a buyer terminal used by the buyer that the purchase conditions determined by the determination means to match the sales conditions are available for purchase. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-139505 [Patent Document 2] Japanese Patent Publication No. 2022-148549 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the prior art documents can only recommend materials in the same field that are close to or similar to the characteristics of the material input by a user in a certain field, and do not take into consideration the recommendation of similar materials in different fields.

[0006] The present invention aims to facilitate the recommendation of similar materials in different fields. [Means for solving the problem]

[0007] A material recommendation device according to one aspect of the invention disclosed in this application is a material recommendation device having a processor that executes a program and a storage device that stores the program, and is capable of accessing a material database that stores, for each material, the field in which the material is used, feature information about the material, and a numerical vector of the material that is an embedded representation of the feature information. The processor executes the following processes: an acquisition process to acquire feature information for a first field; a first conversion process to convert the feature information for the first field acquired by the acquisition process into a first numerical vector; a second conversion process to convert the first numerical vector converted by the first conversion process into a second numerical vector for a second field different from the first field; a determination process to determine recommended materials for the second field from the material database based on the second numerical vector for the second field converted by the second conversion process; and an output process to output the recommended materials for the second field determined by the determination process. [Effects of the Invention]

[0008] According to the exemplary embodiment of the present invention, it is possible to facilitate the recommendation of similar materials in different fields. Objects, configurations and effects other than those described above will become apparent from the following description of the examples. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is an explanatory diagram showing the recommended range of materials. [Figure 2] FIG. 2 is an explanatory diagram illustrating an example of the system configuration of the material recommendation system. [Figure 3] FIG. 3 is a block diagram illustrating an example of the hardware configuration of a computer. [Figure 4] FIG. 4 is an explanatory diagram showing an example of a field DB. [Figure 5] FIG. 5 is an explanatory diagram showing an example of the material DB. [Figure 6] FIG. 6 is an explanatory diagram showing an example of the material data table. [Figure 7] FIG. 7 is an explanatory diagram illustrating an example of the numeric vector table. [Figure 8] FIG. 8 is an explanatory diagram showing an example of text embedding. [Figure 9] FIG. 9 is an explanatory diagram illustrating an example of a function table. [Figure 10] FIG. 10 is a flowchart illustrating an example of a procedure for generating a function. [Figure 11] FIG. 11 is an explanatory diagram showing the relationship between a group of field numeric vectors from a source field and a group of field numeric vectors from a destination field. [Figure 12] FIG. 12 is an explanatory diagram showing the functions fha and fah. [Figure 13] FIG. 13 is a flowchart illustrating an example of a material recommendation process. [Figure 14] FIG. 14 is an explanatory diagram showing an example 1 of a material information input screen. [Figure 15] FIG. 15 is an explanatory diagram showing an example 2 of the material information input screen. [Figure 16] FIG. 16 is an explanatory diagram showing an example of material recommendation shown in S1301 to S1305. [Figure 17] FIG. 17 is an explanatory diagram showing an example of a material recommendation screen. DETAILED DESCRIPTION OF THE INVENTION

[0010] <Figure 1 Recommended range of materials> 1 is an explanatory diagram showing the recommended range of materials. In this embodiment, recycled plastic (hereinafter referred to as recycled material) is used as an example of a material that a user is considering purchasing, but the material is not limited to recycled material.

[0011] (A) When users search for recycled materials by narrowing down the search to the field in which they are considering using recycled materials, they encounter the problem of not being able to find recycled materials that meet their requirements. This is because the supply of recycled materials in the field that the user is aware of is limited compared to the total supply of recycled materials.

[0012] (B) There may be recycled materials suitable for the user's intended use or components outside the range recognized by the user. However, currently, there are only a limited number of systems that recommend materials that are different from the field recognized by the user. In consideration of this situation, this embodiment aims to expand the user's recycled material options by recommending recycled materials from a field that is different from the field the user is considering but that can be used for the user's intended use.

