Information display device and information display method

Abstract

To visualize contribution among nodes.SOLUTION: An information display device including a processor for executing a program and a storage device storing the program, can access to database storing a feature value of each of a plurality of nodes constituting a graph, an estimated value of the node, contribution levels of the node and the other nodes to the estimated value, and a connection relationship between the nodes. The information display device executes designation processing for receiving a designation of a node, acquisition processing for acquiring the contribution levels of the node and the other nodes to the estimated value of the node designated by the designation processing from the database, and output processing for outputting the contribution levels of the node and the other nodes acquired by the acquiring processing in a displayable manner.SELECTED DRAWING: Figure 13

Description

[Technical Field]

[0001] The present invention relates to an information display device and an information display method. [Background technology]

[0002] Patent Document 1 below discloses an information processing device that outputs input / output characteristics of a subnetwork. This information processing device performs processing on a neural network that includes a plurality of subnetworks, each having an input node that accepts input of data and a plurality of intermediate nodes that are connected to the input node in multiple stages, one for each type of data, and an output node that outputs data using output data from the plurality of subnetworks. This information processing device includes an output data acquisition unit that inputs a plurality of sample data to the input node of a first subnetwork, which is one of the plurality of subnetworks, and acquires a plurality of output data from the first subnetwork that correspond to each of the plurality of sample data, and an input / output characteristic output unit that outputs the input / output characteristics of the first subnetwork based on the sample data and the output data.

[0003] The following Patent Document 2 discloses a real-time graph-based embedding construction method for personalized content recommendation. This embedding method includes the steps of generating a graph showing the relationship between content to be recommended, learning the embedding of content using the graph, and performing user embedding by reflecting user activities related to the content in the learned graph. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-39710 [Patent Document 2] Patent Publication No. 2021-26779 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the above-mentioned conventional techniques, the relationship of contributions between nodes is not taken into consideration.

[0006] The present invention aims to visualize the contributions between nodes. [Means for solving the problem]

[0007] An information display device according to one aspect of the invention disclosed in the present application is an information display device having a processor that executes a program and a storage device that stores the program, an input having a plurality of nodes and an edge connecting two of the plurality of nodes; Configure the graph The aforementioned The processor is capable of accessing a database that stores a feature value of each of a plurality of nodes, an estimated value of the node, the contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes, and the processor performs a designation process of accepting designation of the node, an acquisition process of acquiring from the database the contribution of the node and other nodes to the estimated value of the node designated by the designation process, and a calculation process of calculating the contribution of the node and other nodes to the estimated value acquired by the acquisition process. The contributions of the node and the other nodes to the estimated value of the node are calculated using a first graph that shows the relationship between the distance between the node and the other nodes in the input graph and the contributions. and outputting the image data in a displayable manner.

[0008] An information display device according to another aspect of the invention disclosed in the present application includes a processor that executes a program and a storage device that stores the program, an input having a plurality of nodes and an edge connecting two of the plurality of nodes; Configure the graph The aforementioned The processor is capable of accessing a database that stores a feature value of each of a plurality of nodes, an estimated value of the node, the contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes, and the processor performs a designation process to accept designation of the node, and an acquisition process to acquire from the database the contribution of the node and other nodes to the estimated value of the node designated by the designation process; The contribution of the node to the estimated value of the node and the other node acquired by the acquisition process is calculated using a second graph showing the relationship between the distance between the node and the other node in the input graph and the contribution.and outputting the image data in a displayable manner. [Effects of the Invention]

