Information processing device, model generation method, and model generation program

Abstract

In order to make it possible to easily generate a hypothesis model for covariance structure analysis, an information processing device (1) comprises: an inference unit (11) that infers, based on a cause-and-effect search, a relationship between a plurality of elements including latent factors associated with a plurality of observation variables; and a model generation unit (12) that generates a hypothesis model for covariance structure analysis on the basis of the inference results. The hypothesis model generated by the information processing device (1) is useful for user decision making and the like.

Description

Information processing device, model generation method, and model generation program

[0001] The present invention relates to an information processing device that supports covariance structure analysis.

[0002] An analytical technique called covariance structure analysis (CSA) has been known for some time. CSA is a type of multivariate analysis also known as structural equation modeling (SEM). CSA analyzes the structure of covariance between observed variables, which are variables obtained through experiments or observations, to introduce latent factors that cannot be directly observed, and analyze the structure of causal relationships including the latent factors.

[0003] More specifically, in covariance structure analysis, a hypothetical model showing the relationships between multiple elements including latent factors is generated, and the generated hypothetical model is statistically evaluated and verified to analyze the relationships between the elements. Although covariance structure analysis is a useful analytical method, it is not something that anyone can easily perform, and therefore, technologies to support covariance structure analysis are being developed. For example, Patent Document 1 listed below discloses a creation support system that uses accumulated models to extract model structures and enables the extracted structures to be used to improve predictive models.

[0004] Japanese Patent Application Publication No. 2009-266158

[0005] In the creation support system described in Patent Document 1, a prediction model (corresponding to the above-mentioned hypothesis model) needs to be created by an operator, but there is a problem in that generating this model is not easy. One aspect of the present invention has been made in consideration of such a problem, and one example of its purpose is to provide a technology that enables easy generation of a hypothesis model for covariance structure analysis.

[0006] An information processing device according to one aspect of the present invention includes an inference means for inferring relationships between a plurality of elements, including latent factors associated with a plurality of observed variables, through causal exploration, and a model generation means for generating a hypothetical model for covariance structure analysis from the plurality of elements based on the results of the inference.

[0007] A model generation method according to one aspect of the present invention includes at least one processor inferring relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration, and generating a hypothetical model for covariance structure analysis from the multiple elements based on the results of the inference.

[0008] A model generation program according to one aspect of the present invention causes a computer to function as inference means that infers relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration, and model generation means that generates a hypothetical model for covariance structure analysis from the multiple elements based on the results of the inference.

[0009] According to one aspect of the present invention, it becomes possible to easily generate a hypothetical model for covariance structure analysis.

[0010] FIG. 1 is a block diagram showing the configuration of an information processing device according to a first exemplary embodiment of the present invention. FIG. 2 is a flow diagram showing the flow of a model generation method according to the first exemplary embodiment of the present invention. FIG. 3 is a diagram showing an example of a hypothetical model generated by a model generation method according to a second exemplary embodiment of the present invention. FIG. 4 is a block diagram showing the configuration of an information processing device according to a second exemplary embodiment of the present invention. FIG. 5 is a diagram showing an example of a setting screen for latent factors. FIG. 6 is a diagram showing an example of a hierarchical hypothetical model. FIG. 7 is a diagram showing an example of a display of a hypothetical model. FIG. 8 is a flow diagram showing the flow of a model generation method according to a second exemplary embodiment of the present invention. FIG. 9 is a diagram explaining an example of displaying a causal graph and allowing an operator to set latent variables. FIG. 10 is a diagram showing an example of a computer that executes instructions of a program that is software real

[0011] [First Exemplary Embodiment] A first exemplary embodiment of the present invention will be described in detail with reference to the drawings. This exemplary embodiment is a basic form of the exemplary embodiments described below.

[0012] (Configuration of Information Processing Device) The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As shown in the figure, the information processing device 1 includes an inference unit 11 and a model generation unit 12.

[0013] The inference unit 11 infers relationships between a plurality of elements, including latent factors associated with a plurality of observed variables, through causal exploration.

[0014] The model generation unit 12 generates a hypothetical model for covariance structure analysis from the above-mentioned multiple elements based on the result of the inference by the inference unit 11.

[0015] As described above, the information processing device 1 according to this exemplary embodiment includes the inference unit 11 that infers relationships between multiple elements, including latent factors associated with multiple observed variables, through causal search, and the model generation unit 12 that generates a hypothesis model for covariance structure analysis from the multiple elements based on the results of the inference. Therefore, the information processing device 1 according to this exemplary embodiment has the effect of making it possible to easily generate a hypothesis model for covariance structure analysis.

[0016] (Model Generation Program) The functions of the information processing device 1 described above can also be realized by a program. The model generation program according to this exemplary embodiment is configured to cause a computer to function as an inference means that infers relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration, and as a model generation means that generates a hypothetical model for covariance structure analysis from the multiple elements based on the results of the inference. Therefore, the model generation program according to this exemplary embodiment has the effect of making it possible to easily generate a hypothetical model for covariance structure analysis.

