Information processing program, information processing device, and information processing method

Abstract

This method accurately determines the range of variables that exhibit a causal effect of a given sign from a data set containing the values ​​of multiple variables. [Solution] The computer performs causal search using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the target variable, based on multiple conditions regarding the range of values ​​of the multiple explanatory variables, to find the causal relationship between the multiple explanatory variables and the target variable. Based on the causal relationship found based on multiple conditions, the computer finds the causal effect of any of the multiple explanatory variables on the target variable. The computer determines the similarity of the causal effects found based on multiple conditions to obtain a similarity determination result. Based on the similarity determination result, the computer finds a specific range of values ​​for each of the multiple explanatory variables in which the causal effect has a predetermined sign, and outputs range information indicating the specific range.

Description

Technical Field

[0001] The present invention relates to information processing.

Background Art

[0002] Statistical causal discovery is a technique for estimating causal relationships between a plurality of variables from a set of data for each variable.

[0003] Regarding the relationship between variables, a factor analysis method for identifying explanatory variables that are factors determining changes in the values of target variables is known (for example, see Patent Document 1). A method for creating a prediction model that can feedback on the explanatory variables of an operation process is also known (for example, see Patent Document 2).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] When obtaining a numerical range of an explanatory variable for which a causal effect of the same sign, positive or negative, is obtained using the search result of statistical causal discovery for a set of data, the accuracy of the obtained numerical range may decrease.

[0006] Note that such a problem occurs not only in statistical causal discovery but also when obtaining a numerical range of an explanatory variable for which a causal effect of the same sign is obtained using various causal discoveries.

[0007] In one aspect, an object of the present invention is to accurately obtain a range of variables for which a causal effect of a predetermined sign is obtained from a set of data including values of a plurality of variables. [Means for solving the problem]

[0008] One approach involves having the computer perform the following processes as part of the information processing program.

[0009] The computer performs causal search using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the dependent variable, based on multiple conditions regarding the range of values ​​of the multiple explanatory variables, to find causal relationships involving multiple explanatory variables and the dependent variable. Based on the causal relationships found based on multiple conditions, the computer determines the causal effect of any of the multiple explanatory variables on the dependent variable.

[0010] The computer determines a similarity result by assessing the similarity of causal effects based on multiple conditions. Based on the similarity result, the computer determines a specific range of values ​​for each of the multiple explanatory variables for which the causal effect has a predetermined sign, and outputs range information indicating that specific range. [Effects of the Invention]

[0011] From one perspective, it is possible to accurately determine the range of variables that exhibit a causal effect of a given sign from a data set containing the values ​​of multiple variables. [Brief explanation of the drawing]

[0012] [Figure 1] This figure shows the search results of conditional causal search. [Figure 2] This figure shows the data used in conditional causal search. [Figure 3] This is a functional configuration diagram of the information processing device according to the embodiment. [Figure 4] This is a flowchart of information processing. [Figure 5] This is a functional configuration diagram of a variable range identification device. [Figure 6] This is a diagram showing the data set. [Figure 7] It is a diagram showing a condition list. [Figure 8] It is a diagram showing a causal graph. [Figure 9] It is a diagram showing causal effect information. [Figure 10] It is a diagram showing an adjacency matrix. [Figure 11] It is a diagram showing the numerical range of explanatory variables that can obtain causal effects with the same sign. [Figure 12] It is a diagram showing range information. [Figure 13] It is a flowchart of variable range specification processing. [Figure 14] It is a flowchart of condition extraction processing. [Figure 15] It is a flowchart of causal exploration processing. [Figure 16] It is a flowchart of distance calculation processing. [Figure 17] It is a flowchart of change rate calculation processing. [Figure 18] It is a flowchart of similarity determination processing. [Figure 19] It is a flowchart of range specification processing. [Figure 20] It is a flowchart of range update processing. [Figure 21] It is a hardware configuration diagram of an information processing device.

Embodiments for Carrying Out the Invention

[0013] Hereinafter, embodiments will be described in detail with reference to the drawings.

[0014] In conditional causal search, variables other than the dependent variable from a set of data are used as explanatory variables. These explanatory variables are discretized, and multiple conditions that correlate with the dependent variable are extracted. Each condition is expressed, for example, by an inequality that specifies the numerical range of one or more explanatory variables. Subsets of data that satisfy each condition are then extracted from the set of data, and by performing statistical causal search using each subset, the causal relationships between variables for each condition can be determined.

[0015] Causal relationships can be represented, for example, using a causal graph. A causal graph includes multiple nodes representing causes or effects in a causal relationship, and edges pointing from the cause node to the effect node. Each edge is weighted with a causal effect, which is an indicator of the strength of the influence a cause has on the effect. If the causal effect is positive, an increase in the cause variable value also increases the effect variable value. If the causal effect is negative, an increase in the cause variable value decreases the effect variable value.

[0016] By using the causal relationships determined by each condition, the causal effect of a specific explanatory variable on the dependent variable can be determined. Then, by associating the sign of the causal effect determined by each condition with the numerical range of the explanatory variable represented by that condition, the numerical range of explanatory variables that yield a positive or negative causal effect can be determined.

[0017] Figure 1 shows an example of the search results of conditional causal search. The condition represents the numerical range of the explanatory variable. In this example, the explanatory variables included in the condition are time and temperature. The factor variable represents any of the explanatory variables included in the condition. The causal effect represents the causal effect of the factor variable on the dependent variable. In this example, the factor variable is temperature and the dependent variable is energy. Energy is, for example, the amount of energy consumed in a facility such as a university campus or a shopping mall.

[0018] When the temperature rises, air conditioning is used, increasing electricity consumption. Conversely, when the temperature falls, heating is used, also increasing electricity consumption. Therefore, in high-temperature situations, the causal effect of temperature on electricity consumption is positive, and in low-temperature situations, the causal effect of temperature on electricity consumption is negative. In this case, it would be useful to know at what temperature a negative causal effect is obtained and at what temperature a positive causal effect is obtained.

[0019] For example, condition C1 is "time < 10 o'clock and temperature < 10°C". Condition C1 represents a time range before 10 o'clock and a temperature range less than 10°C. The causal effect obtained using conditions C1 to C3 has a negative sign, and the causal effect obtained using condition C4 has a positive sign. In this case, the maximum range of explanatory variables that yield a negative causal effect is "time < 10 o'clock and temperature < 20°C".

[0020] Figure 2 shows an example of the data used in the conditional causal search in Figure 1. The symbol "×" represents a point in three-dimensional space, represented by time, temperature, and energy. The temperature and energy indicated by "×" are measurements taken at the same time indicated by "×". Curve 201 represents an approximation curve that approximates the distribution of multiple "×" points.

[0021] Range 211 corresponds to the numerical range of the explanatory variable represented by condition C1 in Figure 1, and range 212 corresponds to the numerical range of the explanatory variable represented by condition C3. In this example, in the range obtained by subtracting range 211 from range 212, "time < 10 o'clock and 10°C ≤ temperature < 20°C", the causal effect of temperature on energy is 0 or positive. Therefore, the maximum range for an explanatory variable that yields a negative causal effect is "time < 10 o'clock and temperature < 10°C".

[0022] However, the search results in Figure 1 do not reveal the causal effect in the range "time < 10:00 and 10°C ≤ temperature < 20°C". Therefore, range 212, which includes this range, is identified as the maximum range where a negative causal effect can be obtained. There is a discrepancy between the identified range 212 and the actual maximum range, range 211.

[0023] Thus, when using the results of a statistical causal search on a data set to determine the numerical range of explanatory variables that yield positive or negative causal effects of the same sign, the precision of the determined numerical range may decrease.

[0024] Figure 3 shows an example of the functional configuration of an information processing device (computer) according to the embodiment. The information processing device 301 in Figure 3 includes a causal search unit 311, a determination unit 312, a range identification unit 313, and an output unit 314.

