Information processing apparatus, information processing method, and program

Abstract

To provide an information processor, a method and a program for facilitating prediction of a prediction target index even in a complicated experiment / manufacturing process.SOLUTION: Acquiring, for a target process group including at least one process, target process group data including information for identifying a material used in the process, a condition, and / or a prediction target index of a generated article; A conversion unit 261 that converts each piece of process data in the target process group data into graph structure data including an edge indicating a relationship with a node having fixed-length vector information, a feature amount acquisition unit 262 that acquires a feature amount vector representing a feature amount of the target process group by inputting the converted graph structure data to a learned graph neural network, and a predicted value acquisition unit 263 that acquires a predicted value of the prediction target index by inputting the acquired feature amount vector to a learned model that outputs the acquired predicted value of the prediction target index.SELECTED DRAWING: Figure 12

Description

[Technical Field]

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

[0002] Conventionally, various indicators (e.g., physical properties, performance indicators of manufactured products such as automobiles) related to products produced through various scientific experiments or industrial processes have been predicted. For example, Patent Document 1 discloses a technique for predicting the physical properties of recycled materials containing paint film, which are produced by removing the paint film from a resin molded body with a paint film, and of resin compositions containing such recycled materials. Specifically, Patent Document 1 linearly approximates the weight fraction of paint film particles present in multiple recycled resins containing paint film particles with different physical properties and the desired physical property to be predicted to obtain an approximation formula. A single physical property value is substituted into this approximation formula to calculate the weight fraction of multiple paint film particles corresponding to this physical property value. The calculated value and the corresponding size of the paint film particles are then plotted on a graph to obtain an approximation line. This method is repeated to create multiple isoproperty approximation lines corresponding to several physical property values ​​and plot them on the same graph. To predict the physical properties of recycled resins, the weight fraction and size of the resin's paint film particles are measured, and the physical properties are predicted from the position of these values ​​on the isoproperty approximation line diagram. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-060149 [Non-patent literature]

[0004] [Non-Patent Document 1] Hao Yuan, Haiyang Yu, Jie Wang, Kang Li, and Shuiwang Ji. On explainability of graph neural networks via subgraph explorations. In International Conference on Machine Learning, pages 12241-12252. PMLR, 2021 [Non-patent document 2] Phillip E Pope, Soheil Kolouri, Mohammad Rostami, Charles E Martin, and Heiko Hoffmann. Explainability methods for graph convolutional neural networks. In Proceedings of the IEEE / CVF conference on computer vision and pattern recognition, pages 10772-10781, 2019 Summary of the Invention [Problem to be solved by the invention]

[0005] However, existing methods are unable to fully represent the complexity of actual experimental and manufacturing processes, such as multi-step reactions and the further mixing of materials created through different processes, making it difficult to predict the target indicators. In particular, when manufacturing processes differ significantly, it is difficult to incorporate collected data into predictions side by side. Even if the field of data collection were narrowed to a level that allows for side-by-side comparisons, there would be very little experimental data available, making it difficult to fully guarantee prediction accuracy.

[0006] The present invention has been made in consideration of the above-mentioned problems, and aims to provide an information processing device, an information processing method, and a program that make it possible to easily predict a target index even in complex experimental or manufacturing processes. [Means for solving the problem]

[0007] An information processing device according to a first aspect of the present invention includes: at least one processor; The at least one processor: an acquisition step of acquiring target process group data for a target process group including at least one process, the data including information identifying materials used in the process, conditions for the process, and / or predicted target indicators of products produced by the process; a conversion procedure for converting each piece of process data in the target process group data into graph structure data consisting of nodes each having fixed-length vector information and edges indicating relationships between the nodes; a feature acquisition step of acquiring a feature vector representing the feature of the target process group by inputting the converted graph structure data into a trained graph neural network; a predicted value acquisition step of acquiring a predicted value of the indicator to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the acquired indicator to be predicted; Execute.

[0008] According to this configuration, by representing the steps of an experiment or manufacturing process in a graph structure, it is possible to represent a variety of experiment or manufacturing processes in a consistent manner, and data with different formats for different experiment or manufacturing processes can be used uniformly for prediction by training a single model. Accordingly, it is possible to train a large amount of data together. As a result, it is possible to easily predict the target indicator even for complex experiment or manufacturing processes.