[0017] (Flow of Model Generation Method) The flow of the model generation method according to this exemplary embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the model generation method. Note that the execution entity of each step in this model generation method may be a processor provided in the information processing device 1, or a processor provided in another device, or each step may be executed by a processor provided in a different device.

[0018] In S11, at least one processor infers relationships among a plurality of elements, including latent factors associated with a plurality of observed variables, through causal exploration.

[0019] In S12, at least one processor generates a hypothetical model for covariance structure analysis from the multiple factors based on the result of the inference in S11.

[0020] As described above, the model generation method according to this exemplary embodiment includes, by at least one processor, inferring relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration, and generating a hypothetical model for covariance structure analysis from the multiple elements based on the results of the inference. Therefore, the model generation method according to this exemplary embodiment has the advantage of making it possible to easily generate a hypothetical model for covariance structure analysis.

[0021] [Exemplary Embodiment 2] (Overview of Model Generation Method) According to a model generation method (hereinafter referred to as the present method) according to this exemplary embodiment, a hypothesis model for covariance structure analysis can be easily generated. As explained in the "Background Art" section, conventionally, hypothesis models have been generated manually, and it has not been easy to generate a valid hypothesis model. In this regard, according to the present method, a hypothesis model can be easily generated, and therefore, by using the present method, it is possible to easily perform covariance structure analysis.

[0022] An overview of this method will be explained with reference to Fig. 3. Fig. 3 is a diagram showing an example of a hypothetical model generated by this method. The hypothetical model M1 shown in Fig. 3 represents the relationships between multiple elements, including observed variables and latent factors (which may also be called latent variables), using nodes and paths.

[0023] Specifically, the hypothesis model M1 includes rectangular nodes n1 to n9 representing observed variables and oval nodes N1 to N4 representing latent factors, and the relationships between these nodes are represented by arrows, i.e., paths, connecting the nodes. For example, a path is drawn between node N1, the latent factor, and node n1, the observed variable, with node N1 as the starting point and node n1 as the end point, indicating that node N1 explains node n1. Each node is labeled with a name indicating the corresponding observed variable or latent factor. For example, "educational environment" listed in node n1 indicates the name given to the observed variable of node n1. The observed variable of node n1 is a variable related to "educational environment."

[0024] According to this method, the relationships between multiple elements, including latent factors, are inferred through causal exploration. This is equivalent to inferring what kind of paths are established between each node. For example, an inference result that there is a relationship between node N1 and node n7 in which node N1 explains node n7 means that a path is established starting from node N1 and ending at node n7. This makes it possible to automatically generate a hypothetical model M1 for covariance structure analysis, as shown in Figure 3.

[0025] The operator can then analyze the relationships between the elements from the generated hypothesis model M1. For example, in the example of Figure 3, an analysis can be made to determine whether measures to improve "ease of raising children," "improvement of infrastructure," and "living environment" and thereby increase resident "satisfaction" would be effective in making a city a "desirable place to live."

[0026] This method can also be used to analyze data in the medical or healthcare fields. For example, the inference unit 205 may infer, by causal exploration, relationships between multiple elements, including latent factors associated with observed variables related to a subject who wishes to maintain or improve their health. Then, the model generation unit 206 may generate a hypothetical model for covariance structure analysis from the multiple elements based on the inference results of the inference unit 205.

[0027] As a specific example, the inference unit 205 may infer, by causal search, the relationships between multiple elements of multiple subjects, including latent factors (e.g., lack of exercise, unbalanced nutrition, etc.) associated with observed variables indicating various test results, lifestyle habits, etc. By referring to the hypothesis model generated based on the results of such inference, the user of the information processing device 1 can analyze effective measures for maintaining or improving their own health condition.

[0028] (Configuration of Information Processing Device) The configuration of the information processing device 2 according to this exemplary embodiment will be described with reference to Fig. 4. Fig. 4 is a block diagram showing the configuration of the information processing device 2. The information processing device 2 is a device having a function of generating a hypothesis model for covariance structure analysis. Note that the information processing device 2 may be a device whose main function is to generate a hypothesis model, or may be a general-purpose device having other functions as well.

[0029] As shown in the figure, the information processing device 2 includes a control unit 20 that controls each unit of the information processing device 2, and a storage unit 21 that stores various data used by the information processing device 2. The information processing device 2 also includes a communication unit 22 that enables the information processing device 2 to communicate with other devices, an input unit 23 that accepts various data input to the information processing device 2, and an output unit 24 that enables the information processing device 2 to output various data. Note that each of the components from the storage unit 21 to the output unit 24 may be built into the information processing device 2, or may be an external device attached to the information processing device 2.

[0030] As shown in the figure, the control unit 20 of the information processing device 2 includes a data acquisition unit 201, a factor analysis unit 202, an index value calculation unit 203, a latent factor setting unit 204, an inference unit 205, a model generation unit 206, a display control unit 207, a reception unit 208, and a structural analysis unit 209.