[0025] Figure 4 is a flowchart showing an example of information processing performed by the information processing device 301 in Figure 3. First, the causal search unit 311 performs a causal search using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the target variable, based on each of several conditions, to find a causal relationship involving multiple explanatory variables and the target variable (step 401). Each of the several conditions represents a range of values ​​for each of the multiple explanatory variables.

[0026] Next, the causal search unit 311 determines the causal effect of one of the explanatory variables on the dependent variable based on the causal relationships determined based on each of the multiple conditions (step 402).

[0027] Next, the determination unit 312 determines the similarity of the causal effects obtained based on each of the multiple conditions, thereby obtaining a similarity determination result (step 403). Next, the range determination unit 313 determines a specific range of values ​​for each of the multiple explanatory variables in which the causal effect has a predetermined sign, based on the similarity determination result (step 404). Then, the output unit 314 outputs range information indicating the specific range (step 405).

[0028] According to the information processing device 301 in Figure 3, the range of variables from which a predetermined causal effect of a given sign can be obtained can be accurately determined from a data set containing the values ​​of each of multiple variables.

[0029] Figure 5 shows an example of the functional configuration of a variable range identification device corresponding to the information processing device 301 in Figure 3. The variable range identification device 501 in Figure 5 includes a condition extraction unit 511, a causal search unit 512, a distance calculation unit 513, a rate of change calculation unit 514, a similarity determination unit 515, a generation unit 516, an output unit 517, and a storage unit 518.

[0030] The causal search unit 512, the generation unit 516, and the output unit 517 correspond to the causal search unit 311, the range identification unit 313, and the output unit 314 in Figure 3, respectively. The distance calculation unit 513, the rate of change calculation unit 514, and the similarity determination unit 515 correspond to the determination unit 312 in Figure 3.

[0031] The variable range identification device 501 analyzes various data sets that are the subject of data analysis. For example, the data set may be a collection of data on the amount of electricity consumed at facilities such as university campuses or shopping malls.

[0032] The data set may be a collection of data relating to the attributes of intermediate or final products in the manufacturing process of industrial products. Industrial products may be automobiles, electrical products, industrial machinery, or chemical products such as pharmaceuticals. The data set may also be a collection of data relating to the work performance or satisfaction of workers in labor management.

[0033] The storage unit 518 stores the data set 521 that will be analyzed. Figure 6 shows an example of the data set 521.

[0034] Each data point in the data set 521 in Figure 6 represents the amount of electricity consumed at the facility and includes ID, time, temperature, and energy consumption. ID is the data identifier. Time represents the time the data was acquired. Temperature represents the measured temperature at the time of acquisition, and energy consumption represents the measured energy consumption at the time of acquisition. Time and temperature are explanatory variables, and energy consumption is the dependent variable.

[0035] The condition extraction unit 511 sets one or more thresholds for each explanatory variable included in the data set 521 and generates inequalities that include the explanatory variables and the set thresholds. Next, the condition extraction unit 511 generates conditions that represent each of the multiple combinations of inequalities by exhaustively combining the inequalities. Each generated condition represents the range of values ​​for each of the multiple explanatory variables included in the data set 521.

[0036] Next, the condition extraction unit 511 extracts data from the data set 521 that satisfies each of the generated conditions and generates a subset containing the extracted data. Then, using the subset, the condition extraction unit 511 extracts conditions that show correlation from the generated conditions, generates a condition list 522 containing the extracted conditions, and stores it in the storage unit 518.

[0037] The condition extraction unit 511 generates all combinations of two variables from a plurality of variables, including one or more explanatory variables and the dependent variable included in the condition. Next, the condition extraction unit 511 calculates the correlation coefficient CC between the first and second variables included in each combination using the data included in the subset. The correlation coefficient CC is calculated using the covariance CV of the first and second variables, the standard deviation SD1 of the first variable, and the standard deviation SD2 of the second variable by the following formula.

[0038] CC = CV / (SD1 × SD2) (1)

[0039] Then, the condition extraction unit 511 extracts a condition as a condition in which a correlation appears if the absolute value of the correlation coefficient of at least one combination is greater than or equal to a threshold.

[0040] Figure 7 shows an example of a condition list 522 generated from the data set 521 in Figure 6. The condition list 522 in Figure 7 includes conditions C11 through C18. Each condition represents a combination of an inequality relating to time and an inequality relating to temperature. In Figure 7, conditions other than C11 through C18 are omitted.

[0041] For example, condition C11 is "time < 10 o'clock and temperature < 5°C". Condition C11 means that the time range is before 10 o'clock and the temperature range is less than 5°C.

[0042] The range of time and temperature that satisfies condition C12 includes the range of time and temperature that satisfies condition C11. The range of time and temperature that satisfies condition C13 includes the range of time and temperature that satisfies condition C11, and also includes the range of time and temperature that satisfies condition C12.

[0043] The range of time and temperature that satisfies condition C14 includes the range of time and temperature that satisfies condition C11, and also includes the range of time and temperature that satisfies condition C12, and also includes the range of time and temperature that satisfies condition C13. The range of time and temperature that satisfies condition C15 includes the range of time and temperature that satisfies condition C11, and also includes the range of time and temperature that satisfies condition C12, and also includes the range of time and temperature that satisfies condition C13, and also includes the range of time and temperature that satisfies condition C14.

[0044] The range of time and temperature that satisfies condition C16 includes the range of time and temperature that satisfies condition C11. The range of time and temperature that satisfies condition C17 includes the range of time and temperature that satisfies condition C11, and also includes the range of time and temperature that satisfies condition C12, and also includes the range of time and temperature that satisfies condition C16.

[0045] The range of time and temperature that satisfies condition C18 includes the range of time and temperature that satisfies condition C11, and also includes the range of time and temperature that satisfies condition C12, and also includes the range of time and temperature that satisfies condition C13. The range of time and temperature that satisfies condition C18 further includes the range of time and temperature that satisfies condition C16, and also includes the range of time and temperature that satisfies condition C17.

[0046] The causal search unit 512 extracts data from the data set 521 that satisfies each condition included in the condition list 522, and generates a subset containing the extracted data. Then, the causal search unit 512 performs a statistical causal search using the generated subset to generate a causal graph 523 for each condition and stores it in the storage unit 518. For example, LiNGAM (Linear Non-Gaussian Acyclic Model) is used as a statistical causal search.

[0047] Figure 8 shows an example of a causal graph 523 generated from the condition list 522 in Figure 7. The causal graph 523 in Figure 8(a) includes nodes representing time, nodes representing temperature, nodes representing energy, edges from temperature to time, edges from time to energy, and edges from temperature to energy.

[0048] An edge from temperature to time has a causal effect of -0.11. Therefore, when the temperature increases by 1°C, time decreases by 0.11 hours. An edge from time to energy has a causal effect of -4.91. Therefore, when time increases by 1 hour, energy decreases by 4.91 kWh. An edge from temperature to energy has a causal effect of 9.71. Therefore, when the temperature increases by 1°C, energy increases by 9.71 kWh.

[0049] If the causal effect is positive, an increase in the value of the causal variable will also increase the value of the effectal variable. If the causal effect is negative, an increase in the value of the causal variable will decrease the value of the effectal variable.

[0050] The causal graph 523 in Figure 8(b) includes nodes representing time, nodes representing temperature, nodes representing energy, edges from temperature to energy, and edges from time to energy.

[0051] The edge from temperature to energy has a causal effect of 0.74. Therefore, when the temperature increases by 1°C, the amount of energy increases by 0.74 kWh. The edge from time to energy has a causal effect of 3.59. Therefore, when the time increases by 1 hour, the amount of energy increases by 3.59 kWh.

[0052] Next, the causal search unit 512 uses one of the explanatory variables included in the condition list 522 as a factor variable to calculate the causal effect of the factor variable on the target variable from the causal graph 523. Then, the causal search unit 512 generates causal effect information 524, which includes the calculated causal effect, and stores it in the storage unit 518. The causal effect information 524 includes the causal effect associated with the combination of conditions and factor variables used to generate the causal graph 523.