[0009] An information processing device according to a second aspect of the present invention is the information processing device according to the first aspect, The at least one processor: further executing an output procedure for outputting information for providing a graphical user interface that allows input of the material, the condition, and the prediction target index of the object in a graph structure for each step; In the acquisition step, the indexes of the materials, the conditions, and the objects input by the user via the graphical interface are acquired for the target process group.

[0010] This configuration makes it easier for the user to input the materials, conditions, and prediction target indices of the objects for each process.

[0011] An information processing device according to a third aspect of the present invention is the information processing device according to the first or second aspect, an identification step of calculating a contribution using an output value of an intermediate layer of the graph neural network or using information on a gradient of the output value of the graph neural network, and identifying an important material or condition based on the contribution; an output step of outputting information for visualizing the identified important materials or conditions in a graph structure; Further execute the following.

[0012] This configuration makes it easier for the user to check important materials or conditions on the graph structure.

[0013] An information processing device according to a fourth aspect of the present invention is the information processing device according to any one of the first to third aspects, The target process group data acquired by the acquisition procedure is information that includes, for each process, at least one material used in the process, zero or more conditions for the process, and an index of the product produced by the process.

[0014] An information processing device according to a fifth aspect of the present invention is the information processing device according to any one of the first to third aspects, The target process group data acquired by the acquisition procedure is information that includes, for each item produced by a process, at least one material used to produce the item, a process for producing the item and zero or more conditions for the process, and an index of the item produced by the process.

[0015] An information processing device according to a sixth aspect of the present invention is the information processing device according to any one of the first to fifth aspects, The trained graph neural network is a model trained using training data in which graph structure data consisting of nodes having information on fixed-length vectors corresponding to process groups and edges indicating relationships between the nodes and feature vectors is used as input and output, The trained model is a machine learning model trained using training data in which a feature vector is input and a predicted value of an index to be predicted is output.

[0016] An information processing device according to a seventh aspect of the present invention is the information processing device according to any one of the first to sixth aspects, The indicators of the substance are physical property values.

[0017] According to this configuration, it is possible to predict physical property values ​​for data from a wide variety of experimental processes using the same processing.

[0018] An information processing method according to an eighth aspect of the present invention includes: an acquisition step of acquiring target process group data for a target process group including at least one process, the data including information identifying materials used in the process, conditions for the process, and / or a predicted target index of an item generated by the process; a conversion procedure for converting each piece of process data in the target process group data into graph structure data consisting of nodes each having fixed-length vector information and edges indicating relationships between the nodes; The method includes a feature acquisition step of acquiring a feature vector representing the feature of a target process group by inputting the converted graph structure data into a trained graph neural network, and a predicted value acquisition step of acquiring a predicted value of the acquired indicator to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the indicator to be predicted.

[0019] According to this configuration, by representing the steps of an experiment or manufacturing process in a graph structure, it is possible to represent a variety of experiment or manufacturing processes in a consistent manner, and data with different formats for different experiment or manufacturing processes can be used uniformly for prediction by training a single model. Accordingly, it is possible to train a large amount of data together. As a result, it is possible to easily predict the target indicator even for complex experiment or manufacturing processes.

[0020] A program according to a ninth aspect of the present invention comprises: On the computer, an acquisition step of acquiring target process group data for a target process group including at least one process, the data including information identifying materials used in the process, conditions for the process, and / or predicted target indicators of products produced by the process; a conversion procedure for converting each piece of process data in the target process group data into graph structure data consisting of nodes each having fixed-length vector information and edges indicating relationships between the nodes; a feature acquisition step of acquiring a feature vector representing the feature of the target process group by inputting the converted graph structure data into a trained graph neural network; a predicted value acquisition step of acquiring a predicted value of the indicator to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the acquired indicator to be predicted; This is a program for executing the above.