[0009] According to the representative embodiment of the present invention, it is possible to visualize the contributions between nodes. Problems, configurations, and effects other than those described above will become clear from the following description of the embodiment. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing an example of the hardware configuration of an information display device. [Figure 2] FIG. 2 is an explanatory diagram illustrating an example of an input graph. [Figure 3] FIG. 3 is an explanatory diagram illustrating an example of the node DB. [Figure 4] FIG. 4 is an explanatory diagram illustrating an example of the edge DB. [Figure 5] FIG. 5 is an explanatory diagram showing an example of the configuration of a GCN. [Figure 6] FIG. 6 is an explanatory diagram showing an example of the EMBEDDING layer. [Figure 7] FIG. 7 is an explanatory diagram illustrating an example of the RGCN layer. [Figure 8] FIG. 8 is an explanatory diagram illustrating an example of an ReLU layer. [Figure 9] FIG. 9 is an explanatory diagram illustrating an example of a dropout layer. [Figure 10] FIG. 10 is an explanatory diagram showing an example of the max layer. [Figure 11] FIG. 11 is an explanatory diagram showing the connection relationship between the GCN layer and the concat layer. [Figure 12] FIG. 12 is an explanatory diagram illustrating an example of an MLP layer and a softmax layer. [Figure 13] FIG. 13 is an explanatory diagram showing an example of the first graph display. [Figure 14] FIG. 14 is an explanatory diagram showing an example of the second graph display. [Figure 15]FIG. 15 is a flowchart showing an example of a graph display processing procedure performed by the information display device. [Figure 16] FIG. 16 is an explanatory diagram showing another example of the first graph display. [Figure 17] FIG. 17 is an explanatory diagram showing another example of the second graph display. [Figure 18] FIG. 18 is a diagram illustrating an example of a GCN according to the second embodiment. [Figure 19] FIG. 19 is an explanatory diagram illustrating an example of a pooling layer. [Figure 20] FIG. 20 is an explanatory diagram illustrating the connection relationship between the GCN layer and the concat layer 503 according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION [Example]

[0011] <Example of hardware configuration for information display device> FIG. 1 is a block diagram showing an example of the hardware configuration of an information display device. The information display device 100 includes a processor 101, a storage device 102, an input device 103, an output device 104, and a communication interface (communication IF) 105. The processor 101, the storage device 102, the input device 103, the output device 104, and the communication IF 105 are connected via a bus 106. The processor 101 controls the information display device 100. The storage device 102 serves as a working area for the processor 101. The storage device 102 is a non-transitory or temporary recording medium that stores various programs and data. Examples of the storage device 102 include a read-only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), and a flash memory. The input device 103 inputs data. Examples of the input device 103 include a keyboard, a mouse, a touch panel, a numeric keypad, a scanner, a microphone, and a sensor. The output device 104 outputs data. The output device 104 includes, for example, a display, a printer, and a speaker. The communication IF 105 connects to a network and transmits and receives data.

[0012] <Input graph> 2 is an explanatory diagram showing an example of an input graph. The input graph 200 has a plurality of nodes N1 to N10 (10 in FIG. 2) and a plurality of edges E1 to E10 (10 in FIG. 2) connecting two of the nodes N1 to N10. When the nodes N1 to N10 are not distinguished, they are referred to as nodes N. When the edges E1 to E10 are not distinguished, they are referred to as edges E.

[0013] <Node DB> 3 is an explanatory diagram showing an example of a node DB. The node DB 300 is a database that stores information about each of the nodes N1 to N10 of the input graph 200. The node DB 300 is stored in the storage device 102. The node DB 300 has the following fields: a node ID 301, a feature amount 302, an estimated value 303, and a contribution degree 304.

[0014] Node ID 301 is identification information that uniquely identifies node N. Feature 302 is an n-dimensional vector consisting of factors F1 to Fn (n is an integer equal to or greater than 1). Estimated value 303 is a prediction result Y for each node N that is output as a result of inputting input graph 200 into a trained GCN (Graph Convolutional Network).

[0015] The contribution 304 is information that quantifies the contribution of nodes N1 to N10 to node N, calculated for each node N. A known method such as SHAP (Shapley Additive exPlanations) or LIME (Local Interpretable Model-agnostic Explanations) can be used to calculate the contribution. Typically, the sum of the contributions 304 from each of nodes N1 to N10 to a certain node N becomes the estimated value 303 of the node N.

[0016] <Edge DB> 4 is an explanatory diagram showing an example of an edge DB. The edge DB 400 is a database that stores information about each of the edges E1 to E10 of the input graph 200. The edge DB 400 is stored in the storage device 102. The edge DB 400 has, as fields, an edge ID 401, a first node ID 402, and a second node ID 403.