[0053] The causal search unit 512 calculates the product of the causal effects of one or more edges included in each path from the factor variable to the target variable in the causal graph 523 as the causal effect of that path. Then, the causal search unit 512 calculates the sum of the causal effects of all paths from the factor variable to the target variable as the causal effect of the factor variable on the target variable.

[0054] For example, in the case of causal graph 523 in Figure 8(a), the causal effect of temperature on electrical energy is calculated as follows.

[0055] Causal effect of the path from temperature → time → energy: (-0.11) × (-4.91) = 0.54 Causal effect of the temperature → energy path: 9.71 Causal effect of temperature on electricity consumption: 0.54 + 9.71 = 10.25

[0056] In the case of causal graph 523 in Figure 8(b), the causal effect of temperature on electrical energy is calculated as follows:

[0057] Causal effect of the temperature → energy flow path: 0.74 Causal effect of the time → energy flow path: 3.59 Causal effect of temperature on electricity consumption: 0.74 + 3.59 = 4.33

[0058] Figure 9 shows an example of causal effect information 524 generated from the condition list 522 in Figure 7. The causal effect information 524 in Figure 9 includes causal effects associated with each combination of condition and factor variable. In this example, the factor variable is temperature, the dependent variable is energy, and the causal effect represents the causal effect of temperature on energy.

[0059] The causal effects obtained using conditions C11-C14 and C16-C18 have a negative sign, while the causal effects obtained using condition C15 have a positive sign. In Figure 9, information regarding conditions other than C11-C18 is omitted.

[0060] The distance calculation unit 513 selects a combination of condition CA and another condition CB that includes condition CA from the condition list 522. If the numerical range of the explanatory variable represented by condition CA is included in the numerical range of the explanatory variable represented by condition CB, it is determined that condition CB includes condition CA.

[0061] Next, the distance calculation unit 513 calculates the distance D between the causal graph 523 generated using condition CA and the causal graph 523 generated using condition CB, generates distance information 525 including distance D, and stores it in the storage unit 518. The distance information 525 includes the distance D associated with the combination of condition CA and condition CB.

[0062] Condition CA is an example of the first condition, and condition CB is an example of the second condition. The subset generated using condition CA is an example of the first subset, and the subset generated using condition CB is an example of the second subset. The causal graph 523 generated using condition CA is an example of the first causal graph representing the first causal relationship, and the causal graph 523 generated using condition CB is an example of the second causal graph representing the second causal relationship.

[0063] The distance calculation unit 513, for example, converts each causal graph 523 into an adjacency matrix whose elements are causal effects, and calculates the distance between the two adjacency matrices as the distance D between the two causal graphs 523.

[0064] Figure 10 shows an example of an adjacency matrix. The adjacency matrix in Figure 10(a) represents the causal graph 523 in Figure 8(a). Each row of the adjacency matrix corresponds to the starting node of an edge, and each column corresponds to the ending node of an edge. For each combination of starting and ending nodes of an edge, the causal effect of that edge is set in the elements of the adjacency matrix, and 0 is set for the other elements. Therefore, 0 is set for the diagonal elements.

[0065] The value "-0.11" corresponding to the combination of the starting node "temperature" and the ending node "time" represents the causal effect of the edge from temperature to time. The value "9.71" corresponding to the combination of the starting node "temperature" and the ending node "energy" represents the causal effect of the edge from temperature to energy. The value "-4.91" corresponding to the combination of the starting node "time" and the ending node "energy" represents the causal effect of the edge from time to energy.

[0066] The adjacency matrix in Figure 10(b) represents the causal graph 523 in Figure 8(b). "0.74", corresponding to the combination of the starting node "temperature" and the ending node "energy", represents the causal effect of the edge from temperature to energy. "3.59", corresponding to the combination of the starting node "time" and the ending node "energy", represents the causal effect of the edge from time to energy.

[0067] The distance D between an N x N (where N is an integer greater than or equal to 2) adjacency matrix A1 and an N x N (where N is an integer greater than or equal to 2) adjacency matrix A2 is calculated using the following formula.

[0068] D=(Σ(A1(i,j)-A2(i,j)) 2 ) 1 / 2 (2)

[0069] A1(i,j) represents the element in the i-th row and j-th column of the adjacency matrix A1, and A2(i,j) represents the element in the i-th row and j-th column of the adjacency matrix A2. Σ represents the sum over i=1 to N and j=1 to N. Therefore, the distance D in equation (2) represents the square root of the sum of the squares of the differences between A1(i,j) and A2(i,j).

[0070] For example, the distance D between the adjacency matrix in Figure 10(a) and the adjacency matrix in Figure 10(b) is calculated using equation (2) as follows:

[0071] D=((-0.11-0) 2 +(9.71-0.74) 2 +(-4.91-3.59) 2 ) 1 / 2 =12.36 (3)

[0072] The rate of change calculation unit 514 selects a combination of condition CA and another condition CB that includes condition CA from the condition list 522. Next, the rate of change calculation unit 514 obtains the causal effect EA obtained using condition CA and the causal effect EB obtained using condition CB from the causal effect information 524, and calculates the difference between causal effect EA and causal effect EB. Causal effect EA and causal effect EB represent the causal effect of the same factor variable on the dependent variable.

[0073] The rate of change calculation unit 514 calculates the rate of change R of the causal effect from the calculated difference, generates rate of change information 526 including the rate of change R, and stores it in the storage unit 518. The rate of change information 526 includes the rate of change R associated with the combination of condition CA, condition CB, and factor variables. The rate of change R is calculated, for example, by the following formula.

[0074] R = (EB - EA) / EA (4)

[0075] Condition CA is an example of the first condition, and condition CB is an example of the second condition. Causal effect EA is an example of the first causal effect, and causal effect EB is an example of the second causal effect.

[0076] For example, condition C11 in Figure 9 is included in condition C12. Therefore, the rate of change R between the causal effect of condition C11 and the causal effect of condition C12 is calculated using equation (4) as follows.

[0077] R = (-3.0 - (-3.5)) / (-3.5) = -0.14 (5)

[0078] Condition C12 in Figure 9 is included in condition C13. Therefore, the rate of change R between the causal effect of condition C12 and the causal effect of condition C13 is calculated using equation (4) as follows.

[0079] R = (-1.0 - (-3.0)) / (-3.0) = -0.67 (6)

[0080] Condition C12 in Figure 9 is included in condition C17. Therefore, the rate of change R between the causal effect of condition C12 and the causal effect of condition C17 is calculated using equation (4) as follows.

[0081] R = (-1.5 - (-3.0)) / (-3.0) = -0.50 (7)

[0082] The similarity determination unit 515 selects a combination of any condition CA and another condition CB that includes condition CA from the condition list 522.

[0083] Next, the similarity determination unit 515 uses the causal effect information 524, distance information 525, and rate of change information 526 to determine the similarity between the causal effect EA obtained using condition CA and the causal effect EB obtained using condition CB. Causal effect EA and causal effect EB represent the causal effect of the same factor variable on the target variable. Then, the similarity determination unit 515 generates a similarity determination result 527 of causal effect EA and causal effect EB and stores it in the storage unit 518.

[0084] The determination result 527 includes determination information associated with the combination of conditions CA, CB, and factor variables. The determination information indicates whether causal effect EA and causal effect EB are similar or not. The similarity determination unit 515 may determine the similarity between causal effect EA and causal effect EB using causal effect information 524 and either distance information 525 or rate of change information 526.

[0085] The similarity determination unit 515 determines the similarity between causal effect EA and causal effect EB using, for example, determination method M1 using distance information 525, determination method M2 using rate of change information 526, or determination method M3 using distance information 525 and rate of change information 526.