[0021] According to this configuration, by representing the steps of an experiment or manufacturing process in a graph structure, it is possible to represent a variety of experiment or manufacturing processes in a consistent manner, and data with different formats for different experiment or manufacturing processes can be used uniformly for prediction by training a single model. Accordingly, it is possible to train a large amount of data together. As a result, it is possible to easily predict the target indicator even for complex experiment or manufacturing processes. [Effects of the Invention]

[0022] According to one aspect of the present invention, by representing the steps of an experimental or manufacturing process in a graph structure, it is possible to represent a variety of experimental or manufacturing processes in a consistent manner, and data of different experimental or manufacturing processes in various formats can be used uniformly for prediction by training a single model. Accordingly, it is possible to train a large amount of data together. As a result, it is possible to easily predict the target indicator even for complex experimental or manufacturing processes. [Brief explanation of the drawings]

[0023] [Figure 1] 1 is a schematic configuration diagram of an information processing system according to an embodiment of the present invention. [Figure 2] 1 is a schematic configuration diagram of an information processing apparatus according to an embodiment of the present invention; [Figure 3] FIG. 1 is a schematic diagram of a graph structure showing an example of one process. [Figure 4] 10 is an example of an experiment setting screen displayed on a terminal. [Figure 5] FIG. 10 is a diagram for explaining JSON data of a first example of target process group data. [Figure 6A] 10 is an example of condition data included in JSON data of the first example of target process group data. [Figure 6B] 10 is an example of physical property data included in JSON data of the first example of target process group data. [Figure 7A] FIG. 1 is a schematic diagram showing a first example of a target process group. [Figure 7B] 7B is an example of data representing the first example of the target process group of FIG. 7A in JSON format. [Figure 8] FIG. 10 is a diagram for explaining JSON data of a second example of target process group data. [Figure 9A] 10 is an example of data on processes included in JSON data of the first example of target process group data. [Figure 9B] 10 is an example of physical property data included in JSON data of the first example of target process group data. [Figure 10A] FIG. 10 is a schematic diagram showing a second example of a target process group. [Figure 10B] 10B is an example of data representing the second example of the target process group of FIG. 10A in JSON format. [Figure 11A] FIG. 2 is a schematic diagram showing a first transformation mode of a graph structure. [Figure 11B] FIG. 10 is a schematic diagram showing a second transformation mode of a graph structure. [Figure 12] FIG. 2 is a functional block diagram showing functions executed by a processor of the information processing device according to the present embodiment. [Figure 13] 10 is an example of a screen in which important nodes are visualized and displayed on a terminal according to the present embodiment. [Figure 14] FIG. 10 is a sequence diagram showing an example of a processing flow according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0024] Each embodiment will be described below with reference to the drawings. However, more detailed explanation than necessary may be omitted. For example, detailed explanation of already well-known matters or redundant explanation of substantially identical configurations may be omitted. This is to avoid unnecessary redundancy in the following explanation and to facilitate understanding by those skilled in the art. Prediction target indicators of an object produced by a process include physical property values ​​of the object and performance indicators of the object (e.g., a manufactured product such as a car). In this embodiment, physical property values ​​will be used as an example of prediction target indicators of an object produced by a process.

[0025] Fig. 1 is a schematic configuration diagram of an information processing system common to this 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 an information processing device 2. Each of the terminals 1-1, ..., 1-N is communicatively connected to the information processing device 2 via a communication network CN. The terminals 1-1, ..., 1-N are, for example, smartphones, tablet terminals, laptop computers, personal computers, or other computers. Hereinafter, the terminals 1-1, ..., 1-N are also collectively referred to as terminals 1. The information processing device 2 is, for example, a server.

[0026] Fig. 2 is a schematic diagram of an information processing device according to this embodiment. As shown in Fig. 2, the information processing device 2 includes, for 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, although one embodiment of the information processing device 2 is described here as including one processor 26, the information processing device 2 may include multiple processors, i.e., one or more processors. Also, one embodiment of the information processing device 2 is described here as including one storage device 13, but the information processing device 2 may include multiple processors, i.e., one or more storage devices.

[0027] The input interface 21 receives input from an administrator of the information processing device 2 (e.g., an employee of the management organization) and outputs an input signal corresponding to the received 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.