[0017] The edge ID 401 is identification information that uniquely identifies the edge E. The first node ID 402 is identification information that uniquely identifies the first node N. The second node ID 403 is identification information that uniquely identifies the second node N. The first node N is a node N connected to one end of the edge E, and the second node N is a node N connected to the other end of the edge E.

[0018] Although not shown, the edge DB 400 may store the feature amount and edge length of each edge E. The feature amount of the edge E is used in addition to the feature amount 302 of the node in the convolution operation of the two nodes N at both ends of the edge E.

[0019] <gcn> 5 is an explanatory diagram showing an example of the configuration of a GCN. The GCN 500 has an EMBEDDING layer 501, one or more GCN layers 502-1 to 502-m (m is an integer equal to or greater than 1), a concat layer 503, an MLP (Multi-Layer Perceptron) layer 504, and a softmax layer 505. When the GCN layers 502-1 to 502-m are not distinguished, they are referred to as GCN layer 502-j (j is an integer satisfying 1≦j≦m). The output of GCN layer 502-j is input to the next GCN layer 502-j+1 and also to the concat layer 503, forming a JK-Net (Jumping Knowledge Network).

[0020] The GCN layer 502-j includes a relational graph convolution network (RGCN) layer 521-j, a normalization layer 522-j, a ReLU layer 523-j, a dropout layer 524-j, and a max layer 525-j.

[0021] FIG. 6 is an explanatory diagram showing an example of the EMBEDDING layer 501. The EMBEDDING layer 501 performs preprocessing of the features 302 included in node N of the input graph 200. In FIG. 6, a company node Na related to a company is used as an example. The company node Na has features 302 F1 (number of employees), F2 (annual sales), F3 (industry), etc. Transformation C1 is no transformation, and the value of feature F1 is used as is. Transformation C2 is a logarithmic transformation, and the value of feature F2 is logarithmically transformed. Transformation C3 is one-hot transformation, and a "0" or "1" is assigned depending on the type. A product-sum operation result 600 is output by performing a product-sum operation on the transformed company node Na and the weight Wa, which is a learning parameter.

[0022] Fig. 7 is an explanatory diagram showing an example of the RGCN layer 521-j. In Fig. 7, the explanation focuses on node N9, but the same applies to the other nodes N1 to N8 and N10. The adjacent node of node N9 is node N8. The adjacent edge of node N9 is edge E9. The RGCN layer 521-j updates the feature F(N9) of node N9 with the convolution operation result F(N9)ud obtained by convolving the feature F(N9) of node N9, the feature F(N8) of adjacent node N8, and the feature F(E9) of adjacent edge E9 (only when j = 1).

[0023] In Figure 5, in order to suppress variations in the absolute values ​​of each feature 302, the normalization layer 522-j normalizes the convolution calculation results of the RGCN layer 521-j for each node N and edge E so that the mean value = 0 and the variance = 1.

[0024] 8 is an explanatory diagram showing an example of the ReLU layer 523-j. The ReLU layer 523-j inputs the updated feature 302 of node N, which is the normalization result 801 by the normalization layer 522-j, to the ReLU function, and converts negative values ​​in the updated feature 302 of node N to 0. In the converted convolution operation result 802, "-3" and "-1" are converted to "0." Using the ReLU function makes it less likely that gradient vanishing will occur, and prevents the problem of learning from stalling.

[0025] FIG. 9 is an explanatory diagram showing an example of the dropout layer 524-j. In FIG. 9, nodes N8 to N10 are used as an example, but the same applies to nodes N1 to N7. The dropout layer 524-j replaces a certain percentage of the feature values ​​302 of node N with 0. This percentage is one of the hyperparameters. In FIG. 9, the feature F(N8) of node N8, the feature F(N9) of node N9, and the feature F(N10) of node N10 are updated to feature F(N8)ud, feature F(N9)ud, and feature F(N10)ud of node N10, respectively.

[0026] FIG. 10 is an explanatory diagram showing an example of the max layer 525-j. In FIG. 10, node N8 is used as an example. The output results from the dropout layers 524-j of node N8 and its adjacent nodes N7, N9, and N10 are defined as features F(N8), F(N7), F(N9), and F(N10). The max layer 525-j obtains convolution results CR(N8), CR(N7), CR(N9), and CR(N10) by convolution of the learning parameters, weights W(N8), W(N7), W(N9), and W(N10), with the features F(N8), F(N7), F(N9), and F(N10). In FIG. 10, these are represented as three-dimensional vectors.