[0013] <Figure 2 System configuration example> 2 is an explanatory diagram showing an example of the system configuration of a material recommendation system. The material recommendation system 200 includes a material recommendation device 201 and a terminal 202. The material recommendation device 201 and the terminal 202 are communicably connected via a network 203 such as the Internet, a LAN (Local Area Network), or a WAN (Wide Area Network).

[0014] The material recommendation device 201 is a computer that executes a material recommendation process to recommend materials. The material recommendation device 201 has a field DB 211 and a material DB 212. The field DB 211 is a database that stores information about fields. The material DB 212 is a database that stores information about materials. Note that the field DB 211 and the material DB 212 may be provided in a computer different from the material recommendation device 201. In this case, the other computer is connected to the material recommendation device 201 via the network 203 so as to be able to communicate with the material recommendation device 201.

[0015] The terminal 202 is a computer that transmits data to the material recommendation device 201, receives data transmitted from the material recommendation device 201, and displays the data. The terminal 202 executes input and output of data to the material recommendation device 201, but the material recommendation device 201 may also accept input by user operation and display data on a display.

[0016] <Figure 3: Example of hardware configuration of computer (material recommendation device 201, terminal 202)> FIG. 3 is a block diagram showing an example of the hardware configuration of a computer (material recommendation device 201, terminal 202). The computer 300 has a processor 301, a storage device 302, an input device 303, an output device 304, and a communication interface (communication IF) 305. The processor 301, the storage device 302, the input device 303, the output device 304, and the communication IF 305 are connected by a bus 306. The processor 301 controls the computer 300. The storage device 302 serves as a working area for the processor 301. The storage device 302 is a non-transitory or temporary recording medium that stores various programs and data. Examples of the storage device 302 include a read-only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), and a flash memory. The input device 303 inputs data. Examples of the input device 303 include a keyboard, a mouse, a touch panel, a numeric keypad, a scanner, a microphone, and a sensor. The output device 304 outputs data. Examples of the output device 304 include a display, a printer, and a speaker. The communication IF 305 connects to the network 203 and transmits and receives data.

[0017] <Figure 4 Field DB211> FIG. 4 is an explanatory diagram showing an example of the field DB 211. The field DB 211 stores a field table 400. The field table 400 has fields, a field ID 401 and a field 402. The field ID 401 is identification information that uniquely identifies the field 402. The field 402 is an area of ​​industries or products in which recycled materials can be used. The field 402 may be, for example, a JIS (Japanese Industrial Standards) classification or an International Patent Classification (IPC).

[0018] <Figure 5 Material DB212> 5 is an explanatory diagram showing an example of the material DB 212. The material DB 212 has a material data table 501, a numerical vector table 502, and a function table 503.

[0019] [Figure 6 Material data table 501] FIG. 6 is an explanatory diagram showing an example of a material data table 501. The material data table 501 has the following fields: ID 600, field 402, material 601, manufacturer 602, model number 603, use 604, component 605, processing method 606, characteristic A 607, and characteristic B 608. In particular, use 604, component 605, and processing method 606 are referred to as characteristic information. Note that characteristic A 607 and characteristic B 608 may also be included in the characteristic information. A combination of values ​​in each field in the same row forms an entry indicating material data that identifies the material 601.

[0020] Material 601 is the raw material or material that becomes the product. In this example, it is a recycled material. Manufacturer 602 is a company that produces material 601. Model number 603 is a number assigned by manufacturer 602 to uniquely identify material 601. Use 604 is a product to which material 601 is applied. Component 605 is a part that constitutes the product identified by use 604 and to which material 601 is applied. Processing method 606 is a method for processing material 601. Property A 607 and property B 608 are properties of material 601, such as impact strength and dielectric strength. The property is not limited to two, property A 607 and property B 608, and may be one, or may be three or more, such as property C, property D, ...