[0028] 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).

[0029] 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.

[0030] Next, a graph structure (tree structure) representing the process of producing a substance will be explained using Figure 3. Figure 3 is a schematic diagram of a graph structure representing an example of one process. In the graph structure of Figure 3, the node of process A1 is connected to the nodes of material a, material b, material c, condition a, and condition b via edges. This indicates that process A1 is carried out using material a, material b, and material c under conditions a and b. Here, the node of material a, which represents the material, is connected to the nodes of the substance name, amount added, unit of amount added, and character string when the substance is expressed in SMILES (Simplified Molecular Input Line Entry System) notation (also known as SMILES). This identifies the substance name, amount added, unit of amount added, and character string when the substance is expressed in SMILES notation for material a.

[0031] FIG. 4 shows an example of an experiment setting screen displayed on a terminal. As shown in the experiment setting screen G1 in FIG. 4, a user can set up a desired group of target processes (e.g., an experiment) by setting up at least one process in a graph structure consisting of nodes and edges. For example, process A1 is connected to the nodes for material a, material b, and material c via edges, and to the nodes for condition a and condition b via edges. This indicates that process A1 uses materials a, b, and c and is performed under conditions a and b. For example, process A2 is connected to the node for process A1 via edges, and to the node for condition c and the node for physical property a via edges. This indicates that process A2 is performed under condition c for the product generated in process A1. For example, process B is connected to the nodes for material e and material f via edges, and to the node for condition d via edges. This indicates that process B uses materials e and f and is performed under condition d. For example, process C is connected to the nodes of process A2 and process B via edges, to the nodes of condition e and condition f via edges, and to the node of physical property b via edges. This sets process C to be executed under conditions e and f using the objects generated in process A2 and the objects generated in process B.

[0032] For example, when the send button R1 in Fig. 4 is pressed, the predicted value of physical property a and the predicted value of physical property b are displayed on the terminal 1. In order to display the experiment setting screen G1 in Fig. 4 on the terminal 1, the processor 26 of the information processing device 2 executes an output procedure that outputs information for providing a graphical user interface that allows input of prediction target indicators of materials, conditions, and objects in a graph structure for each process, for example. In this case, the processor 26 of the information processing device 2 executes an acquisition procedure for acquiring, for a target process group including at least one process, information identifying the materials used in that process, the conditions of that process, and / or the predicted target index of the product generated by that process (collectively referred to as target process group data). Specifically, for example, in this acquisition procedure, the processor 26 acquires the materials, conditions, and product indexes input by the user via a graphical interface for the target process group. This configuration makes it easier for the user to input the materials, conditions, and product indexes for each process.

[0033] <First example of target process group data> A first example of the target process group data will be described with reference to FIGS. 5 to 7B. The first example of the target process group data is information including, for each process, at least one material used in that process, zero or more conditions for that process, and an index of the product produced by that process. FIG. 5 is a diagram illustrating JSON (JavaScript Object Notation) data of the first example of the target process group data. As shown in FIG. 5, in the first example of the target process group data, materials, conditions, and physical properties are stored in parallel under the process name in association with each other. The materials field contains a list of materials used in that process. Furthermore, the materials field may contain JSON data representing products produced in other processes to represent products produced in those other processes. The JSON data representing products produced in those other processes may include, for example, a list of materials used in those other processes, a list of conditions for those other processes, and a list of physical properties of the products produced in those processes. This allows a process tree such as that shown in FIG. 4 to be represented.

[0034] 6A shows an example of condition data included in the JSON data of the first example of target process group data. As shown in Fig. 6A, this condition includes, for example, the condition name, the value of that condition (e.g., temperature, pressure, etc.), and the unit of that value (e.g., °C for temperature, Pascal for pressure, etc.). 6B is an example of data on physical properties included in the JSON data of the first example of target process group data. As shown in Fig. 6B, the physical property includes, for example, the property name of the property, the measurement value of the property, the unit of the measurement value, the condition name as the measurement condition, the value of the measurement condition (e.g., temperature, pressure, etc.), and the unit of the value (e.g., °C for temperature, Pascal for pressure, etc.).