[0027] The max layer 525-j generates a maximum value vector of factors of the same dimension of the convolution operation results CR(N8), CR(N7), CR(N9), and CR(N10), and sets it as the updated feature F(N8)ud of node N8.

[0028] 11 is an explanatory diagram showing the connection relationship of a JK-Net (Jumping Knowledge Network) composed of GCN layers 502-1 to 502-m and a concat layer 503. GCN layers 502-1 to 502-(m-1) output features F(N)-1 to F(N)-(m-1) of node N obtained by convolutional computation to subsequent GCN layers 502-2 to 502-m and concat layer 503, respectively. GCN layer 502-m outputs feature F(N)-m of node N obtained by convolutional computation to concat layer 503. Concat layer 503 combines features F(N)-1 to F(N)-m.

[0029] 12 is an explanatory diagram showing an example of the MLP layer 504 and the softmax layer 505. The MLP layer 504 calculates a guess value for each node N from the output result from the concat layer 503. The softmax layer 505 inputs the output result from the MLP layer 504 into a softmax function to calculate a probability value between 0 and 1 as the guess value 303.

[0030] In FIG. 12, (A) indicates two-class classification, (B) indicates four-class classification, and (C) indicates regression. In (A) to (C), the MLP layer 504 inputs the concat layer 503's concat result. In (A) and (B), the MLP layer 504 inputs the final calculation result to the softmax layer 505. In (A) two-class classification, the softmax layer 505 outputs a value between 0 and 1 as a predicted value 303(Y) for each node N, and in (B) four-class classification, it outputs a four-dimensional vector as the predicted value Y. Each value in this four-dimensional vector is between 0 and 1, and the sum of the four values ​​is 1. In (C) regression, the softmax layer 505 outputs the predicted value 303(Y) indicating the prediction result without going through the softmax layer 505.

[0031] Here, the estimated value 303 and the contribution 304 of the node DB 300 will be explained in more detail. The information display device 100 uses the GCN model 500 to calculate, for each feature 304 of node N, a SHAP (SHapley Additive exPlanations) value indicating the contribution 304 of each node N to the estimated value 303 of node N. Furthermore, if edge E has a feature, the information display device 100 calculates the contribution for each feature of edge E. Prior to inputting the input graph 200 to the GCN model 500, the information display device 100 receives, through a user operation, a designation of the number of intermediate layers of the GCN model 500; specifically, for example, the layer number j of GCN layer 502-j. The information display device 100 calculates the estimated value 303 and the contribution 304 for each feature of node N using GCN layer 502-1 through GCN layer 502-j with the designated layer number j.

[0032] The information display device 100 calculates the estimated value 303 by specifying the layer number j in advance, and stores the estimated value 303 and the contribution 304 in the node DB 300. Specifically, for example, the information display device 100 stores in the node DB 300 the estimated value 303 and the contribution 304 calculated by the weight filter for the GCN layer 502-1, the estimated value 303 and the contribution 304 calculated by the weight filter for the GCN layers 502-1 to 502-2, the estimated value 303 and the contribution 304 calculated by the weight filter for the GCN layers 502-1 to 502-3, ..., the estimated value 303 and the contribution 304 calculated by the weight filter for the GCN layers 502-1 to 502-j, ..., the estimated value 303 and the contribution 304 calculated by the weight filter for the GCN layers 502-1 to 502-m.

[0033] Thereafter, upon receiving a display instruction from the user regarding the estimated value 303 and contribution 304 of node N, the information display device 100 references the estimated value 303 and contribution 304 of node N stored in the node DB 300, and displays information regarding the estimated value 303 and contribution 304 of node N. Note that, although the contribution 304 of node N is hereinafter described as a one-dimensional numerical value, in reality the contribution 304 of node N has the same number of dimensions as the feature 302. If necessary, the user can specify in advance a method for aggregating the feature 302 of node N into one dimension, such as selecting a numerical value for a specific dimension of the feature 302, or taking the sum of all dimensions.

