Information processing device, information processing method, and recording medium

Abstract

To quickly and accurately analyze a plurality of pieces of data.SOLUTION: An information processing apparatus includes: an objective variable acquisition section that acquires a multi-dimensional objective variable; an objective variable dimension compression section that compresses the number of dimensions of the objective variable; an explanatory variable acquisition section that acquires an explanatory variable; and an influence degree calculation section that sets a coefficient obtained by compressing the number of dimensions of the objective variable as a new objective variable and calculates an influence degree on the new objective variable using the explanatory variable.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] One embodiment of the present invention relates to an information processing apparatus, an information processing method, and a recording medium.

Background Art

[0002] Since it is not easy to manually analyze a huge amount of data, techniques for analyzing data using a computer have been proposed. For example, when analyzing whether there is a defect in a certain product from an image of the product, a plurality of reference images obtained by photographing a plurality of defect forms in advance are prepared, and the analysis target image is compared with the reference images, whereby the analysis target image can be analyzed.

[0003] However, when there are many defect forms, it takes time to identify the defect forms included in the analysis target image. In addition, when the analysis target image includes a defect in which a plurality of defect forms are combined or a defect whose form is slightly different from the reference image, there is a risk that appropriate analysis cannot be performed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Therefore, one embodiment of the present invention provides an information processing apparatus, an information processing method, and a recording medium capable of quickly and accurately analyzing a plurality of data.

Means for Solving the Problems

[0006] In order to solve the above problems, according to one embodiment of the present invention, a target variable acquisition unit that acquires a multi-dimensional target variable, a target variable dimension compression unit that compresses the number of dimensions of the target variable, An explanatory variable acquisition unit that acquires explanatory variables, An information processing device is provided, which includes a coefficient obtained by compressing the dimensionality of the aforementioned objective variable and an influence calculation unit that calculates the degree of influence on the new objective variable using the aforementioned explanatory variables. [Brief explanation of the drawing]

[0007] [Figure 1] A block diagram showing the schematic configuration of an information processing device according to one embodiment. [Figure 2] A block diagram showing a more detailed configuration of the information processing device according to this embodiment. [Figure 3A] A flowchart illustrating the processing operation of the information processing device according to this embodiment. [Figure 3B] Flowchart following Figure 3A. [Figure 4] A schematic diagram illustrating how defects occur during the multi-step processing of semiconductor wafers. [Figure 5] A diagram showing four types of reference images. [Figure 6] A diagram showing an example of a target variable input to the target variable dimensionality reduction unit. [Figure 7] A diagram illustrating an example where the dependent variable is dimensionally reduced to two coefficients and two basis vectors. [Figure 8] A diagram showing an example of coefficients and bases obtained by NMF. [Figure 9] A diagram showing an example of the user interface operation screen. [Figure 10] Detailed flowchart of the processes performed by the impact calculation unit in steps S17 to S20 of Figure 3B. [Figure 11] A diagram showing an example of input data for the multitasking Lasso application. [Figure 12] A diagram illustrating an example of converting categorical variables to numerical data. [Figure 13] A diagram showing an example of output data from the multitasking Lasso application. [Figure 14A] A diagram showing an example of the screen display in the impact level display section. [Figure 14B]A diagram showing an example of a screen display of the target variable information display section. [Figure 15] A diagram showing an example of applying the information processing apparatus according to the present embodiment to the analysis of the relationship between the nucleotide sequence of a gene and a genetic disease.

Embodiment for Carrying Out the Invention

[0008] Hereinafter, embodiments of an information processing apparatus, an information processing method, and a recording medium will be described with reference to the drawings. Hereinafter, the main components of the information processing apparatus will be mainly described, but there may be components and functions that are not shown or described in the information processing apparatus. The following description does not exclude components and functions that are not shown or described.