[0031] The data acquisition unit 201 acquires various data used in the covariance structure analysis. For example, the data acquisition unit 201 acquires observed variables used in the covariance structure analysis. The method of acquiring the data is not particularly limited. For example, the data acquisition unit 201 may acquire data input by an operator of the information processing device 2 via the input unit 23, or may acquire data from another device by communication via the communication unit 22.

[0032] The factor analysis unit 202 classifies multiple observed variables into multiple groups using exploratory factor analysis. Exploratory factor analysis is an analytical method that estimates factors that cause correlations between multiple observed variables based on the correlation coefficients between the observed variables. In other words, the groups correspond to factors that cause correlations between the observed variables included in the group, i.e., latent factors. In other words, the factor analysis unit 202 groups multiple observed variables according to the latent factors to which the observed variables are associated. The number of groups to classify (in other words, the number of latent factors to set) may be determined in advance or may be specified by an operator.

[0033] The index value calculation unit 203 calculates an index value for the grouped observed variables that indicates the appropriateness of associating those observed variables with one latent factor. The method for calculating the index value is not particularly limited. For example, the index value calculation unit 203 may calculate Cronbach's alpha as the index value. Alternatively, the index value calculation unit 203 may calculate a factor loading that indicates the degree of influence that each variable receives from a factor as the index value. Such index values ​​also function as reference information when the operator determines the name of the latent factor.

[0034] The latent factor setting unit 204 sets latent factors for the covariance structure analysis. The setting of latent factors may be performed automatically or manually. When setting latent factors automatically, the latent factor setting unit 204 sets a latent factor corresponding to each group into which the factor analysis unit 202 classifies the observed variables. On the other hand, when setting latent factors manually, the latent factor setting unit 204 sets a latent factor corresponding to each group of observed variables specified by the operator. After the factor analysis unit 202 classifies the observed variables into multiple groups, the display control unit 207 may display each group as a candidate latent factor, allowing the operator to change the observed variables belonging to each group. In this case, the latent factor setting unit 204 sets a latent factor corresponding to each changed group.

[0035] The inference unit 205 infers relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration. The latent factors are set by the latent factor setting unit 204. The elements that are the target of causal exploration may include the latent factors set by the latent factor setting unit 204, the observed variables acquired by the data acquisition unit 201, and composite variables generated from the observed variables.

[0036] Causal exploration is a method of inferring what kind of causal relationship exists between elements (or whether there is no causal relationship) from each element. A specific method of causal exploration is known, for example, a method using a structural equation model, and the inference unit 205 may apply such a method to make the above inference.

[0037] The model generation unit 206 generates a hypothesis model for covariance structure analysis from multiple elements based on the result of inference by the inference unit 205. The generated hypothesis model is a model that shows the result of inference by the inference unit 205, i.e., the relationship between multiple elements including latent factors (what causal relationship exists between which elements).

[0038] The display control unit 207 displays various information related to the covariance structure analysis on the display device. For example, the display control unit 207 displays each of the groups into which the factor analysis unit 202 classifies the observed variables as candidates for latent factors. The display control unit 207 also displays the index values ​​calculated by the index value calculation unit 203. The display device may be included in the information processing device 2 or may be an external device to the information processing device 2.

[0039] The receiving unit 208 receives input from an operator and performs processing in accordance with the input. For example, when the latent factors are manually set, the receiving unit 208 receives specifications from the operator and classifies the observed variables into multiple groups in accordance with the specifications.

[0040] The structural analysis unit 209 performs a structural analysis of the hypothetical model generated by the model generation unit 206 and calculates an index value indicating the goodness of fit of the hypothetical model. The method for calculating the index value is not particularly limited. For example, the structural analysis unit 209 may calculate at least one of the following values ​​as the index value: Root Mean Square Error of Approximation (RMSEA), Goodness of Fit Index (GFI), Comparative Fit Index (CFI), and Akaike's Information Criterion (AIC). The structural analysis unit 209 also calculates path coefficients between elements having a causal relationship in the hypothetical model generated by the model generation unit 206. The path coefficients are calculated using weight values ​​of paths connecting each element. The weight values ​​can be calculated using a known causal inference method. The calculation of the index value and the calculation of the path coefficients may be performed in different processing blocks.

[0041] As described above, the information processing device 2 according to this exemplary embodiment includes an inference unit 205 that infers relationships between multiple elements, including latent factors associated with multiple observed variables, through causal search, and a model generation unit 206 that generates a hypothesis model for covariance structure analysis from the multiple elements based on the results of inference by the inference unit 205. Therefore, the information processing device 2 according to this exemplary embodiment has the advantage of being able to easily generate a hypothesis model for covariance structure analysis. The hypothesis model generated by the information processing device 1 is useful for user decision-making, etc. In other words, the information processing device 1 can facilitate user decision-making.

[0042] Furthermore, the information processing device 2 according to this exemplary embodiment is configured to include a factor analysis unit 202 that classifies a plurality of observed variables into a plurality of groups by exploratory factor analysis, and a display control unit 207 that displays each of the groups as a candidate for a latent factor. Therefore, the information processing device 2 according to this exemplary embodiment has the effect of making it possible to easily set latent factors, in addition to the effect achieved by the information processing device 1 according to the exemplary embodiment 1.