[0034] However, when the information display device 100 receives an instruction to display the estimated value 303 and contribution 304 of node N, it may calculate the estimated value 303 and contribution 304 of node N using GCN layers 502-j up to layer number j specified by the user, and display information regarding the calculated estimated value 303 and contribution 304 of node N.

[0035] <Graph display example> 13 is an explanatory diagram showing an example of the first graph display. The first graph display is an example in which the contribution 304 of each node N to the estimated value 303 of a node (hereinafter referred to as the designated node) N designated by the user is displayed on a display screen 1300. A node N that contributes to the designated node N is called a contributing node N. Note that the distance from a certain node N to another node N connected by a single edge E is called one hop. The larger the layer number j, the larger the number of hops to the node N from which the contribution 304 to the estimated value 303 of the designated node N can be obtained.

[0036] In Fig. 13, the nodes N at the start of the arrows are contribution nodes N2 to N6, and N8 (including the node N4 itself), and the node N at the end of the arrows is the designated node N4. In the example of Fig. 13, the values ​​of contribution 304 of nodes N1, N7, N9, and N10 to the designated node N4 are "0", so no arrows are displayed. Values ​​D42, D43, D44, D45, D46, and D48 of contribution 304 of contribution nodes N2 to N6 and N8 to the designated node N4 are displayed in association with contribution nodes N2 to N6 and N8.

[0037] The sum of the contribution 304 values ​​D42, D43, D44, D45, D46, and D48 becomes the value Y of the estimated value 303 of the specified node N4. In Fig. 13, the contribution 304 values ​​D42, D43, D44, D45, D46, and D48 are displayed in speech bubbles, but the information display device 100 may also display them by highlighting the colors like a heat map.

[0038] 14 is an explanatory diagram showing an example of the second graph display. The second graph display is an example in which the contribution 304 of a specified node N to the estimated value 303 of each node N is displayed on a display screen 1400. A node N that contributes to the specified node N is called a contributed node N. Note that the specified node N is also referred to as node N4 in FIG. 14.

[0039] 14, the node N at the start point of the arrow is the specified node N4, and the node N at the end point of the arrow is the contributed nodes N2 to N6 and N8 (including its own node N4). In the example of FIG. 14, the value of the contribution 304 of the specified node N4 to the nodes N1, N7, N9 and N10 is "0", so no arrow is displayed. The contribution 304 values ​​D24, D34, D44, D54, D64 and D84 of the specified node N4 to the contributed nodes N2 to N6 and N8 are displayed in association with the arrows to the contributed nodes N2 to N6 and N8.

[0040] The values ​​D24, D34, D44, D54, D64, and D84 of the degree of contribution 304 become part of the value Y of the estimated value 303 of each of the contributed nodes N2 to N6 and N8. In Fig. 14, the values ​​D24, D34, D44, D54, D64, and D84 of the degree of contribution 304 are displayed in speech bubbles, but the information display device 100 may also display them by highlighting the colors like a heat map.

[0041] 13 and 14, when the type of node N is registered in the node DB 300, the information display device 100 may display each type of node N in a different color.

[0042] 15 is a flowchart showing an example of a graph display processing procedure by the information display device 100. The information display device 100 accepts a selection of a display method by a user operating the input device 103 (step S1501). The information display device 100 accepts an input of a layer number j of the GCN layer GCN by a user operating the input device 103 (step S1502). The information display device 100 accepts a node designation by a user operating the input device 103 (step S1503). Note that steps S1501 to S1503 may be performed in any order.

[0043] The information processing device determines whether the display method selected in step S1501 is the first graph display (see FIG. 13) or the second graph display (see FIG. 14) (step S1504). If the display method selected in step S1501 is the first graph display (step S1504: first), the information display device 100 acquires the contribution 304 of each node N corresponding to the layer number j input in step S1502 and to the estimated value 303 of the specified node N (step S1505). The acquisition of the contribution 304 may be performed by reading it from the node DB 300 or by calculating it using the GCNN 500. Then, the information display device 100 executes the first graph display as shown in FIG. 13 (step S1506).