[0009] FIG. 1 is a block diagram showing a schematic configuration of an information processing apparatus 1 according to an embodiment, and FIG. 2 is a block diagram showing a more detailed configuration of the information processing apparatus 1 according to the present embodiment. The information processing apparatus 1 according to the present embodiment includes a target variable acquisition unit 2, a target variable dimensional compression unit 3, an explanatory variable acquisition unit 4, and an influence degree calculation unit 5. The information processing apparatus 1 in FIG. 1 can perform the processing operations of the target variable acquisition unit 2, the target variable dimensional compression unit 3, and the influence degree calculation unit 5 by, for example, a computer executing a program stored in a recording medium. Alternatively, as will be described later, a hardware device that executes at least a part of the target variable acquisition unit 2, the target variable dimensional compression unit 3, and the influence degree calculation unit 5 may be provided.

[0010] The target variable acquisition unit 2 acquires a target variable having a plurality of dimensions. The target variable acquisition unit 2 may acquire, for example, a target variable stored in a target variable storage unit (not shown), or may acquire a target variable input from a target variable input unit (not shown). The target variable is a continuous variable that is the target of dimensional compression. The target variable having a plurality of dimensions acquired by the target variable acquisition unit 2 includes, for example, at least one of an electrical characteristic value, a numerical value, and a measured value.

[0011] The target variable dimensionality compression unit 3 compresses the dimensionality of the target variable acquired by the target variable acquisition unit 2. By compressing the dimensionality of the target variable, the processing load of the information processing apparatus 1 can be reduced, and the analysis of the target variable becomes easier. For example, when performing defect analysis, the target variable is compressed to the dimensionality highly related to defects. The target variable dimensionality compression unit 3 compresses the dimensionality of the target variable using, for example, PCA (Principal Component Analysis), clustering, or NMF (Non-negative Matrix Factorization).

[0012] The explanatory variable acquisition unit 4 acquires explanatory variables. The explanatory variables are variables that affect the target variable and include at least one of continuous variables and categorical variables. Continuous variables are numerical data, while categorical variables are non-numerical data. The explanatory variable acquisition unit 4 may acquire the explanatory variables input from an explanatory variable input unit (not shown).

[0013] The influence degree calculation unit 5 uses the coefficient obtained by compressing the dimensionality of the target variable as a new target variable, and calculates the influence degree on the new target variable using the explanatory variables acquired by the explanatory variable acquisition unit 4. The influence degree is, for example, the degree of influence on the occurrence of defects when performing defect analysis. More specifically, when the influence degree calculation unit 5 uses multi-task Lasso as described later, the influence degree is a regression coefficient.

[0014] The information processing apparatus 1 according to the present embodiment may include a preprocessing unit 6 and an explanatory variable storage unit 7 as shown in FIG. 2. The preprocessing unit 6 performs preprocessing for converting the explanatory variables acquired by the explanatory variable acquisition unit 4 into a form capable of data analysis. The explanatory variable storage unit 7 stores the preprocessed explanatory variables.

[0015] More specifically, the preprocessing unit 6 may include an explanatory variable determination unit 8, a numerical conversion processing unit 9, a missing value processing unit 10, and a standardization processing unit 11. The explanatory variable determination unit 8 determines whether the explanatory variable obtained by the explanatory variable acquisition unit 4 is a categorical variable or a continuous variable. The numerical conversion processing unit 9 performs numerical conversion processing for categorical variables if the explanatory variable obtained by the explanatory variable acquisition unit 4 is a categorical variable. The missing value processing unit 10 performs processing to fill in missing values ​​if the explanatory variable obtained by the explanatory variable acquisition unit 4 is a continuous variable. The standardization processing unit 11 performs standardization processing on the explanatory variable after numerical conversion processing or missing value processing. Standardization processing is, for example, processing to normalize by setting the maximum value to 1. The explanatory variable storage unit 7 stores the explanatory variable after standardization processing. The influence calculation unit 5 receives the explanatory variable stored in the explanatory variable storage unit 7 as input.

[0016] The information processing device 1 according to this embodiment may include a parameter input unit 12, as shown in Figure 2. The parameter input unit 12 inputs parameters necessary to compress the dimensionality of the target variable. For example, if the target variable includes multiple tasks, a parameter specifying the number of tasks is input. The parameters input in the parameter input unit 12 are used in the target variable dimensionality compression unit 3. The parameter input unit 12 may also have a user interface unit 12a that has a function to re-input parameters while displaying the result of compressing the dimensionality of the target variable. The screen of the user interface unit 12a displays the result of compressing the dimensionality of the target variable. The operator can check the result of compressing the dimensionality of the target variable and re-input parameters. This allows the operator to input parameters that result in a good compression of the dimensionality of the target variable.