[0086] When using the determination method M1, the similarity determination unit 515 obtains the causal effect EA obtained using condition CA and the causal effect EB obtained using condition CB from the causal effect information 524, and compares the signs of causal effect EA and causal effect EB. Then, the similarity determination unit 515 obtains the distance D between the causal graph 523 generated using condition CA and the causal graph 523 generated using condition CB from the distance information 525.

[0087] If the distance D is less than the threshold T1, the similarity determination unit 515 determines that the causal graph 523 generated using condition CA and the causal graph 523 generated using condition CB are similar. In this case, the subset generated using condition CA and the subset generated using condition CB are determined to be similar.

[0088] By comparing the distance D between two causal graphs 523 with a threshold T1, the similarity between a subset generated using condition CA and a subset generated using condition CB can be easily determined.

[0089] Assume that the signs of causal effect EA and causal effect EB are the same, and the subset generated using condition CA and the subset generated using condition CB are similar. In this case, the similarity determination unit 515 determines that causal effect EA and causal effect EB are similar. Otherwise, the similarity determination unit 515 determines that causal effect EA and causal effect EB are not similar.

[0090] According to determination method M1, based on the similarity between the subset generated using condition CA and the subset generated using condition CB, the similarity between causal effect EA and causal effect EB can be accurately determined.

[0091] When using determination method M2, the similarity determination unit 515 acquires causal effect EA obtained using condition CA and causal effect EB obtained using condition CB from the causal effect information 524, and compares the signs of causal effect EA and causal effect EB. Then, the similarity determination unit 515 acquires the change rate R between causal effect EA and causal effect EB from the change rate information 526.

[0092] When the signs of causal effect EA and causal effect EB are the same and the absolute value of the change rate R is less than the threshold value T2, the similarity determination unit 515 determines that causal effect EA and causal effect EB are similar. Otherwise, the similarity determination unit 515 determines that causal effect EA and causal effect EB are not similar.

[0093] For example, when condition C11 in FIG. 9 is selected as condition CA and condition C12 is selected as condition CB, causal effect EA is -3.5 and causal effect EB is -3.0. Causal effect EA and causal effect EB have the same sign. The change rate R is calculated by Equation (5) and is -0.14. When T2 = 0.2, since the absolute value of the change rate R < T2, it is determined that causal effect EA obtained using condition C11 and causal effect EB obtained using condition C12 are similar.

[0094] If condition C12 in Figure 9 is selected as condition CA and condition C13 is selected as condition CB, the causal effect EA is -3.0 and the causal effect EB is -1.0. Causal effects EA and EB have the same sign. The rate of change R is calculated by equation (6) and is -0.67. When T2 = 0.2, the absolute value of the rate of change R > T2, so it is determined that the causal effect EA obtained using condition C12 and the causal effect EB obtained using condition C13 are not similar.

[0095] If condition C12 in Figure 9 is selected as condition CA and condition C17 is selected as condition CB, the causal effect EA is -3.0 and the causal effect EB is -1.5. Causal effects EA and EB have the same sign. The rate of change R is calculated by equation (7) and is -0.50. When T2 = 0.2, the absolute value of the rate of change R > T2, so it is determined that the causal effect EA obtained using condition C12 and the causal effect EB obtained using condition C17 are not similar.

[0096] According to the determination method M2, the similarity between causal effect EA, obtained using condition CA, and causal effect EB, obtained using condition CB, can be determined with high accuracy.

[0097] When using the determination method M3, the similarity determination unit 515 obtains the causal effect EA obtained using condition CA and the causal effect EB obtained using condition CB from the causal effect information 524, and compares the signs of causal effect EA and causal effect EB. Then, the similarity determination unit 515 obtains the distance D between the causal graph 523 generated using condition CA and the causal graph 523 generated using condition CB from the distance information 525, and obtains the rate of change R between causal effect EA and causal effect EB from the rate of change information 526.

[0098] If the distance D is less than the threshold T1, the similarity determination unit 515 determines that the causal graph 523 generated using condition CA and the causal graph 523 generated using condition CB are similar. In this case, the subset generated using condition CA and the subset generated using condition CB are determined to be similar.

[0099] Assume that the signs of causal effect EA and causal effect EB are the same, and that the subset generated using condition CA and the subset generated using condition CB are similar. In this case, the similarity determination unit 515 determines that causal effect EA and causal effect EB are similar.

[0100] Next, we consider the case where the signs of causal effect EA and causal effect EB are the same, and the subset generated using condition CA and the subset generated using condition CB are not similar. In this case, the similarity determination unit 515 compares the absolute value of the rate of change R with the threshold T2, and if the absolute value of the rate of change R is less than the threshold T2, it determines that causal effect EA and causal effect EB are similar.

[0101] Otherwise, the similarity determination unit 515 determines that causal effect EA and causal effect EB are not similar.

[0102] According to the determination method M3, by using determination methods M1 and M2 in combination, the probability of determining that causal effect EA and causal effect EB are similar can be increased.

[0103] The generation unit 516 uses the determination result 527 to identify the numerical range of explanatory variables that produce a causal effect of the same sign as a specific factor variable on the dependent variable. The generation unit 516 then generates range information 528 indicating the identified numerical range and stores it in the storage unit 518. The output unit 517 outputs the range information 528 stored in the storage unit 518. The numerical range indicated by the range information 528 is an example of a specific range of values ​​for each of the multiple explanatory variables.

[0104] The generation unit 516 selects, for example, one of the conditions CX from the condition list 522. Then, the generation unit 516 selects one of the explanatory variables included in condition CX as explanatory variable V, and selects the unselected condition that is closest to condition CX for explanatory variable V as condition CY. The unselected condition that is closest to condition CX for explanatory variable V is the condition that represents the narrowest numerical range of explanatory variable V among the other unselected conditions that include condition CX.

[0105] Next, the generation unit 516 obtains judgment information from the judgment result 527 that is associated with the combination of condition CX, condition CY, and a specific factor variable, and checks whether the causal effect obtained using condition CX is similar to the causal effect obtained using condition CY.

[0106] If the two causal effects are similar, the generation unit 516 identifies the numerical range of the explanatory variable represented by condition CY as the numerical range of the explanatory variable from which a causal effect of the same sign can be obtained. If the two causal effects are not similar, the generation unit 516 identifies the numerical range of the explanatory variable represented by condition CX as the numerical range of the explanatory variable from which a causal effect of the same sign can be obtained. Condition CX is an example of the first condition, and condition CY is an example of the second condition.

[0107] For example, if condition C11 in Figure 7 is selected as condition CX and temperature is selected as the explanatory variable V, then the unselected conditions that include condition CX are conditions C12 through C18. Of conditions C12 through C18, the condition that represents the narrowest numerical range of the explanatory variable V is condition C16. Therefore, condition C16 is selected as condition CY.

[0108] If condition C16 is selected as condition CY, and the causal effect obtained using condition CX is not similar to the causal effect obtained using condition CY, then the unselected condition that is closest to condition CX for the explanatory variable V is selected as the new condition CY.

[0109] In this case, the unselected conditions, including condition CX, are conditions C12-C15, C17, and C18. Of these, the conditions that best represent the narrowest numerical range of the explanatory variable V are conditions C12 and C17. Therefore, either condition C12 or condition C17 is selected as the new condition CY.

[0110] As an example, consider a case where condition C12 is selected as a new condition CY, and the causal effect obtained using condition CX is similar to the causal effect obtained using condition CY. In this case, the numerical range of the explanatory variable represented by condition C12 is identified as the numerical range of the explanatory variable from which a causal effect of the same sign is obtained. The generation unit 516 then uses condition CY as a new condition CX and repeats the same process until there are no more unselected conditions that include condition CX.

[0111] Figure 11 shows an example of the numerical range of explanatory variables for which causal effects of the same sign are obtained. The symbol "×" represents a point in three-dimensional space represented by time, temperature, and energy. The temperature and energy indicated by "×" are measured values ​​at the same time indicated by "×". Curve 1101 represents an approximation curve that approximates the distribution of multiple "×" points.