[0043] Furthermore, the information processing device 2 according to this exemplary embodiment is configured to include a receiving unit 208 that receives designations from an operator and classifies a plurality of observed variables into a plurality of groups in accordance with the designations, an index value calculation unit 203 that calculates, for each of the groups, an index value that indicates the appropriateness of setting a latent factor corresponding to the group, and a display control unit 207 that displays the index values. Thus, in addition to the effects achieved by the information processing device 1 according to the first exemplary embodiment, the information processing device 2 according to this exemplary embodiment has the effect of enabling appropriate latent factors to be set while reflecting the subjective opinions of the operator.

[0044] (Example of Latent Factor Setting Screen) When covariance structure analysis is supported by the information processing device 2, when setting latent factors, a setting screen such as that shown in Fig. 5 may be displayed to accept input from an operator. Fig. 5 is a diagram showing an example of the latent factor setting screen.

[0045] 5 shows a table of the relationships between multiple observed variables and the latent factors associated with each observed variable. More specifically, in the setting screen A1, the observed variables are shown in the rows below the "index value" row, and the latent factors are shown in multiple columns, with check boxes displayed at the intersections of the rows and columns.

[0046] A check in the checkbox indicates that the observed variable in that row is associated with the latent factor in that column. For example, for the rows of the observed variables "greenness" and "waterside environment," the checkboxes at the intersection with the "Latent Factor 1" column are checked. This indicates that these observed variables are associated with one latent factor, "Latent Factor 1."

[0047] In addition, on the setting screen A1, in the column of each latent factor, the index value calculated by the index value calculation unit 203 is displayed, along with an object A11 for rearranging the observed variables in that column and an object A12 for deleting that column. An object A13 for adding a new latent factor is also displayed.

[0048] When the receiving unit 208 receives an operator's input for object A11, the display control unit 207 rearranges the rows of the observation variables in the column corresponding to object A11 so that the observation variables with checked checkboxes are arranged in order. When the receiving unit 208 receives an operator's input for object A12, the display control unit 207 deletes the column corresponding to object A12. Then, when the receiving unit 208 receives an operator's input for object A13, the display control unit 207 displays a column corresponding to a new latent variable.

[0049] The setting screen A1 can be generated based on the classification results of the observed variables by the factor analysis unit 202. In this case, the operator simply inputs the observed variables to be used into the information processing device 2 and specifies three as the number of latent factors, i.e., the number of groups into which the observed variables are classified. The factor analysis unit 202 then classifies the observed variables into three groups, and the display control unit 207 generates and displays the setting screen A1 in accordance with the classification results. Furthermore, for each group of observed variables defined by the factor analysis unit 202, the index value calculation unit 203 calculates an index value indicating the appropriateness of associating the observed variables with one latent factor, and the display control unit 207 displays the calculated index value in the "Index Value" row. The index value may be, for example, Cronbach's alpha.

[0050] Here, the receiving unit 208 may accept a change in the observation variable associated with the latent factor on the setting screen A1. For example, when an operation to select an already checked checkbox is performed, the display control unit 207 may uncheck the checkbox, and the receiving unit 208 may cancel the association between the observation variable corresponding to the checkbox and the latent factor corresponding to the checkbox. Alternatively, when an operation to select an unchecked checkbox is performed, the display control unit 207 may check the checkbox, and the receiving unit 208 may associate the observation variable corresponding to the checkbox with the latent factor corresponding to the checkbox.

[0051] If the factor analysis unit 202 does not group the observed variables, the display control unit 207 may display the setting screen A1 with none of the check boxes checked. In this case, the operator simply determines the observed variables to associate with each latent factor and checks the check boxes corresponding to those observed variables. The index value calculation unit 203 then calculates an index value indicating the appropriateness of associating those observed variables with one latent factor, and the display control unit 207 displays the calculated index value.

[0052] Once it has been determined as described above which latent factors each observation variable should be associated with, the latent factor setting unit 204 sets each determined latent factor as a latent factor in the covariance structure analysis. At this time, the receiving unit 208 may receive input of the name of each latent factor, and in this case, the latent factor setting unit 204 records the input name in the storage unit 21 or the like together with the observation variables associated with that latent factor. Furthermore, the display control unit 207 may display the input name on the setting screen A1 (for example, instead of "Latent Factor 1" to "Latent Factor 3").

[0053] (Generation of a hierarchical hypothesis model) The information processing device 2 can also generate a hierarchical hypothesis model. This will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of a hierarchical hypothesis model. Note that in Fig. 6, nodes of observed variables and nodes of latent factors are shown as circular objects without distinction.

[0054] The hypothetical model M2 shown in Figure 6 represents the relationships between nine nodes using paths connecting them. The nine nodes are organized into three hierarchical layers: "Layer 1," "Layer 2," and "Layer 3."