[0017] As will be described later, the dependent variable can be calculated based on the product of a basis that characterizes the dependent variable and coefficients that weight the basis. The dependent variable dimensionality reduction unit 3 can calculate the coefficients and basis by inputting the dependent variable into an equation that calculates the dependent variable based on the product of a basis that characterizes the dependent variable and coefficients that weight the basis. The coefficients are used as the dependent variable in the influence calculation unit 5.

[0018] The influence calculation unit 5 calculates the degree of influence on the target variable by performing calculations using, for example, a multi-learning method. More specifically, the influence calculation unit 5 may calculate the degree of influence on the target variable using multi-task Lasso or Bayesian optimization, etc. The calculated degree of influence is stored, for example, in the influence storage unit 13.

[0019] The information processing device 1 according to this embodiment may include an influence display unit 14 and a target variable information display unit 15, as shown in Figure 2.

[0020] The influence display unit 14 displays each of the explanatory variables and their corresponding influence for each coefficient obtained by dimensionality reduction of the dependent variable. The influence calculation unit 5 calculates the regression coefficients as influences, for example using multitasking Lasso. In this case, the influence display unit 14 may rearrange and display each of the explanatory variables in order of the regression coefficient values. The dependent variable information display unit 15 displays information about the dependent variable used in multitasking Lasso.

[0021] The information processing device 1 according to this embodiment may include a target variable selection unit 16 and a target variable setting unit 17, as shown in Figure 2. The target variable selection unit 16 selects the target variables necessary for the influence calculation unit 5 to calculate the influence from among the target variables with the number of dimensions compressed by the target variable dimension compression unit 3. The target variable setting unit 17 sets the parameters selected by the target variable selection unit 16 to the target variables with the number of dimensions compressed.

[0022] The information processing device 1 according to this embodiment may include a data division unit 18, a parameter setting unit 19, a multi-learning calculation unit 20, and an analysis accuracy verification unit 21, as shown in Figure 2.

[0023] The data splitting unit 18 generates a dataset by combining the explanatory variables read from the explanatory variable storage unit 7 and the target variable with reduced dimensionality, in order for the influence calculation unit 5 to calculate the influence, and then splits this dataset into training data and validation data. The parameter setting unit 19 sets the parameters used by the multi-learning calculation unit 20. The multi-learning calculation unit 20 uses the training data and parameters split by the data splitting unit 18 to perform calculations such as multi-task Lasso, a sparse modeling method, or Bayesian optimization, a multi-objective optimization method, to calculate the influence of regression coefficients, etc. The analysis accuracy verification unit 21 calculates the accuracy of the influence calculated by the influence calculation unit 5 based on the validation data split by the data splitting unit 18.

[0024] Figures 3A and 3B are flowcharts showing the processing operation of the information processing device 1 according to this embodiment. Steps S1 to S8 in Figure 3A show the processing operation of explanatory variables, and steps S11 to S20 in Figure 3B show the processing operation of the target variable. Steps S11 to S16 are performed in parallel with the processing of steps S1 to S8, and steps S17 onward are executed using the processing results of steps S8 and S16.

[0025] First, the explanatory variable acquisition unit 4 acquires explanatory variables from the database 22 (step S1). Next, the explanatory variable determination unit 8 determines whether the acquired explanatory variables are categorical variables or not (step S2).

[0026] If the result in step S2 is YES, that is, if the explanatory variable is a categorical variable, the numerical conversion processing unit 9 excludes the unique variable (step S3) and converts the non-numerical data into numerical data (step S4).

[0027] On the other hand, if the result in step S2 is NO, that is, if the explanatory variable is a continuous variable, the missing value processing unit 10 removes the unnecessary variable (step S5) and fills in the missing value (step S6).

[0028] When the processing in step S4 or S6 is completed, the standardization processing unit 11 standardizes the explanatory variables (step S7). The standardized explanatory variables are stored in the explanatory variable storage unit 7 (step S8).