[0112] Range 1111 corresponds to the numerical range of the explanatory variable represented by condition C11 in Figure 9, range 1112 corresponds to the numerical range of the explanatory variable represented by condition C12, and range 1113 corresponds to the numerical range of the explanatory variable represented by condition C13. Range 1114 corresponds to the numerical range of the explanatory variable represented by the condition "time > 15:00 AND temperature > 20°C", and range 1115 corresponds to the numerical range of the explanatory variable represented by the condition "time > 13:00 AND temperature > 15°C". Information regarding these two conditions is omitted in Figure 9.

[0113] In the example in Figure 11, the causal effect obtained using condition C11 is similar to the causal effect obtained using condition C12, and the signs of these causal effects are negative. However, the causal effect obtained using condition C12 is not similar to the causal effect obtained using condition C13, and the causal effect obtained using condition C12 is not similar to the causal effect obtained using condition C17. Therefore, the range 1112, which includes range 1111, is identified as the numerical range in which a negative causal effect is obtained.

[0114] In the example in Figure 11, the causal effect obtained using the conditions "time > 15:00 and temperature > 20°C" is similar to the causal effect obtained using the conditions "time > 13:00 and temperature > 15°C," and the signs of these causal effects are positive. Therefore, the range 1115, which includes range 1114, is identified as a numerical range in which a positive causal effect is obtained.

[0115] Figure 12 shows an example of range information 528. The range information 528 in Figure 12 includes a factor variable, a sign, and a range. In this example, the factor variable is temperature. The sign represents the sign of the causal effect, and the range represents the numerical range in which a causal effect of that sign is obtained. The numerical range in which a positive causal effect is obtained is "time > 13:00 and temperature > 15°C", and the numerical range in which a negative causal effect is obtained is "time < 10:00 and temperature < 10°C". The positive or negative sign is just one example of a given sign.

[0116] The output unit 517 may display the numerical range indicated by the range information 528 on the screen as a region in a three-dimensional space, as shown in Figure 11.

[0117] According to the variable range identification device 501 in Figure 5, one of the conditions CA and another condition CB that includes condition CA are selected from the conditions included in the condition list 522, and the similarity between the causal effect obtained using condition CA and the causal effect obtained using condition CB is determined.

[0118] If the two causal effects are not similar, the numerical range of the explanatory variable represented by condition CB is not adopted as the numerical range in which causal effects of the same sign are obtained. Instead, the numerical range of the explanatory variable represented by condition CA is adopted as the numerical range in which causal effects of the same sign are obtained. This allows for the accurate determination of the numerical range in which causal effects of the same sign are obtained.

[0119] For example, if the data set 521 contains data like that shown in Figure 2, then range 212 is not adopted as the numerical range where a negative causal effect is obtained, and range 211, which is the original maximum range, is adopted as the numerical range where a negative causal effect is obtained.

[0120] Figure 13 is a flowchart showing an example of the variable range identification process performed by the variable range identification device 501 in Figure 5. First, the condition extraction unit 511 performs the condition extraction process (step 1301), and the causal search unit 512 performs the causal search process (step 1302).

[0121] Next, the distance calculation unit 513 performs distance calculation processing (step 1303), the rate of change calculation unit 514 performs rate of change calculation processing (step 1304), and the similarity determination unit 515 performs similarity determination processing (step 1305). Then, the generation unit 516 performs range identification processing (step 1306).

[0122] Figure 14 is a flowchart showing an example of the condition extraction process in step 1301 of Figure 13. First, the condition extraction unit 511 sets one or more thresholds for each explanatory variable included in the data set 521 and generates inequalities that include the explanatory variables and the set thresholds (step 1401). Then, the condition extraction unit 511 generates conditions that represent each of the multiple combinations of inequalities by comprehensively combining the generated inequalities (step 1402).

[0123] Next, the condition extraction unit 511 selects one of the generated conditions (step 1403). Then, the condition extraction unit 511 extracts data that satisfies the selected condition from the data set 521 and generates a subset containing the extracted data (step 1404).

[0124] Next, the condition extraction unit 511 generates all possible combinations of two variables from a plurality of variables, including one or more explanatory variables and the dependent variable, that are included in the selected conditions. Then, the condition extraction unit 511 uses the data included in the generated subset to calculate the correlation coefficient of the two variables included in each combination (step 1405).

[0125] Next, the condition extraction unit 511 compares the number of data points included in the subset with the threshold TA (step 1406). If the number of data points is greater than or equal to the threshold TA (step 1406, YES), the condition extraction unit 511 compares the absolute value of the correlation coefficient calculated for each combination with the threshold TB (step 1407).

[0126] If the absolute value of one or more correlation coefficients is greater than or equal to the threshold TB (Step 1407, YES), the condition extraction unit 511 extracts the selected conditions as conditions in which correlation appears and adds them to the condition list 522 (Step 1408).

[0127] Next, the condition extraction unit 511 checks whether all conditions have been selected (step 1409). If there are any unselected conditions remaining (step 1409, NO), the condition extraction unit 511 repeats the process from step 1403 onwards for the next condition.

[0128] If the number of data points is less than the threshold TA (step 1406, NO), or if the absolute value of all correlation coefficients is less than the threshold TB (step 1407, NO), the condition extraction unit 511 performs the processing from step 1409 onwards. If all conditions are selected (step 1409, YES), the condition extraction unit 511 terminates the processing.

[0129] Figure 15 is a flowchart showing an example of the causal search process in step 1302 of Figure 13. First, the causal search unit 512 selects one condition from the condition list 522 (step 1501). Next, the causal search unit 512 extracts data that satisfies the selected condition from the data set 521 and generates a subset containing the extracted data (step 1502). Then, the causal search unit 512 generates a causal graph 523 by performing a statistical causal search using the generated subset (step 1503).

[0130] Next, the causal search unit 512 selects one of the explanatory variables included in the condition list 522 as a factor variable (step 1504). Then, the causal search unit 512 calculates the causal effect of the factor variable on the target variable from the generated causal graph 523, associates the conditions, factor variables, and causal effects, and adds them to the causal effect information 524 (step 1505).

[0131] Next, the causal search unit 512 checks whether all explanatory variables have been selected (step 1506). If there are any unselected explanatory variables remaining (step 1506, NO), the causal search unit 512 repeats the process from step 1504 onwards for the next explanatory variable.

[0132] If all explanatory variables are selected (Step 1506, YES), the causal search unit 512 checks whether all conditions have been selected (Step 1507). If there are any unselected conditions remaining (Step 1507, NO), the causal search unit 512 repeats the process from Step 1501 onwards for the next condition. If all conditions are selected (Step 1507, YES), the causal search unit 512 terminates the process.

[0133] Figure 16 is a flowchart showing an example of the distance calculation process in step 1303 of Figure 13. First, the distance calculation unit 513 selects a combination of any condition CA and another condition CB that includes condition CA from the condition list 522 (step 1601).

[0134] Next, the distance calculation unit 513 calculates the distance D between the causal graph 523 generated using condition CA and the causal graph 523 generated using condition CB, associates condition CA, condition CB, and distance D, and adds them to the distance information 525 (step 1602).

[0135] Next, the distance calculation unit 513 checks whether all combinations of conditions CA and CB have been selected (step 1603). If there are any unselected combinations remaining (step 1603, NO), the distance calculation unit 513 repeats the process from step 1601 onwards for the next combination. If all combinations have been selected (step 1603, YES), the distance calculation unit 513 terminates the process.

[0136] Figure 17 is a flowchart showing an example of the rate of change calculation process in step 1304 of Figure 13. First, the rate of change calculation unit 514 selects a combination of any condition CA and another condition CB that includes condition CA from the condition list 522 (step 1701). Next, the rate of change calculation unit 514 selects any explanatory variable included in the condition list 522 as a factor variable (step 1702).