[0035] FIG. 7A is a schematic diagram showing a first example of a target process group. As shown in FIG. 7A, in this first example of a target process group, the output of the "polymerization" node is connected to the input of the "kneading" node, indicating that the "kneading" process is performed after the "polymerization" process. The polymerization node is connected to two material nodes, terephthalic acid and ethylene glycol, via an edge, and to two condition nodes, a reaction temperature of 80°C and a reaction time of 1 hour, via an edge. As a result, in this polymerization process, polymerization of the two materials, terephthalic acid and ethylene glycol, is performed under the conditions of a reaction temperature of 80°C and a reaction time of 1 hour. The kneading node is also connected to a polymerization node, a material node specifying the product name and melting point of the material, and a condition node specifying the equipment to be used. As a result, in the kneading process, the material is applied to the product produced by polymerization, and kneading is performed using the specified equipment. The kneading node is connected to a tensile strength node via an edge, and this tensile strength is connected to a condition node, temperature, via an edge. As a result, the physical properties of the product obtained by this kneading are shown to be a tensile strength of 200 MPa at a temperature of 25 degrees.

[0036] 7B is an example of data representing the first example of the target process group of FIG. 7A in JSON format. As shown in FIG. 7B, the first example of the target process group of FIG. 7A can be represented in JSON format in accordance with FIG. 5.

[0037] <Second example of target process group data> A second example of the target process group data will be described with reference to Figures 8 to 10B. The second example of the target process group data is information that includes, for each object produced by a process, at least one material used to produce the object, a process for producing the object and zero or more conditions for the process, and an index of the object produced by the process. FIG. 8 is a diagram illustrating JSON data of a second example of target process group data. As shown in FIG. 8, in the second example of target process group data, the JSON data for the item (also called the final product) finally produced in the target process group contains materials, processes, and physical properties in parallel. The materials are a list of materials used in the process. The materials included in this material list may contain items produced in the previous process stored in JSON format, and the items produced in the previous process are expressed by a process tree. The processes are a list of operations performed in the process of producing this final product. The physical properties are a list of the physical properties of the final product produced in this process.

[0038] Figure 9A shows an example of process data included in the JSON data of the first example of target process group data. In the example of Figure 9A, the process, or process, includes a process name, which is its name, and process conditions, which are the conditions for that process. The process conditions include a condition name, which is its name, the value of that condition (e.g., temperature, pressure, etc.), and the unit of that value (e.g., °C for temperature, Pascal for pressure, etc.).

[0039] 9B is an example of physical property data included in the JSON data of the first example of target process group data. In the example of Fig. 9B, the physical property includes the property name, which is the name of the property, the measurement value of the property, the measurement unit, which is the unit of the measurement value, and the measurement condition, and the measurement condition includes the condition name, which is the name of the property, the value of the condition (e.g., temperature, pressure, etc.), and the unit of this value (e.g., °C for temperature, Pascal for pressure, etc.).

[0040] FIG. 10A is a schematic diagram showing a second example of the target process group. Compared to FIG. 7A and the first example of the target process group, the material added during kneading is excluded. FIG. 10B is an example of data representing the second example of the target process group of FIG. 10A in JSON format. As shown in FIG. 10B, the second example of the target process group of FIG. 10A can be represented in JSON format in accordance with FIG. 8.

[0041] Next, two examples of how process data (for example, JSON data) is converted into a graph structure will be described below.

[0042] <First transformation mode of graph structure> Here, we will explain the first conversion mode of the graph structure. Fig. 11A is a schematic diagram showing the first conversion mode of the graph structure. In the first conversion mode, key information of JSON data, which is an example of target process group data, is represented in nodes. The key information includes, for example, the substance name, product name, value of the condition (e.g., temperature, pressure, etc.), and the unit of this value (e.g., °C for temperature, Pascal for pressure, etc.).

[0043] <Second transformation of graph structure> Next, we will explain the second conversion mode of the graph structure. Figure 11B is a schematic diagram showing the second conversion mode of the graph structure. In the second conversion mode, key information of JSON data, which is an example of target process group data, is assigned to edge attributes.