[0029] In parallel with the processing in steps S1 to S8, the target variable acquisition unit 2 acquires the target variable from the database 22 (step S11 in Figure 3B). Next, the parameter input unit 12 inputs the parameters to be used in the processing of the target variable dimensionality reduction unit 3 (step S12). The input parameters are, for example, the number of dimensions of the target variable. The parameter input unit 12 inputs the parameters using the user interface unit 12a.

[0030] Next, the target variable dimensionality reduction unit 3 compresses the number of dimensions of the target variable using the target variable obtained in step S11 and the parameters entered in step S12 (step S13). Next, it is determined whether the compression accuracy by the target variable dimensionality reduction unit 3 exceeds a reference value (step S14). If it is determined to be NO in step S14, that is, if the compression accuracy falls below the reference value, the process in steps S12 to S14 is repeated. The process in steps S12 to S14 can be performed, for example, by the user interface unit 12a, allowing the operator to re-enter parameters while viewing the screen display.

[0031] If the result in step S14 is YES, that is, if the compression accuracy exceeds the standard value, the target variable selection unit 16 selects a coefficient from among the coefficients obtained by compressing the dimensionality of the target variable that the influence calculation unit 5 will use as the new target variable (step S15). Next, the target variable setting unit 17 sets the target variable selected by the target variable selection unit 16 as the new target variable for influence calculation (step S16).

[0032] Next, the influence calculation unit 5 calculates the degree of influence on the target variable based on the target variable set in step S16 and the explanatory variables stored in the explanatory variable storage unit 7 in step S8 (step S17). The influence calculation unit 5 performs calculations using, for example, a multi-learning method, as will be described later. The degree of influence calculated by the influence calculation unit 5 is stored in, for example, the influence storage unit 13 (step S18).

[0033] Next, the influence levels stored in the influence level storage unit 13 are displayed by the influence level display unit 14 as needed (step S19). Also, if necessary, the target variable information display unit 15 displays the target variable used to calculate the influence level (step S20).

[0034] The following describes an example of performing defect analysis in the semiconductor device manufacturing process using the information processing device 1 according to this embodiment. Semiconductor devices are manufactured by performing numerous processes such as film deposition, exposure, and etching on a semiconductor wafer. Defects may occur during any of these processes, and by analyzing which process caused what type of defect, the cause of the defect can be identified, countermeasures can be taken, and this leads to an improvement in yield.

[0035] Figure 4 schematically illustrates how defects occur during the processing of a semiconductor wafer through multiple steps. Figure 4 shows an example where a semiconductor wafer is fed into a semiconductor manufacturing machine and processed through a total of six steps, from step A to step F, before being discharged as a processed semiconductor wafer. Figure 4 shows an example where a defect occurs in the center of the semiconductor wafer in step B, and a ring-shaped defect occurs in step E.

[0036] Let's assume that by visualizing the electrical characteristics of the semiconductor wear or some measured value as the target variable, it was possible to classify it into four types of reference images as shown in Figure 5. Reference image IM1 is an image without defects, reference image IM2 is an image containing only defects in the central part, reference image IM3 is an image containing only ring-shaped defects, and reference image IM4 is an image containing both defects in the central part and ring-shaped defects. The four types of reference images shown in Figure 5 represent chips with high or low measured values ​​by color or brightness changes, based on the assumption that chips with high or low measured values ​​are defective compared to other chips.

[0037] When performing failure analysis on semiconductor wafers, it is conceivable to use image analysis to determine which of the reference images IM1 to IM4 the image of the semiconductor wafer being analyzed, with its electrical characteristics or measured values ​​as the target variable, corresponds to. However, a single semiconductor wafer may contain multiple different types of defects, and as the number of defect types increases, considering combinations of different defects would result in a very large number of reference images, making the classification of the image being analyzed extremely time-consuming. Furthermore, if the failure rate differs significantly for each defect type, there is a risk that the true cause of defects with low failure rates cannot be accurately determined.

[0038] In contrast, in this embodiment, the number of dimensions of the target variable is reduced by performing the processing operations of the flowcharts in Figures 3A and 3B. The coefficient obtained by reducing the number of dimensions of the target variable is then used as a new target variable, and the degree of influence on the new target variable is calculated using a multi-learning method.