[0137] Next, the rate of change calculation unit 514 obtains the causal effect EA associated with condition CA and factor variable, and the causal effect EB associated with condition CB and factor variable from the causal effect information 524 (step 1703). Then, the rate of change calculation unit 514 calculates the rate of change R of the causal effect using the causal effect EA and causal effect EB, associates condition CA, condition CB, factor variable, and rate of change R, and adds it to the rate of change information 526 (step 1704).

[0138] Next, the rate of change calculation unit 514 checks whether all explanatory variables have been selected (step 1705). If there are any unselected explanatory variables remaining (step 1705, NO), the rate of change calculation unit 514 repeats the process from step 1702 onwards for the next explanatory variable.

[0139] If all explanatory variables are selected (Step 1705, YES), the rate of change calculation unit 514 checks whether all combinations of conditions CA and CB have been selected (Step 1706). If there are any unselected combinations remaining (Step 1706, NO), the rate of change calculation unit 514 repeats the process from Step 1701 onwards for the next combination. If all combinations are selected (Step 1706, YES), the rate of change calculation unit 514 terminates the process.

[0140] Figure 18 is a flowchart showing an example of the similarity determination process in step 1305 of Figure 13. In the similarity determination process in Figure 18, determination method M3 is used.

[0141] First, the similarity determination unit 515 selects a combination of any condition CA and another condition CB that includes condition CA from the condition list 522 (step 1801). Then, the similarity determination unit 515 obtains the distance D associated with condition CA and condition CB from the distance information 525 (step 1802).

[0142] Next, the similarity determination unit 515 selects one of the explanatory variables included in the condition list 522 as a factor variable (step 1803). Next, the similarity determination unit 515 obtains the causal effect EA associated with condition CA and the factor variable, and the causal effect EB associated with condition CB and the factor variable from the causal effect information 524 (step 1804). Then, the similarity determination unit 515 obtains the rate of change R associated with condition CA, condition CB, and the factor variable from the rate of change information 526 (step 1805).

[0143] Next, the similarity determination unit 515 compares the sign of causal effect EA with the sign of causal effect EB (step 1806). If the signs of causal effect EA and causal effect EB are the same (step 1806, YES), the similarity determination unit 515 compares the distance D with the threshold T1 (step 1807).

[0144] If the distance D is less than the threshold T1 (step 1807, YES), the similarity determination unit 515 determines that causal effect EA and causal effect EB are similar. The similarity determination unit 515 then associates condition CA, condition CB, factor variables, and determination information indicating that causal effect EA and causal effect EB are similar, and adds it to the determination result 527 (step 1809).

[0145] If the distance D is greater than or equal to the threshold T1 (step 1807, NO), the similarity determination unit 515 compares the absolute value of the rate of change R with the threshold T2 (step 1808). If the absolute value of the rate of change R is less than the threshold T2 (step 1808, YES), the similarity determination unit 515 determines that causal effect EA and causal effect EB are similar. Then, the similarity determination unit 515 associates condition CA, condition CB, factor variables, and determination information indicating that causal effect EA and causal effect EB are similar, and adds it to the determination result 527 (step 1809).

[0146] If the signs of causal effect EA and causal effect EB are different (step 1806, NO), the similarity determination unit 515 determines that causal effect EA and causal effect EB are not similar. The similarity determination unit 515 then associates condition CA, condition CB, factor variables, and determination information indicating that causal effect EA and causal effect EB are not similar, and adds them to the determination result 527 (step 1812).

[0147] If the absolute value of the rate of change R is greater than or equal to the threshold T2 (step 1808, NO), the similarity determination unit 515 determines that causal effect EA and causal effect EB are not similar. The similarity determination unit 515 then associates condition CA, condition CB, factor variables, and determination information indicating that causal effect EA and causal effect EB are not similar, and adds it to the determination result 527 (step 1812).

[0148] Next, the similarity determination unit 515 checks whether all explanatory variables have been selected (step 1810). If there are any unselected explanatory variables remaining (step 1810, NO), the similarity determination unit 515 repeats the process from step 1803 onwards for the next explanatory variable.

[0149] If all explanatory variables are selected (Step 1810, YES), the similarity determination unit 515 checks whether all combinations of conditions CA and CB have been selected (Step 1811). If there are any unselected combinations remaining (Step 1811, NO), the similarity determination unit 515 repeats the process from Step 1801 onwards for the next combination. If all combinations are selected (Step 1811, YES), the similarity determination unit 515 terminates the process.

[0150] Figure 19 is a flowchart showing an example of the range identification process in step 1306 of Figure 13. First, the generation unit 516 selects a condition from the condition list 522 that represents the narrowest numerical range for the explanatory variable (step 1901). Then, the generation unit 516 generates range information 528 for a specific factor variable by performing a range update process that updates the numerical range represented by the selected condition (step 1902).

[0151] Next, the generation unit 516 refers to the causal effect information 524 and determines the sign of the causal effect associated with the combination of the condition selected in step 1901 and the specific factor variable. Then, the generation unit 516 checks whether there is a causal effect in the causal effect information 524 that has a different sign from the one determined (step 1903).

[0152] If causal effects with different signs exist (Step 1903, YES), the generation unit 516 selects the condition that represents the narrowest numerical range among the conditions corresponding to the causal effects with different signs from the condition list 522 (Step 1904). Then, the generation unit 516 updates the range information 528 by performing a range update process that updates the numerical range represented by the selected condition (Step 1905), and outputs the range information 528 (Step 1906).

[0153] If there are no causal effects with different signs (step 1903, NO), the generation unit 516 performs the process of step 1906.

[0154] Figure 20 is a flowchart illustrating an example of the range update process in steps 1902 and 1905 of Figure 19. First, the generation unit 516 sets the condition selected from the condition list 522 as condition CX (step 2001). Then, the generation unit 516 adds the numerical range of the explanatory variable represented by condition CX to the range information 528 as a numerical range in which a causal effect of the same sign can be obtained for a specific factor variable (step 2002).

[0155] Next, the generation unit 516 selects one of the explanatory variables included in condition CX as explanatory variable V, and selects the unselected condition that is closest to condition CX for explanatory variable V as condition CY (step 2003).

[0156] Next, the generation unit 516 obtains judgment information from the judgment result 527 that is associated with a combination of condition CX, condition CY, and a specific factor variable. The generation unit 516 then checks whether the obtained judgment information indicates that the causal effect obtained using condition CX is similar to the causal effect obtained using condition CY (step 2004).

[0157] If the causal effect obtained using condition CX is similar to the causal effect obtained using condition CY (step 2004, YES), the generation unit 516 performs the process in step 2005. In step 2005, the generation unit 516 updates the range information 528 by overwriting the numerical range of the explanatory variable represented by condition CY with the numerical range of the explanatory variable represented by condition CX (step 2005). Then, the generation unit 516 updates condition CX by setting condition CY to condition CX (step 2006).

[0158] Next, the generation unit 516 checks whether there are any unselected conditions that include condition CX (step 2007). If there are any unselected conditions that include condition CX (step 2007, YES), the generation unit 516 repeats the processing from step 2003 onwards.

[0159] If the causal effect obtained using condition CX and the causal effect obtained using condition CY are not similar (step 2004, NO), the generation unit 516 performs the processing from step 2007 onwards. If there are no unselected conditions including condition CX (step 2007, NO), the generation unit 516 terminates the processing.

[0160] The configuration of the information processing device 301 shown in Figure 3 is merely an example, and some components may be omitted or changed depending on the intended use or conditions of the information processing device 301.

[0161] The configuration of the variable range identification device 501 shown in Figure 5 is merely an example, and some components may be omitted or changed depending on the application or conditions of the variable range identification device 501. For example, when using determination method M1, the rate of change calculation unit 514 can be omitted, and when using determination method M2, the distance calculation unit 513 can be omitted.