[0055] When generating such a hypothesis model M2, it is sufficient to assign hierarchical information indicating the hierarchical level to each element that is the subject of covariance structure analysis in advance. Then, if the inference unit 205 performs a causal search using the hierarchical level information as a constraint, the model generation unit 206 can generate a hierarchical hypothesis model in hierarchical units indicated by the hierarchical level information.

[0056] Specifically, the inference unit 205 may infer the relationship between elements in different hierarchical levels indicated in the hierarchical information by causal search. As a result, the relationship between elements belonging to the same level is not inferred, and therefore a hierarchical hypothesis model can be generated for each level indicated in the hierarchical information. The inference unit 205 may also infer the relationship between elements in adjacent levels indicated in the hierarchical information by causal search. As a result, a hierarchical hypothesis model can be generated in accordance with the hierarchical order indicated in the hierarchical information.

[0057] As described above, each of the multiple elements that are the subject of covariance structure analysis may be assigned hierarchical information indicating the hierarchical level to which the element belongs. In this case, the inference unit 205 may infer the relationship between elements that belong to different hierarchical levels indicated in the hierarchical level information by causal search, and the model generation unit 206 may generate a hierarchical hypothesis model for each hierarchical level indicated in the hierarchical level information. This provides the effect of being able to automatically generate a hierarchical hypothesis model in addition to the effect provided by the information processing device 1 according to the first exemplary embodiment.

[0058] A hierarchical hypothesis model is useful, for example, in analyses related to product manufacturing management. For example, observed variables may be collected for each product manufacturing process, and each observed variable may be associated with hierarchical information indicating the manufacturing process. This makes it possible to generate a hypothesis model that allows easy recognition of the relationships between the observed variables in each manufacturing process and the latent factors associated with those observed variables and the observed variables in other manufacturing processes.

[0059] (Regarding the Display Mode of the Hypothesis Model) As described above, an operator performing covariance structure analysis using the information processing device 2 analyzes the relationships between elements from the generated hypothetical model. At this time, if the number of elements to be analyzed is large, the generated hypothetical model may become complicated, and it may be difficult to smoothly analyze the relationships between the elements. For this reason, it is preferable that the information processing device 2 display the hypothetical model in a display mode that allows the analysis of the relationships between the elements to smoothly proceed, even when the number of elements to be analyzed is large.

[0060] For example, the information processing device 2 may display the hypothesis model in a display format as shown in Fig. 7. Fig. 7 is a diagram showing a display example of the hypothesis model. Fig. 7 shows a hypothesis model M3 generated by the model generation unit 206. The hypothesis model M3 is a graph that represents the relationship between nodes n11 to n15 of the observed variables and nodes N11 to N13 of the latent factors using paths.

[0061] In the hypothesis model M3, paths connecting nodes are displayed with lines of a thickness corresponding to the path coefficients calculated by the structural analysis unit 209, etc. In this way, the display control unit 207 may display paths in a display mode corresponding to the path coefficients. This allows the operator to recognize paths with a stronger influence, allowing for smooth analysis of the relationships between elements.

[0062] 7, paths with path coefficients equal to or greater than a specified threshold and the nodes that are the starting points of those paths are displayed, while paths with path coefficients less than the threshold and the nodes that are the starting points of those paths are not displayed. Object B1 shown in FIG. 7 is an object for specifying this threshold, i.e., a threshold specification object.

[0063] The object B1 is an object in which a slider that can be moved on a slider bar is arranged on the slider bar. The slider bar indicates a changeable range of the pass coefficient threshold, and the position of the slider indicates the pass coefficient threshold. By the display control unit 207 displaying such an object B1, the operator can change the pass coefficient threshold by an intuitive and simple operation of moving the slider on the slider bar.

[0064] As shown in the upper part of Fig. 7, the operator can display all of the nodes and paths that make up the generated hypothesis model M3 by using the threshold specification object B1 to set a smaller threshold value. Also, as shown in the lower part of Fig. 7, the operator can also display a hypothesis model M3' in which paths with path coefficients less than the threshold value and the nodes that are the starting points of those paths are hidden by operating the threshold specification object B1 with the cursor Cu to change the threshold to a larger value. Note that the method of operating the threshold specification object B1 is not limited to using the cursor Cu.

[0065] As described above, the display control unit 207 may display a graph representing a hypothetical model using nodes and paths, and a threshold specification object that accepts specification of a threshold for path coefficients. The display control unit 207 may then display paths with path coefficients equal to or greater than the threshold specified by the threshold specification object and the nodes that are the starting points of those paths, but not display paths with path coefficients less than the threshold and the nodes that are the starting points of those paths. This provides the same advantages as the information processing device 1 according to the first exemplary embodiment, as well as the ability to easily perform analysis limited to paths with path coefficients equal to or greater than the specified threshold and the nodes connected by those paths.

[0066] (Flow of Model Generation Method) The flow of the model generation method according to this exemplary embodiment will be described with reference to Fig. 8. Fig. 8 is a flow diagram showing the flow of the model generation method.