[0021] [Figure 7 Numerical Vector Table 502] 7 is an explanatory diagram showing an example of the numeric vector table 502. The numeric vector table 502 has the following fields: ID 600, field 402, vector name 701, and numeric vector 702. The vector name 701 is a character string indicating the name given to the numeric vector 702. The numeric vector 702 is a vector obtained by converting the material data (entry in FIG. 6) identified by the ID 600 using text embedding as shown in FIG. 8.

[0022] [Figure 8 Text Embedding] 8 is an explanatory diagram showing an example of text embedding. Text embedding is a process in which text is input into a computer program called an embedding model 800 and the input text is converted into a numerical vector that is easy to process by a language model (e.g., a large-scale language model). The embedding model 800 generates the numerical vector by using a distributed representation such as word2vec, GloVe, BERT, or the text-embedding-ada-002 model.

[0023] In FIG. 8 , as an example, the entry with ID 600 "001" is the input material data 801, but the same applies if ID 600 is other than "001". When the input material data 801 is input to the embedding model 800, a numeric vector Vh1 is output from the embedding model 800. The embedding model 800 may be implemented in the material recommendation device 201, or may be implemented in another computer different from the material recommendation device 201. The other computer is communicably connected to the material recommendation device 201 via the network 203.

[0024] When implemented in another computer, the material recommendation device 201 transmits input material data 801 to the other computer. The other computer receives the input material data 801, inputs it to the embedding model 800, and converts it into a numeric vector Vh1. The other computer transmits the numeric vector Vh1 to the material recommendation device 201. The material recommendation device 201 receives the numeric vector Vh1 and stores it in the numeric vector table 502.

[0025] The input material data 801 includes the field 402, material 601, manufacturer 602, application 604, component 605, processing method 606, characteristic A 607, and characteristic B 608, but some of these may be missing. Regardless of whether there are missing data, the numerical vector Vh1 is generated as a vector with the same number of dimensions.

[0026] [Figure 9 Function table 503] 9 is an explanatory diagram showing an example of the function table 503. The function table 503 contains a function f pq The function f pq The "p" in the formula is a lowercase alphabet that indicates the source field 901, and the "q" is a lowercase alphabet that indicates the destination field 902. For example, home appliances are represented by "h", automobiles by "a", IT equipment by "i", construction by "c", and general goods by "g". When the fields are not distinguished, it may be written as function f. For example, the function f ha is a function f that converts a numeric vector of a home appliance (h) into a numeric vector of a car (a). If there are n types of fields 402 (n is an integer of 2 or more), (n-1) functions f are generated for each field 402.

[0027] <Figure 10 Function generation process> FIG. 10 is a flowchart illustrating an example of a procedure for generating a function.

[0028] (Step S1000) 8 for each ingredient data in the ingredient data table 501. The numeric vectors generated by the text embedding are stored in the numeric vector table 502.

[0029] (Step S1001) The material recommendation device 201 acquires a source field numeric vector group and a destination field numeric vector group from the numeric vector table 502. The source field numeric vector group is a set of numeric vectors (source numeric vectors) converted from material data of the source field 402. The destination field material data group is a set of numeric vectors (destination numeric vectors) converted from material data of the destination field 402. The destination field 402 is a field 402 different from the source field 402.

[0030] (Step S1002) The material recommendation device 201 generates a function f that converts a source field numeric vector into a destination numeric vector. Specifically, for example, the material recommendation device 201 generates the function f by linear regression or Gaussian process regression using a group of source field numeric vectors and a group of destination field numeric vectors. If there are n types of fields 402, (n-1) functions f are generated for each field 402. Details of the generation of the function f will be described later with reference to FIG. 11 and FIG. 12.

[0031] The function f can be generated not only by linear regression or Gaussian process regression, but also by applying random forest, which is a type of machine learning that can be used for both unsupervised (classification problems) and supervised (regression problems) learning.

[0032] (Step S1003) The ingredient recommendation device 201 stores the generated function f in the function table 503. In this way, the function table 503 is constructed.