[0039] Figure 6 shows an example of the target variable Y input to the target variable dimensionality reduction unit 3. Figure 6 shows chip information on a semiconductor wafer. Chip information includes, for example, electrical characteristic values, numerical data, and measured values ​​for each chip. Since multiple chips are formed on a semiconductor wafer, the target variable containing the chip information for each of the multiple chips is input to the target variable dimensionality reduction unit 3.

[0040] The target variable dimensionality reduction unit 3 reduces the dimensionality of the target variable, for example, by NMF. NMF is expressed by the following equation (1): Y is the target variable to be reduced in dimensionality, W is the coefficient obtained by dimensionality reduction, H is the basis obtained by dimensionality reduction, and E is the error. For example, Y is an n x d matrix, W is an n x m matrix, and H is an m x d matrix (d, m, and n are integers greater than or equal to 1). Y = W·H + E …(1)

[0041] Figure 7 shows an example where the dependent variable Y is reduced in dimensionality to two coefficients w1 and w2 and two basis vectors h1 and h2. Coefficient w1 is multiplied by basis vector h1 and determines the magnitude of basis vector h1. Coefficient w2 is multiplied by basis vector h2 and determines the magnitude of basis vector h2. Basis vectors h1 and h2 are, for example, defects of different forms.

[0042] Figure 8 shows an example of the coefficients W and basis H obtained by NMF when the objective variable Y from Figure 6 is input to the objective variable dimensionality reduction unit 3. In the example in Figure 8, NMF obtains two tasks as a basis and coefficients W corresponding to each task. A task is, for example, the type of defect.

[0043] In this way, by inputting the target variable Y into the target variable dimensionality reduction unit 3 and arbitrarily setting conditions such as the number of NMF tasks as parameters, the coefficient W, basis H, and error E can be automatically calculated from the target variable Y under the set conditions.

[0044] The objective variable dimensionality reduction unit 3 requires appropriate input of conditions such as the number of tasks. If undesirable conditions are input, it will be impossible to calculate appropriate coefficients W, base H, and error E. In this embodiment, the parameter input unit 12 allows the operator to arbitrarily input conditions such as the number of tasks used in the objective variable dimensionality reduction unit 3 via the user interface unit 12a.

[0045] Figure 9 shows an example of the operation screen of the user interface unit 12a. Screen SC1 in Figure 9 displays the value of the target variable (e.g., electrical characteristic value) before the number of tasks is compressed as a line graph, and screen SC2 shows an example of displaying the value of the target variable when the number of tasks is compressed to 3 as a line graph. In screens SC1 and SC2, the horizontal axis represents the chip type, and the vertical axis represents the value of the target variable. The operator can adjust the number of tasks via the user interface unit 12a to approximate the line graph before compression.

[0046] Figure 10 is a detailed flowchart of the process performed by the influence calculation unit 5 in steps S17 to S20 of Figure 3B. First, a dataset is generated by combining the explanatory variables stored in the explanatory variable storage unit 7 and the target variable obtained in step S16, and this dataset is then split into training data and validation data (step S21).

[0047] Next, the impact calculation unit 5, for example, when calculating the impact using multitasking Lasso, sets the parameters of multitasking Lasso (step S22). The parameters set here relate to the number of divisions for cross-validation, the data pattern during cross-validation, and the convergence determination of the coordinate descent method.

[0048] Next, the multitasking Lasso program is run using the training data that was split in step S21 (step S23).

[0049] Next, the model constructed using multitasking Lasso is given the validation data divided in step S21, and the model's prediction accuracy is calculated. The specific indicator is R 2 (Residual sum squares), MSE (Mean Squared Error), and RMSE (Root Mean Squared Error) are commonly used.