[0067] In S21, the data acquisition unit 201 acquires a plurality of observation variables to be used in the covariance structure analysis. For example, the data acquisition unit 201 may acquire observation variables input by an operator via the input unit 23.

[0068] In S22, the factor analysis unit 202 determines whether or not to automatically generate latent factor candidates. For example, the factor analysis unit 202 may determine to automatically generate latent factor candidates (YES in S22) when the operator performs an input operation to automatically generate latent factor candidates, and may determine not to automatically generate latent factor candidates (NO in S22) when the operator performs an input operation to manually generate latent factor candidates. If the determination in S22 is YES, the process proceeds to S27, and if the determination in S22 is NO, the process proceeds to S23.

[0069] In S23, the display control unit 207 displays the observation variables acquired in S21 on the display device. Furthermore, in S24, the reception unit 208 receives designation of observation variables to be grouped for the observation variables displayed by the processing of S23. For example, the display control unit 207 may present the observation variables to the operator by displaying a setting screen such as that shown in FIG. 5 . In this case, in S24, the reception unit 208 may receive designation of the observation variables via this setting screen.

[0070] Then, in S25, the index value calculation unit 203 calculates an index value for the observed variables grouped in S24, which indicates the appropriateness of associating those observed variables with one latent factor, and the display control unit 207 displays the calculated index value. For example, as in the example of FIG. 5 , the display control unit 207 may display the index value calculated for each latent factor in association with that latent factor.

[0071] In S26, the latent factor setting unit 204 determines whether the groups have been confirmed. For example, the latent factor setting unit 204 may determine that the groups have been confirmed (YES in S26) when, for each of the number of groups designated by the operator, the observed variables to be included in that group are designated and an operation to confirm the designation is performed. If the determination in S26 is NO, the process returns to S24, where the reception unit 208 receives the designation (which may be a change) of the observed variables. On the other hand, if the determination in S26 is YES, the process proceeds to S29.

[0072] In S27, the factor analysis unit 202 classifies the observed variables acquired in S21 into multiple groups through exploratory factor analysis. Then, in S28, the display control unit 207 displays the observed variables acquired in S21 into multiple groups based on the classification results of S27. For example, the display control unit 207 may display a setting screen such as the example of FIG. 5. In this case, index values ​​may be calculated and displayed in the same manner as in S25. Note that, on this setting screen, the reception unit 208 may receive a change to the group of observed variables (which may also be referred to as the observed variables associated with the latent factors). In this case, the processing of S25 and subsequent steps is performed for the changed group.

[0073] In S29, the receiving unit 208 receives input of the name of the latent factor corresponding to each group of observed variables. Note that the process of S29 may be performed at any timing after S23 or S28.

[0074] In S30, the latent factor setting unit 204 sets each of the above-mentioned groups to which one or more observed variables belong as a latent factor in the covariance structure analysis.

[0075] In S31, the inference unit 205 infers the relationships between multiple elements that are the subject of covariance structure analysis through causal exploration. The multiple elements include the latent factors set in S30. The multiple elements may also include other elements, such as the explanatory variables input in S21.

[0076] In S32, the model generation unit 206 generates a hypothetical model for covariance structure analysis based on the result of the inference in S31. Then, in S33, the display control unit 207 displays the hypothetical model generated in S32, and the processing in FIG. 8 is then terminated.

[0077] After S32, the structural analysis unit 209 may perform a structural analysis of the hypothetical model generated in S32, and calculate an index value indicating the goodness of fit of the hypothetical model and a path coefficient of each path included in the hypothetical model. In this case, in S33, the display control unit 207 may also display the calculated index values ​​and path coefficients.

[0078] Furthermore, the accepting unit 208 may accept a modification of the hypothetical model by an operator. When the hypothetical model is modified, the structural analysis unit 209 calculates an index value indicating the goodness of fit of the modified hypothetical model and the path coefficients of each path included in the hypothetical model, and the display control unit 207 displays these.

[0079] As described above, the model generation method according to this exemplary embodiment includes inferring relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration (S31), and generating a hypothetical model for covariance structure analysis from multiple elements to be subjected to covariance structure analysis based on the results of the inference (S32). This provides the advantage of making it possible to easily generate a hypothetical model for covariance structure analysis.

[0080] (Display of Causal Graph) When the information processing device 2 has the operator set a latent variable, the information processing device 2 may display a causal graph showing the causal relationships between the observed variables. This will be described with reference to Fig. 9. Fig. 9 is a diagram illustrating an example in which a causal graph is displayed and the operator is allowed to set a latent variable.

[0081] A causal graph is a graph in which multiple elements subjected to causal analysis are represented by nodes, and the causal relationships between the elements are indicated by directed edges (unidirectional arrows), and is also called a directed acyclic graph. The causal graph G1 shown in Figure 9 indicates the causal relationships between observed variables such as "cultural and artistic activities," which are the subject of covariance structure analysis, using directed edges.

[0082] Such a causal graph can be generated by the inference unit 205 performing a causal search on observed variables that are the targets of covariance structure analysis. In this way, the inference unit 205 can be used not only to generate a hypothetical model but also to generate a causal graph.