[0044] Furthermore, if the display method selected in step S1501 is the second graph display (step S1504: second), the information display device 100 acquires the contribution 304 of the specified node N to the estimated value 303 of each node N, which corresponds to the layer number j input in step S1502 (step S1507). Then, the information display device 100 executes the second graph display as shown in Fig. 14 (step S1508). This ends the graph display process.

[0045] Fig. 16 is an explanatory diagram showing another example of the first graph display, and Fig. 17 is an explanatory diagram showing another example of the second graph display. Fig. 16 is an example of displaying, using a bar graph 1600, values ​​D42, D43, D44, D45, D46, and D48 of the contribution 304 of each of nodes N2 to N6 and N8 to the estimated value 303 of the specified node N4. Fig. 17 is an example of displaying, using a bar graph 1700, values ​​D24, D34, D44, D54, D64, and D84 of the contribution 304 of the specified node N4 to the estimated value 303 of each of nodes N2 to N6 and N8.

[0046] 16 and 17, when the type of node N is registered in the node DB 300, the information display device 100 may display the bar of each type of node N in a different color. [Example]

[0047] Next, a description will be given of Example 2. In Example 2, differences from Example 1 will be described. In Example 2, the input graph 200 is reduced by pooling by grouping a certain node N and its neighboring nodes N into a certain group of nodes.

[0048] 18 is an explanatory diagram illustrating an example of a GCNN according to Example 2. In Example 2, a GCNN 1800 is used that has a GCN layer 1802-j in which a pooling layer 1821-j is arranged between a dropout layer 524-j and a max layer 525-j.

[0049] 19 is an explanatory diagram showing an example of the pooling layer 1821-j. First, the pooling layer 1821-j (A) calculates the feature of edge E for each edge E. Taking edge E10 as an example, the pooling layer 1821-j calculates the feature F(E10) of edge E10 by performing a convolution operation between the feature F(N8) and F(N10) of nodes 8 and N10 at both ends of edge E10 and the learning parameters W(N8) and W(N10). The value of the feature F(E10) is set to "7." Similar calculations are performed for the other edges E1 to E9.

[0050] Next, the pooling layer 1821-j divides the nodes N into groups by dividing them at edges E whose feature value F(E) is equal to or less than a threshold value (for example, "5"), so that the nodes N2 and N3 are grouped into node group G1, the nodes N4 and N5 are grouped into node group G2, the nodes N6 and N7 are grouped into node group G3, and the nodes N8 to N10 are grouped into node group G4.

[0051] Then, the pooling layer 1821-j performs feature pooling for each node group G. Taking node group G4 as an example, it adds up the features F(N8) to F(N10) of nodes N8 to N10 in node group G4 and outputs the feature F(G4) for node group G4. The same applies to the other node groups G1 to G3. Note that node N1 is a node N that does not belong to any of the node groups G1 to G4.

[0052] 20 is an explanatory diagram illustrating the connection relationship between the GCN layers 1802-1 to 1802-m and the concat layer 503 according to the second embodiment. In FIG. 20, as an example, attention is focused on the node N9, and j=4. The processing of the max layer 525-j is omitted. The GCN layers 1802-1 to 1802-4 perform pooling.

[0053] Specifically, for example, the GCN layer 1802-1 outputs the feature F(N9) of node N9 to the concat layer 503, and calculates the feature F(G4) of node group G4, as shown in FIG. 19. The GCN layer 1802-2 outputs the feature F(G4) of node group G4 to the concat layer 503, generates node group G40, and calculates its feature F(G40). The GCN layer 1802-3 outputs the feature F(G40) of node group G40 to the concat layer 503, generates node group G400, and calculates its feature F(G400). The GCN layer 1802-4 outputs the feature F(G400) of node group G400 to the concat layer 503.

[0054] The concat layer 503 combines the feature F(N9), the feature F(G4), the feature F(G40), and the feature F(G400) to generate an updated feature F(N9)ud for node N9.

[0055] The node groups that are the pooling results of each of the GCN layers 1802-1 to 1802-4 are stored in the node DB 300 and the edge DB 400, similar to the entry of node N. Also, affiliation information, such as which GCN layer 1802-j each node N belongs to and to which node group it belongs, is also stored in the node DB 300.