[0050] Figure 11 shows an example of input data for the multitasking Lasso. As shown in Figure 11, the impact calculation unit 5 receives a dataset consisting of an index (identification number), a target variable Y, and explanatory variables X. In the example in Figure 11, the target variable Y includes the defect rate and the class. The class indicates, for example, the type of defect. The defect rate indicates the defect rate for each class. The explanatory variables X include information such as the equipment name, manufacturing conditions, processing temperature, and processing gas pressure for each process. Of the explanatory variables in Figure 11, the equipment name and manufacturing conditions for each process are categorical variables consisting of non-numerical data. The processing temperature and processing gas pressure for each process are continuous variables consisting of numerical data.

[0051] The categorical variables in Figure 11 are converted to numerical data. Figure 12 shows an example of converting categorical variables to numerical data. In Figure 12, the equipment names for process A, for example, are converted to 1 and for example, 0.

[0052] The multitasking Lasso model uses the following equation (2). W is the dependent variable using the coefficient W from equation (1), X is the number of parameters, and B is the regression coefficient. For example, W is represented as n rows and m columns, X as n rows and p columns, and B as p rows and m columns. As shown in Figure 8, if the coefficient W in equation (1) includes Task 1 and Task 2, then the dependent variable W in equation (2) also includes Task 1 and Task 2. W = X × B …(2)

[0053] More specifically, the multitasking Lasso takes the coefficient vector wim that constitutes the dependent variable W and the independent variable x as inputs and calculates the regression coefficient βo and the regression coefficient vector β(β1, β2, ..., βn) based on equation (3) below. Equation (3) is a minimization function, as indicated by min. λ is the regularization parameter and m is the number of operations. || || is the norm and means calculating the square root or maximum value of the sum of squares.

number

[0054] Figure 13 shows an example of output data from the multitasking Lasso. As shown in Figure 13, for each task w1 and w2, which are the dependent variable W, the corresponding explanatory variables are output in descending order of their regression coefficient B. In the example in Figure 13, the dependent variable W includes task 1, which has a central defect, and task 2, which has a ring-shaped defect. For the central defect, it is most likely that equipment A in process B is the cause of the defect, and for the ring-shaped defect, it is most likely that equipment B in process E is the cause of the defect.

[0055] When the processing in step S23 of Figure 10 is completed, the analysis accuracy verification unit 21 then calculates the prediction accuracy of the model constructed with multitask Lasso based on the verification data divided in step S21 (step S24). The prediction accuracy is expressed by the calculation result of equation (3), and the smaller the calculation result of equation (3), the higher the prediction accuracy. The calculation result of equation (3) depends on the regression coefficient β0 and the regression coefficient vector β(β1, β2, ... βn). Next, regression coefficients whose prediction accuracy is equal to or greater than the threshold value are stored in the influence storage unit 13 (step S25).

[0056] The influence display unit 14 displays the regression coefficients stored in the influence storage unit 13 as needed (step S26). The dependent variable information display unit 15 also displays the value of the dependent variable stored in the influence storage unit 13 as needed (step S27).

[0057] Figure 14A shows an example of the screen display of the impact display unit 14. In the example in Figure 14A, the impact (regression coefficient) of Task 1 included in the dependent variable is displayed in correspondence with the corresponding explanatory variable, and the impact (regression coefficient) of Task 3 is displayed in correspondence with the corresponding explanatory variable. In addition, the impact index (for example, R) is displayed. 2 Values ​​are displayed for each MSE.

[0058] Figure 14B shows an example of the screen display of the target variable information display unit. In the example in Figure 14B, the values ​​for each chip of Task 1 and Task 3 included in the target variable are displayed as line graphs. In Figure 14B, the horizontal axis represents the type of chip, and the vertical axis represents the value of the target variable.

[0059] The above description illustrates an example of applying the information processing device 1 according to this embodiment to the analysis of manufacturing defects in semiconductor devices. However, the information processing device 1 according to this embodiment can be applied to the analysis of various types of data. Figure 15 shows an example of applying the information processing device 1 according to this embodiment to the analysis of the relationship between gene base sequences and genetic diseases. Figure 15 shows single nucleotide polymorphism (SNP) mutations. In Figure 15, the mutations of SNP1 to SNP7 and the disease rate are correlated for each individual sample. By inputting SNP1 to SNP7 of the sample to be analyzed into the information processing device 1 according to this embodiment, the disease rate can be predicted with high accuracy.