[0033] <Figure 11 Relationship between the source field numeric vector group and the destination field numeric vector group> 11 is an explanatory diagram showing the relationship between a source field numeric vector group and a destination field numeric vector group. In FIG. 11, the source field 901 is "home appliances" and the destination field 902 is "automobiles."

[0034] The source field numeric vector group 1101 is a set of numeric vectors 702 in the source field 901, and the destination field numeric vector group 1102 is a set of numeric vectors 702 in the destination field 902. The function f ha When generating the material recommendation device 201, the material recommendation device 201 extracts combinations of the numeric vectors 702 of the source field 901 and the destination field 902 that match the components 605 from the source field numeric vector group 1101 and the destination field numeric vector group 1102.

[0035] For example, a combination is extracted of a numeric vector 702(Vh1) in the source field 901 (home appliances) with an ID 600 of "001" and a numeric vector 702(Va1) in the source field 901 (automobiles) with an ID 600 of "101". Also, a combination is extracted of a numeric vector 702(Vh4) in the source field 901 (home appliances) with an ID 600 of "004" and a numeric vector 702(Va4) in the source field 901 (automobiles) with an ID 604 of "104". Also, a combination is extracted of a numeric vector 702(Vh5) in the source field 901 (home appliances) with an ID 600 of "005" and a numeric vector 702(Va5) in the source field 901 (automobiles) with an ID 600 of "105".

[0036] The ingredient recommendation device 201 uses the combinations extracted in this way as a learning data set and calculates the function f ha and function f ah Generate.

[0037] In FIG. 12, a combination of a numerical vector 702 in the source field 901 and a numerical vector 702 in the destination field 902 that match the component 605 is extracted, but a combination of a numerical vector 702 in the source field 901 and a numerical vector 702 in the destination field 902 that match other attributes (for example, either the use 604 or the processing method 606) instead of the component 605 may also be extracted.

[0038] In Figure 12, combinations of numerical vectors 702 in the source field 901 and the destination field 902 that match the component 605 are extracted, but it is also possible to extract combinations of numerical vectors 702 in the source field 901 and the destination field 902 that match at least two of the component 605, use 604, and processing method 606.

[0039] In Figure 12, combinations of numerical vectors 702 in the source field 901 and the destination field 902 that match the component 605 are extracted, but it is also possible to extract combinations of numerical vectors 702 in the source field 901 and the destination field 902 that match the combination of at least one of the component 605, use 604, and processing method 606 and at least one of the characteristics (characteristic A 607, characteristic B 608, ...).

[0040] <Figure 12 Function f ha and function f ah > Figure 12 shows the function f ha and function f ah The numerical vectors Vh1 to Vh5 are plotted in a vector space 1200. The vector space 1200 is a space configured with the number of dimensions of the numerical vector 702. The numerical vectors Vh1 to Vh5 are plotted in a vector space 1200. The vector space 1200 is a space configured with the number of dimensions of the numerical vector 702. ha When input to the function f, it is converted into the numerical vectors Va1 to Va5. Similarly, when the numerical vectors Va1 to Va5 are input to the function f, ah When input, they are converted into numeric vectors Vh1 to Vh5.

[0041] <Figure 13 Material recommendation process> 13 is a flowchart showing an example of the material recommendation process, which will be described below with reference to FIGS.

[0042] (Step S1301) The material recommendation device 201 acquires the input material characteristic information and proceeds to step S1302. Specifically, for example, the input material characteristic information is input into the material information input screen of Fig. 14 or 15. The material information input screen of Fig. 14 or 15 is displayed on a display, which is an example of the output device 304 of the material recommendation device 201 or the terminal 202, for example.

[0043] [Figure 14 Material information input screen example 1] 14 is an explanatory diagram showing material information input screen example 1. The material information input screen 1400 has an application field input field 1401, a material input field 1402, a use input field 1403, a component input field 1404, a processing method input field 1405, a property A input field 1406, a property B input field 1407, and a start button 1408.