[0056] If node N9 is the designated node N, when calculations are performed up to GCN layer 1802-2, the information display device 100 outputs the node N in node group G4 so that it can be displayed as an associated node N that has a stronger relationship with node N9 than other nodes N outside node group G4. Furthermore, when calculations are performed up to GCN layer 1802-3, the information display device 100 outputs the node N in node group G40 so that it can be displayed as an associated node N that has a stronger relationship with node N9 than other nodes N outside node group G40. Furthermore, when calculations are performed up to GCN layer 1802-4, the information display device 100 outputs the node N in node group G400 so that it can be displayed as an associated node N that has a stronger relationship with node N9 than other nodes N outside node group G400.

[0057] In this way, by combining the features of each pooling layer 1821-j in the concat layer 503, the updated feature amount of node N includes the features of node N itself, the features including neighboring nodes N, a wider range of features, and the features of the entire input graph 200. Therefore, compared to the first embodiment, it is possible to identify a wider range of contributing nodes N that contribute to the estimated value 303 of the specified node N, and it is possible to identify a wider range of contributed nodes N that are contributed by the contribution 304 of the specified node N. [Example]

[0058] The third embodiment shows a specific example of the input graph 200 described in the first and second embodiments.

[0059] <Corporate sales activities> For example, in the case of input graph 200, which is a graph of the sales activities of a certain company, node N represents a salesperson, engineer, or customer, and edge E represents the sales activity history, organizational relationships within the company, information sharing, and proposal support. If node N is a salesperson or engineer, the feature amount 302 of node N may be, for example, age, gender, or affiliation, and if node N is a customer, the feature amount 302 may be, for example, the industry, number of employees, unit price of products sold to customers, or sales of the company to customers. In addition, estimated value 303 may be, for example, a predicted value of a KPI (Key Performance Indicator).

[0060] In the case of the first graph display, for example, if the specified node N4 is a customer, it is possible to identify the salespeople and engineers (contributing nodes N2 to N6, N8) who contribute to the customer's KPI. Also, in the case of the second graph display, for example, if the specified node N4 is an engineer, it is possible to identify the customers, salespeople, and other engineers (contributed nodes N2 to N6, N8) to whom the engineer contributes.

[0061] <Map Information AI (Artificial Intelligence)> For example, in the case of an input graph 200 that is a graph of map data, the input graph 200 represents a transportation network. Specifically, for example, node N represents an intersection, and edge E represents a road (including the number of lanes and lane direction). The feature amount 302 of node N is, for example, location information, the presence or absence of traffic lights, the type of building, the presence or absence of bus stops, etc. The estimated values ​​are predicted values ​​of KPIs such as land prices, foot traffic, transportation demand, and livability.

[0062] In the case of the first graph display, for example, it is possible to identify intersections (contributing nodes N2 to N6, N8) that contribute to the foot traffic at the intersection indicated by the specified node N4. Also, in the case of the second graph display, for example, it is possible to identify intersections (contributed nodes N2 to N6, N8) to which the foot traffic at the intersection indicated by the specified node N4 contributes.

[0063] 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.

[0064] 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.

[0065] 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).

[0066] 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]

[0067] 100 Information display device 101 processors 102 Storage Devices 302 Features 303 Estimated Value 304 Contribution N-node E-edge< / gcn>

Claims

1. An information display device having a processor that executes a program and a storage device that stores the program, an input graph including a plurality of nodes and an edge connecting two nodes among the plurality of nodes; a database storing a feature value of each of the plurality of nodes, an estimated value of the node, a contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes; The processor: a designation process for accepting designation of the node; an acquisition process of acquiring, from the database, contributions of the node and other nodes to the estimated value of the node designated by the designation process; an output process of outputting the contributions of the node and the other nodes to the estimated value of the node acquired by the acquisition process so as to be displayable in a first graph showing a relationship between the distances between the node and the other nodes in the input graph and the contributions; An information display device characterized by executing the above.