[0083] Because the causal graph shows the causal relationships between observed variables, displaying the causal graph allows the operator to set latent factors taking into account the causal relationships between the observed variables. For example, in the example of Figure 9, the observed variables "waterside environment" and "abundance of greenery" are connected by a causal relationship. Recognizing this, the operator can group these observed variables. In the example of Figure 9, the fact that these observed variables are grouped is indicated by the frame E1 surrounding the nodes of these observed variables.

[0084] Also, within the frame E1, a name input field E11 for inputting the name of a latent factor to be associated with the observed variable surrounded by the frame E1, and an index value E12 calculated by the index value calculation unit 203 for the group of observed variables indicated by the frame E1 are displayed in association with each other. The name "living environment" input by the operator is displayed in the name input field E11.

[0085] Similarly, box E2 in Figure 9 indicates that a total of three observed variables, such as "adequate public transportation," have been grouped. As with box E1, box E2 also displays a name input field E21 and an index value E22 in association with each other, but no name has yet been entered in the name input field E21. When the operator selects the name input field E21 and enters a desired name, the name of the latent factor corresponding to the group in box E2 is specified, and this name is displayed in the name input field E21.

[0086] 9 shows an example in which observed variables are grouped by enclosing them with a mouse drag operation, and the locus of the drag operation is shown by the curve E3. By enabling observed variables to be grouped with such an intuitive operation, anyone can easily set latent factors.

[0087] When one or more observed variables are surrounded by the locus of the drag operation, the latent factor setting unit 204 may set latent factors corresponding to those observed variables. Then, the index value calculation unit 203 may calculate index values ​​for the latent factors, and the display control unit 207 may display the calculated index values.

[0088] Furthermore, it is preferable that the group of observation variables be easily changeable by an operator. For example, the receiving unit 208 may accept an operation to move an observation variable node from inside to outside a frame representing a group as an operation to remove the observation variable from the group, or may accept an operation to move an observation variable node inside a frame representing a group as an operation to add the observation variable to the group. Furthermore, the receiving unit 208 may accept, for example, an operation to widen or narrow the frame representing a group. In this case, the nodes included within the frame after the shape or size has been changed become nodes belonging to the group corresponding to that frame.

[0089] Note that the type of grouping operation is arbitrary and is not limited to the above example. Furthermore, the specification of the observation variables to be grouped does not necessarily have to be accepted on the causal graph. For example, the display control unit 207 may display a setting screen such as that shown in FIG. 5 while displaying the causal graph, and accept the specification of the observation variables to be grouped on the setting screen.

[0090] As described above, the display control unit 207 may display a causal graph showing the causal relationships between multiple observed variables. The receiving unit 208 may then receive a designation from an operator and classify the observed variables into multiple groups in accordance with the designation, and the latent factor setting unit 204 may set latent factors corresponding to each of the groups. This provides the effect of being able to set latent factors in consideration of the causal relationships between observed variables, in addition to the effect provided by the information processing device 1 according to the first exemplary embodiment.

[0091] [Modification] The execution entity of each process described in the above embodiment is arbitrary and is not limited to the above example. In other words, the functions of the information processing devices 1 and 2 can be realized by multiple devices (which can also be called processors) that can communicate with each other. For example, the execution of each process described in the flowcharts of Figures 2 and 8 can be shared among multiple processors. In other words, the execution entity of the model generation method in the above embodiment may be one processor or multiple processors.

[0092] [Example of Software Implementation] Some or all of the functions of the information processing device 1 or 2 may be implemented by hardware such as an integrated circuit (IC chip), or may be implemented by software.

[0093] In the latter case, the information processing device 1 or 2 is realized, for example, by a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in FIG. 10. The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program (model generation program) P for operating the computer C as the information processing device 1 or 2. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing device 1 or 2.

[0094] The processor C1 may be, for example, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0095] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0096] The program P can also be recorded on a non-transitory, tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0097] [Additional Note 1] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiments are also included in the technical scope of the present invention.

[0098] [Additional Note 2] Part or all of the above-described embodiment can also be described as follows: However, the present invention is not limited to the following described aspects.

[0099] (Supplementary Note 1) An information processing device comprising: an inference means for inferring relationships between a plurality of elements, including latent factors associated with a plurality of observed variables, through causal exploration; and a model generation means for generating a hypothesis model for covariance structure analysis from the plurality of elements based on the results of the inference.

[0100] (Supplementary Note 2) The information processing device according to Supplementary Note 1, comprising: factor analysis means for classifying a plurality of the observed variables into a plurality of groups by exploratory factor analysis; and display control means for displaying each of the groups as a candidate for a latent factor.

[0101] (Supplementary Note 3) The information processing device according to Supplementary Note 1 or 2, comprising: a receiving means for receiving a specification from an operator and classifying a plurality of the observation variables into a plurality of groups in accordance with the specification; an index value calculation means for calculating, for each of the groups, an index value indicating the appropriateness of setting a latent factor corresponding to the group; and a display control means for displaying the index value.