[0044] The field of use input field 1401 is an input field for inputting a character string indicating the field 402. The material input field 1402 is an input field for inputting a character string indicating the material 601. The use input field 1403 is an input field for inputting a character string indicating the use 604. The component input field 1404 is an input field for inputting a character string indicating the component 605. The processing method input field 1405 is an input field for inputting a character string indicating the processing method 606. The characteristic A input field 1406 is an input field for inputting a character string indicating the characteristic A 607. The characteristic B input field 1407 is an input field for inputting a character string indicating the characteristic B 608. The start button 1408 is a user interface that accepts input to start the processing from step S1302 onwards.

[0045] The application field input field 1401 to the characteristic B input field 1407 may be a user interface that allows selection from a plurality of items displayed in a pull-down format.

[0046] [Figure 15 Material information input screen example 2] 15 is an explanatory diagram showing material information input screen example 2. The material information input screen 1500 has an application field selection section 1501, a material selection section 1502, a use selection section 1503, a component selection section 1504, a processing method selection section 1505, a characteristic A selection section 1506, a characteristic B selection section 1507, and a start button 1508.

[0047] The application field selection section 1501 is a user interface for selecting the field 402. The material selection section 1502 is a user interface for selecting the material 601. The use selection section 1503 is a user interface for selecting the use 604. The component selection section 1504 is a user interface for selecting the component 605. The processing method selection section 1505 is a user interface for selecting the processing method 606. The characteristic A selection section 1506 is a user interface for selecting the characteristic A 607. The characteristic B selection section 1507 is a user interface for selecting the characteristic B 608. The start button 1508 is a user interface for receiving input to start the processing from step S1302 onwards.

[0048] In FIGS. 14 and 15, "home appliances" is input as the field 402, "television" as the use 604, "switch" as the component 605, and "1.7" as the characteristic A 607, and the combination of these is acquired as input material characteristic information.

[0049] (Step S1302) 13, the ingredient recommendation device 201 performs text embedding (converts into a numerical vector 702) of the input ingredient feature information as shown in Fig. 8, and proceeds to step S1303. The numerical vector 702 converted from the input ingredient feature information is referred to as an input numerical vector.

[0050] (Step S1303) The ingredient recommendation device 201 extracts a function f for converting the field 402 of the input ingredient characteristic information to another field 402 from the function table 503, and proceeds to step S1304. Specifically, for example, the ingredient recommendation device 201 sets the field 402 of the input ingredient characteristic information as the source field 901, and the other field 402 as the destination field 902, and extracts the corresponding function f from the function table 503. Since the destination field 902 is not specified from the input ingredient characteristic information, a field 402 other than the source field 901 becomes the destination field 902. Therefore, if the source field 901 is "home appliances", the function f ha , f hi , fhc , f hg , ... are extracted. If the corresponding function f has not been generated, the ingredient recommendation device 201 executes the function generation process shown in FIG.

[0051] (Step S1304) The ingredient recommendation device 201 converts the input numeric vector into an output numeric vector using each of the extraction functions f in step S1303, and then proceeds to step S1305.

[0052] (Step S1305) The material recommendation device 201 determines materials to be recommended in other fields 402, and proceeds to step S1306. Specifically, for example, the material recommendation device 201 identifies a numeric vector 702 in other fields 402 that is close to the output numeric vector, and identifies the ID 600 of the numeric vector 702 from the numeric vector table 502. Then, the material recommendation device 201 extracts an entry that becomes material data for the identified ID 600 from the material data table 501. The entry extracted in this way indicates materials to be recommended in other fields.

[0053] Furthermore, the numerical vectors 702 in other fields 402 that are close to the output numerical vector may be the top N (N is a predetermined integer of 1 or greater) numerical vectors 702 in order of shortest Euclidean distance between the vector and the output numerical vector, or may be numerical vectors 702 whose Euclidean distance between the vector and the output numerical vector is equal to or less than a predetermined distance.