[0044] 12 is a functional block diagram showing functions executed by the processor of the information processing device 2 according to this embodiment. As shown in FIG. 12, the processor 26 of the information processing device 2 functions as a conversion unit 261, a feature acquisition unit 262, and a predicted value acquisition unit 263. First, processor 26 is assumed to have acquired target process group data for a target process group that includes at least one process, the data including information identifying the materials used in the process, the conditions of the process, and / or the predicted target indicators of the product produced by the process. The conversion unit 261 converts each piece of process data in the target process group data into graph structure data consisting of nodes with fixed-length vector information and edges indicating the relationships between them. Specifically, for example, the conversion unit 261 converts the molecular structure data included in the target process group data into a fixed-length vector using a molecular structure processing model. Here, this molecular structure processing model is, for example, an arbitrary algorithm that converts data representing a molecular structure (such as a character string when expressed in SMILES notation) into a fixed-length vector, or a model that has been trained using training data, for example, a model that has been trained using training data that receives data representing a molecular structure as input and outputs a fixed-length vector. Alternatively, for example, the conversion unit 261 converts the text data included in the target process group data into a fixed-length vector using a text information processing model. Here, this text information processing model is, for example, an arbitrary algorithm that converts text data into a fixed-length vector, or a model that has been trained using training data that receives a character string of a material name as input and outputs a fixed-length vector.

[0045] The feature acquisition unit 262 acquires feature vectors representing the features of the target process group by inputting graph structure data consisting of nodes having information on fixed-length vectors and edges showing the relationships between them to a trained graph neural network. Here, this graph neural network is a model trained using training data that inputs graph structure data consisting of nodes having information on fixed-length vectors corresponding to the process group and edges showing the relationships between them and outputs feature vectors.

[0046] The predicted value acquisition unit 263 acquires a predicted value of the target index to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the target index to be predicted. Here, the target index to be predicted is, for example, an index of an object (e.g., a physical property value) input by a user via a graphical interface displayed on the terminal 1. Here, this model is, for example, model j (M is a natural number and j is an integer from 1 to M) that outputs the physical property value of predicted physical property j, which is the target index to be predicted, among models 1 to M. Here, the storage device 23 stores trained models 1 to M, and each model j is a machine learning model (e.g., a neural network) trained using training data that inputs the feature vector and outputs the physical property value of predicted physical property j.

[0047] FIG. 13 is an example of a screen in which important nodes are visualized and displayed on a terminal according to this embodiment. Screen G2 in FIG. 13 includes, as an example, predicted values ​​for the physical properties to be predicted. Specifically, for example, predicted values ​​for physical property b are displayed in screen regions R21 and R22 on screen G2 in FIG. 13. On screen G2 shown in FIG. 13, important nodes (e.g., important material nodes or important condition nodes) in the graph structure representing the target process group are visualized in a manner different from other nodes. As an example of this visualization, in FIG. 13, important nodes (e.g., material a nodes and condition b nodes) may be displayed in a different color from other nodes. However, the visualization manner is not limited to this. Important nodes may be displayed larger than other nodes, may have a different shape from other nodes, or may have a different background pattern from other nodes.

[0048] Information for displaying this screen G2 is generated as follows. The processor 26 of the information processing device 2 calculates the contribution using the output values ​​of the intermediate layer of the graph neural network or using information on the gradient of the output values ​​of the graph neural network, and identifies important materials or conditions based on the contribution. As a result, important nodes are identified because the nodes in the graph structure representing the target process group include materials or conditions. This calculation of the contribution may be performed using a method in the field of explainable AI (see, for example, the SubgraphX ​​technology in Non-Patent Document 1 or the GradCam technology in Non-Patent Document 2). The processor 26 then outputs information for visualizing the identified important materials or conditions on the graph structure. Specifically, this information is, for example, information for visualizing nodes representing important materials or conditions in a manner different from other nodes.

[0049] This information is transmitted from the information processing device 2 to the terminal 1, and upon receiving this information, the terminal 1 uses this information to display a screen such as screen G2. This configuration makes it easy for the user to check important materials or conditions on a graph structure.