[0162] The flowcharts in Figures 4 and 13 to 20 are merely examples, and some processes may be omitted or modified depending on the configuration or conditions of the information processing device 301 and the variable range identification device 501. For example, when using determination method M1, steps 1304 in Figure 13 and 1808 in Figure 18 can be omitted, and when using determination method M2, steps 1303 in Figure 13 and 1807 in Figure 18 can be omitted.

[0163] The conditions shown in Figures 1 and 7 are merely examples, and the generated conditions will vary depending on the data set 521. The numerical ranges of the explanatory variables shown in Figures 2 and 11 are also merely examples, and the numerical ranges of the explanatory variables will vary depending on the conditions generated from the data set 521.

[0164] The data set 521 shown in Figure 6 is merely an example, and the data set 521 changes depending on the application of the variable range identification device 501. The causal graph 523 shown in Figure 8, the causal effect information 524 shown in Figure 9, and the adjacency matrix shown in Figure 10 are merely examples, and the causal graph 523, the causal effect information 524, and the adjacency matrix change depending on the conditions generated from the data set 521. The range information 528 shown in Figure 12 is merely an example, and the range information 528 changes depending on the causal graph 523 and the causal effect information 524.

[0165] Equations (1) to (7) are merely examples, and the variable range identification device 501 may perform the variable range identification process using other mathematical formulas.

[0166] Figure 21 shows an example of the hardware configuration of an information processing device used as the information processing device 301 in Figure 3 and the variable range identification device 501 in Figure 5. The information processing device in Figure 21 includes a CPU (Central Processing Unit) 2101, memory 2102, input device 2103, output device 2104, auxiliary storage device 2105, media drive device 2106, and network connection device 2107. These components are hardware and are connected to each other by a bus 2108.

[0167] Memory 2102 is a semiconductor memory such as ROM (Read Only Memory) or RAM (Random Access Memory), and stores the program and data used for processing. Memory 2102 may also operate as the storage unit 518 in Figure 5.

[0168] The CPU 2101 (processor) operates as the causal search unit 311, the determination unit 312, and the range identification unit 313 in Figure 3, for example, by executing a program using the memory 2102. The CPU 2101 also operates as the condition extraction unit 511, the causal search unit 512, the distance calculation unit 513, the rate of change calculation unit 514, the similarity determination unit 515, and the generation unit 516 in Figure 5, by executing a program using the memory 2102.

[0169] The input device 2103 is, for example, a keyboard, a pointing device, etc., and is used for inputting instructions or information from the user or operator. The output device 2104 is, for example, a display device, a printer, etc., and is used for inquiries or instructions to the user or operator, and for outputting processing results. The output device 2104 may operate as the output unit 314 in Figure 3 or the output unit 517 in Figure 5. The processing result may be range information 528.

[0170] The auxiliary storage device 2105 is, for example, a magnetic disk drive, an optical disk drive, a magneto-optical disk drive, a tape drive, etc. The auxiliary storage device 2105 may also be a hard disk drive or an SSD (Solid State Drive). The information processing device can store programs and data in the auxiliary storage device 2105 and load them into the memory 2102 for use. The auxiliary storage device 2105 may also operate as the storage unit 518 in Figure 5.

[0171] The media drive unit 2106 drives the portable recording medium 2109 and accesses its recorded contents. The portable recording medium 2109 is a memory device, flexible disk, optical disk, magneto-optical disk, etc. The portable recording medium 2109 may also be a CD-ROM (Compact Disk Read Only Memory), DVD (Digital Versatile Disk), USB (Universal Serial Bus) memory, etc. The user or operator can store programs and data on the portable recording medium 2109 and load them into the memory 2102 for use.

[0172] Thus, the computer-readable recording medium that stores the programs and data used in the processing is a physical (non-temporary) recording medium such as memory 2102, auxiliary storage device 2105, or portable recording medium 2109.

[0173] The network connection device 2107 is a communication device that connects to a communication network such as a WAN (Wide Area Network) or LAN (Local Area Network) and performs data conversion associated with communication. The information processing device can receive programs and data from external devices via the network connection device 2107 and load them into the memory 2102 for use. The network connection device 2107 may also operate as the output unit 314 in Figure 3 or the output unit 517 in Figure 5.

[0174] Note that the information processing device does not need to include all the components shown in Figure 21, and some components may be omitted or modified depending on the intended use or conditions of the information processing device. For example, if an interface with a user or operator is not required, the input device 2103 and output device 2104 can be omitted. If a portable recording medium 2109 or a communication network is not used, the media drive device 2106 or network connection device 2107 can be omitted.

[0175] While embodiments of the disclosure and their advantages have been described in detail, those skilled in the art will be able to make various modifications, additions, and omissions without departing from the scope of the invention as expressly stated in the claims.