[0054] (Step S1306) The material recommendation device 201 outputs the determination result in step S1305. Specifically, for example, the material recommendation device 201 outputs entries (material data) indicating materials to be recommended in other fields 402. The material recommendation device 201 displays the entries (material data) indicating materials to be recommended in other fields 402 on a display which is an example of the output device 304 of the material recommendation device 201, or outputs them so as to be displayable on a display which is an example of the output device 304 of the terminal 202.

[0055] [Figure 16 Recommended materials example] FIG. 16 is an explanatory diagram showing an example of ingredient recommendation shown in S1301 to S1305. In step S1301, input ingredient characteristic information 1600 is acquired. In step S1302, the input ingredient characteristic information 1600 is converted into an input numerical vector 1601. In step S1302, for example, a function f ha is extracted.

[0056] In step S1304, the input numeric vector Vhx is converted into a function f ha is input to the function f ha In step S1305, a numeric vector 702 that is close to the output numeric vector Vax is identified in the vector space 1200. For example, a numeric vector Va1 that has the shortest inter-vector distance from the output numeric vector Vax is identified.

[0057] [Figure 17 Example of material recommendation screen] 17 is an explanatory diagram showing an example of an ingredient recommendation screen. The ingredient recommendation screen 1700 is displayed on a display, which is an example of the output device 304 of the ingredient recommendation device 201 or the terminal 202, for example.

[0058] The ingredient recommendation screen 1700 displays the input ingredient characteristic information 1600, a first recommendation result 1701, and a second recommendation result 1702. The first recommendation result 1701 is information that is output by acquiring the input ingredient characteristic information 1600 in step S1301 and executing steps S1302 to S1305. The first recommendation result 1701 includes a numeric vector 702 that is close to the output numeric vector Vax identified in step S1305.

[0059] The second recommendation result 1702 is a recommendation result obtained by a conventional method. For example, the material recommendation device 201 extracts from the material data table 501 entries that belong to the same field 402 as the field 402 of the input material characteristic information 1600 and have a characteristic A607 that is the smallest difference from the characteristic A607 of the input material characteristic information 1600, and outputs the extracted entries as the second recommendation result 1702. This allows the user to compare the first recommendation result 1701 and the second recommendation result 1702.

[0060] In this way, according to this embodiment, it is possible to facilitate the recommendation of similar materials 601 in different fields 402. That is, it is possible to actively recommend recycled materials in fields 402 different from the field 402 input by the user. In other words, it is possible to realize an inference that if the component 605 in which a certain material 601 is used is the same, it can be used regardless of the field 402 or the purpose 604, or that even if the component 605 is not exactly the same, it may be usable if it has similar dimensions or a similar shape.

[0061] Determining what constitutes similar or dissimilar is like creating a thesaurus, and is therefore extremely time-consuming. Since it requires human judgment, it takes time and can lead to inconsistencies in the judgment. In the above-described embodiment, the determination of similarity or dissimilarity is made by converting at least one of the use 604, material 605, and processing method 606 into a numerical vector using text embedding, and the above estimation can be realized by calculating the Euclidean distance between the numerical vectors.

[0062] The present invention is not limited to the above-described embodiments, and includes various modifications and equivalent configurations within the spirit and scope of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to configurations including all of the described configurations. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added to, deleted from, or replaced with other configurations.

[0063] Furthermore, the aforementioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by having a processor interpret and execute a program that realizes each function.

[0064] Information such as programs, tables, files, etc. that realize each function can be stored in storage devices such as memory, hard disks, SSDs (Solid State Drives), or recording media such as IC (Integrated Circuit) cards, SD cards, and DVDs (Digital Versatile Discs).