[0060] Thus, in this embodiment, when calculating the influence of a multidimensional target variable, the number of dimensions of the target variable is compressed. This significantly reduces the processing time for calculating the influence. By simply inputting the target variable into the target variable dimensionality compression unit 3, the basis that characterizes the target variable and the coefficients that weight the basis can be calculated, thus improving computational efficiency. PCA, clustering, or NMF can be used to compress the number of dimensions of the target variable. Compressing the number of dimensions of the target variable requires inputting parameters such as the number of tasks, but in this embodiment, a user interface unit 12a is provided to make it easier for the operator to input parameters, so that the number of dimensions of the target variable can be reduced by inputting optimal parameters.

[0061] Furthermore, since the coefficient obtained by reducing the dimensionality of the target variable is input to the influence calculation unit 5 as a new target variable, the influence can be calculated simply and accurately. The influence calculation unit 5 calculates the influence using, for example, a multi-learning method, so the influence can be calculated accurately in a short processing time.

[0062] At least a part of the information processing device 1 described in the above-described embodiment may be configured as hardware or as software. If configured as software, a program that realizes at least a part of the functions of the information processing device 1 may be stored on a recording medium such as a flexible disk or CD-ROM, and loaded into a computer for execution. The recording medium is not limited to removable ones such as magnetic disks or optical disks, but may also be a fixed recording medium such as a hard disk drive or memory.

[0063] Furthermore, a program that implements at least some of the functions of the information processing device 1 may be distributed via communication lines such as the Internet (including wireless communication). In addition, the program may be encrypted, modulated, or compressed and distributed via wired or wireless lines such as the Internet, or stored on a recording medium.

[0064] The aspects of this disclosure are not limited to the individual embodiments described above, but include various modifications that a person skilled in the art could conceive, and the effects of this disclosure are not limited to those described above. In other words, various additions, modifications, and partial deletions are possible, as long as they do not depart from the conceptual idea and spirit of this disclosure derived from the claims and their equivalents. [Explanation of Symbols]

[0065] 1 Information processing device, 2 Target variable acquisition unit, 3 Target variable dimensionality reduction unit, 4 Explanatory variable acquisition unit, 5 Influence calculation unit, 6 Preprocessing unit, 7 Explanatory variable storage unit, 8 Explanatory variable determination unit, 9 Numerical conversion processing unit, 10 Missing value processing unit, 11 Standardization processing unit, 12 Parameter input unit, 12a User interface unit, 13 Influence storage unit, 14 Influence display unit, 15 Target variable information display unit, 16 Target variable selection unit, 17 Target variable setting unit, 18 Data splitting unit, 19 Parameter setting unit, 20 Multi-learning calculation unit, 21 Analysis accuracy verification unit, 21 Verification unit, 22 Database

Claims

1. A target variable acquisition unit that acquires multi-dimensional target variables, A target variable dimensionality reduction unit that compresses the number of dimensions of the target variable calculated based on the result of multiplying a plurality of basis sets that characterize the target variable by a plurality of coefficients that weight the plurality of basis sets, An explanatory variable acquisition unit that acquires explanatory variables, An influence calculation unit that selects two or more new target variables from at least one of the plurality of basis variables and the plurality of coefficients, and calculates the degree of influence on the two or more new target variables by performing calculations using a multi-learning method with the selected two or more new target variables and the explanatory variables, A target variable selection unit selects a coefficient from among the plurality of coefficients to be used as the new target variable by the influence calculation unit, An information processing apparatus comprising: an objective variable setting unit that sets the coefficient selected by the objective variable selection unit to the new objective variable.

2. The system includes a preprocessing unit that performs preprocessing to convert the explanatory variables acquired by the explanatory variable acquisition unit into a format that can be analyzed for data analysis. The information processing apparatus according to claim 1, wherein the explanatory variables after the preprocessing are input to the influence calculation unit.