2. An information display device having a processor that executes a program and a storage device that stores the program, an input graph including a plurality of nodes and an edge connecting two nodes among the plurality of nodes; a database storing a feature value of each of the plurality of nodes, an estimated value of the node, a contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes; The processor: a designation process for accepting designation of the node; an acquisition process of acquiring, from the database, contributions of the node and other nodes to the estimated value of the node designated by the designation process; an output process of outputting the contribution of the node to the estimated value of the node and the other nodes acquired by the acquisition process so as to be displayable in a second graph showing the relationship between the distance between the node and the other nodes in the input graph and the contribution; An information display device characterized by executing the above.

3. An information display device according to claim 1, The processor: execute a selection process to accept a selection of either a first display that displays the contribution of the node and the other nodes to the estimated value of the node, or a second display that displays the contribution of the node to the estimated value of the node and the other nodes; In the acquisition process, when the first display is selected, the processor acquires the contributions of the node and the other nodes to the estimated value of the node, and when the second display is selected, the processor acquires the contributions of the node to the estimated values ​​of the node and the other nodes from the database; In the output process, when the first display is selected, the processor outputs the contributions of the node and the other nodes so as to be displayable in the first graph, and when the second display is selected, the processor outputs the contributions of the node to the estimated values ​​of the node and the other nodes so as to be displayable in a second graph showing the relationship between the distances between the node and the other nodes in the input graph and the contributions. An information display device characterized by:

4. An information display device having a processor that executes a program and a storage device that stores the program, an input graph including a plurality of nodes and an edge connecting two nodes among the plurality of nodes; a database storing a feature value of each of the plurality of nodes, an estimated value of the node, a contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes; The processor: an input process for receiving input of layer numbers of multiple intermediate layers of the graph convolutional neural network to which the input graph has been input; a designation process for accepting designation of the node; an acquisition process of acquiring, from the database, contributions of the node and the other nodes to the estimated value of the node designated by the designation process corresponding to the layer number input by the input process; an output process of displayably outputting the contributions of the node and the other nodes acquired by the acquisition process; An information display device characterized by executing the above.

5. An information display device according to claim 4, the intermediate layer includes a pooling layer; In the output process, the processor displays and outputs a related node related to the node based on the pooling result of the pooling layer corresponding to the layer number input in the input process. An information display device characterized by:

6. An information display method executed by an information display device having a processor that executes a program and a storage device that stores the program, comprising: an input graph including a plurality of nodes and an edge connecting two nodes among the plurality of nodes; a database storing a feature value of each of the plurality of nodes, an estimated value of the node, a contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes; The processor: a designation process for accepting designation of the node; an acquisition process of acquiring, from the database, contributions of the node and other nodes to the estimated value of the node designated by the designation process; an output process of outputting the contributions of the node and the other nodes to the estimated value of the node acquired by the acquisition process so as to be displayable in a first graph showing a relationship between the distances between the node and the other nodes in the input graph and the contributions; An information display method comprising:

7. An information display method executed by an information display device having a processor that executes a program and a storage device that stores the program, comprising: an input graph including a plurality of nodes and an edge connecting two nodes among the plurality of nodes; a database storing a feature value of each of the plurality of nodes, an estimated value of the node, a contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes; The processor: a designation process for accepting designation of the node; an acquisition process of acquiring, from the database, contributions of the node and other nodes to the estimated value of the node designated by the designation process; an output process of outputting the contribution of the node to the estimated value of the node and the other nodes acquired by the acquisition process so as to be displayable in a second graph showing the relationship between the distance between the node and the other nodes in the input graph and the contribution; An information display method comprising:

8. An information display method executed by an information display device having a processor that executes a program and a storage device that stores the program, comprising: an input graph including a plurality of nodes and an edge connecting two nodes among the plurality of nodes; a database storing a feature value of each of the plurality of nodes, an estimated value of the node, a contribution of the node and other nodes to the estimated value, and a connection relationship between the nodes; The processor: an input process for receiving input of layer numbers of multiple intermediate layers of the graph convolutional neural network to which the input graph has been input; a designation process for accepting designation of the node; an acquisition process of acquiring, from the database, contributions of the node and the other nodes to the estimated value of the node designated by the designation process corresponding to the layer number input by the input process; an output process of displayably outputting the contributions of the node and the other nodes acquired by the acquisition process; An information display method comprising:

Images