[0065] In addition, the control lines and information lines shown are those that are considered necessary for explanation, and do not necessarily represent all the control lines and information lines that are necessary for implementation. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0066] 211 Field DB 212 Material DB 200 Material Recommendation System 201 Material recommendation device 202 terminals 301 processor 302 Storage Devices 402 fields 601 Material 604 Purpose 605 Materials 606 Processing method 701 Vector Names 702 Numeric Vector 800 Embedding Model 801 Input material data 901 Source field 902 Destination field 1101 Transformation source field numerical vector group 1102 Destination field numeric vector group 1200 Vector Space

Claims

1. An ingredient recommendation device having a processor that executes a program and a storage device that stores the program, a materials database is accessible that stores, for each material, the field in which the material is used, characteristic information of the material, and a numerical vector of the material that is an embedded representation of the characteristic information; The processor: an acquisition process for acquiring feature information of a first field; a first conversion process for converting the feature information of the first field acquired by the acquisition process into a first numerical vector; a second conversion process for converting the first numerical vector converted by the first conversion process into a second numerical vector in a second field different from the first field; a determination process for determining a recommended material for the second category from the material database based on the second numerical vector for the second category converted by the second conversion process; an output process for outputting the recommended material in the second field determined by the determination process; The material recommendation device is characterized by executing the above.

2. The material recommendation device according to claim 1, The material database stores, for each combination of a source field and a destination field, a set of functions for converting a numerical vector in the source field into a numerical vector in the destination field; The processor: execute an extraction process to extract a function from the function group, the function having the first field as the source field and the second field as the destination field; In the second conversion process, the processor converts the first numerical vector into a second numerical vector of the second field by inputting the first numerical vector into the function. The material recommendation device is characterized by:

3. The material recommendation device according to claim 2, The processor: A generation process for generating the function for each combination of the source field and the destination field by solving a regression problem between a set of numerical vectors that are embedded representations of materials in the source field and a set of numerical vectors that are embedded representations of materials in the destination field; a storing process of storing the function generated by the generating process in the material database; The material recommendation device is characterized by executing the above.

4. The material recommendation device according to claim 3, In the generation process, the processor generates the function by solving a regression problem based on a set of combinations of the numerical vectors in the destination field and the numerical vectors in the destination field that match at least a part of the feature information. The material recommendation device is characterized by:

5. The material recommendation device according to claim 1, The characteristic information includes at least one of the use, components, processing method, and properties of the material. The material recommendation device is characterized by:

6. The material recommendation device according to claim 1, In the determination process, the processor identifies a third numerical vector of the second field that is within a predetermined distance from the second numerical vector of the second field, and extracts feature information in which the third numerical vector is an embedded representation from the material database. The material recommendation device is characterized by:

7. The material recommendation device according to claim 1, In the determination process, the processor identifies a third numerical vector in the second field that is the shortest distance from the second numerical vector in the second field, and extracts feature information in which the third numerical vector is an embedded representation from the material database. The material recommendation device is characterized by:

8. An ingredient recommendation method executed by an ingredient recommendation device having a processor that executes a program and a storage device that stores the program, a materials database is accessible that stores, for each material, the field in which the material is used, characteristic information of the material, and a numerical vector of the material that is an embedded representation of the characteristic information; The processor: an acquisition process for acquiring feature information of a first field; a first conversion process for converting the feature information of the first field acquired by the acquisition process into a first numerical vector; a second conversion process for converting the first numerical vector converted by the first conversion process into a second numerical vector in a second field different from the first field; a determination process for determining a recommended material for the second category from the material database based on the second numerical vector for the second category converted by the second conversion process; an output process for outputting the recommended material in the second field determined by the determination process; A material recommendation method characterized by executing the above.

9. a processor that can access a materials database that stores, for each material, a field in which the material is used, characteristic information of the material, and a numerical vector of the material that is an embedded representation of the characteristic information; an acquisition process for acquiring feature information of a first field; a first conversion process for converting the feature information of the first field acquired by the acquisition process into a first numerical vector; a second conversion process for converting the first numerical vector converted by the first conversion process into a second numerical vector in a second field different from the first field; a determination process for determining a recommended material for the second category from the material database based on the second numerical vector for the second category converted by the second conversion process; an output process for outputting the recommended material in the second field determined by the determination process; A material recommendation program characterized by executing the above.

Images