[0050] 14 is a sequence diagram showing an example of the flow of processing according to this embodiment. The following processing is performed by a processor, but the processor will not be explicitly shown to avoid redundant explanation. (Step S110) The terminal 1 accepts an operation to access the experiment setting screen.

[0051] (Step S120) Then, the terminal 1 transmits an HTTP request for the experiment setting screen to the information processing device 2.

[0052] (Step S210) The information processing device 2 transmits the information of the experiment setting screen to the terminal 1 in response to the HTTP request.

[0053] (Step S130) When the terminal 1 receives the information for the experiment setting screen, it displays the experiment setting screen using this information, thereby displaying, for example, screen G1 in FIG.

[0054] (Step S140) Next, the terminal 1 accepts from the user an operation to set a target process group and a press of the send button on this experiment setting screen.

[0055] (Step S150) Next, the terminal 1 transmits the target process group data obtained by the target process group setting operation to the information processing device 2. This transmitted target process group data may be the JSON data or a portion of the JSON data (e.g., numerical values, text, etc.). In the case of a portion of the JSON data (e.g., numerical values, text, etc.), the JSON data itself may be generated by the information processing device 2 based on that portion of the data, as described below.

[0056] (Step S220) When the information processing device 2 receives the target process group data, it converts each piece of process data in this target process group data into a fixed-length vector as necessary, and then converts it into graph structure data consisting of nodes having information about the fixed-length vectors and edges indicating the relationships between them. Note that when the target process group data received by the information processing device 2 is data that is a portion of the JSON data (for example, numerical values, text, etc.), the information processing device 2 may generate the JSON data from that portion of the data and convert the JSON data into graph structure data.

[0057] (Step S230) Next, the information processing device 2 inputs the graph structure data into the graph neural network to acquire a feature vector.

[0058] (Step S240) Next, the information processing device 2 inputs the feature vector into the model j to obtain the predicted value of the physical property to be predicted.

[0059] (Step S250) Next, the information processing device 2 identifies important materials or conditions as needed.

[0060] (Step S260) Next, the information processing device 2 transmits to the terminal 1 information for visualizing the predicted values ​​of the predicted object properties and the nodes of important materials or conditions.

[0061] (Step S160) When the terminal 1 receives the information for visualizing the predicted values ​​of the predicted object properties and the nodes of important materials or conditions, it displays the predicted values ​​of the predicted object properties and visualizes the important nodes of important materials or conditions. As a result, for example, screen G2 in FIG. 13 is displayed.

[0062] As described above, the information processing device 2 according to this embodiment includes at least one processor, and the at least one processor executes the following steps: an acquisition step of acquiring, for a target process group including at least one process, a process data group including information identifying the materials used in the process, the conditions for the process, and / or the target indicator of the product generated by the process; a conversion step of converting each of the process data in the process data group into graph structure data consisting of nodes having fixed-length vector information and edges indicating the relationships between them; a feature acquisition step of acquiring a feature vector representing the feature of the target process group by inputting the converted graph structure data into a trained graph neural network; and a prediction value acquisition step of acquiring a prediction value of the target indicator by inputting the acquired feature vector into a trained model that outputs a prediction value of the acquired prediction target indicator.

[0063] According to this configuration, by representing the steps of an experiment or manufacturing process in a graph structure, it is possible to represent a variety of experiment or manufacturing processes in a consistent manner, and data with different formats for different experiment or manufacturing processes can be used uniformly for prediction by training a single model. Accordingly, it is possible to train a large amount of data together. As a result, it is possible to easily predict the target indicator even for complex experiment or manufacturing processes.

[0064] At least a part of the information processing device 2 described in the above embodiment may be configured with hardware or software. If configured with software, a program that realizes at least a part of the functions of the information processing device 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 media such as magnetic disks and optical disks, but may also be fixed recording media such as hard disk drives and memories.

[0065] In addition, a program that realizes at least a part of the functions of the information processing device 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.

[0066] Furthermore, the information processing device 2 may be functioned by one or more information devices. When multiple information devices are used, at least one of the devices may be a computer, and the computer may execute a predetermined program to realize the functions as at least one means of the information processing device 2.