3. The system includes an explanatory variable storage unit that stores the explanatory variables after the aforementioned preprocessing, The aforementioned pre-processing unit, An explanatory variable determination unit determines whether the explanatory variable obtained by the explanatory variable acquisition unit is a categorical variable or a continuous variable. If the explanatory variable obtained by the explanatory variable acquisition unit is a categorical variable, a numerical conversion processing unit performs numerical conversion processing of the categorical variable, If the explanatory variable obtained by the explanatory variable acquisition unit is a continuous variable, the missing value processing unit performs a process to impute the missing value, The system includes a standardization processing unit which performs standardization processing on explanatory variables that have undergone the numerical conversion processing or the processing to impute missing values, The explanatory variable storage unit stores the explanatory variables after the standardization process, The information processing apparatus according to claim 2, wherein the explanatory variables stored in the explanatory variable storage unit are input to the influence calculation unit.

4. The information processing apparatus according to any one of claims 1 to 3, wherein the objective variable dimensionality reduction unit compresses the number of dimensions of the objective variable using PCA (Principal Component Analysis), clustering, or NMF (Non-negative Matrix Factorization).

5. The information processing apparatus according to any one of claims 1 to 4, comprising a parameter input unit used in the processing of the objective variable dimension compression unit for inputting a parameter that specifies the number of dimensions to compress the objective variable.

6. The information processing apparatus according to claim 5, wherein the parameter input unit has a user interface unit that has a function to re-input the parameters while displaying the result of compressing the number of dimensions of the target variable.

7. The aforementioned target variable dimensionality reduction unit compresses the number of dimensions of the target variable using NMF, The information processing apparatus according to any one of claims 1 to 6, wherein the influence calculation unit performs multitask Lasso or Bayesian optimization processing on the new target variable using the coefficient obtained by compressing the dimensionality of the target variable using NMF, and calculates the influence on the new target variable.

8. The information processing apparatus according to any one of claims 1 to 7, further comprising a verification unit for verifying the accuracy of the aforementioned degree of influence.

9. The dataset, which includes the aforementioned new target variable and explanatory variables, is further divided into training data and validation data by a data splitting unit. The influence calculation unit calculates the influence using the training data, The information processing apparatus according to claim 8, wherein the verification unit verifies the accuracy of the degree of influence using the verification data.

10. The information processing apparatus according to any one of claims 1 to 9, comprising an influence display unit for displaying the degree of influence.

11. The information processing apparatus according to any one of claims 1 to 10, further comprising a target variable information display unit that displays information regarding the target variable of the compressed number of dimensions.

12. The information processing apparatus according to claim 10, wherein the influence display unit displays each of the explanatory variables and the corresponding influence for each of the new objective variables.

13. The influence calculation unit uses multitasking Lasso to calculate the regression coefficient as the influence, The information processing apparatus according to claim 12, wherein the influence display unit rearranges and displays each of the explanatory variables in the order of the regression coefficient values.

14. The information processing apparatus according to any one of claims 1 to 13, wherein the multi-dimensional target variable acquired by the target variable acquisition unit includes at least one of an electrical characteristic value, a numerical value, and a measured value.

15. The computer is By obtaining a multi-dimensional target variable, The number of dimensions of the objective variable, calculated based on the product of a plurality of basis sets that characterize the objective variable and a plurality of coefficients that weight the plurality of basis sets, is compressed. Obtain the explanatory variables and Two or more new target variables are selected from at least one of the aforementioned bases and coefficients, and the degree of influence on the two or more new target variables is calculated by performing calculations using a multi-learning method with the two or more selected new target variables and the explanatory variables. From the aforementioned plurality of coefficients, select the coefficient to be used as the new dependent variable when calculating the degree of influence. An information processing method that sets the selected coefficient as the new target variable.

16. Steps to obtain multi-dimensional target variables, A step of compressing the dimensionality of the objective variable, which is calculated based on the result of multiplying a plurality of basis sets that characterize the objective variable by a plurality of coefficients that weight the plurality of basis sets, Steps to obtain explanatory variables, The steps include selecting two or more new target variables from at least one of the plurality of basis variables and the plurality of coefficients, and calculating the degree of influence on the two or more new target variables by performing calculations using a multi-learning method with the two or more selected new target variables and the explanatory variables, The steps include selecting from the plurality of coefficients to be used as the new dependent variable when calculating the degree of influence, A recording medium for recording a computer-readable program that performs the steps of setting the selected coefficient to the new target variable.

Images