[0102] (Supplementary Note 4) An information processing device according to any one of Supplementary Notes 1 to 3, wherein each of the plurality of elements is assigned hierarchical information indicating the hierarchical level to which the element belongs, the inference means infers the relationship between elements at different hierarchical levels indicated in the hierarchical information by causal search, and the model generation means generates the hypothesis model hierarchically organized in hierarchical units indicated in the hierarchical information.

[0103] (Supplementary Note 5) An information processing device according to any of Supplementary Notes 1 to 4, comprising a display control means for displaying a graph representing the hypothesis model using nodes and paths, and a threshold designation object for accepting designation of a threshold value for path coefficients, wherein the display control means displays paths with path coefficients equal to or greater than the threshold value designated by the threshold designation object and the nodes that are the starting points of the paths, but does not display paths with path coefficients less than the threshold value and the nodes that are the starting points of the paths.

[0104] (Supplementary Note 6) The information processing device according to Supplementary Note 1, comprising: a display control means for displaying a causal graph showing the causal relationships between a plurality of the observation variables; a receiving means for receiving a designation by an operator and classifying the observation variables into a plurality of groups in accordance with the designation; and a latent factor setting means for setting a latent factor corresponding to each of the groups.

[0105] (Supplementary Note 7) A model generation method including: at least one processor inferring relationships between a plurality of elements, including latent factors associated with a plurality of observed variables, through causal exploration; and generating a hypothetical model for covariance structure analysis from the plurality of elements based on results of the inference.

[0106] (Supplementary Note 8) A model generation program that causes a computer to function as an inference means that infers relationships between multiple elements, including latent factors associated with multiple observed variables, through causal exploration, and a model generation means that generates a hypothetical model for covariance structure analysis from the multiple elements based on the results of the inference.

[0107] [Additional Note 3] Some or all of the above-described embodiments can also be expressed as follows: An information processing device including at least one processor, the processor executing an inference process for inferring relationships between a plurality of elements, including a latent factor associated with a plurality of observed variables, through causal exploration, and a model generation process for generating a hypothetical model for covariance structure analysis from the plurality of elements based on the result of the inference.

[0108] The information processing device may further include a memory that stores a model generation program for causing the processor to execute the inference process and the model generation process. The model generation program may be recorded on a computer-readable, non-transitory, tangible recording medium.

[0109] REFERENCE SIGNS LIST 1 Information processing device 11 Inference unit (inference means) 12 Model generation unit (model generation means) 2 Information processing device 202 Factor analysis unit (factor analysis means) 203 Index value calculation unit (index value calculation means) 204 Latent factor setting unit (latent factor setting means) 205 Inference unit (inference means) 206 Model generation unit (model generation means) 207 Display control unit (display control means) 208 Reception unit (reception means)

Claims

1. an inference means for inferring relationships among multiple elements, including latent factors associated with multiple observed variables, through causal exploration; and model generation means for generating a hypothetical model for covariance structure analysis from the plurality of elements based on the result of the inference.

2. a factor analysis means for classifying the plurality of observed variables into a plurality of groups by exploratory factor analysis; The information processing apparatus according to claim 1 , further comprising: a display control means for displaying each of the groups as a candidate for a latent factor.

3. a receiving means for receiving a specification from an operator and classifying the plurality of observation variables into a plurality of groups in accordance with the specification; an index value calculation means for calculating, for each of the groups, an index value indicating the appropriateness of setting a latent factor corresponding to the group; The information processing apparatus according to claim 1 , further comprising: a display control means for displaying the index value.

4. each of the plurality of elements is assigned with hierarchical information indicating a hierarchical level to which the element belongs; the inference means infers a relationship between elements in different hierarchical levels indicated in the hierarchical information by causal search; The information processing apparatus according to claim 1 , wherein the model generating means generates the hypothesis model layered in layer units indicated in the layer information.

5. a display control means for displaying a graph in which the hypothetical model is represented by nodes and paths, and a threshold value designation object for receiving designation of a threshold value for a path coefficient; 3. The information processing device according to claim 1, wherein the display control means displays paths with path coefficients equal to or greater than the threshold specified by the threshold specifying object and the nodes that are the starting points of the paths, but does not display paths with path coefficients less than the threshold and the nodes that are the starting points of the paths.

6. a display control means for displaying a causal graph showing the causal relationships between the plurality of observed variables; a receiving means for receiving a specification from an operator and classifying the observation variables into a plurality of groups in accordance with the specification; The information processing apparatus according to claim 1 , further comprising: latent factor setting means for setting latent factors corresponding to each of said groups.

7. At least one processor Inferring relationships among multiple factors, including latent factors associated with multiple observed variables, through causal exploration; generating a hypothetical model for covariance structure analysis from the plurality of elements based on the result of the inference.

8. Computer, An inference method for inferring relationships among multiple elements, including latent factors associated with multiple observed variables, through causal exploration; and a model generation program that functions as a model generation means for generating a hypothetical model for covariance structure analysis from the plurality of elements based on the result of the inference;