[0067] In the method invention, all processes (steps) may be realized by automatic control using a computer. Alternatively, each process may be performed by a computer, with progress control between processes being performed manually. Furthermore, at least some of the processes may be performed manually.

[0068] As described above, the present invention is not limited to the above-described embodiments, and the components can be modified and embodied in practice without departing from the spirit of the invention. Furthermore, various inventions can be formed by appropriately combining multiple components disclosed in the above-described embodiments. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components from different embodiments may be appropriately combined. [Explanation of symbols]

[0069] 1-1, ..., 1-N terminals 2. Information processing equipment 21 Input Interface 22 Communication Module 23 Storage device 24 memory 25 Output Interface 26 processors S Information Processing System

Claims

1. at least one processor; The at least one processor: an acquisition step of acquiring target process group data for a target process group including at least one process, the data including information identifying materials used in the process, conditions for the process, and / or a predicted target index of an item produced by the process; a conversion procedure for converting each piece of process data in the target process group data into graph structure data consisting of nodes each having fixed-length vector information and edges indicating relationships between the nodes; a feature acquisition step of acquiring a feature vector representing the feature of the target process group by inputting the converted graph structure data into a trained graph neural network; a predicted value acquisition step of acquiring a predicted value of the indicator to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the acquired indicator to be predicted; An information processing device that executes the above.

2. The at least one processor: further executing an output procedure for outputting information for providing a graphical user interface that allows input of the material, the condition, and the prediction target index of the object in a graph structure for each step; In the acquisition step, the indexes of the materials, the conditions, and the objects input by the user through the graphical interface are acquired for the target process group. The information processing device according to claim 1 .

3. an identification step of calculating a contribution using an output value of an intermediate layer of the graph neural network or using information on a gradient of the output value of the graph neural network, and identifying an important material or condition based on the contribution; an output step of outputting information for visualizing the identified important materials or conditions in a graph structure; The information processing apparatus according to claim 1 or 2, further comprising:

4. The target process group data acquired by the acquisition procedure is information including, for each process, at least one material used in the process, zero or more conditions for the process, and an index of an object produced by the process. The information processing device according to claim 1 .

5. The target process group data acquired by the acquisition procedure is information including, for each object produced by a process, at least one material used to produce the object, a process for producing the object and zero or more conditions for the process, and an index of the object produced by the process. The information processing device according to claim 1 .

6. The trained graph neural network is a model trained using training data in which graph structure data consisting of nodes having information on fixed-length vectors corresponding to process groups and edges indicating relationships between the nodes and feature vectors is used as input and output, The trained model is a machine learning model trained using training data in which a feature vector is input and a predicted value of an index to be predicted is output. The information processing device according to claim 1 .

7. The indicator of the substance is a physical property value. The information processing device according to claim 1 .

8. an acquisition step of acquiring target process group data for a target process group including at least one process, the data including information identifying materials used in the process, conditions for the process, and / or a predicted target index of an item produced by the process; a conversion procedure for converting each piece of process data in the target process group data into graph structure data consisting of nodes each having fixed-length vector information and edges indicating relationships between the nodes; a feature acquisition step of acquiring a feature vector representing the feature of the target process group by inputting the converted graph structure data into a trained graph neural network; a predicted value acquisition step of acquiring a predicted value of the indicator to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the acquired indicator to be predicted; An information processing method comprising:

9. On the computer, an acquisition step of acquiring target process group data for a target process group including at least one process, the data including information identifying materials used in the process, conditions for the process, and / or a predicted target index of an item produced by the process; a conversion procedure for converting each piece of process data in the target process group data into graph structure data consisting of nodes each having fixed-length vector information and edges indicating relationships between the nodes; a feature acquisition step of acquiring a feature vector representing the feature of the target process group by inputting the converted graph structure data into a trained graph neural network; a predicted value acquisition step of acquiring a predicted value of the indicator to be predicted by inputting the acquired feature vector into a trained model that outputs a predicted value of the acquired indicator to be predicted; A program to execute.

Images