[0176] With reference to the embodiments described with reference to Figures 1 to 21, the following additional information is disclosed. (Note 1) By performing causal exploration using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the dependent variable, based on multiple conditions relating to the range of values ​​of the multiple explanatory variables, the causal relationship between the multiple explanatory variables and the dependent variable is determined. Based on the causal relationships determined based on each of the aforementioned conditions, the causal effect of any of the aforementioned explanatory variables on the dependent variable is determined. By determining the similarity of the causal effects obtained based on each of the aforementioned conditions, the similarity determination result is obtained. Based on the similarity determination result, a specific range of values ​​for each of the multiple explanatory variables is determined such that the causal effect has a predetermined sign. Outputting range information indicating the aforementioned specific range, An information processing program characterized by having a computer perform the processing. (Note 2) The process for determining the similarity result is as follows: A process for determining the similarity between a first subset extracted based on the first condition among the aforementioned multiple conditions and a second subset extracted based on the second condition among the aforementioned multiple conditions, The information processing program according to Appendix 1, characterized in that it includes a process for determining that the first causal effect and the second causal effect are similar if the first subset and the second subset are similar, and the sign of the first causal effect determined based on the first condition is the same as the sign of the second causal effect determined based on the second condition. (Note 3) The information processing program according to Appendix 2 is characterized in that the process for determining the similarity between the first subset and the second subset includes the process of determining that the first subset and the second subset are similar if the distance between the first causal graph representing the first causal relationship obtained using the first subset and the second causal graph representing the second causal relationship obtained using the second subset is less than a threshold. (Note 4) The process for determining the similarity result is as follows: A process to calculate the difference between the first causal effect obtained based on the first condition among the aforementioned multiple conditions and the second causal effect obtained based on the second condition among the aforementioned multiple conditions, The information processing program according to Appendix 1, characterized by including a process for determining the similarity between the first causal effect and the second causal effect based on the difference, the sign of the first causal effect, and the sign of the second causal effect. (Note 5) The information processing program according to any one of Appendix 2 to Appendix 4, characterized in that the process of determining the specific range includes, if the range of values ​​of each of the multiple explanatory variables that satisfy the second condition includes the range of values ​​of each of the multiple explanatory variables that satisfy the first condition, and if it is determined that the first causal effect and the second causal effect are similar, the process of determining the specific range based on the range of values ​​of each of the multiple explanatory variables that satisfy the second condition. (Note 6) A causal search unit performs causal search using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the target variable, based on multiple conditions relating to the range of values ​​of the multiple explanatory variables, to find the causal relationship between the multiple explanatory variables and the target variable, and then finds the causal effect of any of the multiple explanatory variables on the target variable based on the causal relationship found based on each of the multiple conditions, A determination unit that determines the similarity of the causal effects obtained based on each of the aforementioned multiple conditions, thereby determining the similarity determination result, A range determination unit that determines a specific range of values ​​for each of the plurality of explanatory variables in which the causal effect has a predetermined sign, based on the similarity determination result, An output unit that outputs range information indicating the aforementioned specific range, An information processing device characterized by comprising: (Note 7) The information processing apparatus according to Appendix 6, characterized in that the determination unit determines the similarity between a first subset extracted from the plurality of conditions based on a first condition and a second subset extracted from the plurality of conditions based on a second condition, and determines that the first subset and the second subset are similar and the sign of the first causal effect obtained based on the first condition is the same as the sign of the second causal effect obtained based on the second condition, and that the first causal effect and the second causal effect are similar. (Note 8) The information processing apparatus according to Appendix 7, characterized in that the determination unit determines that the first subset and the second subset are similar if the distance between the first causal graph representing the first causal relationship obtained using the first subset and the second causal graph representing the second causal relationship obtained using the second subset is less than a threshold. (Note 9) The information processing apparatus according to Appendix 6, characterized in that the determination unit determines the difference between a first causal effect obtained based on a first condition among the plurality of conditions and a second causal effect obtained based on a second condition among the plurality of conditions, and determines the similarity between the first causal effect and the second causal effect based on the difference, the sign of the first causal effect, and the sign of the second causal effect. (Note 10) The information processing apparatus according to any one of Appendix 7 to Appendix 9, characterized in that the range-specification unit determines the specific range based on the range of values ​​of each of the multiple explanatory variables that satisfy the second condition, when it is determined that the range of values ​​of each of the multiple explanatory variables that satisfy the second condition includes the range of values ​​of each of the multiple explanatory variables that satisfy the first condition, and the first causal effect and the second causal effect are similar. (Note 11) By performing causal exploration using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the dependent variable, based on multiple conditions relating to the range of values ​​of the multiple explanatory variables, the causal relationship between the multiple explanatory variables and the dependent variable is determined. Based on the causal relationships determined based on each of the aforementioned conditions, the causal effect of any of the aforementioned explanatory variables on the dependent variable is determined. By determining the similarity of the causal effects obtained based on each of the aforementioned conditions, the similarity determination result is obtained. Based on the similarity determination result, a specific range of values ​​for each of the multiple explanatory variables is determined such that the causal effect has a predetermined sign. Outputting range information indicating the aforementioned specific range, An information processing method characterized in that the processing is performed by a computer. (Note 12) The process for determining the similarity result is as follows: A process for determining the similarity between a first subset extracted based on the first condition among the aforementioned multiple conditions and a second subset extracted based on the second condition among the aforementioned multiple conditions, The information processing method according to Appendix 11, characterized in that it includes a process of determining that the first causal effect and the second causal effect are similar if the first subset and the second subset are similar, and the sign of the first causal effect determined based on the first condition is the same as the sign of the second causal effect determined based on the second condition. (Note 13) The information processing method according to Appendix 12, characterized in that the process for determining the similarity between the first subset and the second subset includes the process of determining that the first subset and the second subset are similar if the distance between the first causal graph representing the first causal relationship obtained using the first subset and the second causal graph representing the second causal relationship obtained using the second subset is less than a threshold. (Note 14) The process for determining the similarity result is as follows: A process to calculate the difference between the first causal effect obtained based on the first condition among the aforementioned multiple conditions and the second causal effect obtained based on the second condition among the aforementioned multiple conditions, The information processing method according to Appendix 11, characterized by including a process to determine the similarity between the first causal effect and the second causal effect based on the difference, the sign of the first causal effect, and the sign of the second causal effect. (Note 15) The information processing method according to any one of Appendix 12 to Appendix 14, characterized in that the process of determining the specific range includes, if it is determined that the range of values ​​of each of the multiple explanatory variables that satisfy the second condition includes the range of values ​​of each of the multiple explanatory variables that satisfy the first condition, and the first causal effect and the second causal effect are similar, the process of determining the specific range based on the range of values ​​of each of the multiple explanatory variables that satisfy the second condition. [Explanation of Symbols]

[0177] 201, 1101 curve Ranges 211, 212, 1111-1115 301 Information Processing Equipment 311, 512 Causal Exploration Department 312 Judgment section 313 Range Identification Unit 314, 517 Output section 501 Variable Range Identification Device 511 Condition extraction part 513 Distance calculation part 514 Change Rate Calculation Unit 515 Similarity Judgment Unit 516 Generation part 518 Storage section 521 Datasets 522 Condition List 523 Causal graph 524 Causal Effect Information 525 Distance Information 526 Rate of Change Information 527 Judgment result 528 Range Information 2101 CPU 2102 memory 2103 Input device 2104 Output device 2105 Auxiliary storage device 2106 Media drive device 2107 Network Connection Device 2108 Bus 2109 Portable recording media

Claims

1. By performing causal exploration using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the dependent variable, based on multiple conditions relating to the range of values ​​of the multiple explanatory variables, the causal relationship between the multiple explanatory variables and the dependent variable is determined. Based on the causal relationships determined based on each of the aforementioned conditions, the causal effect of any of the aforementioned explanatory variables on the dependent variable is determined. By determining the similarity of the causal effects obtained based on each of the aforementioned conditions, the similarity determination result is obtained. Based on the similarity determination result, a specific range of values ​​for each of the multiple explanatory variables is determined such that the causal effect has a predetermined sign. Outputting range information indicating the aforementioned specific range, An information processing program characterized by having a computer perform the processing.

2. The process for determining the similarity result is as follows: A process for determining the similarity between a first subset extracted based on a first condition from among the aforementioned multiple conditions and a second subset extracted based on a second condition from among the aforementioned multiple conditions, The information processing program according to claim 1, characterized in that it includes a process for determining that the first causal effect and the second causal effect are similar if the first subset and the second subset are similar, and the sign of the first causal effect determined based on the first condition is the same as the sign of the second causal effect determined based on the second condition.

3. The information processing program according to claim 2, characterized in that the process for determining the similarity between the first subset and the second subset includes a process for determining that the first subset and the second subset are similar if the distance between the first causal graph representing the first causal relationship obtained using the first subset and the second causal graph representing the second causal relationship obtained using the second subset is less than a threshold.

4. The process for determining the similarity result is as follows: A process to calculate the difference between the first causal effect obtained based on the first condition among the aforementioned multiple conditions and the second causal effect obtained based on the second condition among the aforementioned multiple conditions, The information processing program according to claim 1, characterized in that it includes a process for determining the similarity between the first causal effect and the second causal effect based on the difference, the sign of the first causal effect, and the sign of the second causal effect.

5. The information processing program according to any one of claims 2 to 4, characterized in that the process of determining the specific range includes, if it is determined that the range of values ​​of each of the plurality of explanatory variables that satisfy the second condition includes the range of values ​​of each of the plurality of explanatory variables that satisfy the first condition, and the first causal effect and the second causal effect are similar, the process of determining the specific range based on the range of values ​​of each of the plurality of explanatory variables that satisfy the second condition.

6. A causal search unit performs causal search using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the target variable, based on multiple conditions relating to the range of values ​​of the multiple explanatory variables, to find the causal relationship between the multiple explanatory variables and the target variable, and then finds the causal effect of any of the multiple explanatory variables on the target variable based on the causal relationship found based on each of the multiple conditions, A determination unit that determines the similarity of the causal effects obtained based on each of the aforementioned multiple conditions, and obtains the similarity determination result, A range determination unit that determines a specific range of values ​​for each of the plurality of explanatory variables in which the causal effect has a predetermined sign, based on the similarity determination result, An output unit that outputs range information indicating the aforementioned specific range, An information processing device characterized by comprising:

7. By performing causal exploration using subsets extracted from a data set containing the values ​​of multiple explanatory variables and the value of the dependent variable, based on multiple conditions relating to the range of values ​​of the multiple explanatory variables, the causal relationship between the multiple explanatory variables and the dependent variable is determined. Based on the causal relationships determined based on each of the aforementioned conditions, the causal effect of any of the aforementioned explanatory variables on the dependent variable is determined. By determining the similarity of the causal effects obtained based on each of the aforementioned conditions, the similarity determination result is obtained. Based on the similarity determination result, a specific range of values ​​for each of the multiple explanatory variables is determined such that the causal effect has a predetermined sign. Outputting range information indicating the aforementioned specific range, An information processing method characterized in that the processing is performed by a computer.

Images